hexsha
stringlengths 40
40
| size
int64 6
14.9M
| ext
stringclasses 1
value | lang
stringclasses 1
value | max_stars_repo_path
stringlengths 6
260
| max_stars_repo_name
stringlengths 6
119
| max_stars_repo_head_hexsha
stringlengths 40
41
| max_stars_repo_licenses
list | max_stars_count
int64 1
191k
⌀ | max_stars_repo_stars_event_min_datetime
stringlengths 24
24
⌀ | max_stars_repo_stars_event_max_datetime
stringlengths 24
24
⌀ | max_issues_repo_path
stringlengths 6
260
| max_issues_repo_name
stringlengths 6
119
| max_issues_repo_head_hexsha
stringlengths 40
41
| max_issues_repo_licenses
list | max_issues_count
int64 1
67k
⌀ | max_issues_repo_issues_event_min_datetime
stringlengths 24
24
⌀ | max_issues_repo_issues_event_max_datetime
stringlengths 24
24
⌀ | max_forks_repo_path
stringlengths 6
260
| max_forks_repo_name
stringlengths 6
119
| max_forks_repo_head_hexsha
stringlengths 40
41
| max_forks_repo_licenses
list | max_forks_count
int64 1
105k
⌀ | max_forks_repo_forks_event_min_datetime
stringlengths 24
24
⌀ | max_forks_repo_forks_event_max_datetime
stringlengths 24
24
⌀ | avg_line_length
float64 2
1.04M
| max_line_length
int64 2
11.2M
| alphanum_fraction
float64 0
1
| cells
list | cell_types
list | cell_type_groups
list |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
cbd7fc55bfe58777d2909686a62bf88b50bef640
| 95,095 |
ipynb
|
Jupyter Notebook
|
tests/neural_networks.ipynb
|
AlejandroSantorum/Gas_sensors_home_activity_monitoring
|
118d10c5663c5e4c78c99d5c145250125f1da66c
|
[
"MIT"
] | 1 |
2021-02-06T11:22:18.000Z
|
2021-02-06T11:22:18.000Z
|
tests/neural_networks.ipynb
|
AlejandroSantorum/Gas_sensors_home_activity_monitoring
|
118d10c5663c5e4c78c99d5c145250125f1da66c
|
[
"MIT"
] | null | null | null |
tests/neural_networks.ipynb
|
AlejandroSantorum/Gas_sensors_home_activity_monitoring
|
118d10c5663c5e4c78c99d5c145250125f1da66c
|
[
"MIT"
] | null | null | null | 39.556988 | 935 | 0.473348 |
[
[
[
"%load_ext autoreload\n%autoreload 2\nimport warnings \nwarnings.filterwarnings('ignore')",
"_____no_output_____"
],
[
"import math\nfrom time import time\nimport pickle\nimport pandas as pd\nimport numpy as np\nfrom time import time\n\nfrom sklearn.neural_network import MLPClassifier\nfrom sklearn.ensemble import BaggingClassifier\nfrom sklearn.metrics import accuracy_score, f1_score",
"_____no_output_____"
],
[
"import sys\nsys.path.append('../src')\nfrom preprocessing import *\nfrom utils import *\nfrom plotting import *",
"_____no_output_____"
]
],
[
[
"# Splitting the dataset",
"_____no_output_____"
]
],
[
[
"features = ['R1', 'R2', 'R3', 'R4', 'R5', 'R6', 'R7', 'R8', 'Temp.', 'Humidity',\n 'R1_mean', 'R2_mean', 'R3_mean', 'R4_mean', 'R5_mean', 'R6_mean', 'R7_mean',\n 'R8_mean', 'Temp._mean', 'Humidity_mean', 'R1_std', 'R2_std', 'R3_std', 'R4_std',\n 'R5_std', 'R6_std', 'R7_std', 'R8_std', 'Temp._std', 'Humidity_std']\n\ndf_db = group_datafiles_byID('../datasets/preprocessed/HT_Sensor_prep_metadata.dat', '../datasets/preprocessed/HT_Sensor_prep_dataset.dat')\ndf_db = reclassify_series_samples(df_db)\ndf_db.head()",
"_____no_output_____"
],
[
"df_train, df_test = split_series_byID(0.75, df_db)\ndf_train, df_test = norm_train_test(df_train, df_test)\nfeatures = ['R1', 'R2', 'R3', 'R4', 'R5', 'R5', 'R6', 'R7', 'R8', 'Temp.', 'Humidity']\nxtrain, ytrain = df_train[features].values, df_train['class'].values\nxtest, ytest = df_test[features].values, df_test['class'].values",
"../src\\preprocessing.py:210: SettingWithCopyWarning: \nA value is trying to be set on a copy of a slice from a DataFrame.\nTry using .loc[row_indexer,col_indexer] = value instead\n\nSee the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n df_train[features_to_norm] = scaler.transform(df_train[features_to_norm])\nC:\\Users\\David\\anaconda3\\lib\\site-packages\\pandas\\core\\indexing.py:1734: SettingWithCopyWarning: \nA value is trying to be set on a copy of a slice from a DataFrame.\nTry using .loc[row_indexer,col_indexer] = value instead\n\nSee the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n isetter(loc, value[:, i].tolist())\n../src\\preprocessing.py:211: SettingWithCopyWarning: \nA value is trying to be set on a copy of a slice from a DataFrame.\nTry using .loc[row_indexer,col_indexer] = value instead\n\nSee the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n df_test[features_to_norm] = scaler.transform(df_test[features_to_norm])\nC:\\Users\\David\\anaconda3\\lib\\site-packages\\pandas\\core\\indexing.py:1734: SettingWithCopyWarning: \nA value is trying to be set on a copy of a slice from a DataFrame.\nTry using .loc[row_indexer,col_indexer] = value instead\n\nSee the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n isetter(loc, value[:, i].tolist())\n"
]
],
[
[
"# Basic Neural Network",
"_____no_output_____"
]
],
[
[
"def printResults(n_hid_layers,n_neur,accuracy,elapsed):\n print('========================================')\n print('Number of hidden layers:', n_hid_layers)\n print('Number of neurons per layer:', n_neur)\n print('Accuracy:', accuracy)\n print('Time (minutes):', (elapsed)/60)\n \ndef printScores(xtest,ytest,clf):\n xback, yback = xtest[ytest=='background'], ytest[ytest=='background']\n print('Score del background:', clf.score(xback,yback))\n xrest, yrest = xtest[ytest!='background'], ytest[ytest!='background']\n print('Score del resto:', clf.score(xrest,yrest))\n num_back = len(yback)\n num_wine = len(yrest[yrest=='wine'])\n num_banana = len(yrest[yrest=='banana'])\n func = lambda x: 1/num_back if x=='background' else (1/num_wine if x=='wine' else 1/num_banana)\n weights = np.array([func(x) for x in ytest])\n # Score donde las tres clases ponderan igual\n print('Score con pesos:', clf.score(xtest,ytest,weights))\n print('========================================')",
"_____no_output_____"
],
[
"# NN with 2 hidden layers and 15 neurons per layer\nxtrain, ytrain, xtest, ytest = split_train_test(df_db,0.75)\nstart = time.time()\n\nclf = MLPClassifier(hidden_layer_sizes=(15,15))\nclf.fit(xtrain,ytrain)\nscore = clf.score(xtest,ytest)\n\nfinal = time.time()\n\nprintResults(2,15,score,final-start)",
"C:\\Users\\David\\anaconda3\\lib\\site-packages\\sklearn\\neural_network\\_multilayer_perceptron.py:582: ConvergenceWarning: Stochastic Optimizer: Maximum iterations (200) reached and the optimization hasn't converged yet.\n warnings.warn(\n"
],
[
"# Adding early stopping and more iterations\nxtrain, ytrain, xtest, ytest = split_train_test(df_db,0.75)\nstart = time.time()\n\nclf = MLPClassifier(hidden_layer_sizes=(15,15),early_stopping=True,max_iter=2000)\nclf.fit(xtrain,ytrain)\nscore = clf.score(xtest,ytest)\n\nfinal = time.time()\n\nprintResults(2,15,score,final-start)",
"========================================\nNumber of hidden layers: 2\nNumber of neurons per layer: 15\nAccuracy: 0.77063573442707\nTime (minutes): 4.929224785168966\n"
],
[
"# Análisis del score\nprint('Proporcion de background:',len(ytest[ytest=='background'])/len(ytest))\nprintScores(xtest,ytest,clf)",
"Proporcion de background: 0.7930098051781594\nScore del background: 0.9041396866445102\nScore del resto: 0.25916250628894305\nScore con pesos: 0.4761530892089738\n========================================\n"
]
],
[
[
"Demasiado sesgo hacia el background, hay que reducirlo aunque el score baje",
"_____no_output_____"
],
[
"# Removing excess of background",
"_____no_output_____"
]
],
[
[
"# prop: ejemplos que no son background que habrá por cada ejemplo de background\ndef remove_bg(df,prop=2):\n new_df = df[df['class']!='background'].copy()\n useful_samples = new_df.shape[0]\n new_df = new_df.append(df[df['class']=='background'].sample(n=int(useful_samples/2)).copy())\n return new_df",
"_____no_output_____"
],
[
"# Para evitar el sesgo quitamos elementos clasificados como background, pero solo en el train set\ndf_train, df_test = split_series_byID(0.75, df_db)\ndf_train, df_test = norm_train_test(df_train, df_test)\ndf_train = remove_bg(df_train)",
"../src\\preprocessing.py:210: SettingWithCopyWarning: \nA value is trying to be set on a copy of a slice from a DataFrame.\nTry using .loc[row_indexer,col_indexer] = value instead\n\nSee the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n df_train[features_to_norm] = scaler.transform(df_train[features_to_norm])\nC:\\Users\\David\\anaconda3\\lib\\site-packages\\pandas\\core\\indexing.py:1734: SettingWithCopyWarning: \nA value is trying to be set on a copy of a slice from a DataFrame.\nTry using .loc[row_indexer,col_indexer] = value instead\n\nSee the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n isetter(loc, value[:, i].tolist())\n../src\\preprocessing.py:211: SettingWithCopyWarning: \nA value is trying to be set on a copy of a slice from a DataFrame.\nTry using .loc[row_indexer,col_indexer] = value instead\n\nSee the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n df_test[features_to_norm] = scaler.transform(df_test[features_to_norm])\nC:\\Users\\David\\anaconda3\\lib\\site-packages\\pandas\\core\\indexing.py:1734: SettingWithCopyWarning: \nA value is trying to be set on a copy of a slice from a DataFrame.\nTry using .loc[row_indexer,col_indexer] = value instead\n\nSee the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n isetter(loc, value[:, i].tolist())\n"
],
[
"features = ['R1', 'R2', 'R3', 'R4', 'R5', 'R5', 'R6', 'R7', 'R8', 'Temp.', 'Humidity']\nxtrain, ytrain = df_train[features].values, df_train['class'].values\nxtest, ytest = df_test[features].values, df_test['class'].values",
"_____no_output_____"
],
[
"start = time()\n\nclf = MLPClassifier(hidden_layer_sizes=(15,15),early_stopping=True,max_iter=2000)\nclf.fit(xtrain,ytrain)\nscore = clf.score(xtest,ytest)\n\nfinal = time()\n\nprintResults(2,15,score,final-start)",
"========================================\nNumber of hidden layers: 2\nNumber of neurons per layer: 15\nAccuracy: 0.6561803382403519\nTime (minutes): 1.2229466478029887\n"
],
[
"# Análisis del score\nprintScores(xtest,ytest,clf)",
"Score del background: 0.7097241472743\nScore del resto: 0.42049129567191906\nScore con pesos: 0.502403749329069\n========================================\n"
]
],
[
[
"Aunque se ponga la misma cantidad de background que de bananas o wine sigue habiendo un sesgo hacia el background.",
"_____no_output_____"
],
[
"# Hyperparameter analysis",
"_____no_output_____"
]
],
[
[
"xtrain, ytrain, xtest, ytest = split_train_test(df_db,0.75)\n\nstart_total = time.time()\n\nfor n_hid_layers in range(2,5):\n for n_neur in [10,20,40]:\n tup = []\n for i in range(n_hid_layers):\n tup.append(n_neur)\n tup = tuple(tup)\n\n start = time.time()\n\n clf_nn = MLPClassifier(\n hidden_layer_sizes = tup,\n max_iter=2000,\n early_stopping=True\n )\n\n clf_nn.fit(xtrain, ytrain)\n ypred = clf_nn.predict(xtest)\n final = time.time()\n metric_report(ytest, ypred)\n print('\\n====>Tiempo transcurrido (minutos):', (final-start)/(60))\n print('Number of hidden layers:', n_hid_layers)\n print('Number of neurons per layer:', n_neur)\n \n\nend_total = time()\nprint('\\n====> Total tiempo transcurrido (horas):', (end_total-start_total)/(60*60))",
"==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.8247409627112922\n\tReal banana percentage: 0.0280836516344145\n\tReal wine percentage: 0.14717538565429336\n------------------------------------------\nAccuracy: 0.8468778356933336\nRecall on background: 0.924579090118213\nRecall on banana: 0.35106702735196876\nRecall on wine: 0.5060652117805626\nF1-score: 0.8474634761377202\n==============================================\n\n====>Tiempo transcurrido (minutos): 1.5095600644747416\nNumber of hidden layers: 2\nNumber of neurons per layer: 10\n==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.8247409627112922\n\tReal banana percentage: 0.0280836516344145\n\tReal wine percentage: 0.14717538565429336\n------------------------------------------\nAccuracy: 0.852157764787811\nRecall on background: 0.9345529911468196\nRecall on banana: 0.23399458972046888\nRecall on wine: 0.508388058845459\nF1-score: 0.8486246905930726\n==============================================\n\n====>Tiempo transcurrido (minutos): 4.483641095956167\nNumber of hidden layers: 2\nNumber of neurons per layer: 20\n==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.8247409627112922\n\tReal banana percentage: 0.0280836516344145\n\tReal wine percentage: 0.14717538565429336\n------------------------------------------\nAccuracy: 0.8189081393631165\nRecall on background: 0.8987411084386674\nRecall on banana: 0.058611361587015326\nRecall on wine: 0.516618393507499\nF1-score: 0.8227555206040718\n==============================================\n\n====>Tiempo transcurrido (minutos): 9.807576270898183\nNumber of hidden layers: 2\nNumber of neurons per layer: 40\n==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.8247409627112922\n\tReal banana percentage: 0.0280836516344145\n\tReal wine percentage: 0.14717538565429336\n------------------------------------------\nAccuracy: 0.8398505919344968\nRecall on background: 0.8981526022209713\nRecall on banana: 0.35617673579801623\nRecall on wine: 0.6054314473344613\nF1-score: 0.8553256677723242\n==============================================\n\n====>Tiempo transcurrido (minutos): 2.541603982448578\nNumber of hidden layers: 3\nNumber of neurons per layer: 10\n==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.8247409627112922\n\tReal banana percentage: 0.0280836516344145\n\tReal wine percentage: 0.14717538565429336\n------------------------------------------\nAccuracy: 0.794053221347627\nRecall on background: 0.8658615219282534\nRecall on banana: 0.2690111211301473\nRecall on wine: 0.49184135814860486\nF1-score: 0.8141114941796247\n==============================================\n\n====>Tiempo transcurrido (minutos): 5.634796448548635\nNumber of hidden layers: 3\nNumber of neurons per layer: 20\n==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.8247409627112922\n\tReal banana percentage: 0.0280836516344145\n\tReal wine percentage: 0.14717538565429336\n------------------------------------------\nAccuracy: 0.7752379344545972\nRecall on background: 0.8487897241696944\nRecall on banana: 0.27923053802224224\nRecall on wine: 0.45771558028160936\nF1-score: 0.8031823284122757\n==============================================\n\n====>Tiempo transcurrido (minutos): 13.947646057605743\nNumber of hidden layers: 3\nNumber of neurons per layer: 40\n==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.8247409627112922\n\tReal banana percentage: 0.0280836516344145\n\tReal wine percentage: 0.14717538565429336\n------------------------------------------\nAccuracy: 0.8588600249013443\nRecall on background: 0.9172611432372959\nRecall on banana: 0.40562067929065226\nRecall on wine: 0.6180780591322302\nF1-score: 0.8692803863319071\n==============================================\n\n====>Tiempo transcurrido (minutos): 3.4188662211100262\nNumber of hidden layers: 4\nNumber of neurons per layer: 10\n==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.8247409627112922\n\tReal banana percentage: 0.0280836516344145\n\tReal wine percentage: 0.14717538565429336\n------------------------------------------\nAccuracy: 0.8009833920695549\nRecall on background: 0.8717056445422445\nRecall on banana: 0.06056507363991584\nRecall on wine: 0.545955091623412\nF1-score: 0.8135246981115328\n==============================================\n\n====>Tiempo transcurrido (minutos): 5.86545577843984\nNumber of hidden layers: 4\nNumber of neurons per layer: 20\n==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.8247409627112922\n\tReal banana percentage: 0.0280836516344145\n\tReal wine percentage: 0.14717538565429336\n------------------------------------------\nAccuracy: 0.7724354780847068\nRecall on background: 0.8489432475308326\nRecall on banana: 0.016230838593327322\nRecall on wine: 0.48799862349803563\nF1-score: 0.7924387268365047\n==============================================\n\n====>Tiempo transcurrido (minutos): 9.918685428301494\nNumber of hidden layers: 4\nNumber of neurons per layer: 40\n"
]
],
[
[
"# Two Neural Networks",
"_____no_output_____"
],
[
"## 1. Classify background",
"_____no_output_____"
]
],
[
[
"def printScoresBack(xtest,ytest,clf):\n xback, yback = xtest[ytest=='background'], ytest[ytest=='background']\n print('Score del background:', clf.score(xback,yback))\n xrest, yrest = xtest[ytest!='background'], ytest[ytest!='background']\n print('Score del resto:', clf.score(xrest,yrest))\n num_back = len(yback)\n num_rest = len(ytest)-num_back\n func = lambda x: 1/num_back if x=='background' else 1/num_rest\n weights = np.array([func(x) for x in ytest])\n # Score donde las tres clases ponderan igual\n print('Score con pesos:', clf.score(xtest,ytest,weights))\n print('========================================')",
"_____no_output_____"
],
[
"df_db = group_datafiles_byID('../datasets/raw/HT_Sensor_metadata.dat', '../datasets/raw/HT_Sensor_dataset.dat')\ndf_db = reclassify_series_samples(df_db)\ndf_db.loc[df_db['class']!='background','class'] = 'not-background'\ndf_db[df_db['class']!='background'].head()",
"_____no_output_____"
],
[
"# Primero probamos a no quitar el exceso de background\n\ndf_train, df_test = split_series_byID(0.75, df_db)\ndf_train, df_test = norm_train_test(df_train, df_test)\nfeatures = ['R1', 'R2', 'R3', 'R4', 'R5', 'R5', 'R6', 'R7', 'R8', 'Temp.', 'Humidity']\nxtrain, ytrain = df_train[features].values, df_train['class'].values\nxtest, ytest = df_test[features].values, df_test['class'].values\n\nstart_total = time.time()\n\nfor n_hid_layers in range(2,5):\n for n_neur in [10,20,40]:\n tup = []\n for i in range(n_hid_layers):\n tup.append(n_neur)\n tup = tuple(tup)\n\n start = time.time()\n\n clf_nn = MLPClassifier(\n hidden_layer_sizes = tup,\n max_iter=2000,\n early_stopping=True\n )\n\n clf_nn.fit(xtrain, ytrain)\n ypred = clf_nn.predict(xtest)\n final = time.time()\n metric_report(ytest, ypred)\n print('\\n====>Tiempo transcurrido (minutos):', (end_total-start_total)/(60))\n \n \n\nend_total = time.time()\nprint('\\n====> Total tiempo transcurrido (horas):', (end_total-start_total)/(60*60))\n# En más de la mitad de ocasiones aquellos datos que no son background son clasificados erroneamente.\n# Veamos si es cuestión de quitar background.",
"_____no_output_____"
],
[
"# Ahora, lo mismo quitando el exceso de background\ndf_train, df_test = split_series_byID(0.75, df_db)\ndf_train = remove_bg(df_train,prop=1)\ndf_train, df_test = norm_train_test(df_train, df_test)\nfeatures = ['R1', 'R2', 'R3', 'R4', 'R5', 'R5', 'R6', 'R7', 'R8', 'Temp.', 'Humidity']\nxtrain, ytrain = df_train[features].values, df_train['class'].values\nxtest, ytest = df_test[features].values, df_test['class'].values\n\nstart_total = time()\n\nfor n_hid_layers in range(2,5):\n for n_neur in [10,20,40]:\n tup = []\n for i in range(n_hid_layers):\n tup.append(n_neur)\n tup = tuple(tup)\n\n start = time()\n\n clf_nn = MLPClassifier(\n hidden_layer_sizes = tup,\n max_iter=2000,\n early_stopping=True,\n shuffle=True\n )\n\n clf_nn.fit(xtrain, ytrain)\n score = clf_nn.score(xtest, ytest)\n final = time()\n \n printResults(n_hid_layers,n_neur,score,final-start)\n printScoresBack(xtest,ytest,clf_nn)\n \n\nend_total = time()\nprint('\\n====> Total tiempo transcurrido (horas):', (end_total-start_total)/(60*60))",
"========================================\nNumber of hidden layers: 1\nNumber of neurons per layer: 5\nAccuracy: 0.5801633486905403\nTime (minutes): 0.4300204555193583\nScore del background: 0.5223014632760336\nScore del resto: 0.7499749841566326\nScore con pesos: 0.6361382237163332\n========================================\n========================================\nNumber of hidden layers: 1\nNumber of neurons per layer: 10\nAccuracy: 0.566574975523127\nTime (minutes): 0.21988988320032757\nScore del background: 0.5181815598806648\nScore del resto: 0.7085987792268437\nScore con pesos: 0.6133901695537544\n========================================\n========================================\nNumber of hidden layers: 1\nNumber of neurons per layer: 15\nAccuracy: 0.5998508076308506\nTime (minutes): 0.598010802268982\nScore del background: 0.5622503196476772\nScore del resto: 0.7101997932023615\nScore con pesos: 0.6362250564250194\n========================================\n========================================\nNumber of hidden layers: 1\nNumber of neurons per layer: 20\nAccuracy: 0.6004441863717856\nTime (minutes): 0.21585826476415\nScore del background: 0.5434919732916608\nScore del resto: 0.7675861378873287\nScore con pesos: 0.6555390555894948\n========================================\n========================================\nNumber of hidden layers: 1\nNumber of neurons per layer: 40\nAccuracy: 0.6325120688997486\nTime (minutes): 0.822665023803711\nScore del background: 0.5955561869583748\nScore del resto: 0.7409692805443447\nScore con pesos: 0.6682627337513597\n========================================\n========================================\nNumber of hidden layers: 2\nNumber of neurons per layer: 5\nAccuracy: 0.6060219465365755\nTime (minutes): 0.3326169490814209\nScore del background: 0.5761557039352181\nScore del resto: 0.6936726593509223\nScore con pesos: 0.6349141816430702\n========================================\n========================================\nNumber of hidden layers: 2\nNumber of neurons per layer: 10\nAccuracy: 0.6315880934317212\nTime (minutes): 0.3337648669878642\nScore del background: 0.6161727518113368\nScore del resto: 0.6768286581501618\nScore con pesos: 0.6465007049807494\n========================================\n========================================\nNumber of hidden layers: 2\nNumber of neurons per layer: 15\nAccuracy: 0.6457274611442885\nTime (minutes): 0.5144515752792358\nScore del background: 0.5996533598522518\nScore del resto: 0.7809445982455555\nScore con pesos: 0.6902989790489039\n========================================\n========================================\nNumber of hidden layers: 2\nNumber of neurons per layer: 20\nAccuracy: 0.6502752853515981\nTime (minutes): 0.9772834459940593\nScore del background: 0.6326523653928114\nScore del resto: 0.7019945965778326\nScore con pesos: 0.6673234809853219\n========================================\n========================================\nNumber of hidden layers: 2\nNumber of neurons per layer: 40\nAccuracy: 0.6782615698258433\nTime (minutes): 1.8018514831860861\nScore del background: 0.6653217786617417\nScore del resto: 0.7162369500683766\nScore con pesos: 0.6907793643650592\n========================================\n========================================\nNumber of hidden layers: 3\nNumber of neurons per layer: 5\nAccuracy: 0.6380898290645384\nTime (minutes): 0.6979097723960876\nScore del background: 0.6323341383719279\nScore del resto: 0.6549814882759081\nScore con pesos: 0.6436578133239181\n========================================\n========================================\nNumber of hidden layers: 3\nNumber of neurons per layer: 10\nAccuracy: 0.6396368522105478\nTime (minutes): 0.5433210690816244\nScore del background: 0.6452450632192073\nScore del resto: 0.6231780127414029\nScore con pesos: 0.6342115379803049\n========================================\n========================================\nNumber of hidden layers: 3\nNumber of neurons per layer: 15\nAccuracy: 0.7160428419450955\nTime (minutes): 1.3348949591318766\nScore del background: 0.7452820002841313\nScore del resto: 0.6302324805710283\nScore con pesos: 0.6877572404275798\n========================================\n========================================\nNumber of hidden layers: 3\nNumber of neurons per layer: 20\nAccuracy: 0.670848573983733\nTime (minutes): 0.8219222148259481\nScore del background: 0.6446427049296775\nScore del resto: 0.7477569127113839\nScore con pesos: 0.6961998088205306\n========================================\n========================================\nNumber of hidden layers: 3\nNumber of neurons per layer: 40\nAccuracy: 0.7024629456168384\nTime (minutes): 2.3735576113065084\nScore del background: 0.7749396221054127\nScore del resto: 0.48976018144825056\nScore con pesos: 0.6323499017768315\n========================================\n========================================\nNumber of hidden layers: 4\nNumber of neurons per layer: 5\nAccuracy: 0.632681605682873\nTime (minutes): 0.4298840324083964\nScore del background: 0.5667793720698963\nScore del resto: 0.826089856909376\nScore con pesos: 0.6964346144896361\n========================================\n========================================\nNumber of hidden layers: 4\nNumber of neurons per layer: 10\nAccuracy: 0.5993082899248529\nTime (minutes): 1.998073144753774\nScore del background: 0.5898621963347066\nScore del resto: 0.6270304526199927\nScore con pesos: 0.6084463244773497\n========================================\n========================================\nNumber of hidden layers: 4\nNumber of neurons per layer: 15\nAccuracy: 0.6333597528153702\nTime (minutes): 1.6846755266189575\nScore del background: 0.6412899559596533\nScore del resto: 0.6100863880457623\nScore con pesos: 0.6256881720027078\n========================================\n========================================\nNumber of hidden layers: 4\nNumber of neurons per layer: 20\nAccuracy: 0.6450832213684161\nTime (minutes): 1.398350465297699\nScore del background: 0.6647080551214661\nScore del resto: 0.5874887428704847\nScore con pesos: 0.6260983989959753\n========================================\n========================================\nNumber of hidden layers: 4\nNumber of neurons per layer: 40\nAccuracy: 0.6633550481696385\nTime (minutes): 2.475974464416504\nScore del background: 0.6868589288251172\nScore del resto: 0.5943764384109936\nScore con pesos: 0.6406176836180556\n========================================\n\n====> Total tiempo transcurrido (horas): 0.3439235324992074\n"
]
],
[
[
"## 2. Classify wine and bananas",
"_____no_output_____"
]
],
[
[
"df_db = group_datafiles_byID('../datasets/raw/HT_Sensor_metadata.dat', '../datasets/raw/HT_Sensor_dataset.dat')\ndf_db = reclassify_series_samples(df_db)\ndf_db = df_db[df_db['class']!='background']\ndf_db.head()",
"_____no_output_____"
],
[
"xtrain, ytrain, xtest, ytest = split_train_test(df_db,0.75)\n\nstart_total = time()\n\nfor n_hid_layers in range(1,5):\n for n_neur in [5,10,15,20,40]:\n tup = []\n for i in range(n_hid_layers):\n tup.append(n_neur)\n tup = tuple(tup)\n\n start = time()\n\n clf_nn = MLPClassifier(\n hidden_layer_sizes = tup,\n max_iter=2000,\n early_stopping=True,\n shuffle=True\n )\n\n clf_nn.fit(xtrain, ytrain)\n score = clf_nn.score(xtest, ytest)\n final = time()\n \n printResults(n_hid_layers,n_neur,score,final-start)\n \n\nend_total = time()\nprint('\\n====> Total tiempo transcurrido (horas):', (end_total-start_total)/(60*60))",
"========================================\nNumber of hidden layers: 1\nNumber of neurons per layer: 5\nAccuracy: 0.7908191826833364\nTime (minutes): 0.12984011173248292\n========================================\nNumber of hidden layers: 1\nNumber of neurons per layer: 10\nAccuracy: 0.7634353424146296\nTime (minutes): 0.47473315795262655\n========================================\nNumber of hidden layers: 1\nNumber of neurons per layer: 15\nAccuracy: 0.7635053181563725\nTime (minutes): 0.5810888727506002\n========================================\nNumber of hidden layers: 1\nNumber of neurons per layer: 20\nAccuracy: 0.7582804627729054\nTime (minutes): 0.370017139116923\n========================================\nNumber of hidden layers: 1\nNumber of neurons per layer: 40\nAccuracy: 0.7931517074080985\nTime (minutes): 1.0173205216725667\n========================================\nNumber of hidden layers: 2\nNumber of neurons per layer: 5\nAccuracy: 0.6941593580891957\nTime (minutes): 0.43400735457738243\n========================================\nNumber of hidden layers: 2\nNumber of neurons per layer: 10\nAccuracy: 0.6807473409218138\nTime (minutes): 0.7764243205388387\n========================================\nNumber of hidden layers: 2\nNumber of neurons per layer: 15\nAccuracy: 0.7795297630154879\nTime (minutes): 1.0063043196996053\n========================================\nNumber of hidden layers: 2\nNumber of neurons per layer: 20\nAccuracy: 0.7281209180817316\nTime (minutes): 1.430491860707601\n========================================\nNumber of hidden layers: 2\nNumber of neurons per layer: 40\nAccuracy: 0.7259749953349506\nTime (minutes): 1.7645756681760152\n========================================\nNumber of hidden layers: 3\nNumber of neurons per layer: 5\nAccuracy: 0.7197938048143311\nTime (minutes): 0.6930012027422587\n========================================\nNumber of hidden layers: 3\nNumber of neurons per layer: 10\nAccuracy: 0.6850858369098712\nTime (minutes): 0.9055802067120869\n========================================\nNumber of hidden layers: 3\nNumber of neurons per layer: 15\nAccuracy: 0.7200970330285501\nTime (minutes): 0.4312401374181112\n========================================\nNumber of hidden layers: 3\nNumber of neurons per layer: 20\nAccuracy: 0.7579772345586864\nTime (minutes): 1.8903615673383076\n========================================\nNumber of hidden layers: 3\nNumber of neurons per layer: 40\nAccuracy: 0.7981433103190894\nTime (minutes): 1.1435235063234965\n========================================\nNumber of hidden layers: 4\nNumber of neurons per layer: 5\nAccuracy: 0.7782468744168688\nTime (minutes): 0.4978533426920573\n========================================\nNumber of hidden layers: 4\nNumber of neurons per layer: 10\nAccuracy: 0.6765954469117372\nTime (minutes): 2.114302909374237\n========================================\nNumber of hidden layers: 4\nNumber of neurons per layer: 15\nAccuracy: 0.7747480873297257\nTime (minutes): 0.8926546216011048\n========================================\nNumber of hidden layers: 4\nNumber of neurons per layer: 20\nAccuracy: 0.7583271132674007\nTime (minutes): 1.4074350674947103\n========================================\nNumber of hidden layers: 4\nNumber of neurons per layer: 40\nAccuracy: 0.7580705355476768\nTime (minutes): 2.1244267861048383\n\n====> Total tiempo transcurrido (horas): 0.33475332227018145\n"
]
],
[
[
"# 3. Merge the 2 NN",
"_____no_output_____"
]
],
[
[
"class doubleNN:\n def __init__(self, n_hid_layers, n_neur):\n self.hid_layers = n_hid_layers\n self.neur = n_neur\n \n tup = []\n for i in range(n_hid_layers):\n tup.append(n_neur)\n tup = tuple(tup)\n self.backNN = MLPClassifier(\n hidden_layer_sizes = tup,\n max_iter=2000,\n early_stopping=True,\n shuffle=True\n )\n self.wineNN = MLPClassifier(\n hidden_layer_sizes = tup,\n max_iter=2000,\n early_stopping=True,\n shuffle=True\n )\n \n def fit_bg(self, xtrain, ytrain):\n ytrain_copy = np.array([x if x=='background' else 'not-background' for x in ytrain])\n self.backNN.fit(xtrain, ytrain_copy)\n \n def fit_wine(self,xtrain,ytrain):\n self.wineNN.fit(xtrain, ytrain)\n \n def predict(self,xtest):\n ypred = self.backNN.predict(xtest)\n ypred[ypred=='not-background'] = self.wineNN.predict(xtest[ypred=='not-background'])\n return ypred\n \n def score(self,xtest,ytest):\n ypred = self.predict(ytest)\n score = np.sum(np.equal(ypred,ytest))/len(ytest)\n return score",
"_____no_output_____"
],
[
"# With all the background\nxtrain, ytrain, xtest, ytest = split_train_test(df_db,0.75)\n\nstart_total = time.time()\n\nfor n_hid_layers in range(2,4):\n for n_neur in [10,20]:\n tup = []\n for i in range(n_hid_layers):\n tup.append(n_neur)\n tup = tuple(tup)\n\n start = time.time()\n\n clf_nn = doubleNN(2,20)\n clf_nn.fit_bg(xtrain, ytrain)\n\n xtrain_notbg = xtrain[ytrain != 'background']\n ytrain_notbg = ytrain[ytrain != 'background']\n clf_nn.fit_wine(xtrain_notbg, ytrain_notbg)\n ypred = clf_nn.predict(xtest)\n \n final = time.time()\n metric_report(ytest, ypred)\n print('\\n====>Tiempo transcurrido (minutos):', (final-start)/(60))\n print('Number of hidden layers:', n_hid_layers)\n print('Number of neurons per layer:', n_neur)\n \n \n\nend_total = time.time()\nprint('\\n====> Total tiempo transcurrido (horas):', (end_total-start_total)/(60*60))",
"==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.7869137667040765\n\tReal banana percentage: 0.03354791874402867\n\tReal wine percentage: 0.17953831455189476\n------------------------------------------\nAccuracy: 0.782324448134442\nRecall on background: 0.8975362249685864\nRecall on banana: 0.1738589727645815\nRecall on wine: 0.39104873004438456\nF1-score: 0.7820318012758023\n==============================================\n\n====>Tiempo transcurrido (minutos): 3.209248713652293\nNumber of hidden layers: 2\nNumber of neurons per layer: 10\n==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.7869137667040765\n\tReal banana percentage: 0.03354791874402867\n\tReal wine percentage: 0.17953831455189476\n------------------------------------------\nAccuracy: 0.7321673682761268\nRecall on background: 0.8322544045214275\nRecall on banana: 0.14376321353065538\nRecall on wine: 0.4034345734668743\nF1-score: 0.741516705842115\n==============================================\n\n====>Tiempo transcurrido (minutos): 4.176293102900187\nNumber of hidden layers: 2\nNumber of neurons per layer: 20\n==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.7869137667040765\n\tReal banana percentage: 0.03354791874402867\n\tReal wine percentage: 0.17953831455189476\n------------------------------------------\nAccuracy: 0.7523144767968225\nRecall on background: 0.8608738528097215\nRecall on banana: 0.1573187414500684\nRecall on wine: 0.3876792229219436\nF1-score: 0.7645714445349678\n==============================================\n\n====>Tiempo transcurrido (minutos): 4.866243561108907\nNumber of hidden layers: 3\nNumber of neurons per layer: 10\n==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.7869137667040765\n\tReal banana percentage: 0.03354791874402867\n\tReal wine percentage: 0.17953831455189476\n------------------------------------------\nAccuracy: 0.7015607855244548\nRecall on background: 0.7748193390699475\nRecall on banana: 0.07548812336774033\nRecall on wine: 0.49745544117305324\nF1-score: 0.7293895255531202\n==============================================\n\n====>Tiempo transcurrido (minutos): 3.0275081912676494\nNumber of hidden layers: 3\nNumber of neurons per layer: 20\n\n====> Total tiempo transcurrido (horas): 0.2565733293029997\n"
],
[
"# Removing background\ndf_train, df_test = split_series_byID(0.75, df_db)\ndf_train = remove_bg(df_train,prop=1)\nfeatures = ['R1', 'R2', 'R3', 'R4', 'R5', 'R5', 'R6', 'R7', 'R8', 'Temp.', 'Humidity']\nxtrain, ytrain = df_train[features].values, df_train['class'].values\nxtest, ytest = df_test[features].values, df_test['class'].values\n\nstart_total = time.time()\n\nfor n_hid_layers in range(2,4):\n for n_neur in [10,20]:\n tup = []\n for i in range(n_hid_layers):\n tup.append(n_neur)\n tup = tuple(tup)\n\n start = time.time()\n\n clf_nn = doubleNN(2,20)\n clf_nn.fit_bg(xtrain, ytrain)\n\n xtrain_notbg = xtrain[ytrain != 'background']\n ytrain_notbg = ytrain[ytrain != 'background']\n clf_nn.fit_wine(xtrain_notbg, ytrain_notbg)\n ypred = clf_nn.predict(xtest)\n \n final = time.time()\n metric_report(ytest, ypred)\n print('\\n====>Tiempo transcurrido (minutos):', (final-start)/(60))\n print('Number of hidden layers:', n_hid_layers)\n print('Number of neurons per layer:', n_neur)\n \n \n\nend_total = time.time()\nprint('\\n====> Total tiempo transcurrido (horas):', (end_total-start_total)/(60*60))",
"==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.8057390890907302\n\tReal banana percentage: 0.09305833360658425\n\tReal wine percentage: 0.1012025773026855\n------------------------------------------\nAccuracy: 0.7004500442666491\nRecall on background: 0.7943239682370117\nRecall on banana: 0.12482381959126145\nRecall on wine: 0.4823619942489166\nF1-score: 0.7126424791959668\n==============================================\n\n====>Tiempo transcurrido (minutos): 1.6882099628448486\nNumber of hidden layers: 2\nNumber of neurons per layer: 10\n==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.8057390890907302\n\tReal banana percentage: 0.09305833360658425\n\tReal wine percentage: 0.1012025773026855\n------------------------------------------\nAccuracy: 0.7359658327048562\nRecall on background: 0.8590555547077287\nRecall on banana: 0.163715644820296\nRecall on wine: 0.28216759143007575\nF1-score: 0.7346641638156696\n==============================================\n\n====>Tiempo transcurrido (minutos): 1.2407544692357382\nNumber of hidden layers: 2\nNumber of neurons per layer: 20\n==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.8057390890907302\n\tReal banana percentage: 0.09305833360658425\n\tReal wine percentage: 0.1012025773026855\n------------------------------------------\nAccuracy: 0.744343705938289\nRecall on background: 0.8532411575889837\nRecall on banana: 0.10606060606060606\nRecall on wine: 0.4642582317443603\nF1-score: 0.7467180671178155\n==============================================\n\n====>Tiempo transcurrido (minutos): 2.2511873960494997\nNumber of hidden layers: 3\nNumber of neurons per layer: 10\n==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.8057390890907302\n\tReal banana percentage: 0.09305833360658425\n\tReal wine percentage: 0.1012025773026855\n------------------------------------------\nAccuracy: 0.724042528773322\nRecall on background: 0.8218240826936479\nRecall on banana: 0.1727448907681466\nRecall on wine: 0.45247256085213233\nF1-score: 0.7411701496183302\n==============================================\n\n====>Tiempo transcurrido (minutos): 1.373010245958964\nNumber of hidden layers: 3\nNumber of neurons per layer: 20\n\n====> Total tiempo transcurrido (horas): 0.11125850253634982\n"
]
],
[
[
"# Creating Windows",
"_____no_output_____"
]
],
[
[
"# with open('../datasets/preprocessed/window120_dataset.pkl', 'wb') as f:\n# pickle.dump(win_df, f)\n \nwin_df = pd.read_pickle('../datasets/preprocessed/window120_dataset.pkl')",
"_____no_output_____"
],
[
"xtrain, ytrain, xtest, ytest = split_train_test(win_df,0.75)\n\nstart = time.time()\n\nclf_nn = MLPClassifier(\n hidden_layer_sizes = (32,16),\n max_iter=2000,\n early_stopping=True,\n shuffle=True,\n alpha=0.01,\n learning_rate_init=0.01\n )\n\nclf_nn.fit(xtrain, ytrain)\nypred = clf_nn.predict(xtest)\nfinal = time.time()\nmetric_report(ytest, ypred)\nprint('\\n====>Tiempo transcurrido (minutos):', (final-start)/(60))",
"==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.7348271825570399\n\tReal banana percentage: 0.09707520485230221\n\tReal wine percentage: 0.1680976125906578\n------------------------------------------\nAccuracy: 0.7699655489875068\nRecall on background: 0.8834440690265085\nRecall on banana: 0.24589108698450843\nRecall on wine: 0.5765513481363996\nF1-score: 0.7671143839504504\n==============================================\n\n====>Tiempo transcurrido (minutos): 4.107463216781616\n"
],
[
"features = ['R1', 'R2', 'R3', 'R4', 'R5', 'R6', 'R7', 'R8', 'Temp.', 'Humidity',\n 'R1_mean', 'R2_mean', 'R3_mean', 'R4_mean', 'R5_mean', 'R6_mean', 'R7_mean',\n 'R8_mean', 'Temp._mean', 'Humidity_mean', 'R1_std', 'R2_std', 'R3_std', 'R4_std',\n 'R5_std', 'R6_std', 'R7_std', 'R8_std', 'Temp._std', 'Humidity_std']\n\n# Varía ciertos hiperparámetros con ventanas e imprime los resultados más relevantes\ndef hyper_sim(win_df,num_val,n_hid_layers,n_neur,alpha):\n errs_acc = []\n errs_f1 = []\n rec_ban = []\n loss = []\n for i in range(num_val):\n df_train, df_test = split_series_byID(0.75, win_df)\n df_train, df_test = norm_train_test(df_train,df_test,features_to_norm=features)\n xtrain, ytrain = df_train[features].values, df_train['class'].values\n xtest, ytest = df_test[features].values, df_test['class'].values\n \n tup = []\n for i in range(n_hid_layers):\n tup.append(n_neur)\n tup = tuple(tup)\n\n clf_nn = MLPClassifier(\n hidden_layer_sizes=tup,\n max_iter=2000,\n early_stopping=True,\n shuffle=True,\n alpha=alpha,\n learning_rate='adaptive'\n )\n\n clf_nn.fit(xtrain, ytrain)\n ypred = clf_nn.predict(xtest)\n errs_acc.append(accuracy_score(ytest,ypred))\n errs_f1.append(f1_score(ytest,ypred,average='weighted'))\n rec_ban.append(np.sum(np.logical_and(ytest=='banana',ypred=='banana'))/np.sum(ytest=='banana'))\n loss.append(clf_nn.loss_)\n\n errs_acc = np.array(errs_acc)\n errs_f1 = np.array(errs_f1)\n rec_ban = np.array(rec_ban)\n loss = np.array(loss)\n print('Train loss:',np.mean(loss),'+-',np.std(loss))\n print('Accuracy:',np.mean(errs_acc),'+-',np.std(errs_acc))\n print('F1-score:',np.mean(errs_f1),'+-',np.std(errs_f1))\n print('Recall bananas:',np.mean(rec_ban),'+-',np.std(rec_ban))",
"_____no_output_____"
],
[
"for alpha in [0.1,0.01,0.001]:\n print('<<<<<<<<<<<<<<<<<<<<<<<<<<<<<>>>>>>>>>>>>>>>>>')\n print('Alpha:',alpha)\n for n_hid_layers in range(1,4):\n print('##############################################')\n print('\\t Hidden layers:',n_hid_layers)\n for n_neur in [4,8,16]:\n print('==============================================')\n print('\\t \\t Neurons per layer:',n_neur)\n hyper_sim(win_df,3,n_hid_layers,n_neur,alpha)\n print('==============================================')",
"<<<<<<<<<<<<<<<<<<<<<<<<<<<<<>>>>>>>>>>>>>>>>>\nAlpha: 0.1\n##############################################\n\t Hidden layers: 1\n==============================================\n\t \t Neurons per layer: 4\nTrain loss: 0.30631837585880617 +- 0.023258576145879454\nAccuracy: 0.8651469754197983 +- 0.028237533883041985\nF1-score: 0.8449596309548207 +- 0.037541891251813155\nRecall bananas: 0.24503317567532323 +- 0.07087260076560559\n==============================================\n==============================================\n\t \t Neurons per layer: 8\nTrain loss: 0.24434164283115822 +- 0.010833188926485271\nAccuracy: 0.845557828927527 +- 0.01047009214209502\nF1-score: 0.8361074906063378 +- 0.007906986730144373\nRecall bananas: 0.24727647351281004 +- 0.02763976878187987\n==============================================\n==============================================\n\t \t Neurons per layer: 16\nTrain loss: 0.1686303778607473 +- 0.009096902594708849\nAccuracy: 0.8361197586130197 +- 0.03367222583611265\nF1-score: 0.815334460775949 +- 0.04026998400344681\nRecall bananas: 0.23011923008334897 +- 0.14783895790099125\n==============================================\n##############################################\n\t Hidden layers: 2\n==============================================\n\t \t Neurons per layer: 4\nTrain loss: 0.2683550733826729 +- 0.015092405102973596\nAccuracy: 0.8613628565857244 +- 0.011758195527331325\nF1-score: 0.8556709869925396 +- 0.01124000132324505\nRecall bananas: 0.4302045070747062 +- 0.06929185865224505\n==============================================\n==============================================\n\t \t Neurons per layer: 8\nTrain loss: 0.18413087397477268 +- 0.012413919682108013\nAccuracy: 0.8085851758592136 +- 0.03583793010995266\nF1-score: 0.8048862683352359 +- 0.03184637520315497\nRecall bananas: 0.27325889817440957 +- 0.0620316324261499\n==============================================\n==============================================\n\t \t Neurons per layer: 16\nTrain loss: 0.10524684957713497 +- 0.007843617944864193\nAccuracy: 0.7888237260774348 +- 0.0577437412876078\nF1-score: 0.767102845342813 +- 0.07054430236105699\nRecall bananas: 0.2255360538031643 +- 0.07690426988019564\n==============================================\n##############################################\n\t Hidden layers: 3\n==============================================\n\t \t Neurons per layer: 4\nTrain loss: 0.25262325831367693 +- 0.007818908778971026\nAccuracy: 0.8438201628355048 +- 0.03401956167829461\nF1-score: 0.8294578212401523 +- 0.034700711249015616\nRecall bananas: 0.30880505120089285 +- 0.05323353609541329\n==============================================\n==============================================\n\t \t Neurons per layer: 8\nTrain loss: 0.161916495085799 +- 0.00867867000757553\nAccuracy: 0.807212954685998 +- 0.033964184988541105\nF1-score: 0.804746514783271 +- 0.03971342592911781\nRecall bananas: 0.18984754434008302 +- 0.03812141227432121\n==============================================\n==============================================\n\t \t Neurons per layer: 16\nTrain loss: 0.08685706016419507 +- 0.003973613154774003\nAccuracy: 0.7630209096997321 +- 0.014822082668194214\nF1-score: 0.7779699742305017 +- 0.0023246453190704282\nRecall bananas: 0.3725600160176812 +- 0.17559022042569752\n==============================================\n<<<<<<<<<<<<<<<<<<<<<<<<<<<<<>>>>>>>>>>>>>>>>>\nAlpha: 0.01\n##############################################\n\t Hidden layers: 1\n==============================================\n\t \t Neurons per layer: 4\nTrain loss: 0.26798739757639617 +- 0.018537094326734678\nAccuracy: 0.8379045835274447 +- 0.03332120426870318\nF1-score: 0.8217012694107337 +- 0.0429520988601411\nRecall bananas: 0.30688738289786693 +- 0.11252089398632736\n==============================================\n==============================================\n\t \t Neurons per layer: 8\nTrain loss: 0.1814170028845931 +- 0.021060214047325016\nAccuracy: 0.8092544059644798 +- 0.05398035343357868\nF1-score: 0.7929234487438119 +- 0.05236955298917366\nRecall bananas: 0.2932916755271471 +- 0.01954715140566428\n==============================================\n==============================================\n\t \t Neurons per layer: 16\nTrain loss: 0.13668053136327707 +- 0.013475685864011289\nAccuracy: 0.8405045494928179 +- 0.022376009932510873\nF1-score: 0.8308206853765445 +- 0.024229269699032937\nRecall bananas: 0.2055767780048334 +- 0.0492918087690434\n==============================================\n##############################################\n\t Hidden layers: 2\n==============================================\n\t \t Neurons per layer: 4\nTrain loss: 0.2444411658614257 +- 0.02433236024842642\nAccuracy: 0.85068785362034 +- 0.029554298250406515\nF1-score: 0.8428660971042526 +- 0.03895132104639477\nRecall bananas: 0.38630114072282473 +- 0.12244346126605067\n==============================================\n==============================================\n\t \t Neurons per layer: 8\nTrain loss: 0.13968794445032995 +- 0.012551494974895635\nAccuracy: 0.7968768935192893 +- 0.027962284232071526\nF1-score: 0.7955946602604994 +- 0.032130913132534765\nRecall bananas: 0.32251206538127786 +- 0.07476341558117484\n==============================================\n==============================================\n\t \t Neurons per layer: 16\nTrain loss: 0.0479406185157547 +- 0.0029359049035783914\nAccuracy: 0.8156473412320598 +- 0.0433465199972356\nF1-score: 0.8064372038788009 +- 0.04624959668451317\nRecall bananas: 0.32856839856004466 +- 0.13373641502165712\n==============================================\n##############################################\n\t Hidden layers: 3\n==============================================\n\t \t Neurons per layer: 4\nTrain loss: 0.2849449330870414 +- 0.01832422415956819\nAccuracy: 0.8770988360219517 +- 0.030511917778153107\nF1-score: 0.8711000506145737 +- 0.031108297863556168\nRecall bananas: 0.2900750885280794 +- 0.11125115101190192\n==============================================\n==============================================\n\t \t Neurons per layer: 8\nTrain loss: 0.11969453406816709 +- 0.00853642026827579\nAccuracy: 0.8222128141454604 +- 0.01532099120245045\nF1-score: 0.8164753549359 +- 0.01696744666017833\nRecall bananas: 0.3911346320042795 +- 0.03977812420167948\n==============================================\n==============================================\n\t \t Neurons per layer: 16\nTrain loss: 0.03158684098722545 +- 0.0023338330090645525\nAccuracy: 0.7395066220852052 +- 0.03684442013668107\nF1-score: 0.7502997053494179 +- 0.035429672917204606\nRecall bananas: 0.30485940885045887 +- 0.008909926284970159\n==============================================\n<<<<<<<<<<<<<<<<<<<<<<<<<<<<<>>>>>>>>>>>>>>>>>\nAlpha: 0.001\n##############################################\n\t Hidden layers: 1\n==============================================\n\t \t Neurons per layer: 4\nTrain loss: 0.27065198866025036 +- 0.013683998682346538\nAccuracy: 0.8684824004217083 +- 0.016174244789373647\nF1-score: 0.8549382271531831 +- 0.021985629645875318\nRecall bananas: 0.2617675499080439 +- 0.08473599188828813\n==============================================\n==============================================\n\t \t Neurons per layer: 8\nTrain loss: 0.20261987627551958 +- 0.008233150176233774\nAccuracy: 0.847857949057785 +- 0.01079989328158993\nF1-score: 0.8223007194836226 +- 0.019003741315711472\nRecall bananas: 0.1774231093998612 +- 0.14986804678255072\n==============================================\n==============================================\n\t \t Neurons per layer: 16\nTrain loss: 0.0959903765639923 +- 0.013925263485351265\nAccuracy: 0.8011973212742222 +- 0.05086756960301938\nF1-score: 0.7989846549035309 +- 0.04262519564804963\nRecall bananas: 0.3101920378251784 +- 0.03864370847392516\n==============================================\n##############################################\n\t Hidden layers: 2\n==============================================\n\t \t Neurons per layer: 4\nTrain loss: 0.2484555992294145 +- 0.00957010965505604\nAccuracy: 0.8488773104590376 +- 0.03326078649202252\nF1-score: 0.8394453205554976 +- 0.03035146840430138\nRecall bananas: 0.257917883837867 +- 0.09998035958137114\n==============================================\n==============================================\n\t \t Neurons per layer: 8\n"
],
[
"# Nos quedamos con:\n# alpha: 0.01\n# hidden_layers: 3\n# n_neurons: 4\nfeatures = ['R1', 'R2', 'R3', 'R4', 'R5', 'R6', 'R7', 'R8', 'Temp.', 'Humidity',\n 'R1_mean', 'R2_mean', 'R3_mean', 'R4_mean', 'R5_mean', 'R6_mean', 'R7_mean',\n 'R8_mean', 'Temp._mean', 'Humidity_mean', 'R1_std', 'R2_std', 'R3_std', 'R4_std',\n 'R5_std', 'R6_std', 'R7_std', 'R8_std', 'Temp._std', 'Humidity_std']\n\n\nerrs_acc = []\nerrs_f1 = []\nrec_ban = []\nfor i in range(5):\n df_train, df_test = split_series_byID(0.75, win_df)\n df_train, df_test = norm_train_test(df_train,df_test,features_to_norm=features)\n xtrain, ytrain = df_train[features].values, df_train['class'].values\n xtest, ytest = df_test[features].values, df_test['class'].values\n\n clf_nn = MLPClassifier(\n hidden_layer_sizes=(4,4,4),\n max_iter=2000,\n early_stopping=True,\n shuffle=True,\n alpha=0.01,\n learning_rate='adaptive'\n )\n\n bag = BaggingClassifier(base_estimator=clf_nn,n_estimators=100,n_jobs=3)\n\n bag.fit(xtrain, ytrain)\n ypred = bag.predict(xtest)\n metric_report(ytest, ypred)\n errs_acc.append(accuracy_score(ytest,ypred))\n errs_f1.append(f1_score(ytest,ypred,average='weighted'))\n rec_ban.append(np.sum(np.logical_and(ytest=='banana',ypred=='banana'))/np.sum(ytest=='banana'))\n\nerrs_acc = np.array(errs_acc)\nerrs_f1 = np.array(errs_f1)\nrec_ban = np.array(rec_ban)\nprint('Accuracy:',np.mean(errs_acc),'+-',np.std(errs_acc))\nprint('F1-score:',np.mean(errs_f1),'+-',np.std(errs_f1))\nprint('Recall bananas:',np.mean(rec_ban),'+-',np.std(rec_ban))",
"==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.753916922093646\n\tReal banana percentage: 0.1666419116290165\n\tReal wine percentage: 0.07944116627733744\n------------------------------------------\nAccuracy: 0.7947987415439042\nRecall on background: 0.9468905034536695\nRecall on banana: 0.17899200518582542\nRecall on wine: 0.6431728045325779\nF1-score: 0.7606228511303869\n==============================================\n==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.7767477738985424\n\tReal banana percentage: 0.11219462168537171\n\tReal wine percentage: 0.11105760441608589\n------------------------------------------\nAccuracy: 0.8900125294844184\nRecall on background: 0.9690876109344324\nRecall on banana: 0.33550592162785153\nRecall on wine: 0.8971373509455173\nF1-score: 0.8764919643302215\n==============================================\n==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.8233239114555213\n\tReal banana percentage: 0.087355779482689\n\tReal wine percentage: 0.08932030906178967\n------------------------------------------\nAccuracy: 0.8873276499598474\nRecall on background: 0.967702294619437\nRecall on banana: 0.2249402721512413\nRecall on wine: 0.7942804896632295\nF1-score: 0.8714271160647824\n==============================================\n==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.6937389698070827\n\tReal banana percentage: 0.0935663892948048\n\tReal wine percentage: 0.21269464089811246\n------------------------------------------\nAccuracy: 0.8593208653308299\nRecall on background: 0.9598094857114283\nRecall on banana: 0.28441004727036795\nRecall on wine: 0.7844692259077287\nF1-score: 0.8474163528877009\n==============================================\n==============================================\nTEST SET PROPORTIONS:\n\tReal background percentage: 0.6997085833623948\n\tReal banana percentage: 0.11173643429496816\n\tReal wine percentage: 0.188554982342637\n------------------------------------------\nAccuracy: 0.872341875408364\nRecall on background: 0.9461207513877988\nRecall on banana: 0.5237668161434977\nRecall on wine: 0.8051191564446523\nF1-score: 0.8675737584057791\n==============================================\nAccuracy: 0.8607603323454729 +- 0.03478117039076238\nF1-score: 0.8447064085637741 +- 0.04318477019141248\nRecall bananas: 0.30952301247575675 +- 0.11949791950018472\n"
],
[
"with open('../datasets/preprocessed/nn_optimal.pkl', 'wb') as f:\n pickle.dump(bag, f)",
"_____no_output_____"
]
]
] |
[
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code",
"code",
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code"
]
] |
cbd80eb2181ed10dc159f7ef411cf49837fdd833
| 21,939 |
ipynb
|
Jupyter Notebook
|
chatbot/12.gru-birnn-seq2seq-greedy.ipynb
|
huseinzol05/Tensorflow-NLP-Models
|
0741216aa8235e1228b3de7903cc36d73f8f2b45
|
[
"MIT"
] | 1,705 |
2018-11-03T17:34:22.000Z
|
2022-03-29T04:30:01.000Z
|
chatbot/12.gru-birnn-seq2seq-greedy.ipynb
|
eridgd/NLP-Models-Tensorflow
|
d46e746cd038f25e8ee2df434facbe12e31576a1
|
[
"MIT"
] | 26 |
2019-03-16T17:23:00.000Z
|
2021-10-08T08:06:09.000Z
|
chatbot/12.gru-birnn-seq2seq-greedy.ipynb
|
eridgd/NLP-Models-Tensorflow
|
d46e746cd038f25e8ee2df434facbe12e31576a1
|
[
"MIT"
] | 705 |
2018-11-03T17:34:25.000Z
|
2022-03-24T02:29:14.000Z
| 37.695876 | 141 | 0.531109 |
[
[
[
"import numpy as np\nimport tensorflow as tf\nfrom sklearn.utils import shuffle\nimport re\nimport time\nimport collections\nimport os",
"_____no_output_____"
],
[
"def build_dataset(words, n_words, atleast=1):\n count = [['PAD', 0], ['GO', 1], ['EOS', 2], ['UNK', 3]]\n counter = collections.Counter(words).most_common(n_words)\n counter = [i for i in counter if i[1] >= atleast]\n count.extend(counter)\n dictionary = dict()\n for word, _ in count:\n dictionary[word] = len(dictionary)\n data = list()\n unk_count = 0\n for word in words:\n index = dictionary.get(word, 0)\n if index == 0:\n unk_count += 1\n data.append(index)\n count[0][1] = unk_count\n reversed_dictionary = dict(zip(dictionary.values(), dictionary.keys()))\n return data, count, dictionary, reversed_dictionary",
"_____no_output_____"
],
[
"lines = open('movie_lines.txt', encoding='utf-8', errors='ignore').read().split('\\n')\nconv_lines = open('movie_conversations.txt', encoding='utf-8', errors='ignore').read().split('\\n')\n\nid2line = {}\nfor line in lines:\n _line = line.split(' +++$+++ ')\n if len(_line) == 5:\n id2line[_line[0]] = _line[4]\n \nconvs = [ ]\nfor line in conv_lines[:-1]:\n _line = line.split(' +++$+++ ')[-1][1:-1].replace(\"'\",\"\").replace(\" \",\"\")\n convs.append(_line.split(','))\n \nquestions = []\nanswers = []\n\nfor conv in convs:\n for i in range(len(conv)-1):\n questions.append(id2line[conv[i]])\n answers.append(id2line[conv[i+1]])\n \ndef clean_text(text):\n text = text.lower()\n text = re.sub(r\"i'm\", \"i am\", text)\n text = re.sub(r\"he's\", \"he is\", text)\n text = re.sub(r\"she's\", \"she is\", text)\n text = re.sub(r\"it's\", \"it is\", text)\n text = re.sub(r\"that's\", \"that is\", text)\n text = re.sub(r\"what's\", \"that is\", text)\n text = re.sub(r\"where's\", \"where is\", text)\n text = re.sub(r\"how's\", \"how is\", text)\n text = re.sub(r\"\\'ll\", \" will\", text)\n text = re.sub(r\"\\'ve\", \" have\", text)\n text = re.sub(r\"\\'re\", \" are\", text)\n text = re.sub(r\"\\'d\", \" would\", text)\n text = re.sub(r\"\\'re\", \" are\", text)\n text = re.sub(r\"won't\", \"will not\", text)\n text = re.sub(r\"can't\", \"cannot\", text)\n text = re.sub(r\"n't\", \" not\", text)\n text = re.sub(r\"n'\", \"ng\", text)\n text = re.sub(r\"'bout\", \"about\", text)\n text = re.sub(r\"'til\", \"until\", text)\n text = re.sub(r\"[-()\\\"#/@;:<>{}`+=~|.!?,]\", \"\", text)\n return ' '.join([i.strip() for i in filter(None, text.split())])\n\nclean_questions = []\nfor question in questions:\n clean_questions.append(clean_text(question))\n \nclean_answers = [] \nfor answer in answers:\n clean_answers.append(clean_text(answer))\n \nmin_line_length = 2\nmax_line_length = 5\nshort_questions_temp = []\nshort_answers_temp = []\n\ni = 0\nfor question in clean_questions:\n if len(question.split()) >= min_line_length and len(question.split()) <= max_line_length:\n short_questions_temp.append(question)\n short_answers_temp.append(clean_answers[i])\n i += 1\n\nshort_questions = []\nshort_answers = []\n\ni = 0\nfor answer in short_answers_temp:\n if len(answer.split()) >= min_line_length and len(answer.split()) <= max_line_length:\n short_answers.append(answer)\n short_questions.append(short_questions_temp[i])\n i += 1\n\nquestion_test = short_questions[500:550]\nanswer_test = short_answers[500:550]\nshort_questions = short_questions[:500]\nshort_answers = short_answers[:500]",
"_____no_output_____"
],
[
"concat_from = ' '.join(short_questions+question_test).split()\nvocabulary_size_from = len(list(set(concat_from)))\ndata_from, count_from, dictionary_from, rev_dictionary_from = build_dataset(concat_from, vocabulary_size_from)\nprint('vocab from size: %d'%(vocabulary_size_from))\nprint('Most common words', count_from[4:10])\nprint('Sample data', data_from[:10], [rev_dictionary_from[i] for i in data_from[:10]])\nprint('filtered vocab size:',len(dictionary_from))\nprint(\"% of vocab used: {}%\".format(round(len(dictionary_from)/vocabulary_size_from,4)*100))",
"vocab from size: 657\nMost common words [('you', 132), ('is', 78), ('i', 68), ('what', 51), ('it', 50), ('that', 49)]\nSample data [7, 28, 129, 35, 61, 42, 12, 22, 82, 225] ['what', 'good', 'stuff', 'she', 'okay', 'they', 'do', 'to', 'hey', 'sweet']\nfiltered vocab size: 661\n% of vocab used: 100.61%\n"
],
[
"concat_to = ' '.join(short_answers+answer_test).split()\nvocabulary_size_to = len(list(set(concat_to)))\ndata_to, count_to, dictionary_to, rev_dictionary_to = build_dataset(concat_to, vocabulary_size_to)\nprint('vocab from size: %d'%(vocabulary_size_to))\nprint('Most common words', count_to[4:10])\nprint('Sample data', data_to[:10], [rev_dictionary_to[i] for i in data_to[:10]])\nprint('filtered vocab size:',len(dictionary_to))\nprint(\"% of vocab used: {}%\".format(round(len(dictionary_to)/vocabulary_size_to,4)*100))",
"vocab from size: 660\nMost common words [('i', 97), ('you', 91), ('is', 62), ('it', 58), ('not', 47), ('what', 39)]\nSample data [12, 216, 5, 4, 94, 25, 59, 10, 8, 79] ['the', 'real', 'you', 'i', 'hope', 'so', 'they', 'do', 'not', 'hi']\nfiltered vocab size: 664\n% of vocab used: 100.61%\n"
],
[
"GO = dictionary_from['GO']\nPAD = dictionary_from['PAD']\nEOS = dictionary_from['EOS']\nUNK = dictionary_from['UNK']",
"_____no_output_____"
],
[
"for i in range(len(short_answers)):\n short_answers[i] += ' EOS'",
"_____no_output_____"
],
[
"class Chatbot:\n def __init__(self, size_layer, num_layers, embedded_size, \n from_dict_size, to_dict_size, learning_rate, \n batch_size, dropout = 0.5, beam_width = 15):\n \n def lstm_cell(size, reuse=False):\n return tf.nn.rnn_cell.GRUCell(size, reuse=reuse)\n \n self.X = tf.placeholder(tf.int32, [None, None])\n self.Y = tf.placeholder(tf.int32, [None, None])\n self.X_seq_len = tf.count_nonzero(self.X, 1, dtype=tf.int32)\n self.Y_seq_len = tf.count_nonzero(self.Y, 1, dtype=tf.int32)\n batch_size = tf.shape(self.X)[0]\n # encoder\n encoder_embeddings = tf.Variable(tf.random_uniform([from_dict_size, embedded_size], -1, 1))\n encoder_embedded = tf.nn.embedding_lookup(encoder_embeddings, self.X)\n for n in range(num_layers):\n (out_fw, out_bw), (state_fw, state_bw) = tf.nn.bidirectional_dynamic_rnn(\n cell_fw = lstm_cell(size_layer // 2),\n cell_bw = lstm_cell(size_layer // 2),\n inputs = encoder_embedded,\n sequence_length = self.X_seq_len,\n dtype = tf.float32,\n scope = 'bidirectional_rnn_%d'%(n))\n encoder_embedded = tf.concat((out_fw, out_bw), 2)\n \n bi_state = tf.concat((state_fw, state_bw), -1)\n self.encoder_state = tuple([bi_state] * num_layers)\n \n self.encoder_state = tuple(self.encoder_state[-1] for _ in range(num_layers))\n main = tf.strided_slice(self.Y, [0, 0], [batch_size, -1], [1, 1])\n decoder_input = tf.concat([tf.fill([batch_size, 1], GO), main], 1)\n # decoder\n decoder_embeddings = tf.Variable(tf.random_uniform([to_dict_size, embedded_size], -1, 1))\n decoder_cells = tf.nn.rnn_cell.MultiRNNCell([lstm_cell(size_layer) for _ in range(num_layers)])\n dense_layer = tf.layers.Dense(to_dict_size)\n training_helper = tf.contrib.seq2seq.TrainingHelper(\n inputs = tf.nn.embedding_lookup(decoder_embeddings, decoder_input),\n sequence_length = self.Y_seq_len,\n time_major = False)\n training_decoder = tf.contrib.seq2seq.BasicDecoder(\n cell = decoder_cells,\n helper = training_helper,\n initial_state = self.encoder_state,\n output_layer = dense_layer)\n training_decoder_output, _, _ = tf.contrib.seq2seq.dynamic_decode(\n decoder = training_decoder,\n impute_finished = True,\n maximum_iterations = tf.reduce_max(self.Y_seq_len))\n predicting_helper = tf.contrib.seq2seq.GreedyEmbeddingHelper(\n embedding = decoder_embeddings,\n start_tokens = tf.tile(tf.constant([GO], dtype=tf.int32), [batch_size]),\n end_token = EOS)\n predicting_decoder = tf.contrib.seq2seq.BasicDecoder(\n cell = decoder_cells,\n helper = predicting_helper,\n initial_state = self.encoder_state,\n output_layer = dense_layer)\n predicting_decoder_output, _, _ = tf.contrib.seq2seq.dynamic_decode(\n decoder = predicting_decoder,\n impute_finished = True,\n maximum_iterations = 2 * tf.reduce_max(self.X_seq_len))\n self.training_logits = training_decoder_output.rnn_output\n self.predicting_ids = predicting_decoder_output.sample_id\n masks = tf.sequence_mask(self.Y_seq_len, tf.reduce_max(self.Y_seq_len), dtype=tf.float32)\n self.cost = tf.contrib.seq2seq.sequence_loss(logits = self.training_logits,\n targets = self.Y,\n weights = masks)\n self.optimizer = tf.train.AdamOptimizer(learning_rate).minimize(self.cost)\n y_t = tf.argmax(self.training_logits,axis=2)\n y_t = tf.cast(y_t, tf.int32)\n self.prediction = tf.boolean_mask(y_t, masks)\n mask_label = tf.boolean_mask(self.Y, masks)\n correct_pred = tf.equal(self.prediction, mask_label)\n correct_index = tf.cast(correct_pred, tf.float32)\n self.accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))",
"_____no_output_____"
],
[
"size_layer = 256\nnum_layers = 2\nembedded_size = 128\nlearning_rate = 0.001\nbatch_size = 16\nepoch = 20",
"_____no_output_____"
],
[
"tf.reset_default_graph()\nsess = tf.InteractiveSession()\nmodel = Chatbot(size_layer, num_layers, embedded_size, len(dictionary_from), \n len(dictionary_to), learning_rate,batch_size)\nsess.run(tf.global_variables_initializer())",
"_____no_output_____"
],
[
"def str_idx(corpus, dic):\n X = []\n for i in corpus:\n ints = []\n for k in i.split():\n ints.append(dic.get(k,UNK))\n X.append(ints)\n return X",
"_____no_output_____"
],
[
"X = str_idx(short_questions, dictionary_from)\nY = str_idx(short_answers, dictionary_to)\nX_test = str_idx(question_test, dictionary_from)\nY_test = str_idx(answer_test, dictionary_from)",
"_____no_output_____"
],
[
"def pad_sentence_batch(sentence_batch, pad_int):\n padded_seqs = []\n seq_lens = []\n max_sentence_len = max([len(sentence) for sentence in sentence_batch])\n for sentence in sentence_batch:\n padded_seqs.append(sentence + [pad_int] * (max_sentence_len - len(sentence)))\n seq_lens.append(len(sentence))\n return padded_seqs, seq_lens",
"_____no_output_____"
],
[
"for i in range(epoch):\n total_loss, total_accuracy = 0, 0\n for k in range(0, len(short_questions), batch_size):\n index = min(k+batch_size, len(short_questions))\n batch_x, seq_x = pad_sentence_batch(X[k: index], PAD)\n batch_y, seq_y = pad_sentence_batch(Y[k: index ], PAD)\n predicted, accuracy,loss, _ = sess.run([model.predicting_ids, \n model.accuracy, model.cost, model.optimizer], \n feed_dict={model.X:batch_x,\n model.Y:batch_y})\n total_loss += loss\n total_accuracy += accuracy\n total_loss /= (len(short_questions) / batch_size)\n total_accuracy /= (len(short_questions) / batch_size)\n print('epoch: %d, avg loss: %f, avg accuracy: %f'%(i+1, total_loss, total_accuracy))",
"epoch: 1, avg loss: 5.602181, avg accuracy: 0.223574\nepoch: 2, avg loss: 4.793142, avg accuracy: 0.256463\nepoch: 3, avg loss: 4.563922, avg accuracy: 0.268960\nepoch: 4, avg loss: 4.292581, avg accuracy: 0.287157\nepoch: 5, avg loss: 3.923540, avg accuracy: 0.314106\nepoch: 6, avg loss: 3.470127, avg accuracy: 0.341917\nepoch: 7, avg loss: 3.034022, avg accuracy: 0.391601\nepoch: 8, avg loss: 2.671629, avg accuracy: 0.458257\nepoch: 9, avg loss: 2.142785, avg accuracy: 0.557814\nepoch: 10, avg loss: 1.724149, avg accuracy: 0.663938\nepoch: 11, avg loss: 1.417766, avg accuracy: 0.748157\nepoch: 12, avg loss: 1.268694, avg accuracy: 0.756069\nepoch: 13, avg loss: 0.978809, avg accuracy: 0.838773\nepoch: 14, avg loss: 0.711267, avg accuracy: 0.899787\nepoch: 15, avg loss: 0.479471, avg accuracy: 0.970754\nepoch: 16, avg loss: 0.385906, avg accuracy: 0.979794\nepoch: 17, avg loss: 0.302871, avg accuracy: 0.996912\nepoch: 18, avg loss: 0.232685, avg accuracy: 1.001252\nepoch: 19, avg loss: 0.186021, avg accuracy: 1.005424\nepoch: 20, avg loss: 0.150257, avg accuracy: 1.008791\n"
],
[
"for i in range(len(batch_x)):\n print('row %d'%(i+1))\n print('QUESTION:',' '.join([rev_dictionary_from[n] for n in batch_x[i] if n not in [0,1,2,3]]))\n print('REAL ANSWER:',' '.join([rev_dictionary_to[n] for n in batch_y[i] if n not in[0,1,2,3]]))\n print('PREDICTED ANSWER:',' '.join([rev_dictionary_to[n] for n in predicted[i] if n not in[0,1,2,3]]),'\\n')",
"row 1\nQUESTION: i am a werewolf\nREAL ANSWER: a werewolf\nPREDICTED ANSWER: a werewolf \n\nrow 2\nQUESTION: i was dreaming again\nREAL ANSWER: i would think so\nPREDICTED ANSWER: i would think so \n\nrow 3\nQUESTION: the kitchen\nREAL ANSWER: very nice\nPREDICTED ANSWER: very nice \n\nrow 4\nQUESTION: the bedroom\nREAL ANSWER: there is only one bed\nPREDICTED ANSWER: there is only one bed \n\n"
],
[
"batch_x, seq_x = pad_sentence_batch(X_test[:batch_size], PAD)\nbatch_y, seq_y = pad_sentence_batch(Y_test[:batch_size], PAD)\npredicted = sess.run(model.predicting_ids, feed_dict={model.X:batch_x,model.X_seq_len:seq_x})\n\nfor i in range(len(batch_x)):\n print('row %d'%(i+1))\n print('QUESTION:',' '.join([rev_dictionary_from[n] for n in batch_x[i] if n not in [0,1,2,3]]))\n print('REAL ANSWER:',' '.join([rev_dictionary_to[n] for n in batch_y[i] if n not in[0,1,2,3]]))\n print('PREDICTED ANSWER:',' '.join([rev_dictionary_to[n] for n in predicted[i] if n not in[0,1,2,3]]),'\\n')",
"row 1\nQUESTION: but david\nREAL ANSWER: is here that\nPREDICTED ANSWER: i do not know \n\nrow 2\nQUESTION: hopeless it is hopeless\nREAL ANSWER: tell ballet then back\nPREDICTED ANSWER: stick god \n\nrow 3\nQUESTION: miss price\nREAL ANSWER: yes learning\nPREDICTED ANSWER: yes i am \n\nrow 4\nQUESTION: mr kessler wake up please\nREAL ANSWER: is here are\nPREDICTED ANSWER: wah it \n\nrow 5\nQUESTION: there were witnesses\nREAL ANSWER: why she out\nPREDICTED ANSWER: well i heard \n\nrow 6\nQUESTION: what about it\nREAL ANSWER: not you are\nPREDICTED ANSWER: what about him \n\nrow 7\nQUESTION: go on ask them\nREAL ANSWER: i just home\nPREDICTED ANSWER: you have a good \n\nrow 8\nQUESTION: beware the moon\nREAL ANSWER: seen hi is he\nPREDICTED ANSWER: how here \n\nrow 9\nQUESTION: did you hear that\nREAL ANSWER: is down what\nPREDICTED ANSWER: the sound again \n\nrow 10\nQUESTION: i heard that\nREAL ANSWER: it here not\nPREDICTED ANSWER: then you believe it \n\nrow 11\nQUESTION: the hound of the baskervilles\nREAL ANSWER: heard\nPREDICTED ANSWER: did here \n\nrow 12\nQUESTION: it is moving\nREAL ANSWER: not you hear\nPREDICTED ANSWER: it is harcourt \n\nrow 13\nQUESTION: nice doggie good boy\nREAL ANSWER: bill stupid\nPREDICTED ANSWER: thank you \n\nrow 14\nQUESTION: it sounds far away\nREAL ANSWER: that pecos baby seen hi\nPREDICTED ANSWER: what is that \n\nrow 15\nQUESTION: debbie klein cried a lot\nREAL ANSWER: is will srai not\nPREDICTED ANSWER: how could it \n\nrow 16\nQUESTION: what are you doing here\nREAL ANSWER: is know look i\nPREDICTED ANSWER: you have got a call \n\n"
]
]
] |
[
"code"
] |
[
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
]
] |
cbd813c0034d30ef3a3a6546bf2654e4b2b355d8
| 786,521 |
ipynb
|
Jupyter Notebook
|
t81_558_class_12_04_atari.ipynb
|
englyndaomanu/t81_558_deep_learning
|
e48517fbcb434021b52033da88bdbd03f10bef60
|
[
"Apache-2.0"
] | 1 |
2020-10-13T06:58:58.000Z
|
2020-10-13T06:58:58.000Z
|
t81_558_class_12_04_atari.ipynb
|
joaquinmorenoa/t81_558_deep_learning
|
569ed623cb225a5d410fda6f49e1a15073b247ea
|
[
"Apache-2.0"
] | null | null | null |
t81_558_class_12_04_atari.ipynb
|
joaquinmorenoa/t81_558_deep_learning
|
569ed623cb225a5d410fda6f49e1a15073b247ea
|
[
"Apache-2.0"
] | null | null | null | 802.572449 | 578,274 | 0.953465 |
[
[
[
"<a href=\"https://colab.research.google.com/github/jeffheaton/t81_558_deep_learning/blob/master/t81_558_class_12_04_atari.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>",
"_____no_output_____"
],
[
"# T81-558: Applications of Deep Neural Networks\n**Module 12: Reinforcement Learning**\n* Instructor: [Jeff Heaton](https://sites.wustl.edu/jeffheaton/), McKelvey School of Engineering, [Washington University in St. Louis](https://engineering.wustl.edu/Programs/Pages/default.aspx)\n* For more information visit the [class website](https://sites.wustl.edu/jeffheaton/t81-558/).",
"_____no_output_____"
],
[
"# Module 12 Video Material\n\n* Part 12.1: Introduction to the OpenAI Gym [[Video]](https://www.youtube.com/watch?v=_KbUxgyisjM&list=PLjy4p-07OYzulelvJ5KVaT2pDlxivl_BN) [[Notebook]](t81_558_class_12_01_ai_gym.ipynb)\n* Part 12.2: Introduction to Q-Learning [[Video]](https://www.youtube.com/watch?v=A3sYFcJY3lA&list=PLjy4p-07OYzulelvJ5KVaT2pDlxivl_BN) [[Notebook]](t81_558_class_12_02_qlearningreinforcement.ipynb)\n* Part 12.3: Keras Q-Learning in the OpenAI Gym [[Video]](https://www.youtube.com/watch?v=qy1SJmsRhvM&list=PLjy4p-07OYzulelvJ5KVaT2pDlxivl_BN) [[Notebook]](t81_558_class_12_03_keras_reinforce.ipynb)\n* **Part 12.4: Atari Games with Keras Neural Networks** [[Video]](https://www.youtube.com/watch?v=co0SwPWoZh0&list=PLjy4p-07OYzulelvJ5KVaT2pDlxivl_BN) [[Notebook]](t81_558_class_12_04_atari.ipynb)\n* Part 12.5: Application of Reinforcement Learning [[Video]](https://www.youtube.com/watch?v=1jQPP3RfwMI&list=PLjy4p-07OYzulelvJ5KVaT2pDlxivl_BN) [[Notebook]](t81_558_class_12_05_apply_rl.ipynb)\n",
"_____no_output_____"
],
[
"# Google CoLab Instructions\n\nThe following code ensures that Google CoLab is running the correct version of TensorFlow, and has the necessary Python libraries installed.",
"_____no_output_____"
]
],
[
[
"try:\n from google.colab import drive\n %tensorflow_version 2.x\n COLAB = True\n print(\"Note: using Google CoLab\")\nexcept:\n print(\"Note: not using Google CoLab\")\n COLAB = False\n\nif COLAB:\n !sudo apt-get install -y xvfb ffmpeg\n !pip install -q 'gym==0.10.11'\n !pip install -q 'imageio==2.4.0'\n !pip install -q PILLOW\n !pip install -q 'pyglet==1.3.2'\n !pip install -q pyvirtualdisplay\n !pip install -q --upgrade tensorflow-probability\n !pip install -q tf-agents",
"Note: using Google CoLab\nReading package lists... Done\nBuilding dependency tree \nReading state information... Done\nffmpeg is already the newest version (7:3.4.6-0ubuntu0.18.04.1).\nxvfb is already the newest version (2:1.19.6-1ubuntu4.4).\n0 upgraded, 0 newly installed, 0 to remove and 31 not upgraded.\n"
]
],
[
[
"# Part 12.4: Atari Games with Keras Neural Networks\n\n\nThe Atari 2600 is a home video game console from Atari, Inc. Released on September 11, 1977. It is credited with popularizing the use of microprocessor-based hardware and games stored on ROM cartridges instead of dedicated hardware with games physically built into the unit. The 2600 was bundled with two joystick controllers, a conjoined pair of paddle controllers, and a game cartridge: initially [Combat](https://en.wikipedia.org/wiki/Combat_(Atari_2600)), and later [Pac-Man](https://en.wikipedia.org/wiki/Pac-Man_(Atari_2600)).\n\nAtari emulators are popular and allow many of the old Atari video games to be played on modern computers. They are even available as JavaScript.\n\n* [Virtual Atari](http://www.virtualatari.org/listP.html)\n\nAtari games have become popular benchmarks for AI systems, particularly reinforcement learning. OpenAI Gym internally uses the [Stella Atari Emulator](https://stella-emu.github.io/). The Atari 2600 is shown in Figure 12.ATARI.\n\n**Figure 12.ATARI: The Atari 2600**\n![Atari 2600 Console](https://raw.githubusercontent.com/jeffheaton/t81_558_deep_learning/master/images/atari-1.png \"Atari 2600 Console\")\n\n### Actual Atari 2600 Specs\n\n* CPU: 1.19 MHz MOS Technology 6507\n* Audio + Video processor: Television Interface Adapter (TIA)\n* Playfield resolution: 40 x 192 pixels (NTSC). Uses a 20-pixel register that is mirrored or copied, left side to right side, to achieve the width of 40 pixels.\n* Player sprites: 8 x 192 pixels (NTSC). Player, ball, and missile sprites use pixels that are 1/4 the width of playfield pixels (unless stretched).\n* Ball and missile sprites: 1 x 192 pixels (NTSC).\n* Maximum resolution: 160 x 192 pixels (NTSC). Max resolution is only somewhat achievable with programming tricks that combine sprite pixels with playfield pixels.\n* 128 colors (NTSC). 128 possible on screen. Max of 4 per line: background, playfield, player0 sprite, and player1 sprite. Palette switching between lines is common. Palette switching mid line is possible but not common due to resource limitations.\n* 2 channels of 1-bit monaural sound with 4-bit volume control.\n\n### OpenAI Lab Atari Pong\n\nOpenAI Gym can be used with Windows; however, it requires a special [installation procedure].(https://towardsdatascience.com/how-to-install-openai-gym-in-a-windows-environment-338969e24d30) \n\nThis chapter demonstrates playing [Atari Pong](https://github.com/wau/keras-rl2/blob/master/examples/dqn_atari.py). Pong is a two-dimensional sports game that simulates table tennis. The player controls an in-game paddle by moving it vertically across the left or right side of the screen. They can compete against another player controlling a second paddle on the opposing side. Players use the paddles to hit a ball back and forth. The goal is for each player to reach eleven points before the opponent; you earn points when one fails to return it to the other. For the Atari 2600 version of Pong, a computer player (controlled by the 2600) is the opposing player.\n\nThis section shows how to adapt TF-Agents to an Atari game. Some changes are necessary when compared to the pole-cart game presented earlier in this chapter. You can quickly adapt this example to any Atari game by simply changing the environment name. However, I tuned the code presented here for Pong, and it may not perform as well for other games. Some tuning will likely be necessary to produce a good agent for other games.\n\nWe begin by importing the needed Python packages.",
"_____no_output_____"
]
],
[
[
"import base64\nimport imageio\nimport IPython\nimport matplotlib\nimport matplotlib.pyplot as plt\nimport numpy as np\nimport PIL.Image\nimport pyvirtualdisplay\n\nimport tensorflow as tf\n\nfrom tf_agents.agents.dqn import dqn_agent\nfrom tf_agents.drivers import dynamic_step_driver\nfrom tf_agents.environments import suite_gym, suite_atari\nfrom tf_agents.environments import tf_py_environment, batched_py_environment\nfrom tf_agents.eval import metric_utils\nfrom tf_agents.metrics import tf_metrics\nfrom tf_agents.networks import q_network\nfrom tf_agents.policies import random_tf_policy\nfrom tf_agents.replay_buffers import tf_uniform_replay_buffer\nfrom tf_agents.trajectories import trajectory\nfrom tf_agents.utils import common\n\nfrom tf_agents.specs import tensor_spec\nfrom tf_agents.trajectories import time_step as ts",
"_____no_output_____"
],
[
"# Set up a virtual display for rendering OpenAI gym environments.\ndisplay = pyvirtualdisplay.Display(visible=0, size=(1400, 900)).start()",
"_____no_output_____"
]
],
[
[
"## Hyperparameters\n\nThe hyperparameter names are the same as the previous DQN example; however, I tuned the numeric values for the more complex Atari game.",
"_____no_output_____"
]
],
[
[
"num_iterations = 250000 \n\ninitial_collect_steps = 200 \ncollect_steps_per_iteration = 10 \nreplay_buffer_max_length = 100000\n\nbatch_size = 32\nlearning_rate = 2.5e-3\nlog_interval = 5000\n\nnum_eval_episodes = 5 \neval_interval = 25000 ",
"_____no_output_____"
]
],
[
[
"The algorithm needs more iterations for an Atari game. I also found that increasing the number of collection steps helped the algorithm to train.",
"_____no_output_____"
],
[
"## Atari Environment's\n\nYou must handle Atari environments differently than games like cart-poll. Atari games typically use their 2D displays as the environment state. AI Gym represents Atari games as either a 3D (height by width by color) state spaced based on their screens, or a vector representing the state of the gam's computer RAM. To preprocess Atari games for greater computational efficiency, we generally skip several frames, decrease the resolution, and discard color information. The following code shows how we can set up an Atari environment.",
"_____no_output_____"
]
],
[
[
"#env_name = 'Breakout-v4'\nenv_name = 'Pong-v0'\n#env_name = 'BreakoutDeterministic-v4'\n#env = suite_gym.load(env_name)\n\n# AtariPreprocessing runs 4 frames at a time, max-pooling over the last 2\n# frames. We need to account for this when computing things like update\n# intervals.\nATARI_FRAME_SKIP = 4\n\nmax_episode_frames=108000 # ALE frames\n\nenv = suite_atari.load(\n env_name,\n max_episode_steps=max_episode_frames / ATARI_FRAME_SKIP,\n gym_env_wrappers=suite_atari.DEFAULT_ATARI_GYM_WRAPPERS_WITH_STACKING)\n#env = batched_py_environment.BatchedPyEnvironment([env])",
"_____no_output_____"
]
],
[
[
"We can now reset the environment and display one step. The following image shows how the Pong game environment appears to a user.",
"_____no_output_____"
]
],
[
[
"env.reset()\nPIL.Image.fromarray(env.render())",
"_____no_output_____"
]
],
[
[
"We are now ready to load and wrap the two environments for TF-Agents. The algorithm uses the first environment for evaluation, and the second to train.",
"_____no_output_____"
]
],
[
[
"train_py_env = suite_atari.load(\n env_name,\n max_episode_steps=max_episode_frames / ATARI_FRAME_SKIP,\n gym_env_wrappers=suite_atari.DEFAULT_ATARI_GYM_WRAPPERS_WITH_STACKING)\n\neval_py_env = suite_atari.load(\n env_name,\n max_episode_steps=max_episode_frames / ATARI_FRAME_SKIP,\n gym_env_wrappers=suite_atari.DEFAULT_ATARI_GYM_WRAPPERS_WITH_STACKING)\n\ntrain_env = tf_py_environment.TFPyEnvironment(train_py_env)\neval_env = tf_py_environment.TFPyEnvironment(eval_py_env)",
"_____no_output_____"
]
],
[
[
"## Agent\n\nI used the following class, from TF-Agents examples, to wrap the regular Q-network class. The AtariQNetwork class ensures that the pixel values from the Atari screen are divided by 255. This division assists the neural network by normalizing the pixel values to between 0 and 1.",
"_____no_output_____"
]
],
[
[
"class AtariQNetwork(q_network.QNetwork):\n \"\"\"QNetwork subclass that divides observations by 255.\"\"\"\n\n def call(self,\n observation,\n step_type=None,\n network_state=(),\n training=False):\n state = tf.cast(observation, tf.float32)\n # We divide the grayscale pixel values by 255 here rather than storing\n # normalized values beause uint8s are 4x cheaper to store than float32s.\n state = state / 255\n return super(AtariQNetwork, self).call(\n state, step_type=step_type, network_state=network_state,\n training=training)",
"_____no_output_____"
]
],
[
[
"Next, we introduce two hyperparameters that are specific to the neural network we are about to define.",
"_____no_output_____"
]
],
[
[
"fc_layer_params = (512,)\nconv_layer_params=((32, (8, 8), 4), (64, (4, 4), 2), (64, (3, 3), 1))\n\nq_net = AtariQNetwork(\n train_env.observation_spec(),\n train_env.action_spec(),\n conv_layer_params=conv_layer_params,\n fc_layer_params=fc_layer_params)",
"_____no_output_____"
]
],
[
[
"Convolutional neural networks usually are made up of several alternating pairs of convolution and max-pooling layers, ultimately culminating in one or more dense layers. These layers are the same types as previously seen in this course. The QNetwork accepts two parameters that define the convolutional neural network structure. \n\nThe more simple of the two parameters is **fc_layer_params**. This parameter specifies the size of each of the dense layers. A tuple specifies the size of each of the layers in a list. \n\nThe second parameter, named **conv_layer_params**, is a list of convolution layers parameters, where each item is a length-three tuple indicating (filters, kernel_size, stride). This implementation of QNetwork supports only convolution layers. If you desire a more complex convolutional neural network, you must define your variant of the QNetwork.\n\nThe QNetwork defined here is not the agent, instead, the QNetwork is used by the DQN agent to implement the actual neural network. This allows flexibility as you can set your own class if needed.\n\nNext, we define the optimizer. For this example, I used RMSPropOptimizer. However, AdamOptimizer is another popular choice. We also create the DQN and reference the Q-network we just created.",
"_____no_output_____"
]
],
[
[
"optimizer = tf.compat.v1.train.RMSPropOptimizer(\n learning_rate=learning_rate,\n decay=0.95,\n momentum=0.0,\n epsilon=0.00001,\n centered=True)\n\ntrain_step_counter = tf.Variable(0)\n\nobservation_spec = tensor_spec.from_spec(train_env.observation_spec())\ntime_step_spec = ts.time_step_spec(observation_spec)\n\naction_spec = tensor_spec.from_spec(train_env.action_spec())\ntarget_update_period=32000 # ALE frames\nupdate_period=16 # ALE frames\n_update_period = update_period / ATARI_FRAME_SKIP\n_global_step = tf.compat.v1.train.get_or_create_global_step()\n\nagent = dqn_agent.DqnAgent(\n time_step_spec,\n action_spec,\n q_network=q_net,\n optimizer=optimizer,\n epsilon_greedy=0.01,\n n_step_update=1.0,\n target_update_tau=1.0,\n target_update_period=(\n target_update_period / ATARI_FRAME_SKIP / _update_period),\n td_errors_loss_fn=common.element_wise_huber_loss,\n gamma=0.99,\n reward_scale_factor=1.0,\n gradient_clipping=None,\n debug_summaries=False,\n summarize_grads_and_vars=False,\n train_step_counter=_global_step)\n\n\n\nagent.initialize()",
"_____no_output_____"
]
],
[
[
"## Metrics and Evaluation\n\nThere are many different ways to measure the effectiveness of a model trained with reinforcement learning. The loss function of the internal Q-network is not a good measure of the entire DQN algorithm's overall fitness. The network loss function measures how close the Q-network was fit to the collected data and did not indicate how effective the DQN is in maximizing rewards. The method used for this example tracks the average reward received over several episodes.",
"_____no_output_____"
]
],
[
[
"def compute_avg_return(environment, policy, num_episodes=10):\n\n total_return = 0.0\n for _ in range(num_episodes):\n\n time_step = environment.reset()\n episode_return = 0.0\n\n while not time_step.is_last():\n action_step = policy.action(time_step)\n time_step = environment.step(action_step.action)\n episode_return += time_step.reward\n total_return += episode_return\n\n avg_return = total_return / num_episodes\n return avg_return.numpy()[0]\n\n\n# See also the metrics module for standard implementations of different metrics.\n# https://github.com/tensorflow/agents/tree/master/tf_agents/metrics",
"_____no_output_____"
]
],
[
[
"## Replay Buffer\n\nDQN works by training a neural network to predict the Q-values for every possible environment-state. A neural network needs training data, so the algorithm accumulates this training data as it runs episodes. The replay buffer is where this data is stored. Only the most recent episodes are stored, older episode data rolls off the queue as the queue accumulates new data.",
"_____no_output_____"
]
],
[
[
"replay_buffer = tf_uniform_replay_buffer.TFUniformReplayBuffer(\n data_spec=agent.collect_data_spec,\n batch_size=train_env.batch_size,\n max_length=replay_buffer_max_length)\n\n# Dataset generates trajectories with shape [Bx2x...]\ndataset = replay_buffer.as_dataset(\n num_parallel_calls=3, \n sample_batch_size=batch_size, \n num_steps=2).prefetch(3)",
"_____no_output_____"
]
],
[
[
"## Random Collection\n\nThe algorithm must prime the pump. Training cannot begin on an empty replay buffer. The following code performs a predefined number of steps to generate initial training data.",
"_____no_output_____"
]
],
[
[
"random_policy = random_tf_policy.RandomTFPolicy(train_env.time_step_spec(),\n train_env.action_spec())\n\ndef collect_step(environment, policy, buffer):\n time_step = environment.current_time_step()\n action_step = policy.action(time_step)\n next_time_step = environment.step(action_step.action)\n traj = trajectory.from_transition(time_step, action_step, next_time_step)\n\n # Add trajectory to the replay buffer\n buffer.add_batch(traj)\n\ndef collect_data(env, policy, buffer, steps):\n for _ in range(steps):\n collect_step(env, policy, buffer)\n\ncollect_data(train_env, random_policy, replay_buffer, steps=initial_collect_steps)",
"_____no_output_____"
]
],
[
[
"## Training the agent\n\nWe are now ready to train the DQN. This process can take many hours, depending on how many episodes you wish to run through. As training occurs, this code will update on both the loss and average return. As training becomes more successful, the average return should increase. The losses reported reflecting the average loss for individual training batches.",
"_____no_output_____"
]
],
[
[
"iterator = iter(dataset)\n\n# (Optional) Optimize by wrapping some of the code in a graph using TF function.\nagent.train = common.function(agent.train)\n\n# Reset the train step\nagent.train_step_counter.assign(0)\n\n# Evaluate the agent's policy once before training.\navg_return = compute_avg_return(eval_env, agent.policy, num_eval_episodes)\nreturns = [avg_return]\n\nfor _ in range(num_iterations):\n\n # Collect a few steps using collect_policy and save to the replay buffer.\n for _ in range(collect_steps_per_iteration):\n collect_step(train_env, agent.collect_policy, replay_buffer)\n\n # Sample a batch of data from the buffer and update the agent's network.\n experience, unused_info = next(iterator)\n train_loss = agent.train(experience).loss\n\n step = agent.train_step_counter.numpy()\n\n if step % log_interval == 0:\n print('step = {0}: loss = {1}'.format(step, train_loss))\n\n if step % eval_interval == 0:\n avg_return = compute_avg_return(eval_env, agent.policy, num_eval_episodes)\n print('step = {0}: Average Return = {1}'.format(step, avg_return))\n returns.append(avg_return)",
"step = 5000: loss = 0.005372311919927597\nstep = 10000: loss = 0.029342571273446083\nstep = 15000: loss = 0.023372460156679153\nstep = 20000: loss = 0.012967261485755444\nstep = 25000: loss = 0.03114483878016472\nstep = 25000: Average Return = -20.0\nstep = 30000: loss = 0.015883663669228554\nstep = 35000: loss = 0.022952664643526077\nstep = 40000: loss = 0.024018988013267517\nstep = 45000: loss = 0.015258202329277992\nstep = 50000: loss = 0.01642722450196743\nstep = 50000: Average Return = -18.399999618530273\nstep = 55000: loss = 0.024171829223632812\nstep = 60000: loss = 0.010190263390541077\nstep = 65000: loss = 0.005736709106713533\nstep = 70000: loss = 0.01117132231593132\nstep = 75000: loss = 0.005509796552360058\nstep = 75000: Average Return = -12.800000190734863\nstep = 80000: loss = 0.009709298610687256\nstep = 85000: loss = 0.009705539792776108\nstep = 90000: loss = 0.006236877758055925\nstep = 95000: loss = 0.017611663788557053\nstep = 100000: loss = 0.00873786024749279\nstep = 100000: Average Return = -10.800000190734863\nstep = 105000: loss = 0.019388657063245773\nstep = 110000: loss = 0.0040118759498000145\nstep = 115000: loss = 0.006819932255893946\nstep = 120000: loss = 0.028965750709176064\nstep = 125000: loss = 0.015978489071130753\nstep = 125000: Average Return = -9.600000381469727\nstep = 130000: loss = 0.023571692407131195\nstep = 135000: loss = 0.006761073134839535\nstep = 140000: loss = 0.005080501548945904\nstep = 145000: loss = 0.013759403489530087\nstep = 150000: loss = 0.02108653262257576\nstep = 150000: Average Return = -5.599999904632568\nstep = 155000: loss = 0.01754268817603588\nstep = 160000: loss = 0.008789192885160446\nstep = 165000: loss = 0.012145541608333588\nstep = 170000: loss = 0.00911545380949974\nstep = 175000: loss = 0.008846037089824677\nstep = 175000: Average Return = -5.199999809265137\nstep = 180000: loss = 0.020279696211218834\nstep = 185000: loss = 0.012781327590346336\nstep = 190000: loss = 0.01562594249844551\nstep = 195000: loss = 0.015836259350180626\nstep = 200000: loss = 0.017415495589375496\nstep = 200000: Average Return = 3.5999999046325684\nstep = 205000: loss = 0.007518010213971138\nstep = 210000: loss = 0.028996415436267853\nstep = 215000: loss = 0.01371004804968834\nstep = 220000: loss = 0.007023532874882221\nstep = 225000: loss = 0.004790903069078922\nstep = 225000: Average Return = -4.400000095367432\nstep = 230000: loss = 0.006244136951863766\nstep = 235000: loss = 0.025019707158207893\nstep = 240000: loss = 0.02555653266608715\nstep = 245000: loss = 0.012253865599632263\nstep = 250000: loss = 0.004736536182463169\nstep = 250000: Average Return = 2.4000000953674316\n"
]
],
[
[
"## Visualization\n\nThe notebook can plot the average return over training iterations. The average return should increase as the program performs more training iterations.",
"_____no_output_____"
]
],
[
[
"iterations = range(0, num_iterations + 1, eval_interval)\nplt.plot(iterations, returns)\nplt.ylabel('Average Return')\nplt.xlabel('Iterations')\nplt.ylim(top=10)",
"_____no_output_____"
]
],
[
[
"### Videos\n\nWe now have a trained model and observed its training progress on a graph. Perhaps the most compelling way to view an Atari game's results is a video that allows us to see the agent play the game. The following functions are defined so that we can watch the agent play the game in the notebook.",
"_____no_output_____"
]
],
[
[
"def embed_mp4(filename):\n \"\"\"Embeds an mp4 file in the notebook.\"\"\"\n video = open(filename,'rb').read()\n b64 = base64.b64encode(video)\n tag = '''\n <video width=\"640\" height=\"480\" controls>\n <source src=\"data:video/mp4;base64,{0}\" type=\"video/mp4\">\n Your browser does not support the video tag.\n </video>'''.format(b64.decode())\n\n return IPython.display.HTML(tag)\n\ndef create_policy_eval_video(policy, filename, num_episodes=5, fps=30):\n filename = filename + \".mp4\"\n with imageio.get_writer(filename, fps=fps) as video:\n for _ in range(num_episodes):\n time_step = eval_env.reset()\n video.append_data(eval_py_env.render())\n while not time_step.is_last():\n action_step = policy.action(time_step)\n time_step = eval_env.step(action_step.action)\n video.append_data(eval_py_env.render())\n return embed_mp4(filename)",
"_____no_output_____"
]
],
[
[
"First, we will observe the trained agent play the game.",
"_____no_output_____"
]
],
[
[
"create_policy_eval_video(agent.policy, \"trained-agent\")",
"WARNING:root:IMAGEIO FFMPEG_WRITER WARNING: input image is not divisible by macro_block_size=16, resizing from (210, 160) to (224, 160) to ensure video compatibility with most codecs and players. To prevent resizing, make your input image divisible by the macro_block_size or set the macro_block_size to None (risking incompatibility). You may also see a FFMPEG warning concerning speedloss due to data not being aligned.\n"
]
],
[
[
"For comparison, we observe a random agent play. While the trained agent is far from perfect, it does outperform the random agent by a considerable amount.",
"_____no_output_____"
]
],
[
[
"create_policy_eval_video(random_policy, \"random-agent\")",
"WARNING:root:IMAGEIO FFMPEG_WRITER WARNING: input image is not divisible by macro_block_size=16, resizing from (210, 160) to (224, 160) to ensure video compatibility with most codecs and players. To prevent resizing, make your input image divisible by the macro_block_size or set the macro_block_size to None (risking incompatibility). You may also see a FFMPEG warning concerning speedloss due to data not being aligned.\n"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
]
] |
cbd8234c990dcfc46412f90421c5f79c5547aa1b
| 36,550 |
ipynb
|
Jupyter Notebook
|
creators_summary.ipynb
|
icappello/ml-predict-text-author
|
38e059164fef0276e79c6f391a9485abc64d3923
|
[
"CC-BY-4.0"
] | 1 |
2021-05-21T13:06:32.000Z
|
2021-05-21T13:06:32.000Z
|
creators_summary.ipynb
|
salhasalman/ml-predict-text-author
|
38e059164fef0276e79c6f391a9485abc64d3923
|
[
"CC-BY-4.0"
] | null | null | null |
creators_summary.ipynb
|
salhasalman/ml-predict-text-author
|
38e059164fef0276e79c6f391a9485abc64d3923
|
[
"CC-BY-4.0"
] | 1 |
2021-05-07T18:44:00.000Z
|
2021-05-07T18:44:00.000Z
| 32.54675 | 662 | 0.565308 |
[
[
[
"# Disclaimer\n\nReleased under the CC BY 4.0 License (https://creativecommons.org/licenses/by/4.0/)\n\n# Purpose of this notebook\n\nThe purpose of this document is to show how I approached the presented problem and to record my learning experience in how to use Tensorflow 2 and CatBoost to perform a classification task on text data.\n\nIf, while reading this document, you think _\"Why didn't you do `<this>` instead of `<that>`?\"_, the answer could be simply because I don't know about `<this>`. Comments, questions and constructive criticism are of course welcome.\n\n# Intro\n\nThis simple classification task has been developed to get familiarized with Tensorflow 2 and CatBoost handling of text data. In summary, the task is to predict the author of a short text.\n\nTo get a number of train/test examples, it is enough to create a twitter app and, using the python client library for twitter, read the user timeline of multiple accounts. This process is not covered here. If you are interested in this topic, feel free to contact me.\n\n\n## Features\n\nIt is assumed the collected raw data consists of:\n\n1. The author handle (the label that will be predicted)\n2. The timestamp of the post\n3. The raw text of the post\n\n### Preparing the dataset\n\nWhen preparing the dataset, the content of the post is preprocessed using these rules:\n\n1. Newlines are replaced with a space\n2. Links are replaced with a placeholder (e.g. `<link>`)\n3. For each possible unicode char category, the number of chars in that category is added as a feature\n4. The number of words for each tweet is added as a feature\n5. Retweets (even retweets with comment) are discarded. Only responses and original tweets are taken into account\n\nThe dataset has been randomly split into three different files for train (70%), validation (10%) and test (20%). For each label, it has been verified that the same percentages hold in all three files.\n\nBefore fitting the data and before evaluation on the test dataset, the timestamp values are normalized, using the mean and standard deviation computed on the train dataset.",
"_____no_output_____"
],
[
"# TensorFlow 2 model\n\nThe model has four different input features:\n\n1. The normalized timestamp.\n2. The input text, represented as the whole sentence. This will be transformed in a 128-dimensional vector by an embedding layer.\n3. The input text, this time represented as a sequence of words, expressed as indexes of tokens. This representation will be used by a LSTM layer to try to extract some meaning from the actual sequence of the used words.\n4. The unicode character category usage. This should help in identify handles that use emojis, a lot of punctuation or unusual chars.\n\nThe resulting layers are concatenated, then after a sequence of two dense layers (with an applied dropout) the final layer computes the logits for the different classes. The used loss function is *sparse categorical crossentropy*, since the labels are represented as indexes of a list of twitter handles.",
"_____no_output_____"
],
[
"## Imports for the TensorFlow 2 model",
"_____no_output_____"
]
],
[
[
"import functools\nimport os\n\nfrom tensorflow.keras import Input, layers\nimport tensorflow as tf\nimport tensorflow_hub as hub\nfrom tensorflow.keras import regularizers\n\nimport pandas as pd\nimport numpy as np\n\nimport copy\n\nimport calendar\nimport datetime\nimport re\n\n\nfrom tensorflow.keras.preprocessing.text import Tokenizer\n\nimport unicodedata\n#masking layers and GPU don't mix\nos.environ[\"CUDA_VISIBLE_DEVICES\"] = \"-1\"",
"_____no_output_____"
]
],
[
[
"## Definitions for the TensorFlow 2 model\n",
"_____no_output_____"
]
],
[
[
"#Download size: ~446MB\nhub_layer = hub.KerasLayer(\n \"https://tfhub.dev/google/tf2-preview/nnlm-en-dim128/1\",\n output_shape=[512],\n input_shape=[],\n dtype=tf.string,\n trainable=False\n)\n\nembed = hub.load(\"https://tfhub.dev/google/tf2-preview/nnlm-en-dim128/1\")\n\nunicode_data_categories = [\n \"Cc\",\n \"Cf\",\n \"Cn\",\n \"Co\",\n \"Cs\",\n \"LC\",\n \"Ll\",\n \"Lm\",\n \"Lo\",\n \"Lt\",\n \"Lu\",\n \"Mc\",\n \"Me\",\n \"Mn\",\n \"Nd\",\n \"Nl\",\n \"No\",\n \"Pc\",\n \"Pd\",\n \"Pe\",\n \"Pf\",\n \"Pi\",\n \"Po\",\n \"Ps\",\n \"Sc\",\n \"Sk\",\n \"Sm\",\n \"So\",\n \"Zl\",\n \"Zp\",\n \"Zs\"\n]\n\ncolumn_names = [\n \"handle\",\n \"timestamp\",\n \"text\"\n]\n\ncolumn_names.extend(unicode_data_categories)\n\ntrain_file = os.path.realpath(\"input.csv\")\n\nn_tokens = 100000\n\ntokenizer = Tokenizer(n_tokens, oov_token='<OOV>')\n\n#List of handles (labels)\n#Fill with the handles you want to consider in your dataset\nhandles = [\n\n]\n\nend_token = \"XEND\"\n\ntrain_file = os.path.realpath(\"data/train.csv\")\nval_file = os.path.realpath(\"data/val.csv\")\ntest_file = os.path.realpath(\"data/test.csv\")",
"_____no_output_____"
]
],
[
[
"## Preprocessing and computing dataset features",
"_____no_output_____"
]
],
[
[
"def get_pandas_dataset(input_file, fit_tokenizer=False, timestamp_mean=None, timestamp_std=None, pad_sequence=None):\n\n pd_dat = pd.read_csv(input_file, names=column_names)\n \n pd_dat = pd_dat[pd_dat.handle.isin(handles)]\n\n if(timestamp_mean is None):\n timestamp_mean = pd_dat.timestamp.mean()\n \n if(timestamp_std is None):\n timestamp_std = pd_dat.timestamp.std()\n\n pd_dat.timestamp = (pd_dat.timestamp - timestamp_mean) / timestamp_std\n\n pd_dat[\"handle_index\"] = pd_dat['handle'].map(lambda x: handles.index(x))\n\n if(fit_tokenizer):\n tokenizer.fit_on_texts(pd_dat[\"text\"])\n pad_sequence = tokenizer.texts_to_sequences([[end_token]])[0][0]\n \n pd_dat[\"sequence\"] = tokenizer.texts_to_sequences(pd_dat[\"text\"])\n\n\n max_seq_length = 30\n pd_dat = pd_dat.reset_index(drop=True)\n\n #max length\n pd_dat[\"sequence\"] = pd.Series(el[0:max_seq_length] for el in pd_dat[\"sequence\"])\n\n #padding\n pd_dat[\"sequence\"] = pd.Series([el + ([pad_sequence] * (max_seq_length - len(el))) for el in pd_dat[\"sequence\"]])\n \n pd_dat[\"words_in_tweet\"] = pd_dat[\"text\"].str.strip().str.split(\" \").str.len() + 1\n \n return pd_dat, timestamp_mean, timestamp_std, pad_sequence\n\ntrain_dataset, timestamp_mean, timestamp_std, pad_sequence = get_pandas_dataset(train_file, fit_tokenizer=True)\n\ntest_dataset, _, _, _= get_pandas_dataset(test_file, timestamp_mean=timestamp_mean, timestamp_std=timestamp_std, pad_sequence=pad_sequence)\n\nval_dataset, _, _, _ = get_pandas_dataset(val_file, timestamp_mean=timestamp_mean, timestamp_std=timestamp_std, pad_sequence=pad_sequence)\n\n#selecting as features only the unicode categories that are used in the train dataset\nnon_null_unicode_categories = []\nfor unicode_data_category in unicode_data_categories:\n category_name = unicode_data_category\n category_sum = train_dataset[category_name].sum()\n\n if(category_sum > 0):\n non_null_unicode_categories.append(category_name) \n\nprint(\"Bucketized unicode categories used as features: \" + repr(non_null_unicode_categories))",
"_____no_output_____"
]
],
[
[
"## Defining input/output features from the datasets",
"_____no_output_____"
]
],
[
[
"def split_inputs_and_outputs(pd_dat):\n\n\n labels = pd_dat['handle_index'].values\n\n icolumns = pd_dat.columns\n\n timestamps = pd_dat.loc[:, \"timestamp\"].astype(np.float32)\n text = pd_dat.loc[:, \"text\"]\n sequence = np.asarray([np.array(el) for el in pd_dat.loc[:, \"sequence\"]])\n #unicode_char_ratios = pd_dat[unicode_data_categories].astype(np.float32)\n unicode_char_categories = {\n category_name: pd_dat[category_name] for category_name in non_null_unicode_categories\n }\n \n words_in_tweet = pd_dat['words_in_tweet']\n\n return timestamps, text, sequence, unicode_char_categories, words_in_tweet, labels\n\ntimestamps_train, text_train, sequence_train, unicode_char_categories_train, words_in_tweet_train, labels_train = split_inputs_and_outputs(train_dataset)\ntimestamps_val, text_val, sequence_val, unicode_char_categories_val, words_in_tweet_val, labels_val = split_inputs_and_outputs(val_dataset)\ntimestamps_test, text_test, sequence_test, unicode_char_categories_test, words_in_tweet_test, labels_test = split_inputs_and_outputs(test_dataset)",
"_____no_output_____"
]
],
[
[
"## Input tensors",
"_____no_output_____"
]
],
[
[
"input_timestamp = Input(shape=(1, ), name='input_timestamp', dtype=tf.float32)\ninput_text = Input(shape=(1, ), name='input_text', dtype=tf.string)\ninput_sequence = Input(shape=(None, 1 ), name=\"input_sequence\", dtype=tf.float32)\ninput_unicode_char_categories = [\n Input(shape=(1, ), name=\"input_\"+category_name, dtype=tf.float32) for category_name in non_null_unicode_categories\n]\ninput_words_in_tweet = Input(shape=(1, ), name=\"input_words_in_tweet\", dtype=tf.float32)\n\ninputs_train = {\n 'input_timestamp': timestamps_train,\n \"input_text\": text_train,\n \"input_sequence\": sequence_train,\n 'input_words_in_tweet': words_in_tweet_train,\n}\n\ninputs_train.update({\n 'input_' + category_name: unicode_char_categories_train[category_name] for category_name in non_null_unicode_categories\n})\n\noutputs_train = labels_train\n\ninputs_val = {\n 'input_timestamp': timestamps_val,\n \"input_text\": text_val,\n \"input_sequence\": sequence_val,\n 'input_words_in_tweet': words_in_tweet_val\n}\n\ninputs_val.update({\n 'input_' + category_name: unicode_char_categories_val[category_name] for category_name in non_null_unicode_categories\n})\n \noutputs_val = labels_val\n\ninputs_test = {\n 'input_timestamp': timestamps_test,\n \"input_text\": text_test,\n \"input_sequence\": sequence_test,\n 'input_words_in_tweet': words_in_tweet_test\n}\n\ninputs_test.update({\n 'input_' + category_name: unicode_char_categories_test[category_name] for category_name in non_null_unicode_categories\n})\n\noutputs_test = labels_test",
"_____no_output_____"
]
],
[
[
"## TensorFlow 2 model definition",
"_____no_output_____"
]
],
[
[
"def get_model():\n\n reg = None\n activation = 'relu'\n\n reshaped_text = layers.Reshape(target_shape=())(input_text)\n embedded = hub_layer(reshaped_text)\n x = layers.Dense(256, activation=activation)(embedded)\n\n masking = layers.Masking(mask_value=pad_sequence)(input_sequence)\n\n lstm_layer = layers.Bidirectional(layers.LSTM(32))(masking)\n\n flattened_lstm_layer = layers.Flatten()(lstm_layer)\n\n x = layers.concatenate([\n input_timestamp, \n flattened_lstm_layer,\n *input_unicode_char_categories,\n input_words_in_tweet,\n x\n ])\n\n x = layers.Dense(n_tokens // 30, activation=activation, kernel_regularizer=reg)(x)\n\n x = layers.Dropout(0.1)(x)\n\n x = layers.Dense(n_tokens // 50, activation=activation, kernel_regularizer=reg)(x)\n \n x = layers.Dropout(0.1)(x)\n\n x = layers.Dense(256, activation=activation, kernel_regularizer=reg)(x)\n\n y = layers.Dense(len(handles), activation='linear')(x)\n\n model = tf.keras.Model(\n inputs=[\n input_timestamp, \n input_text, \n input_sequence,\n *input_unicode_char_categories,\n input_words_in_tweet\n ],\n outputs=[y]\n )\n\n cce = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)\n\n model.compile(\n optimizer='adam',\n loss=cce,\n metrics=['sparse_categorical_accuracy']\n )\n\n return model\n\nmodel = get_model()\n\ntf.keras.utils.plot_model(model, to_file='twitstar.png', show_shapes=True)",
"_____no_output_____"
]
],
[
[
"## TensorFlow 2 model fitting",
"_____no_output_____"
]
],
[
[
"history = model.fit(\n inputs_train,\n outputs_train,\n epochs=15,\n batch_size=64,\n verbose=True,\n validation_data=(inputs_val, outputs_val),\n callbacks=[\n tf.keras.callbacks.ModelCheckpoint(\n os.path.realpath(\"weights.h5\"),\n monitor=\"val_sparse_categorical_accuracy\", \n save_best_only=True,\n verbose=2\n ),\n tf.keras.callbacks.EarlyStopping(\n patience=3,\n monitor=\"val_sparse_categorical_accuracy\"\n ),\n ]\n)",
"_____no_output_____"
]
],
[
[
"## TensorFlow 2 model plots for train loss and accuracy\n",
"_____no_output_____"
]
],
[
[
"import matplotlib.pyplot as plt\n%matplotlib inline\nplt.plot(history.history['loss'])\nplt.plot(history.history['val_loss'])\nplt.title('Loss vs. epochs')\nplt.ylabel('Loss')\nplt.xlabel('Epoch')\nplt.legend(['Training', 'Validation'], loc='upper right')\nplt.show()\n\n\nplt.plot(history.history['sparse_categorical_accuracy'])\nplt.plot(history.history['val_sparse_categorical_accuracy'])\nplt.title('Accuracy vs. epochs')\nplt.ylabel('Accuracy')\nplt.xlabel('Epoch')\nplt.legend(['Training', 'Validation'], loc='upper right')\nplt.show()",
"_____no_output_____"
]
],
[
[
"## TensorFlow 2 model evaluation",
"_____no_output_____"
]
],
[
[
"#loading the \"best\" weights\nmodel.load_weights(os.path.realpath(\"weights.h5\"))\n\nmodel.evaluate(inputs_test, outputs_test)",
"_____no_output_____"
]
],
[
[
"### TensorFlow 2 model confusion matrix\n\nUsing predictions on the test set, a confusion matrix is produced",
"_____no_output_____"
]
],
[
[
"def tf2_confusion_matrix(inputs, outputs):\n predictions = model.predict(inputs)\n wrong_labelled_counter = np.zeros((len(handles), len(handles)))\n\n wrong_labelled_sequences = np.empty((len(handles), len(handles)), np.object)\n\n for i in range(len(handles)):\n for j in range(len(handles)):\n wrong_labelled_sequences[i][j] = []\n\n tot_wrong = 0\n\n for i in range(len(predictions)):\n predicted = int(predictions[i].argmax())\n \n true_value = int(outputs[i])\n\n wrong_labelled_counter[true_value][predicted] += 1\n wrong_labelled_sequences[true_value][predicted].append(inputs.get('input_text')[i])\n \n ok = (int(true_value) == int(predicted))\n if(not ok):\n tot_wrong += 1\n\n return wrong_labelled_counter, wrong_labelled_sequences, predictions\n\ndef print_confusion_matrix(wrong_labelled_counter):\n the_str = \"\\t\"\n for handle in handles:\n the_str += handle + \"\\t\"\n print(the_str)\n\n ctr = 0\n for row in wrong_labelled_counter:\n the_str = handles[ctr] + '\\t'\n ctr+=1\n for i in range(len(row)):\n the_str += str(int(row[i]))\n if(i != len(row) -1):\n the_str += \"\\t\"\n print(the_str)\n\nwrong_labelled_counter, wrong_labelled_sequences, predictions = tf2_confusion_matrix(inputs_test, outputs_test)\n\n\nprint_confusion_matrix(wrong_labelled_counter)",
"_____no_output_____"
]
],
[
[
"# CatBoost model\n\nThis CatBoost model instance was developed reusing the ideas presented in these tutorials from the official repository: [classification](https://github.com/catboost/tutorials/blob/master/classification/classification_tutorial.ipynb) and [text features](https://github.com/catboost/tutorials/blob/master/text_features/text_features_in_catboost.ipynb)",
"_____no_output_____"
],
[
"## Imports for the CatBoost model",
"_____no_output_____"
]
],
[
[
"import functools\nimport os\n\nimport pandas as pd\nimport numpy as np\n\nimport copy\n\nimport calendar\nimport datetime\nimport re\n\nimport unicodedata\nfrom catboost import Pool, CatBoostClassifier",
"_____no_output_____"
]
],
[
[
"## Definitions for the CatBoost model",
"_____no_output_____"
]
],
[
[
"\nunicode_data_categories = [\n \"Cc\",\n \"Cf\",\n \"Cn\",\n \"Co\",\n \"Cs\",\n \"LC\",\n \"Ll\",\n \"Lm\",\n \"Lo\",\n \"Lt\",\n \"Lu\",\n \"Mc\",\n \"Me\",\n \"Mn\",\n \"Nd\",\n \"Nl\",\n \"No\",\n \"Pc\",\n \"Pd\",\n \"Pe\",\n \"Pf\",\n \"Pi\",\n \"Po\",\n \"Ps\",\n \"Sc\",\n \"Sk\",\n \"Sm\",\n \"So\",\n \"Zl\",\n \"Zp\",\n \"Zs\"\n]\n\n\n\ncolumn_names = [\n \"handle\",\n \"timestamp\",\n \"text\"\n]\n\ncolumn_names.extend(unicode_data_categories)\n\n#List of handles (labels)\n#Fill with the handles you want to consider in your dataset\nhandles = [\n\n]\n\ntrain_file = os.path.realpath(\"./data/train.csv\")\nval_file = os.path.realpath(\"./data/val.csv\")\ntest_file = os.path.realpath(\"./data/test.csv\")\n",
"_____no_output_____"
]
],
[
[
"## Preprocessing and computing dataset features",
"_____no_output_____"
]
],
[
[
"def get_pandas_dataset(input_file, timestamp_mean=None, timestamp_std=None):\n\n pd_dat = pd.read_csv(input_file, names=column_names)\n\n pd_dat = pd_dat[pd_dat.handle.isin(handles)]\n\n if(timestamp_mean is None):\n timestamp_mean = pd_dat.timestamp.mean()\n \n if(timestamp_std is None):\n timestamp_std = pd_dat.timestamp.std()\n\n pd_dat.timestamp = (pd_dat.timestamp - timestamp_mean) / timestamp_std\n\n pd_dat[\"handle_index\"] = pd_dat['handle'].map(lambda x: handles.index(x))\n\n pd_dat = pd_dat.reset_index(drop=True)\n\n return pd_dat, timestamp_mean, timestamp_std\n\ntrain_dataset, timestamp_mean, timestamp_std = get_pandas_dataset(train_file)\n\ntest_dataset, _, _ = get_pandas_dataset(test_file, timestamp_mean=timestamp_mean, timestamp_std=timestamp_std)\n\nval_dataset, _, _ = get_pandas_dataset(val_file, timestamp_mean=timestamp_mean, timestamp_std=timestamp_std)\n\ndef split_inputs_and_outputs(pd_dat):\n\n labels = pd_dat['handle_index'].values\n\n del(pd_dat['handle'])\n del(pd_dat['handle_index'])\n\n return pd_dat, labels\n\nX_train, labels_train = split_inputs_and_outputs(train_dataset)\nX_val, labels_val = split_inputs_and_outputs(val_dataset)\nX_test, labels_test = split_inputs_and_outputs(test_dataset)",
"_____no_output_____"
]
],
[
[
"## CatBoost model definition",
"_____no_output_____"
]
],
[
[
"def get_model(catboost_params={}):\n cat_features = []\n text_features = ['text']\n\n catboost_default_params = {\n 'iterations': 1000,\n 'learning_rate': 0.03,\n 'eval_metric': 'Accuracy',\n 'task_type': 'GPU',\n 'early_stopping_rounds': 20\n }\n \n catboost_default_params.update(catboost_params)\n \n model = CatBoostClassifier(**catboost_default_params)\n\n return model, cat_features, text_features\n\n\nmodel, cat_features, text_features = get_model()",
"_____no_output_____"
]
],
[
[
"## CatBoost model fitting",
"_____no_output_____"
]
],
[
[
"def fit_model(X_train, X_val, y_train, y_val, model, cat_features, text_features, verbose=100):\n\n learn_pool = Pool(\n X_train, \n y_train, \n cat_features=cat_features,\n text_features=text_features,\n feature_names=list(X_train)\n )\n\n val_pool = Pool(\n X_val, \n y_val, \n cat_features=cat_features,\n text_features=text_features,\n feature_names=list(X_val)\n )\n\n model.fit(learn_pool, eval_set=val_pool, verbose=verbose)\n\n return model\n\nmodel = fit_model(X_train, X_val, labels_train, labels_val, model, cat_features, text_features)",
"_____no_output_____"
]
],
[
[
"## CatBoost model evaluation\n\nAlso for the CatBoost model, predictions on the test set, a confusion matrix is produced",
"_____no_output_____"
]
],
[
[
"def predict(X, model, cat_features, text_features):\n pool = Pool(\n data=X,\n cat_features=cat_features,\n text_features=text_features,\n feature_names=list(X)\n )\n\n probs = model.predict_proba(pool)\n\n return probs\n\ndef check_predictions_on(inputs, outputs, model, cat_features, text_features, handles):\n predictions = predict(inputs, model, cat_features, text_features)\n labelled_counter = np.zeros((len(handles), len(handles)))\n\n labelled_sequences = np.empty((len(handles), len(handles)), np.object)\n\n for i in range(len(handles)):\n for j in range(len(handles)):\n labelled_sequences[i][j] = []\n\n tot_wrong = 0\n\n for i in range(len(predictions)):\n predicted = int(predictions[i].argmax())\n \n true_value = int(outputs[i])\n\n labelled_counter[true_value][predicted] += 1\n labelled_sequences[true_value][predicted].append(inputs.get('text').values[i])\n \n ok = (int(true_value) == int(predicted))\n if(not ok):\n tot_wrong += 1\n\n return labelled_counter, labelled_sequences, predictions\n\ndef confusion_matrix(labelled_counter, handles):\n the_str = \"\\t\"\n for handle in handles:\n the_str += handle + \"\\t\"\n\n the_str += \"\\n\"\n\n ctr = 0\n for row in labelled_counter:\n the_str += handles[ctr] + '\\t'\n ctr+=1\n for i in range(len(row)):\n the_str += str(int(row[i]))\n if(i != len(row) -1):\n the_str += \"\\t\"\n the_str += \"\\n\"\n\n return the_str\n\nlabelled_counter, labelled_sequences, predictions = check_predictions_on(\n X_test,\n labels_test,\n model,\n cat_features,\n text_features,\n handles\n)\n\nconfusion_matrix_string = confusion_matrix(labelled_counter, handles)\n\nprint(confusion_matrix_string)",
"_____no_output_____"
]
],
[
[
"# Evaluation\n\nTo perform some experiments and evaluate the two models, 18 Twitter users were selected and, for each user, a number of tweets and responses to other users' tweets were collected. In total 39786 tweets were collected. The difference in class representation could be eliminated, for example limiting the number of tweets for each label to the number of tweets in the less represented class. This difference, however, was not eliminated, in order to test if it represents an issue for the accuracy of the two trained models.\n\nThe division of the tweets corresponding to each twitter handle for each file (train, test, validation) is reported in the following table. To avoid policy issues (better safe than sorry), the actual user handle is masked using C_x placeholders and a brief description of the twitter user is presented instead.\n\n|Description|Handle|Train|Test|Validation|Sum|\n|-------|-------|-------|-------|-------|-------|\n|UK-based labour politician|C_1|1604|492|229|2325|\n|US-based democratic politician|C_2|1414|432|195|2041|\n|US-based democratic politician|C_3|1672|498|273|2443|\n|US-based actor|C_4|1798|501|247|2546|\n|UK-based actress|C_5|847|243|110|1200|\n|US-based democratic politician|C_6|2152|605|304|3061|\n|US-based singer|C_7|2101|622|302|3025|\n|US-based singer|C_8|1742|498|240|2480|\n|Civil rights activist|C_9|314|76|58|448|\n|US-based republican politician|C_10|620|159|78|857|\n|US-based TV host|C_11|2022|550|259|2831|\n|Parody account of C_15 |C_12|2081|624|320|3025|\n|US-based democratic politician|C_13|1985|557|303|2845|\n|US-based actor/director|C_14|1272|357|183|1812|\n|US-based republican politician|C_15|1121|298|134|1553|\n|US-based writer|C_16|1966|502|302|2770|\n|US-based writer|C_17|1095|305|153|1553|\n|US-based entrepreneur|C_18|2084|581|306|2971|\n|Sum||27890|7900|3996|39786|\n\n\n\n## TensorFlow 2 model\n\nThe following charts show loss and accuracy vs epochs for train and validation for a typical run of the TF2 model:\n\n\n\n\nIf the images do not show correctly, they can be found at these links: [loss](https://github.com/icappello/ml-predict-text-author/blob/master/img/tf2_train_val_loss.png) [accuracy](https://github.com/icappello/ml-predict-text-author/blob/master/img/tf2_train_val_accuracy.png)\n\nAfter a few epochs, the model starts overfitting on the train data, and the accuracy for the validation set quickly reaches a plateau.\n\nThe obtained accuracy on the test set is 0.672\n\n## CatBoost model\n\nThe fit procedure stopped after 303 iterations. The obtained accuracy on the test set is 0.808\n\n## Confusion matrices\n\nThe confusion matrices for the two models are reported [here](https://docs.google.com/spreadsheets/d/17JGDXYRajnC4THrBnZrbcqQbgzgjo0Jb7KAvPYenr-w/edit?usp=sharing), since large tables are not displayed correctly in the embedded github viewer for jupyter notebooks. Rows represent the actual classes, while columns represent the predicted ones.\n\n## Summary\n\nThe CatBoost model obtained a better accuracy overall, as well as a better accuracy on all but one label. No particular optimization was done on the definition of the CatBoost model. The TF2 model could need more data, as well as some changes to its definition, to perform better (comments and pointers on this are welcome). Some variants of the TF2 model were tried: a deeper model with more dense layers, higher dropout rate, more/less units in layers, using only a subset of features, regularization methods (L1, L2, batch regularization), different activation functions (sigmoid, tanh) but none performed significantly better than the one presented.\n\nLooking at the results summarized in the confusion matrices, tweets from C_9 clearly represented a problem, either for the under-representation relative to the other classes or for the actual content of the tweets (some were not written in english). Also, tweets from handles C_5 and C_14 were hard to correctly classify for both models, even if they were not under-represented w.r.t other labels.",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown"
] |
[
[
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
]
] |
cbd823b3c396d1fe67b83dc3232fc18f4e920718
| 8,441 |
ipynb
|
Jupyter Notebook
|
secao5 - Python para Analise de dados utizando Numpy/indexacao e fatiamento de arrays.ipynb
|
IgoPereiraBarros/data-science-and-machine-learning
|
e0430171b5597766da5112b48cbdbfd278fbb0bb
|
[
"Apache-2.0"
] | 1 |
2020-09-19T02:01:26.000Z
|
2020-09-19T02:01:26.000Z
|
secao5 - Python para Analise de dados utizando Numpy/indexacao e fatiamento de arrays.ipynb
|
IgoPereiraBarros/data-science-and-machine-learning
|
e0430171b5597766da5112b48cbdbfd278fbb0bb
|
[
"Apache-2.0"
] | null | null | null |
secao5 - Python para Analise de dados utizando Numpy/indexacao e fatiamento de arrays.ipynb
|
IgoPereiraBarros/data-science-and-machine-learning
|
e0430171b5597766da5112b48cbdbfd278fbb0bb
|
[
"Apache-2.0"
] | null | null | null | 20.945409 | 83 | 0.429215 |
[
[
[
"import numpy as np",
"_____no_output_____"
],
[
"vet = np.arange(50).reshape((5,10))",
"_____no_output_____"
],
[
"vet",
"_____no_output_____"
],
[
"vet.shape",
"_____no_output_____"
],
[
"vet[0:2][:1]",
"_____no_output_____"
],
[
"vet2 = vet[:4]",
"_____no_output_____"
],
[
"vet2",
"_____no_output_____"
],
[
"vet2[:] = 100",
"_____no_output_____"
],
[
"vet2",
"_____no_output_____"
],
[
"vet",
"_____no_output_____"
],
[
"vet2 = vet[:3].copy()",
"_____no_output_____"
],
[
"vet2",
"_____no_output_____"
],
[
"vet[1:4,5:7]",
"_____no_output_____"
],
[
"bol = vet > 50",
"_____no_output_____"
],
[
"bol",
"_____no_output_____"
],
[
"vet[bol]",
"_____no_output_____"
],
[
"vet3 = np.linspace(0,100,30)",
"_____no_output_____"
],
[
"vet3.shape",
"_____no_output_____"
],
[
"vet31 = vet3.reshape(3, 10)",
"_____no_output_____"
],
[
"vet31",
"_____no_output_____"
],
[
"vet31[0:2, 2]",
"_____no_output_____"
]
]
] |
[
"code"
] |
[
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
]
] |
cbd82634efb228ba115e884cac337c7910680757
| 99,824 |
ipynb
|
Jupyter Notebook
|
Deploying a Sentiment Analysis Model/SageMaker Project.ipynb
|
Yasel-Garces/deep-learning-v2-pytorch
|
95283d87062fcba594d11a881a0cbf2bfe835b4b
|
[
"MIT"
] | null | null | null |
Deploying a Sentiment Analysis Model/SageMaker Project.ipynb
|
Yasel-Garces/deep-learning-v2-pytorch
|
95283d87062fcba594d11a881a0cbf2bfe835b4b
|
[
"MIT"
] | null | null | null |
Deploying a Sentiment Analysis Model/SageMaker Project.ipynb
|
Yasel-Garces/deep-learning-v2-pytorch
|
95283d87062fcba594d11a881a0cbf2bfe835b4b
|
[
"MIT"
] | null | null | null | 50.365288 | 1,439 | 0.615974 |
[
[
[
"# Creating a Sentiment Analysis Web App\n## Using PyTorch and SageMaker\n\n_Deep Learning Nanodegree Program | Deployment_\n\n---\n\nNow that we have a basic understanding of how SageMaker works we will try to use it to construct a complete project from end to end. Our goal will be to have a simple web page which a user can use to enter a movie review. The web page will then send the review off to our deployed model which will predict the sentiment of the entered review.\n\n## Instructions\n\nSome template code has already been provided for you, and you will need to implement additional functionality to successfully complete this notebook. You will not need to modify the included code beyond what is requested. Sections that begin with '**TODO**' in the header indicate that you need to complete or implement some portion within them. Instructions will be provided for each section and the specifics of the implementation are marked in the code block with a `# TODO: ...` comment. Please be sure to read the instructions carefully!\n\nIn addition to implementing code, there will be questions for you to answer which relate to the task and your implementation. Each section where you will answer a question is preceded by a '**Question:**' header. Carefully read each question and provide your answer below the '**Answer:**' header by editing the Markdown cell.\n\n> **Note**: Code and Markdown cells can be executed using the **Shift+Enter** keyboard shortcut. In addition, a cell can be edited by typically clicking it (double-click for Markdown cells) or by pressing **Enter** while it is highlighted.\n\n## General Outline\n\nRecall the general outline for SageMaker projects using a notebook instance.\n\n1. Download or otherwise retrieve the data.\n2. Process / Prepare the data.\n3. Upload the processed data to S3.\n4. Train a chosen model.\n5. Test the trained model (typically using a batch transform job).\n6. Deploy the trained model.\n7. Use the deployed model.\n\nFor this project, you will be following the steps in the general outline with some modifications. \n\nFirst, you will not be testing the model in its own step. You will still be testing the model, however, you will do it by deploying your model and then using the deployed model by sending the test data to it. One of the reasons for doing this is so that you can make sure that your deployed model is working correctly before moving forward.\n\nIn addition, you will deploy and use your trained model a second time. In the second iteration you will customize the way that your trained model is deployed by including some of your own code. In addition, your newly deployed model will be used in the sentiment analysis web app.",
"_____no_output_____"
],
[
"## Step 1: Downloading the data\n\nAs in the XGBoost in SageMaker notebook, we will be using the [IMDb dataset](http://ai.stanford.edu/~amaas/data/sentiment/)\n\n> Maas, Andrew L., et al. [Learning Word Vectors for Sentiment Analysis](http://ai.stanford.edu/~amaas/data/sentiment/). In _Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies_. Association for Computational Linguistics, 2011.",
"_____no_output_____"
]
],
[
[
"%mkdir ../data\n!wget -O ../data/aclImdb_v1.tar.gz http://ai.stanford.edu/~amaas/data/sentiment/aclImdb_v1.tar.gz\n!tar -zxf ../data/aclImdb_v1.tar.gz -C ../data",
"mkdir: cannot create directory ‘../data’: File exists\n--2020-05-25 14:17:51-- http://ai.stanford.edu/~amaas/data/sentiment/aclImdb_v1.tar.gz\nResolving ai.stanford.edu (ai.stanford.edu)... 171.64.68.10\nConnecting to ai.stanford.edu (ai.stanford.edu)|171.64.68.10|:80... connected.\nHTTP request sent, awaiting response... 200 OK\nLength: 84125825 (80M) [application/x-gzip]\nSaving to: ‘../data/aclImdb_v1.tar.gz’\n\n../data/aclImdb_v1. 100%[===================>] 80.23M 24.1MB/s in 4.1s \n\n2020-05-25 14:17:55 (19.6 MB/s) - ‘../data/aclImdb_v1.tar.gz’ saved [84125825/84125825]\n\n"
]
],
[
[
"## Step 2: Preparing and Processing the data\n\nAlso, as in the XGBoost notebook, we will be doing some initial data processing. The first few steps are the same as in the XGBoost example. To begin with, we will read in each of the reviews and combine them into a single input structure. Then, we will split the dataset into a training set and a testing set.",
"_____no_output_____"
]
],
[
[
"import os\nimport glob\n\ndef read_imdb_data(data_dir='../data/aclImdb'):\n data = {}\n labels = {}\n \n for data_type in ['train', 'test']:\n data[data_type] = {}\n labels[data_type] = {}\n \n for sentiment in ['pos', 'neg']:\n data[data_type][sentiment] = []\n labels[data_type][sentiment] = []\n \n path = os.path.join(data_dir, data_type, sentiment, '*.txt')\n files = glob.glob(path)\n \n for f in files:\n with open(f) as review:\n data[data_type][sentiment].append(review.read())\n # Here we represent a positive review by '1' and a negative review by '0'\n labels[data_type][sentiment].append(1 if sentiment == 'pos' else 0)\n \n assert len(data[data_type][sentiment]) == len(labels[data_type][sentiment]), \\\n \"{}/{} data size does not match labels size\".format(data_type, sentiment)\n \n return data, labels",
"_____no_output_____"
],
[
"data, labels = read_imdb_data()\nprint(\"IMDB reviews: train = {} pos / {} neg, test = {} pos / {} neg\".format(\n len(data['train']['pos']), len(data['train']['neg']),\n len(data['test']['pos']), len(data['test']['neg'])))",
"IMDB reviews: train = 12500 pos / 12500 neg, test = 12500 pos / 12500 neg\n"
]
],
[
[
"Now that we've read the raw training and testing data from the downloaded dataset, we will combine the positive and negative reviews and shuffle the resulting records.",
"_____no_output_____"
]
],
[
[
"from sklearn.utils import shuffle\n\ndef prepare_imdb_data(data, labels):\n \"\"\"Prepare training and test sets from IMDb movie reviews.\"\"\"\n \n #Combine positive and negative reviews and labels\n data_train = data['train']['pos'] + data['train']['neg']\n data_test = data['test']['pos'] + data['test']['neg']\n labels_train = labels['train']['pos'] + labels['train']['neg']\n labels_test = labels['test']['pos'] + labels['test']['neg']\n \n #Shuffle reviews and corresponding labels within training and test sets\n data_train, labels_train = shuffle(data_train, labels_train)\n data_test, labels_test = shuffle(data_test, labels_test)\n \n # Return a unified training data, test data, training labels, test labets\n return data_train, data_test, labels_train, labels_test",
"_____no_output_____"
],
[
"train_X, test_X, train_y, test_y = prepare_imdb_data(data, labels)\nprint(\"IMDb reviews (combined): train = {}, test = {}\".format(len(train_X), len(test_X)))",
"IMDb reviews (combined): train = 25000, test = 25000\n"
]
],
[
[
"Now that we have our training and testing sets unified and prepared, we should do a quick check and see an example of the data our model will be trained on. This is generally a good idea as it allows you to see how each of the further processing steps affects the reviews and it also ensures that the data has been loaded correctly.",
"_____no_output_____"
]
],
[
[
"print(train_X[100])\nprint(train_y[100])",
"Mulva is put in a sugercoma by Teen Ape. When she awakens she's considerably hotter (the parts played by Debbie Rochon in this sequel), and out for revenge on those that did her wrong. As the sub-title and box art implies this is indeed a take off on the \"Kill Bill\" films, but this being a Chris Seaver's film, it's a wildly incompetent satire (and I use that last term extremely loosely) I'd love to say this is better than the first film, but truth be told I was so impossibly drunk off my ass when I saw the previous film that I can't even hope to compare the two at this point in time. But I must have enjoyed it since I bought the sequel (see when I'm drunk I like, for lack of a better word, complete and total crappy movies) There are a few laugh out loud moments (very few) but I do remember Bonejack being funnier though. Well at least at slightly over an hour, it IS over mighty quickly for what it's worth.<br /><br />My Grade: D+ <br /><br />DVD Extras: Audio commentary by Director Chris Seaver, actress Debbie Rochon, and the Lbp gang; Second commentary by Seaver; 31 minute Making of featurette; Lloyd Kaufman's 6 minute tribute to low budget pictures; Fake 2 minute syrup commercial; stills gallery; Promo trailer; and trailers for :Mulva\", 'Quest for the Egg Salad\", \"Fiilthy McNasty\"1, 2, & 3; \"the Feral Man\", \"the Bonesetter\", \"Midnight Skater\", \"Demon Summer\" & \"Splatter Rampage Wrestling\"\n0\n"
]
],
[
[
"The first step in processing the reviews is to make sure that any html tags that appear should be removed. In addition we wish to tokenize our input, that way words such as *entertained* and *entertaining* are considered the same with regard to sentiment analysis.",
"_____no_output_____"
]
],
[
[
"import nltk\nfrom nltk.corpus import stopwords\nfrom nltk.stem.porter import *\n\nimport re\nfrom bs4 import BeautifulSoup\n\ndef review_to_words(review):\n nltk.download(\"stopwords\", quiet=True)\n stemmer = PorterStemmer()\n \n text = BeautifulSoup(review, \"html.parser\").get_text() # Remove HTML tags\n text = re.sub(r\"[^a-zA-Z0-9]\", \" \", text.lower()) # Convert to lower case\n words = text.split() # Split string into words\n words = [w for w in words if w not in stopwords.words(\"english\")] # Remove stopwords\n words = [PorterStemmer().stem(w) for w in words] # stem\n \n return words",
"_____no_output_____"
]
],
[
[
"The `review_to_words` method defined above uses `BeautifulSoup` to remove any html tags that appear and uses the `nltk` package to tokenize the reviews. As a check to ensure we know how everything is working, try applying `review_to_words` to one of the reviews in the training set.",
"_____no_output_____"
]
],
[
[
"# TODO: Apply review_to_words to a review (train_X[100] or any other review)\nreview_to_words(train_X[100])",
"_____no_output_____"
]
],
[
[
"**Question:** Above we mentioned that `review_to_words` method removes html formatting and allows us to tokenize the words found in a review, for example, converting *entertained* and *entertaining* into *entertain* so that they are treated as though they are the same word. What else, if anything, does this method do to the input?",
"_____no_output_____"
],
[
"**Answer:** This method also converts to lowercase all the words and remove stop words. Stop words are common language words such as \"the\", \"a\", \"an\" or \"in\". ",
"_____no_output_____"
],
[
"The method below applies the `review_to_words` method to each of the reviews in the training and testing datasets. In addition it caches the results. This is because performing this processing step can take a long time. This way if you are unable to complete the notebook in the current session, you can come back without needing to process the data a second time.",
"_____no_output_____"
]
],
[
[
"import pickle\n\ncache_dir = os.path.join(\"../cache\", \"sentiment_analysis\") # where to store cache files\nos.makedirs(cache_dir, exist_ok=True) # ensure cache directory exists\n\ndef preprocess_data(data_train, data_test, labels_train, labels_test,\n cache_dir=cache_dir, cache_file=\"preprocessed_data.pkl\"):\n \"\"\"Convert each review to words; read from cache if available.\"\"\"\n\n # If cache_file is not None, try to read from it first\n cache_data = None\n if cache_file is not None:\n try:\n with open(os.path.join(cache_dir, cache_file), \"rb\") as f:\n cache_data = pickle.load(f)\n print(\"Read preprocessed data from cache file:\", cache_file)\n except:\n pass # unable to read from cache, but that's okay\n \n # If cache is missing, then do the heavy lifting\n if cache_data is None:\n # Preprocess training and test data to obtain words for each review\n #words_train = list(map(review_to_words, data_train))\n #words_test = list(map(review_to_words, data_test))\n words_train = [review_to_words(review) for review in data_train]\n words_test = [review_to_words(review) for review in data_test]\n \n # Write to cache file for future runs\n if cache_file is not None:\n cache_data = dict(words_train=words_train, words_test=words_test,\n labels_train=labels_train, labels_test=labels_test)\n with open(os.path.join(cache_dir, cache_file), \"wb\") as f:\n pickle.dump(cache_data, f)\n print(\"Wrote preprocessed data to cache file:\", cache_file)\n else:\n # Unpack data loaded from cache file\n words_train, words_test, labels_train, labels_test = (cache_data['words_train'],\n cache_data['words_test'], cache_data['labels_train'], cache_data['labels_test'])\n \n return words_train, words_test, labels_train, labels_test",
"_____no_output_____"
],
[
"# Preprocess data\ntrain_X, test_X, train_y, test_y = preprocess_data(train_X, test_X, train_y, test_y)",
"Read preprocessed data from cache file: preprocessed_data.pkl\n"
]
],
[
[
"## Transform the data\n\nIn the XGBoost notebook we transformed the data from its word representation to a bag-of-words feature representation. For the model we are going to construct in this notebook we will construct a feature representation which is very similar. To start, we will represent each word as an integer. Of course, some of the words that appear in the reviews occur very infrequently and so likely don't contain much information for the purposes of sentiment analysis. The way we will deal with this problem is that we will fix the size of our working vocabulary and we will only include the words that appear most frequently. We will then combine all of the infrequent words into a single category and, in our case, we will label it as `1`.\n\nSince we will be using a recurrent neural network, it will be convenient if the length of each review is the same. To do this, we will fix a size for our reviews and then pad short reviews with the category 'no word' (which we will label `0`) and truncate long reviews.",
"_____no_output_____"
],
[
"### (TODO) Create a word dictionary\n\nTo begin with, we need to construct a way to map words that appear in the reviews to integers. Here we fix the size of our vocabulary (including the 'no word' and 'infrequent' categories) to be `5000` but you may wish to change this to see how it affects the model.\n\n> **TODO:** Complete the implementation for the `build_dict()` method below. Note that even though the vocab_size is set to `5000`, we only want to construct a mapping for the most frequently appearing `4998` words. This is because we want to reserve the special labels `0` for 'no word' and `1` for 'infrequent word'.",
"_____no_output_____"
]
],
[
[
"import numpy as np\n\ndef build_dict(data, vocab_size = 5000):\n \"\"\"Construct and return a dictionary mapping each of the most frequently appearing words to a unique integer.\"\"\"\n \n # TODO: Determine how often each word appears in `data`. Note that `data` is a list of sentences and that a\n # sentence is a list of words.\n \n word_count = {} # A dict storing the words that appear in the reviews along with how often they occur\n for sentence in data:\n for word in sentence:\n if word in word_count.keys():\n word_count[word] += 1\n else:\n word_count[word]=1 \n \n # TODO: Sort the words found in `data` so that sorted_words[0] is the most frequently appearing word and\n # sorted_words[-1] is the least frequently appearing word.\n sorted_words = [k for k, _ in sorted(word_count.items(), key=lambda item: item[1],reverse=True)]\n \n word_dict = {} # This is what we are building, a dictionary that translates words into integers\n for idx, word in enumerate(sorted_words[:vocab_size - 2]): # The -2 is so that we save room for the 'no word'\n word_dict[word] = idx + 2 # 'infrequent' labels\n \n return word_dict",
"_____no_output_____"
],
[
"word_dict = build_dict(train_X)",
"_____no_output_____"
]
],
[
[
"**Question:** What are the five most frequently appearing (tokenized) words in the training set? Does it makes sense that these words appear frequently in the training set?",
"_____no_output_____"
],
[
"**Answer:** The most frequently words in the training set are {\"hostil\", \"assort\", \"handicap\", \"monti\", \"sparkl\"}. Well, if we take into account that the training set has positive and negative reviews, we can associate \"hostil\" to negative movies and \"sparkl\" to positive ones, but honestly, I have not a clear idea if they have a sense in this context. My intuition tell me that this is the idea because the words are tokenized but maybe this is because of my English level :/. ",
"_____no_output_____"
]
],
[
[
"# TODO: Use this space to determine the five most frequently appearing words in the training set.\ntop_wors_training=[k for k, _ in sorted(word_dict.items(), key=lambda item: item[1],reverse=True)]\ntop_wors_training[:5]",
"_____no_output_____"
]
],
[
[
"### Save `word_dict`\n\nLater on when we construct an endpoint which processes a submitted review we will need to make use of the `word_dict` which we have created. As such, we will save it to a file now for future use.",
"_____no_output_____"
]
],
[
[
"data_dir = '../data/pytorch' # The folder we will use for storing data\nif not os.path.exists(data_dir): # Make sure that the folder exists\n os.makedirs(data_dir)",
"_____no_output_____"
],
[
"with open(os.path.join(data_dir, 'word_dict.pkl'), \"wb\") as f:\n pickle.dump(word_dict, f)",
"_____no_output_____"
]
],
[
[
"### Transform the reviews\n\nNow that we have our word dictionary which allows us to transform the words appearing in the reviews into integers, it is time to make use of it and convert our reviews to their integer sequence representation, making sure to pad or truncate to a fixed length, which in our case is `500`.",
"_____no_output_____"
]
],
[
[
"def convert_and_pad(word_dict, sentence, pad=500):\n NOWORD = 0 # We will use 0 to represent the 'no word' category\n INFREQ = 1 # and we use 1 to represent the infrequent words, i.e., words not appearing in word_dict\n \n working_sentence = [NOWORD] * pad\n \n for word_index, word in enumerate(sentence[:pad]):\n if word in word_dict:\n working_sentence[word_index] = word_dict[word]\n else:\n working_sentence[word_index] = INFREQ\n \n return working_sentence, min(len(sentence), pad)\n\ndef convert_and_pad_data(word_dict, data, pad=500):\n result = []\n lengths = []\n \n for sentence in data:\n converted, leng = convert_and_pad(word_dict, sentence, pad)\n result.append(converted)\n lengths.append(leng)\n \n return np.array(result), np.array(lengths)",
"_____no_output_____"
],
[
"train_X, train_X_len = convert_and_pad_data(word_dict, train_X)\ntest_X, test_X_len = convert_and_pad_data(word_dict, test_X)",
"_____no_output_____"
]
],
[
[
"As a quick check to make sure that things are working as intended, check to see what one of the reviews in the training set looks like after having been processeed. Does this look reasonable? What is the length of a review in the training set?",
"_____no_output_____"
]
],
[
[
"# Use this cell to examine one of the processed reviews to make sure everything is working as intended.\nprint('One of the review in the training data after codification is:\\n {}'.format(train_X[1]))\nprint('This review have a length of:\\n {}'.format(train_X_len[1]))",
"One of the review in the training data after codification is:\n [2108 498 3 36 1 2423 520 36 2829 1799 96 70 2377 4910\n 90 1 24 1 237 132 70 9 350 34 3 58 12 138\n 333 64 49 2793 2644 239 663 101 103 23 498 1870 2108 30\n 1 1671 3264 713 398 2 5 139 93 9 3 2349 10 253\n 2554 1313 167 363 976 1 74 9 4506 153 1 14 316 37\n 2236 63 2236 1122 547 438 757 465 1313 1313 447 2394 248 181\n 9 2349 64 42 550 18 3 136 114 119 557 233 103 2108\n 434 23 82 13 117 10 253 2554 520 68 176 11 15 1\n 2423 42 138 773 458 61 1 773 70 1 68 531 601 1\n 4058 110 78 1 2388 14 374 249 31 1 2388 74 15 5\n 11 32 1083 9 2004 2793 36 70 1 1685 725 443 2 248\n 18 3560 11 1 451 77 2108 498 27 857 532 12 595 101\n 186 248 225 7 90 2793 68 748 69 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n 0 0 0 0 0 0 0 0 0 0]\nThis review have a length of:\n 177\n"
]
],
[
[
"**Question:** In the cells above we use the `preprocess_data` and `convert_and_pad_data` methods to process both the training and testing set. Why or why not might this be a problem?",
"_____no_output_____"
],
[
"**Answer:** In my opinion, the most important point here is to remember that the dictionary was constructed only on the training data, so, don't exist data leakage. Also, these functions take as argument the dictionary and a dataset, so, they are totally independent when processing the training or testing data. In conclusion, I don't see a problem to process the data using this approach.",
"_____no_output_____"
],
[
"## Step 3: Upload the data to S3\n\nAs in the XGBoost notebook, we will need to upload the training dataset to S3 in order for our training code to access it. For now we will save it locally and we will upload to S3 later on.\n\n### Save the processed training dataset locally\n\nIt is important to note the format of the data that we are saving as we will need to know it when we write the training code. In our case, each row of the dataset has the form `label`, `length`, `review[500]` where `review[500]` is a sequence of `500` integers representing the words in the review.",
"_____no_output_____"
]
],
[
[
"import pandas as pd\n \npd.concat([pd.DataFrame(train_y), pd.DataFrame(train_X_len), pd.DataFrame(train_X)], axis=1) \\\n .to_csv(os.path.join(data_dir, 'train.csv'), header=False, index=False)",
"_____no_output_____"
]
],
[
[
"### Uploading the training data\n\n\nNext, we need to upload the training data to the SageMaker default S3 bucket so that we can provide access to it while training our model.",
"_____no_output_____"
]
],
[
[
"import sagemaker\n\nsagemaker_session = sagemaker.Session()\n\nbucket = sagemaker_session.default_bucket()\nprefix = 'sagemaker/sentiment_rnn'\n\nrole = sagemaker.get_execution_role()",
"_____no_output_____"
],
[
"input_data = sagemaker_session.upload_data(path=data_dir, bucket=bucket, key_prefix=prefix)",
"_____no_output_____"
]
],
[
[
"**NOTE:** The cell above uploads the entire contents of our data directory. This includes the `word_dict.pkl` file. This is fortunate as we will need this later on when we create an endpoint that accepts an arbitrary review. For now, we will just take note of the fact that it resides in the data directory (and so also in the S3 training bucket) and that we will need to make sure it gets saved in the model directory.",
"_____no_output_____"
],
[
"## Step 4: Build and Train the PyTorch Model\n\nIn the XGBoost notebook we discussed what a model is in the SageMaker framework. In particular, a model comprises three objects\n\n - Model Artifacts,\n - Training Code, and\n - Inference Code,\n \neach of which interact with one another. In the XGBoost example we used training and inference code that was provided by Amazon. Here we will still be using containers provided by Amazon with the added benefit of being able to include our own custom code.\n\nWe will start by implementing our own neural network in PyTorch along with a training script. For the purposes of this project we have provided the necessary model object in the `model.py` file, inside of the `train` folder. You can see the provided implementation by running the cell below.",
"_____no_output_____"
]
],
[
[
"!pygmentize train/model.py",
"\u001b[34mimport\u001b[39;49;00m \u001b[04m\u001b[36mtorch.nn\u001b[39;49;00m \u001b[34mas\u001b[39;49;00m \u001b[04m\u001b[36mnn\u001b[39;49;00m\r\n\r\n\u001b[34mclass\u001b[39;49;00m \u001b[04m\u001b[32mLSTMClassifier\u001b[39;49;00m(nn.Module):\r\n \u001b[33m\"\"\"\u001b[39;49;00m\r\n\u001b[33m This is the simple RNN model we will be using to perform Sentiment Analysis.\u001b[39;49;00m\r\n\u001b[33m \"\"\"\u001b[39;49;00m\r\n\r\n \u001b[34mdef\u001b[39;49;00m \u001b[32m__init__\u001b[39;49;00m(\u001b[36mself\u001b[39;49;00m, embedding_dim, hidden_dim, vocab_size):\r\n \u001b[33m\"\"\"\u001b[39;49;00m\r\n\u001b[33m Initialize the model by settingg up the various layers.\u001b[39;49;00m\r\n\u001b[33m \"\"\"\u001b[39;49;00m\r\n \u001b[36msuper\u001b[39;49;00m(LSTMClassifier, \u001b[36mself\u001b[39;49;00m).\u001b[32m__init__\u001b[39;49;00m()\r\n\r\n \u001b[36mself\u001b[39;49;00m.embedding = nn.Embedding(vocab_size, embedding_dim, padding_idx=\u001b[34m0\u001b[39;49;00m)\r\n \u001b[36mself\u001b[39;49;00m.lstm = nn.LSTM(embedding_dim, hidden_dim)\r\n \u001b[36mself\u001b[39;49;00m.dense = nn.Linear(in_features=hidden_dim, out_features=\u001b[34m1\u001b[39;49;00m)\r\n \u001b[36mself\u001b[39;49;00m.sig = nn.Sigmoid()\r\n \r\n \u001b[36mself\u001b[39;49;00m.word_dict = \u001b[36mNone\u001b[39;49;00m\r\n\r\n \u001b[34mdef\u001b[39;49;00m \u001b[32mforward\u001b[39;49;00m(\u001b[36mself\u001b[39;49;00m, x):\r\n \u001b[33m\"\"\"\u001b[39;49;00m\r\n\u001b[33m Perform a forward pass of our model on some input.\u001b[39;49;00m\r\n\u001b[33m \"\"\"\u001b[39;49;00m\r\n x = x.t()\r\n lengths = x[\u001b[34m0\u001b[39;49;00m,:]\r\n reviews = x[\u001b[34m1\u001b[39;49;00m:,:]\r\n embeds = \u001b[36mself\u001b[39;49;00m.embedding(reviews)\r\n lstm_out, _ = \u001b[36mself\u001b[39;49;00m.lstm(embeds)\r\n out = \u001b[36mself\u001b[39;49;00m.dense(lstm_out)\r\n out = out[lengths - \u001b[34m1\u001b[39;49;00m, \u001b[36mrange\u001b[39;49;00m(\u001b[36mlen\u001b[39;49;00m(lengths))]\r\n \u001b[34mreturn\u001b[39;49;00m \u001b[36mself\u001b[39;49;00m.sig(out.squeeze())\r\n"
]
],
[
[
"The important takeaway from the implementation provided is that there are three parameters that we may wish to tweak to improve the performance of our model. These are the embedding dimension, the hidden dimension and the size of the vocabulary. We will likely want to make these parameters configurable in the training script so that if we wish to modify them we do not need to modify the script itself. We will see how to do this later on. To start we will write some of the training code in the notebook so that we can more easily diagnose any issues that arise.\n\nFirst we will load a small portion of the training data set to use as a sample. It would be very time consuming to try and train the model completely in the notebook as we do not have access to a gpu and the compute instance that we are using is not particularly powerful. However, we can work on a small bit of the data to get a feel for how our training script is behaving.",
"_____no_output_____"
]
],
[
[
"import torch\nimport torch.utils.data\n\n# Read in only the first 250 rows\ntrain_sample = pd.read_csv(os.path.join(data_dir, 'train.csv'), header=None, names=None, nrows=250)\n\n# Turn the input pandas dataframe into tensors\ntrain_sample_y = torch.from_numpy(train_sample[[0]].values).float().squeeze()\ntrain_sample_X = torch.from_numpy(train_sample.drop([0], axis=1).values).long()\n\n# Build the dataset\ntrain_sample_ds = torch.utils.data.TensorDataset(train_sample_X, train_sample_y)\n# Build the dataloader\ntrain_sample_dl = torch.utils.data.DataLoader(train_sample_ds, batch_size=50)",
"_____no_output_____"
]
],
[
[
"### (TODO) Writing the training method\n\nNext we need to write the training code itself. This should be very similar to training methods that you have written before to train PyTorch models. We will leave any difficult aspects such as model saving / loading and parameter loading until a little later.",
"_____no_output_____"
]
],
[
[
"def train(model, train_loader, epochs, optimizer, loss_fn, device):\n for epoch in range(1, epochs + 1):\n model.train()\n total_loss = 0\n for batch in train_loader: \n # Extract the samples\n batch_X, batch_y = batch\n # Import the samples to the available device\n batch_X = batch_X.to(device)\n batch_y = batch_y.to(device)\n \n # TODO: Complete this train method to train the model provided.\n # zero gradients\n model.zero_grad()\n # Forward propagation (return prediction and hidden)\n prediction = model(batch_X)\n # Calculate the loss and perform backpropagation and optimization\n loss = loss_fn(prediction,batch_y)\n loss.backward()\n # Optimizer step\n optimizer.step()\n \n total_loss += loss.data.item()\n print(\"Epoch: {}, BCELoss: {}\".format(epoch, total_loss / len(train_loader)))",
"_____no_output_____"
]
],
[
[
"Supposing we have the training method above, we will test that it is working by writing a bit of code in the notebook that executes our training method on the small sample training set that we loaded earlier. The reason for doing this in the notebook is so that we have an opportunity to fix any errors that arise early when they are easier to diagnose.",
"_____no_output_____"
]
],
[
[
"import torch.optim as optim\nfrom train.model import LSTMClassifier\n\ndevice = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n\nmodel = LSTMClassifier(32, 100, 5000).to(device)\noptimizer = optim.Adam(model.parameters())\nloss_fn = torch.nn.BCELoss()\n\ntrain(model, train_sample_dl, 5, optimizer, loss_fn, device)",
"cpu\nEpoch: 1, BCELoss: 0.6923757076263428\nEpoch: 2, BCELoss: 0.6818168997764588\nEpoch: 3, BCELoss: 0.6724243640899659\nEpoch: 4, BCELoss: 0.6616947054862976\nEpoch: 5, BCELoss: 0.6478506565093994\n"
]
],
[
[
"In order to construct a PyTorch model using SageMaker we must provide SageMaker with a training script. We may optionally include a directory which will be copied to the container and from which our training code will be run. When the training container is executed it will check the uploaded directory (if there is one) for a `requirements.txt` file and install any required Python libraries, after which the training script will be run.",
"_____no_output_____"
],
[
"### (TODO) Training the model\n\nWhen a PyTorch model is constructed in SageMaker, an entry point must be specified. This is the Python file which will be executed when the model is trained. Inside of the `train` directory is a file called `train.py` which has been provided and which contains most of the necessary code to train our model. The only thing that is missing is the implementation of the `train()` method which you wrote earlier in this notebook.\n\n**TODO**: Copy the `train()` method written above and paste it into the `train/train.py` file where required.\n\nThe way that SageMaker passes hyperparameters to the training script is by way of arguments. These arguments can then be parsed and used in the training script. To see how this is done take a look at the provided `train/train.py` file.",
"_____no_output_____"
]
],
[
[
"from sagemaker.pytorch import PyTorch\n\nestimator = PyTorch(entry_point=\"train.py\",\n source_dir=\"train\",\n role=role,\n framework_version='0.4.0',\n train_instance_count=1,\n train_instance_type='ml.m4.xlarge',#'ml.p2.xlarge'\n hyperparameters={\n 'epochs': 10,\n 'hidden_dim': 200,\n })",
"_____no_output_____"
],
[
"estimator.fit({'training': input_data})",
"2020-05-25 14:25:47 Starting - Starting the training job...\n2020-05-25 14:25:49 Starting - Launching requested ML instances......\n2020-05-25 14:26:54 Starting - Preparing the instances for training......\n2020-05-25 14:27:54 Downloading - Downloading input data...\n2020-05-25 14:28:40 Training - Training image download completed. Training in progress.\u001b[34mbash: cannot set terminal process group (-1): Inappropriate ioctl for device\u001b[0m\n\u001b[34mbash: no job control in this shell\u001b[0m\n\u001b[34m2020-05-25 14:28:41,718 sagemaker-containers INFO Imported framework sagemaker_pytorch_container.training\u001b[0m\n\u001b[34m2020-05-25 14:28:41,721 sagemaker-containers INFO No GPUs detected (normal if no gpus installed)\u001b[0m\n\u001b[34m2020-05-25 14:28:41,734 sagemaker_pytorch_container.training INFO Block until all host DNS lookups succeed.\u001b[0m\n\u001b[34m2020-05-25 14:28:41,738 sagemaker_pytorch_container.training INFO Invoking user training script.\u001b[0m\n\u001b[34m2020-05-25 14:28:42,003 sagemaker-containers INFO Module train does not provide a setup.py. \u001b[0m\n\u001b[34mGenerating setup.py\u001b[0m\n\u001b[34m2020-05-25 14:28:42,004 sagemaker-containers INFO Generating setup.cfg\u001b[0m\n\u001b[34m2020-05-25 14:28:42,004 sagemaker-containers INFO Generating MANIFEST.in\u001b[0m\n\u001b[34m2020-05-25 14:28:42,004 sagemaker-containers INFO Installing module with the following command:\u001b[0m\n\u001b[34m/usr/bin/python -m pip install -U . -r requirements.txt\u001b[0m\n\u001b[34mProcessing /opt/ml/code\u001b[0m\n\u001b[34mCollecting pandas (from -r requirements.txt (line 1))\u001b[0m\n\u001b[34m Downloading https://files.pythonhosted.org/packages/74/24/0cdbf8907e1e3bc5a8da03345c23cbed7044330bb8f73bb12e711a640a00/pandas-0.24.2-cp35-cp35m-manylinux1_x86_64.whl (10.0MB)\u001b[0m\n\u001b[34mCollecting numpy (from -r requirements.txt (line 2))\n Downloading https://files.pythonhosted.org/packages/38/92/fa5295d9755c7876cb8490eab866e1780154033fa45978d9cf74ffbd4c68/numpy-1.18.4-cp35-cp35m-manylinux1_x86_64.whl (20.0MB)\u001b[0m\n\u001b[34mCollecting nltk (from -r requirements.txt (line 3))\n Downloading https://files.pythonhosted.org/packages/92/75/ce35194d8e3022203cca0d2f896dbb88689f9b3fce8e9f9cff942913519d/nltk-3.5.zip (1.4MB)\u001b[0m\n\u001b[34mCollecting beautifulsoup4 (from -r requirements.txt (line 4))\n Downloading https://files.pythonhosted.org/packages/66/25/ff030e2437265616a1e9b25ccc864e0371a0bc3adb7c5a404fd661c6f4f6/beautifulsoup4-4.9.1-py3-none-any.whl (115kB)\u001b[0m\n\u001b[34mCollecting html5lib (from -r requirements.txt (line 5))\n Downloading https://files.pythonhosted.org/packages/a5/62/bbd2be0e7943ec8504b517e62bab011b4946e1258842bc159e5dfde15b96/html5lib-1.0.1-py2.py3-none-any.whl (117kB)\u001b[0m\n\u001b[34mRequirement already satisfied, skipping upgrade: python-dateutil>=2.5.0 in /usr/local/lib/python3.5/dist-packages (from pandas->-r requirements.txt (line 1)) (2.7.5)\u001b[0m\n\u001b[34mCollecting pytz>=2011k (from pandas->-r requirements.txt (line 1))\n Downloading https://files.pythonhosted.org/packages/4f/a4/879454d49688e2fad93e59d7d4efda580b783c745fd2ec2a3adf87b0808d/pytz-2020.1-py2.py3-none-any.whl (510kB)\u001b[0m\n\u001b[34mRequirement already satisfied, skipping upgrade: click in /usr/local/lib/python3.5/dist-packages (from nltk->-r requirements.txt (line 3)) (7.0)\u001b[0m\n\u001b[34mCollecting joblib (from nltk->-r requirements.txt (line 3))\n Downloading https://files.pythonhosted.org/packages/28/5c/cf6a2b65a321c4a209efcdf64c2689efae2cb62661f8f6f4bb28547cf1bf/joblib-0.14.1-py2.py3-none-any.whl (294kB)\u001b[0m\n\u001b[34mCollecting regex (from nltk->-r requirements.txt (line 3))\u001b[0m\n\u001b[34m Downloading https://files.pythonhosted.org/packages/14/8d/d44863d358e9dba3bdfb06099bbbeddbac8fb360773ba73250a849af4b01/regex-2020.5.14.tar.gz (696kB)\u001b[0m\n\u001b[34mCollecting tqdm (from nltk->-r requirements.txt (line 3))\u001b[0m\n\u001b[34m Downloading https://files.pythonhosted.org/packages/c9/40/058b12e8ba10e35f89c9b1fdfc2d4c7f8c05947df2d5eb3c7b258019fda0/tqdm-4.46.0-py2.py3-none-any.whl (63kB)\u001b[0m\n\u001b[34mCollecting soupsieve>1.2 (from beautifulsoup4->-r requirements.txt (line 4))\n Downloading https://files.pythonhosted.org/packages/6f/8f/457f4a5390eeae1cc3aeab89deb7724c965be841ffca6cfca9197482e470/soupsieve-2.0.1-py3-none-any.whl\u001b[0m\n\u001b[34mCollecting webencodings (from html5lib->-r requirements.txt (line 5))\n Downloading https://files.pythonhosted.org/packages/f4/24/2a3e3df732393fed8b3ebf2ec078f05546de641fe1b667ee316ec1dcf3b7/webencodings-0.5.1-py2.py3-none-any.whl\u001b[0m\n\u001b[34mRequirement already satisfied, skipping upgrade: six>=1.9 in /usr/local/lib/python3.5/dist-packages (from html5lib->-r requirements.txt (line 5)) (1.11.0)\u001b[0m\n\u001b[34mBuilding wheels for collected packages: nltk, train, regex\n Running setup.py bdist_wheel for nltk: started\n Running setup.py bdist_wheel for nltk: finished with status 'done'\n Stored in directory: /root/.cache/pip/wheels/ae/8c/3f/b1fe0ba04555b08b57ab52ab7f86023639a526d8bc8d384306\u001b[0m\n\u001b[34m Running setup.py bdist_wheel for train: started\n Running setup.py bdist_wheel for train: finished with status 'done'\n Stored in directory: /tmp/pip-ephem-wheel-cache-1ltkub8v/wheels/35/24/16/37574d11bf9bde50616c67372a334f94fa8356bc7164af8ca3\n Running setup.py bdist_wheel for regex: started\u001b[0m\n\u001b[34m Running setup.py bdist_wheel for regex: finished with status 'done'\n Stored in directory: /root/.cache/pip/wheels/ee/3a/5c/1f0ce151d6ddeee56e03e933603e21b5b8dcc727989fde82f5\u001b[0m\n\u001b[34mSuccessfully built nltk train regex\u001b[0m\n\u001b[34mInstalling collected packages: numpy, pytz, pandas, joblib, regex, tqdm, nltk, soupsieve, beautifulsoup4, webencodings, html5lib, train\n Found existing installation: numpy 1.15.4\n Uninstalling numpy-1.15.4:\u001b[0m\n\u001b[34m Successfully uninstalled numpy-1.15.4\u001b[0m\n\u001b[34mSuccessfully installed beautifulsoup4-4.9.1 html5lib-1.0.1 joblib-0.14.1 nltk-3.5 numpy-1.18.4 pandas-0.24.2 pytz-2020.1 regex-2020.5.14 soupsieve-2.0.1 tqdm-4.46.0 train-1.0.0 webencodings-0.5.1\u001b[0m\n\u001b[34mYou are using pip version 18.1, however version 20.2b1 is available.\u001b[0m\n\u001b[34mYou should consider upgrading via the 'pip install --upgrade pip' command.\u001b[0m\n\u001b[34m2020-05-25 14:29:05,183 sagemaker-containers INFO No GPUs detected (normal if no gpus installed)\u001b[0m\n\u001b[34m2020-05-25 14:29:05,197 sagemaker-containers INFO Invoking user script\n\u001b[0m\n\u001b[34mTraining Env:\n\u001b[0m\n\u001b[34m{\n \"num_cpus\": 4,\n \"job_name\": \"sagemaker-pytorch-2020-05-25-14-25-46-965\",\n \"framework_module\": \"sagemaker_pytorch_container.training:main\",\n \"user_entry_point\": \"train.py\",\n \"resource_config\": {\n \"current_host\": \"algo-1\",\n \"network_interface_name\": \"eth0\",\n \"hosts\": [\n \"algo-1\"\n ]\n },\n \"module_dir\": \"s3://sagemaker-us-east-2-935865098957/sagemaker-pytorch-2020-05-25-14-25-46-965/source/sourcedir.tar.gz\",\n \"input_data_config\": {\n \"training\": {\n \"TrainingInputMode\": \"File\",\n \"RecordWrapperType\": \"None\",\n \"S3DistributionType\": \"FullyReplicated\"\n }\n },\n \"hosts\": [\n \"algo-1\"\n ],\n \"log_level\": 20,\n \"model_dir\": \"/opt/ml/model\",\n \"output_intermediate_dir\": \"/opt/ml/output/intermediate\",\n \"input_config_dir\": \"/opt/ml/input/config\",\n \"output_data_dir\": \"/opt/ml/output/data\",\n \"module_name\": \"train\",\n \"channel_input_dirs\": {\n \"training\": \"/opt/ml/input/data/training\"\n },\n \"additional_framework_parameters\": {},\n \"network_interface_name\": \"eth0\",\n \"input_dir\": \"/opt/ml/input\",\n \"current_host\": \"algo-1\",\n \"output_dir\": \"/opt/ml/output\",\n \"hyperparameters\": {\n \"hidden_dim\": 200,\n \"epochs\": 10\n },\n \"num_gpus\": 0\u001b[0m\n\u001b[34m}\n\u001b[0m\n\u001b[34mEnvironment variables:\n\u001b[0m\n\u001b[34mSM_HPS={\"epochs\":10,\"hidden_dim\":200}\u001b[0m\n\u001b[34mSM_HP_HIDDEN_DIM=200\u001b[0m\n\u001b[34mSM_MODEL_DIR=/opt/ml/model\u001b[0m\n\u001b[34mSM_INPUT_DATA_CONFIG={\"training\":{\"RecordWrapperType\":\"None\",\"S3DistributionType\":\"FullyReplicated\",\"TrainingInputMode\":\"File\"}}\u001b[0m\n\u001b[34mSM_INPUT_CONFIG_DIR=/opt/ml/input/config\u001b[0m\n\u001b[34mSM_LOG_LEVEL=20\u001b[0m\n\u001b[34mSM_MODULE_NAME=train\u001b[0m\n\u001b[34mSM_HOSTS=[\"algo-1\"]\u001b[0m\n\u001b[34mSM_NUM_CPUS=4\u001b[0m\n\u001b[34mSM_OUTPUT_DATA_DIR=/opt/ml/output/data\u001b[0m\n\u001b[34mSM_CHANNEL_TRAINING=/opt/ml/input/data/training\u001b[0m\n\u001b[34mSM_TRAINING_ENV={\"additional_framework_parameters\":{},\"channel_input_dirs\":{\"training\":\"/opt/ml/input/data/training\"},\"current_host\":\"algo-1\",\"framework_module\":\"sagemaker_pytorch_container.training:main\",\"hosts\":[\"algo-1\"],\"hyperparameters\":{\"epochs\":10,\"hidden_dim\":200},\"input_config_dir\":\"/opt/ml/input/config\",\"input_data_config\":{\"training\":{\"RecordWrapperType\":\"None\",\"S3DistributionType\":\"FullyReplicated\",\"TrainingInputMode\":\"File\"}},\"input_dir\":\"/opt/ml/input\",\"job_name\":\"sagemaker-pytorch-2020-05-25-14-25-46-965\",\"log_level\":20,\"model_dir\":\"/opt/ml/model\",\"module_dir\":\"s3://sagemaker-us-east-2-935865098957/sagemaker-pytorch-2020-05-25-14-25-46-965/source/sourcedir.tar.gz\",\"module_name\":\"train\",\"network_interface_name\":\"eth0\",\"num_cpus\":4,\"num_gpus\":0,\"output_data_dir\":\"/opt/ml/output/data\",\"output_dir\":\"/opt/ml/output\",\"output_intermediate_dir\":\"/opt/ml/output/intermediate\",\"resource_config\":{\"current_host\":\"algo-1\",\"hosts\":[\"algo-1\"],\"network_interface_name\":\"eth0\"},\"user_entry_point\":\"train.py\"}\u001b[0m\n\u001b[34mSM_FRAMEWORK_MODULE=sagemaker_pytorch_container.training:main\u001b[0m\n\u001b[34mSM_INPUT_DIR=/opt/ml/input\u001b[0m\n\u001b[34mSM_OUTPUT_INTERMEDIATE_DIR=/opt/ml/output/intermediate\u001b[0m\n\u001b[34mSM_RESOURCE_CONFIG={\"current_host\":\"algo-1\",\"hosts\":[\"algo-1\"],\"network_interface_name\":\"eth0\"}\u001b[0m\n\u001b[34mSM_USER_ARGS=[\"--epochs\",\"10\",\"--hidden_dim\",\"200\"]\u001b[0m\n\u001b[34mSM_HP_EPOCHS=10\u001b[0m\n\u001b[34mSM_CHANNELS=[\"training\"]\u001b[0m\n\u001b[34mSM_OUTPUT_DIR=/opt/ml/output\u001b[0m\n\u001b[34mSM_NETWORK_INTERFACE_NAME=eth0\u001b[0m\n\u001b[34mSM_CURRENT_HOST=algo-1\u001b[0m\n\u001b[34mSM_FRAMEWORK_PARAMS={}\u001b[0m\n\u001b[34mSM_USER_ENTRY_POINT=train.py\u001b[0m\n\u001b[34mSM_NUM_GPUS=0\u001b[0m\n\u001b[34mPYTHONPATH=/usr/local/bin:/usr/lib/python35.zip:/usr/lib/python3.5:/usr/lib/python3.5/plat-x86_64-linux-gnu:/usr/lib/python3.5/lib-dynload:/usr/local/lib/python3.5/dist-packages:/usr/lib/python3/dist-packages\u001b[0m\n\u001b[34mSM_MODULE_DIR=s3://sagemaker-us-east-2-935865098957/sagemaker-pytorch-2020-05-25-14-25-46-965/source/sourcedir.tar.gz\n\u001b[0m\n\u001b[34mInvoking script with the following command:\n\u001b[0m\n\u001b[34m/usr/bin/python -m train --epochs 10 --hidden_dim 200\n\n\u001b[0m\n\u001b[34mUsing device cpu.\u001b[0m\n\u001b[34mGet train data loader.\u001b[0m\n\u001b[34mModel loaded with embedding_dim 32, hidden_dim 200, vocab_size 5000.\u001b[0m\n"
]
],
[
[
"## Step 5: Testing the model\n\nAs mentioned at the top of this notebook, we will be testing this model by first deploying it and then sending the testing data to the deployed endpoint. We will do this so that we can make sure that the deployed model is working correctly.\n\n## Step 6: Deploy the model for testing\n\nNow that we have trained our model, we would like to test it to see how it performs. Currently our model takes input of the form `review_length, review[500]` where `review[500]` is a sequence of `500` integers which describe the words present in the review, encoded using `word_dict`. Fortunately for us, SageMaker provides built-in inference code for models with simple inputs such as this.\n\nThere is one thing that we need to provide, however, and that is a function which loads the saved model. This function must be called `model_fn()` and takes as its only parameter a path to the directory where the model artifacts are stored. This function must also be present in the python file which we specified as the entry point. In our case the model loading function has been provided and so no changes need to be made.\n\n**NOTE**: When the built-in inference code is run it must import the `model_fn()` method from the `train.py` file. This is why the training code is wrapped in a main guard ( ie, `if __name__ == '__main__':` )\n\nSince we don't need to change anything in the code that was uploaded during training, we can simply deploy the current model as-is.\n\n**NOTE:** When deploying a model you are asking SageMaker to launch an compute instance that will wait for data to be sent to it. As a result, this compute instance will continue to run until *you* shut it down. This is important to know since the cost of a deployed endpoint depends on how long it has been running for.\n\nIn other words **If you are no longer using a deployed endpoint, shut it down!**\n\n**TODO:** Deploy the trained model.",
"_____no_output_____"
]
],
[
[
"# TODO: Deploy the trained model\nestimator_predictor = estimator.deploy(initial_instance_count=1, instance_type='ml.m4.xlarge')",
"Using already existing model: sagemaker-pytorch-2020-05-25-14-25-46-965\n"
]
],
[
[
"## Step 7 - Use the model for testing\n\nOnce deployed, we can read in the test data and send it off to our deployed model to get some results. Once we collect all of the results we can determine how accurate our model is.",
"_____no_output_____"
]
],
[
[
"test_X = pd.concat([pd.DataFrame(test_X_len), pd.DataFrame(test_X)], axis=1)",
"_____no_output_____"
],
[
"# We split the data into chunks and send each chunk seperately, accumulating the results.\n\ndef predict(data, rows=512):\n split_array = np.array_split(data, int(data.shape[0] / float(rows) + 1))\n predictions = np.array([])\n for array in split_array:\n predictions = np.append(predictions, estimator_predictor.predict(array))\n \n return predictions",
"_____no_output_____"
],
[
"predictions = predict(test_X.values)\npredictions = [round(num) for num in predictions]",
"_____no_output_____"
],
[
"from sklearn.metrics import accuracy_score\naccuracy_score(test_y, predictions)",
"_____no_output_____"
]
],
[
[
"**Question:** How does this model compare to the XGBoost model you created earlier? Why might these two models perform differently on this dataset? Which do *you* think is better for sentiment analysis?",
"_____no_output_____"
],
[
"**Answer:** With the XGBoost model I obtained an accuracy of $\\approx 0.82$, while with the RNN the accuracy is around 0.85, which suggest that the RNN performance is better than XGBoost. Well, in my opinion this difference could be explained because two main factors:\n1. The model itself: On one hand, XGBoost is a boosting tree approach that uses the information about the residuals to assign weights to the wrongly classified instances. These kinds of approaches are good to reduce variance (like any other ensemble method) but don't have memory. On the other hand, RNN has a memory component that makes predictions more suitable in these kinds of scenarios. Specifically, the LSTM cell included in the architecture of the current RNN is widely known for its great performance in these kinds of NLP problems.\n2. The data preparation: For the XGBoost model, we use a Bag of Words approach to count the number of times that words in our dictionary occur in a given review, while in the case of RNN we assign a unique tag (integer) to each word in the dictionary and creates the input vectors using these tags. The difference here remains in that with the first alternative we only consider the frequency information, while with the second we merge the frequency and the spatial (place, moment, location) information in a single input vector. This subtle difference provides, in my opinion, room to the RNN for \"understand\" the relationship between the variables. \n\nIn my opinion, the better option for sentimental analysis is RNN. With this model, we are plenty of options to test differents architectures, besides the \"memory\" factor, which has vital importance in these problems.",
"_____no_output_____"
],
[
"### (TODO) More testing\n\nWe now have a trained model which has been deployed and which we can send processed reviews to and which returns the predicted sentiment. However, ultimately we would like to be able to send our model an unprocessed review. That is, we would like to send the review itself as a string. For example, suppose we wish to send the following review to our model.",
"_____no_output_____"
]
],
[
[
"test_review = 'The simplest pleasures in life are the best, and this film is one of them. Combining a rather basic storyline of love and adventure this movie transcends the usual weekend fair with wit and unmitigated charm.'",
"_____no_output_____"
]
],
[
[
"The question we now need to answer is, how do we send this review to our model?\n\nRecall in the first section of this notebook we did a bunch of data processing to the IMDb dataset. In particular, we did two specific things to the provided reviews.\n - Removed any html tags and stemmed the input\n - Encoded the review as a sequence of integers using `word_dict`\n \nIn order process the review we will need to repeat these two steps.\n\n**TODO**: Using the `review_to_words` and `convert_and_pad` methods from section one, convert `test_review` into a numpy array `test_data` suitable to send to our model. Remember that our model expects input of the form `review_length, review[500]`.",
"_____no_output_____"
]
],
[
[
"# TODO: Convert test_review into a form usable by the model and save the results in test_data\nsentence_X, sentence_X_len = convert_and_pad(word_dict, review_to_words(test_review))\ntest_data = np.array([sentence_X_len] + sentence_X,ndmin=2)",
"_____no_output_____"
]
],
[
[
"Now that we have processed the review, we can send the resulting array to our model to predict the sentiment of the review.",
"_____no_output_____"
]
],
[
[
"estimator_predictor.predict(test_data)",
"_____no_output_____"
]
],
[
[
"Since the return value of our model is close to `1`, we can be certain that the review we submitted is positive.",
"_____no_output_____"
],
[
"### Delete the endpoint\n\nOf course, just like in the XGBoost notebook, once we've deployed an endpoint it continues to run until we tell it to shut down. Since we are done using our endpoint for now, we can delete it.",
"_____no_output_____"
]
],
[
[
"estimator_predictor.delete_endpoint()",
"_____no_output_____"
]
],
[
[
"## Step 6 (again) - Deploy the model for the web app\n\nNow that we know that our model is working, it's time to create some custom inference code so that we can send the model a review which has not been processed and have it determine the sentiment of the review.\n\nAs we saw above, by default the estimator which we created, when deployed, will use the entry script and directory which we provided when creating the model. However, since we now wish to accept a string as input and our model expects a processed review, we need to write some custom inference code.\n\nWe will store the code that we write in the `serve` directory. Provided in this directory is the `model.py` file that we used to construct our model, a `utils.py` file which contains the `review_to_words` and `convert_and_pad` pre-processing functions which we used during the initial data processing, and `predict.py`, the file which will contain our custom inference code. Note also that `requirements.txt` is present which will tell SageMaker what Python libraries are required by our custom inference code.\n\nWhen deploying a PyTorch model in SageMaker, you are expected to provide four functions which the SageMaker inference container will use.\n - `model_fn`: This function is the same function that we used in the training script and it tells SageMaker how to load our model.\n - `input_fn`: This function receives the raw serialized input that has been sent to the model's endpoint and its job is to de-serialize and make the input available for the inference code.\n - `output_fn`: This function takes the output of the inference code and its job is to serialize this output and return it to the caller of the model's endpoint.\n - `predict_fn`: The heart of the inference script, this is where the actual prediction is done and is the function which you will need to complete.\n\nFor the simple website that we are constructing during this project, the `input_fn` and `output_fn` methods are relatively straightforward. We only require being able to accept a string as input and we expect to return a single value as output. You might imagine though that in a more complex application the input or output may be image data or some other binary data which would require some effort to serialize.\n\n### (TODO) Writing inference code\n\nBefore writing our custom inference code, we will begin by taking a look at the code which has been provided.",
"_____no_output_____"
]
],
[
[
"!pygmentize serve/predict.py",
"\u001b[34mimport\u001b[39;49;00m \u001b[04m\u001b[36margparse\u001b[39;49;00m\r\n\u001b[34mimport\u001b[39;49;00m \u001b[04m\u001b[36mjson\u001b[39;49;00m\r\n\u001b[34mimport\u001b[39;49;00m \u001b[04m\u001b[36mos\u001b[39;49;00m\r\n\u001b[34mimport\u001b[39;49;00m \u001b[04m\u001b[36mpickle\u001b[39;49;00m\r\n\u001b[34mimport\u001b[39;49;00m \u001b[04m\u001b[36msys\u001b[39;49;00m\r\n\u001b[34mimport\u001b[39;49;00m \u001b[04m\u001b[36msagemaker_containers\u001b[39;49;00m\r\n\u001b[34mimport\u001b[39;49;00m \u001b[04m\u001b[36mpandas\u001b[39;49;00m \u001b[34mas\u001b[39;49;00m \u001b[04m\u001b[36mpd\u001b[39;49;00m\r\n\u001b[34mimport\u001b[39;49;00m \u001b[04m\u001b[36mnumpy\u001b[39;49;00m \u001b[34mas\u001b[39;49;00m \u001b[04m\u001b[36mnp\u001b[39;49;00m\r\n\u001b[34mimport\u001b[39;49;00m \u001b[04m\u001b[36mtorch\u001b[39;49;00m\r\n\u001b[34mimport\u001b[39;49;00m \u001b[04m\u001b[36mtorch.nn\u001b[39;49;00m \u001b[34mas\u001b[39;49;00m \u001b[04m\u001b[36mnn\u001b[39;49;00m\r\n\u001b[34mimport\u001b[39;49;00m \u001b[04m\u001b[36mtorch.optim\u001b[39;49;00m \u001b[34mas\u001b[39;49;00m \u001b[04m\u001b[36moptim\u001b[39;49;00m\r\n\u001b[34mimport\u001b[39;49;00m \u001b[04m\u001b[36mtorch.utils.data\u001b[39;49;00m\r\n\r\n\u001b[34mfrom\u001b[39;49;00m \u001b[04m\u001b[36mmodel\u001b[39;49;00m \u001b[34mimport\u001b[39;49;00m LSTMClassifier\r\n\r\n\u001b[34mfrom\u001b[39;49;00m \u001b[04m\u001b[36mutils\u001b[39;49;00m \u001b[34mimport\u001b[39;49;00m review_to_words, convert_and_pad\r\n\r\n\u001b[34mdef\u001b[39;49;00m \u001b[32mmodel_fn\u001b[39;49;00m(model_dir):\r\n \u001b[33m\"\"\"Load the PyTorch model from the `model_dir` directory.\"\"\"\u001b[39;49;00m\r\n \u001b[34mprint\u001b[39;49;00m(\u001b[33m\"\u001b[39;49;00m\u001b[33mLoading model.\u001b[39;49;00m\u001b[33m\"\u001b[39;49;00m)\r\n\r\n \u001b[37m# First, load the parameters used to create the model.\u001b[39;49;00m\r\n model_info = {}\r\n model_info_path = os.path.join(model_dir, \u001b[33m'\u001b[39;49;00m\u001b[33mmodel_info.pth\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m)\r\n \u001b[34mwith\u001b[39;49;00m \u001b[36mopen\u001b[39;49;00m(model_info_path, \u001b[33m'\u001b[39;49;00m\u001b[33mrb\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m) \u001b[34mas\u001b[39;49;00m f:\r\n model_info = torch.load(f)\r\n\r\n \u001b[34mprint\u001b[39;49;00m(\u001b[33m\"\u001b[39;49;00m\u001b[33mmodel_info: {}\u001b[39;49;00m\u001b[33m\"\u001b[39;49;00m.format(model_info))\r\n\r\n \u001b[37m# Determine the device and construct the model.\u001b[39;49;00m\r\n device = torch.device(\u001b[33m\"\u001b[39;49;00m\u001b[33mcuda\u001b[39;49;00m\u001b[33m\"\u001b[39;49;00m \u001b[34mif\u001b[39;49;00m torch.cuda.is_available() \u001b[34melse\u001b[39;49;00m \u001b[33m\"\u001b[39;49;00m\u001b[33mcpu\u001b[39;49;00m\u001b[33m\"\u001b[39;49;00m)\r\n model = LSTMClassifier(model_info[\u001b[33m'\u001b[39;49;00m\u001b[33membedding_dim\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m], model_info[\u001b[33m'\u001b[39;49;00m\u001b[33mhidden_dim\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m], model_info[\u001b[33m'\u001b[39;49;00m\u001b[33mvocab_size\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m])\r\n\r\n \u001b[37m# Load the store model parameters.\u001b[39;49;00m\r\n model_path = os.path.join(model_dir, \u001b[33m'\u001b[39;49;00m\u001b[33mmodel.pth\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m)\r\n \u001b[34mwith\u001b[39;49;00m \u001b[36mopen\u001b[39;49;00m(model_path, \u001b[33m'\u001b[39;49;00m\u001b[33mrb\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m) \u001b[34mas\u001b[39;49;00m f:\r\n model.load_state_dict(torch.load(f))\r\n\r\n \u001b[37m# Load the saved word_dict.\u001b[39;49;00m\r\n word_dict_path = os.path.join(model_dir, \u001b[33m'\u001b[39;49;00m\u001b[33mword_dict.pkl\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m)\r\n \u001b[34mwith\u001b[39;49;00m \u001b[36mopen\u001b[39;49;00m(word_dict_path, \u001b[33m'\u001b[39;49;00m\u001b[33mrb\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m) \u001b[34mas\u001b[39;49;00m f:\r\n model.word_dict = pickle.load(f)\r\n\r\n model.to(device).eval()\r\n\r\n \u001b[34mprint\u001b[39;49;00m(\u001b[33m\"\u001b[39;49;00m\u001b[33mDone loading model.\u001b[39;49;00m\u001b[33m\"\u001b[39;49;00m)\r\n \u001b[34mreturn\u001b[39;49;00m model\r\n\r\n\u001b[34mdef\u001b[39;49;00m \u001b[32minput_fn\u001b[39;49;00m(serialized_input_data, content_type):\r\n \u001b[34mprint\u001b[39;49;00m(\u001b[33m'\u001b[39;49;00m\u001b[33mDeserializing the input data.\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m)\r\n \u001b[34mif\u001b[39;49;00m content_type == \u001b[33m'\u001b[39;49;00m\u001b[33mtext/plain\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m:\r\n data = serialized_input_data.decode(\u001b[33m'\u001b[39;49;00m\u001b[33mutf-8\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m)\r\n \u001b[34mreturn\u001b[39;49;00m data\r\n \u001b[34mraise\u001b[39;49;00m \u001b[36mException\u001b[39;49;00m(\u001b[33m'\u001b[39;49;00m\u001b[33mRequested unsupported ContentType in content_type: \u001b[39;49;00m\u001b[33m'\u001b[39;49;00m + content_type)\r\n\r\n\u001b[34mdef\u001b[39;49;00m \u001b[32moutput_fn\u001b[39;49;00m(prediction_output, accept):\r\n \u001b[34mprint\u001b[39;49;00m(\u001b[33m'\u001b[39;49;00m\u001b[33mSerializing the generated output.\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m)\r\n \u001b[34mreturn\u001b[39;49;00m \u001b[36mstr\u001b[39;49;00m(prediction_output)\r\n\r\n\u001b[34mdef\u001b[39;49;00m \u001b[32mpredict_fn\u001b[39;49;00m(input_data, model):\r\n \u001b[34mprint\u001b[39;49;00m(\u001b[33m'\u001b[39;49;00m\u001b[33mInferring sentiment of input data.\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m)\r\n\r\n device = torch.device(\u001b[33m\"\u001b[39;49;00m\u001b[33mcuda\u001b[39;49;00m\u001b[33m\"\u001b[39;49;00m \u001b[34mif\u001b[39;49;00m torch.cuda.is_available() \u001b[34melse\u001b[39;49;00m \u001b[33m\"\u001b[39;49;00m\u001b[33mcpu\u001b[39;49;00m\u001b[33m\"\u001b[39;49;00m)\r\n \r\n \u001b[34mif\u001b[39;49;00m model.word_dict \u001b[35mis\u001b[39;49;00m \u001b[36mNone\u001b[39;49;00m:\r\n \u001b[34mraise\u001b[39;49;00m \u001b[36mException\u001b[39;49;00m(\u001b[33m'\u001b[39;49;00m\u001b[33mModel has not been loaded properly, no word_dict.\u001b[39;49;00m\u001b[33m'\u001b[39;49;00m)\r\n \r\n \u001b[37m# TODO: Process input_data so that it is ready to be sent to our model.\u001b[39;49;00m\r\n \u001b[37m# You should produce two variables:\u001b[39;49;00m\r\n \u001b[37m# data_X - A sequence of length 500 which represents the converted review\u001b[39;49;00m\r\n \u001b[37m# data_len - The length of the review\u001b[39;49;00m\r\n data_X, data_len = convert_and_pad(model.word_dict, review_to_words(input_data))\r\n \u001b[37m#data_X = None\u001b[39;49;00m\r\n \u001b[37m#data_len = None\u001b[39;49;00m\r\n\r\n \u001b[37m# Using data_X and data_len we construct an appropriate input tensor. Remember\u001b[39;49;00m\r\n \u001b[37m# that our model expects input data of the form 'len, review[500]'.\u001b[39;49;00m\r\n data_pack = np.hstack((data_len, data_X))\r\n data_pack = data_pack.reshape(\u001b[34m1\u001b[39;49;00m, -\u001b[34m1\u001b[39;49;00m)\r\n \r\n data = torch.from_numpy(data_pack)\r\n data = data.to(device)\r\n\r\n \u001b[37m# Make sure to put the model into evaluation mode\u001b[39;49;00m\r\n model.eval()\r\n\r\n \u001b[37m# TODO: Compute the result of applying the model to the input data. The variable `result` should\u001b[39;49;00m\r\n \u001b[37m# be a numpy array which contains a single integer which is either 1 or 0\u001b[39;49;00m\r\n \r\n predictions=model.predict(data)\r\n\r\n result = np.array(\u001b[36mint\u001b[39;49;00m(predictions>\u001b[34m0.5\u001b[39;49;00m))\r\n\r\n \u001b[34mreturn\u001b[39;49;00m result\r\n"
]
],
[
[
"As mentioned earlier, the `model_fn` method is the same as the one provided in the training code and the `input_fn` and `output_fn` methods are very simple and your task will be to complete the `predict_fn` method. Make sure that you save the completed file as `predict.py` in the `serve` directory.\n\n**TODO**: Complete the `predict_fn()` method in the `serve/predict.py` file.",
"_____no_output_____"
],
[
"### Deploying the model\n\nNow that the custom inference code has been written, we will create and deploy our model. To begin with, we need to construct a new PyTorchModel object which points to the model artifacts created during training and also points to the inference code that we wish to use. Then we can call the deploy method to launch the deployment container.\n\n**NOTE**: The default behaviour for a deployed PyTorch model is to assume that any input passed to the predictor is a `numpy` array. In our case we want to send a string so we need to construct a simple wrapper around the `RealTimePredictor` class to accomodate simple strings. In a more complicated situation you may want to provide a serialization object, for example if you wanted to sent image data.",
"_____no_output_____"
]
],
[
[
"from sagemaker.predictor import RealTimePredictor\nfrom sagemaker.pytorch import PyTorchModel\n\nclass StringPredictor(RealTimePredictor):\n def __init__(self, endpoint_name, sagemaker_session):\n super(StringPredictor, self).__init__(endpoint_name, sagemaker_session, content_type='text/plain')\n\nmodel = PyTorchModel(model_data=estimator.model_data,\n role = role,\n framework_version='0.4.0',\n entry_point='predict.py',\n source_dir='serve',\n predictor_cls=StringPredictor)\n\npredictor = model.deploy(initial_instance_count=1, instance_type='ml.m4.xlarge')",
"---------------!"
]
],
[
[
"### Testing the model\n\nNow that we have deployed our model with the custom inference code, we should test to see if everything is working. Here we test our model by loading the first `250` positive and negative reviews and send them to the endpoint, then collect the results. The reason for only sending some of the data is that the amount of time it takes for our model to process the input and then perform inference is quite long and so testing the entire data set would be prohibitive.",
"_____no_output_____"
]
],
[
[
"import glob\n\ndef test_reviews(data_dir='../data/aclImdb', stop=250):\n \n results = []\n ground = []\n \n # We make sure to test both positive and negative reviews \n for sentiment in ['pos', 'neg']:\n \n path = os.path.join(data_dir, 'test', sentiment, '*.txt')\n files = glob.glob(path)\n \n files_read = 0\n \n print('Starting ', sentiment, ' files')\n \n # Iterate through the files and send them to the predictor\n for f in files:\n with open(f) as review:\n # First, we store the ground truth (was the review positive or negative)\n if sentiment == 'pos':\n ground.append(1)\n else:\n ground.append(0)\n # Read in the review and convert to 'utf-8' for transmission via HTTP\n review_input = review.read().encode('utf-8')\n # Send the review to the predictor and store the results\n results.append(int(predictor.predict(review_input)))\n \n # Sending reviews to our endpoint one at a time takes a while so we\n # only send a small number of reviews\n files_read += 1\n if files_read == stop:\n break\n \n return ground, results",
"_____no_output_____"
],
[
"ground, results = test_reviews()",
"Starting pos files\nStarting neg files\n"
],
[
"from sklearn.metrics import accuracy_score\naccuracy_score(ground, results)",
"_____no_output_____"
]
],
[
[
"As an additional test, we can try sending the `test_review` that we looked at earlier.",
"_____no_output_____"
]
],
[
[
"predictor.predict(test_review)",
"_____no_output_____"
]
],
[
[
"Now that we know our endpoint is working as expected, we can set up the web page that will interact with it. If you don't have time to finish the project now, make sure to skip down to the end of this notebook and shut down your endpoint. You can deploy it again when you come back.",
"_____no_output_____"
],
[
"## Step 7 (again): Use the model for the web app\n\n> **TODO:** This entire section and the next contain tasks for you to complete, mostly using the AWS console.\n\nSo far we have been accessing our model endpoint by constructing a predictor object which uses the endpoint and then just using the predictor object to perform inference. What if we wanted to create a web app which accessed our model? The way things are set up currently makes that not possible since in order to access a SageMaker endpoint the app would first have to authenticate with AWS using an IAM role which included access to SageMaker endpoints. However, there is an easier way! We just need to use some additional AWS services.\n\n<img src=\"Web App Diagram.svg\">\n\nThe diagram above gives an overview of how the various services will work together. On the far right is the model which we trained above and which is deployed using SageMaker. On the far left is our web app that collects a user's movie review, sends it off and expects a positive or negative sentiment in return.\n\nIn the middle is where some of the magic happens. We will construct a Lambda function, which you can think of as a straightforward Python function that can be executed whenever a specified event occurs. We will give this function permission to send and recieve data from a SageMaker endpoint.\n\nLastly, the method we will use to execute the Lambda function is a new endpoint that we will create using API Gateway. This endpoint will be a url that listens for data to be sent to it. Once it gets some data it will pass that data on to the Lambda function and then return whatever the Lambda function returns. Essentially it will act as an interface that lets our web app communicate with the Lambda function.\n\n### Setting up a Lambda function\n\nThe first thing we are going to do is set up a Lambda function. This Lambda function will be executed whenever our public API has data sent to it. When it is executed it will receive the data, perform any sort of processing that is required, send the data (the review) to the SageMaker endpoint we've created and then return the result.\n\n#### Part A: Create an IAM Role for the Lambda function\n\nSince we want the Lambda function to call a SageMaker endpoint, we need to make sure that it has permission to do so. To do this, we will construct a role that we can later give the Lambda function.\n\nUsing the AWS Console, navigate to the **IAM** page and click on **Roles**. Then, click on **Create role**. Make sure that the **AWS service** is the type of trusted entity selected and choose **Lambda** as the service that will use this role, then click **Next: Permissions**.\n\nIn the search box type `sagemaker` and select the check box next to the **AmazonSageMakerFullAccess** policy. Then, click on **Next: Review**.\n\nLastly, give this role a name. Make sure you use a name that you will remember later on, for example `LambdaSageMakerRole`. Then, click on **Create role**.\n\n#### Part B: Create a Lambda function\n\nNow it is time to actually create the Lambda function.\n\nUsing the AWS Console, navigate to the AWS Lambda page and click on **Create a function**. When you get to the next page, make sure that **Author from scratch** is selected. Now, name your Lambda function, using a name that you will remember later on, for example `sentiment_analysis_func`. Make sure that the **Python 3.6** runtime is selected and then choose the role that you created in the previous part. Then, click on **Create Function**.\n\nOn the next page you will see some information about the Lambda function you've just created. If you scroll down you should see an editor in which you can write the code that will be executed when your Lambda function is triggered. In our example, we will use the code below. \n\n```python\n# We need to use the low-level library to interact with SageMaker since the SageMaker API\n# is not available natively through Lambda.\nimport boto3\n\ndef lambda_handler(event, context):\n\n # The SageMaker runtime is what allows us to invoke the endpoint that we've created.\n runtime = boto3.Session().client('sagemaker-runtime')\n\n # Now we use the SageMaker runtime to invoke our endpoint, sending the review we were given\n response = runtime.invoke_endpoint(EndpointName = '**ENDPOINT NAME HERE**', # The name of the endpoint we created\n ContentType = 'text/plain', # The data format that is expected\n Body = event['body']) # The actual review\n\n # The response is an HTTP response whose body contains the result of our inference\n result = response['Body'].read().decode('utf-8')\n\n return {\n 'statusCode' : 200,\n 'headers' : { 'Content-Type' : 'text/plain', 'Access-Control-Allow-Origin' : '*' },\n 'body' : result\n }\n```\n\nOnce you have copy and pasted the code above into the Lambda code editor, replace the `**ENDPOINT NAME HERE**` portion with the name of the endpoint that we deployed earlier. You can determine the name of the endpoint using the code cell below.",
"_____no_output_____"
]
],
[
[
"predictor.endpoint",
"_____no_output_____"
]
],
[
[
"Once you have added the endpoint name to the Lambda function, click on **Save**. Your Lambda function is now up and running. Next we need to create a way for our web app to execute the Lambda function.\n\n### Setting up API Gateway\n\nNow that our Lambda function is set up, it is time to create a new API using API Gateway that will trigger the Lambda function we have just created.\n\nUsing AWS Console, navigate to **Amazon API Gateway** and then click on **Get started**.\n\nOn the next page, make sure that **New API** is selected and give the new api a name, for example, `sentiment_analysis_api`. Then, click on **Create API**.\n\nNow we have created an API, however it doesn't currently do anything. What we want it to do is to trigger the Lambda function that we created earlier.\n\nSelect the **Actions** dropdown menu and click **Create Method**. A new blank method will be created, select its dropdown menu and select **POST**, then click on the check mark beside it.\n\nFor the integration point, make sure that **Lambda Function** is selected and click on the **Use Lambda Proxy integration**. This option makes sure that the data that is sent to the API is then sent directly to the Lambda function with no processing. It also means that the return value must be a proper response object as it will also not be processed by API Gateway.\n\nType the name of the Lambda function you created earlier into the **Lambda Function** text entry box and then click on **Save**. Click on **OK** in the pop-up box that then appears, giving permission to API Gateway to invoke the Lambda function you created.\n\nThe last step in creating the API Gateway is to select the **Actions** dropdown and click on **Deploy API**. You will need to create a new Deployment stage and name it anything you like, for example `prod`.\n\nYou have now successfully set up a public API to access your SageMaker model. Make sure to copy or write down the URL provided to invoke your newly created public API as this will be needed in the next step. This URL can be found at the top of the page, highlighted in blue next to the text **Invoke URL**.",
"_____no_output_____"
],
[
"## Step 4: Deploying our web app\n\nNow that we have a publicly available API, we can start using it in a web app. For our purposes, we have provided a simple static html file which can make use of the public api you created earlier.\n\nIn the `website` folder there should be a file called `index.html`. Download the file to your computer and open that file up in a text editor of your choice. There should be a line which contains **\\*\\*REPLACE WITH PUBLIC API URL\\*\\***. Replace this string with the url that you wrote down in the last step and then save the file.\n\nNow, if you open `index.html` on your local computer, your browser will behave as a local web server and you can use the provided site to interact with your SageMaker model.\n\nIf you'd like to go further, you can host this html file anywhere you'd like, for example using github or hosting a static site on Amazon's S3. Once you have done this you can share the link with anyone you'd like and have them play with it too!\n\n> **Important Note** In order for the web app to communicate with the SageMaker endpoint, the endpoint has to actually be deployed and running. This means that you are paying for it. Make sure that the endpoint is running when you want to use the web app but that you shut it down when you don't need it, otherwise you will end up with a surprisingly large AWS bill.\n\n**TODO:** Make sure that you include the edited `index.html` file in your project submission.",
"_____no_output_____"
],
[
"Now that your web app is working, trying playing around with it and see how well it works.\n\n**Question**: Give an example of a review that you entered into your web app. What was the predicted sentiment of your example review?",
"_____no_output_____"
],
[
"**Answer:** I tried multiple examples of reviews. I am include only one of the positive and another negative.\n* This movie was very bad, I hate it! --> Your review was NEGATIVE!\n* This movie is adorable! I want to see it again. --> Your review was POSITIVE!",
"_____no_output_____"
],
[
"### Delete the endpoint\n\nRemember to always shut down your endpoint if you are no longer using it. You are charged for the length of time that the endpoint is running so if you forget and leave it on you could end up with an unexpectedly large bill.",
"_____no_output_____"
]
],
[
[
"predictor.delete_endpoint()",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code"
]
] |
cbd82a2faa42f3981bde8d6c980f6267509851a7
| 17,255 |
ipynb
|
Jupyter Notebook
|
Ro_Davies_LS_DS_431_RNN_and_LSTM_Assignment.ipynb
|
120Davies/DS-Unit-4-Sprint-3-Deep-Learning
|
0b58d1d7982208886071fd2dbe43a5462eba4678
|
[
"MIT"
] | null | null | null |
Ro_Davies_LS_DS_431_RNN_and_LSTM_Assignment.ipynb
|
120Davies/DS-Unit-4-Sprint-3-Deep-Learning
|
0b58d1d7982208886071fd2dbe43a5462eba4678
|
[
"MIT"
] | null | null | null |
Ro_Davies_LS_DS_431_RNN_and_LSTM_Assignment.ipynb
|
120Davies/DS-Unit-4-Sprint-3-Deep-Learning
|
0b58d1d7982208886071fd2dbe43a5462eba4678
|
[
"MIT"
] | null | null | null | 43.463476 | 717 | 0.489887 |
[
[
[
"<a href=\"https://colab.research.google.com/github/120Davies/DS-Unit-4-Sprint-3-Deep-Learning/blob/master/Ro_Davies_LS_DS_431_RNN_and_LSTM_Assignment.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>",
"_____no_output_____"
],
[
"<img align=\"left\" src=\"https://lever-client-logos.s3.amazonaws.com/864372b1-534c-480e-acd5-9711f850815c-1524247202159.png\" width=200>\n<br></br>\n<br></br>\n\n## *Data Science Unit 4 Sprint 3 Assignment 1*\n\n# Recurrent Neural Networks and Long Short Term Memory (LSTM)\n\n\n\nIt is said that [infinite monkeys typing for an infinite amount of time](https://en.wikipedia.org/wiki/Infinite_monkey_theorem) will eventually type, among other things, the complete works of Wiliam Shakespeare. Let's see if we can get there a bit faster, with the power of Recurrent Neural Networks and LSTM.\n\nThis text file contains the complete works of Shakespeare: https://www.gutenberg.org/files/100/100-0.txt\n\nUse it as training data for an RNN - you can keep it simple and train character level, and that is suggested as an initial approach.\n\nThen, use that trained RNN to generate Shakespearean-ish text. Your goal - a function that can take, as an argument, the size of text (e.g. number of characters or lines) to generate, and returns generated text of that size.\n\nNote - Shakespeare wrote an awful lot. It's OK, especially initially, to sample/use smaller data and parameters, so you can have a tighter feedback loop when you're trying to get things running. Then, once you've got a proof of concept - start pushing it more!",
"_____no_output_____"
]
],
[
[
"# TODO - Words, words, mere words, no matter from the heart.\nimport numpy as np\n\nwith open('./shakespear/100-0.txt', 'r') as f:\n text = f.read()",
"_____no_output_____"
],
[
"text = \" \".join(text[:250000])\nchars = list(set(text))\n\nnum_chars = len(chars)\ntxt_size = len(text)\nprint('Number of unique characters:', num_chars)\nprint('Total characters in text:', txt_size)",
"Number of unique characters: 78\nTotal characters in text: 499999\n"
],
[
"char_to_int = dict((c,i) for i,c in enumerate(chars))\nint_to_char = dict((i,c) for i,c in enumerate(chars))\n\nprint(char_to_int)\nprint('-'*50)\nprint(int_to_char)\n\ninteger_encoded = [char_to_int[i] for i in text]\n\nprint(len(integer_encoded))",
"{'J': 0, 'A': 1, '!': 2, '\\ufeff': 3, 't': 4, 'R': 5, 'X': 6, '1': 7, ';': 8, 'q': 9, 'x': 10, '#': 11, 'v': 12, '?': 13, 'D': 14, '2': 15, 'p': 16, 'u': 17, 'a': 18, 'd': 19, 'I': 20, 'L': 21, 'H': 22, '(': 23, 'i': 24, '9': 25, 'P': 26, 'e': 27, '5': 28, 'h': 29, '3': 30, 'M': 31, 'C': 32, 's': 33, 'B': 34, '7': 35, 'n': 36, 'f': 37, 'T': 38, '\\n': 39, '.': 40, '6': 41, 'w': 42, 'b': 43, 'm': 44, ')': 45, 'r': 46, '’': 47, 'l': 48, 'S': 49, 'U': 50, 'c': 51, '[': 52, 'F': 53, ']': 54, 'o': 55, '8': 56, '0': 57, 'K': 58, ':': 59, 'V': 60, '4': 61, 'y': 62, 'Y': 63, ',': 64, 'E': 65, 'N': 66, 'j': 67, 'G': 68, 'O': 69, '*': 70, ' ': 71, 'z': 72, 'k': 73, 'W': 74, 'g': 75, '‘': 76, '-': 77}\n--------------------------------------------------\n{0: 'J', 1: 'A', 2: '!', 3: '\\ufeff', 4: 't', 5: 'R', 6: 'X', 7: '1', 8: ';', 9: 'q', 10: 'x', 11: '#', 12: 'v', 13: '?', 14: 'D', 15: '2', 16: 'p', 17: 'u', 18: 'a', 19: 'd', 20: 'I', 21: 'L', 22: 'H', 23: '(', 24: 'i', 25: '9', 26: 'P', 27: 'e', 28: '5', 29: 'h', 30: '3', 31: 'M', 32: 'C', 33: 's', 34: 'B', 35: '7', 36: 'n', 37: 'f', 38: 'T', 39: '\\n', 40: '.', 41: '6', 42: 'w', 43: 'b', 44: 'm', 45: ')', 46: 'r', 47: '’', 48: 'l', 49: 'S', 50: 'U', 51: 'c', 52: '[', 53: 'F', 54: ']', 55: 'o', 56: '8', 57: '0', 58: 'K', 59: ':', 60: 'V', 61: '4', 62: 'y', 63: 'Y', 64: ',', 65: 'E', 66: 'N', 67: 'j', 68: 'G', 69: 'O', 70: '*', 71: ' ', 72: 'z', 73: 'k', 74: 'W', 75: 'g', 76: '‘', 77: '-'}\n499999\n"
],
[
"# hyperparameters\niteration = 10\nsequence_length = 40\nbatch_size = round((txt_size /sequence_length)+0.5) # = math.ceil\nhidden_size = 500 # size of hidden layer of neurons. \nlearning_rate = 1e-1\n\n\n# model parameters\n\nW_xh = np.random.randn(hidden_size, num_chars)*0.01 # weight input -> hidden. \nW_hh = np.random.randn(hidden_size, hidden_size)*0.01 # weight hidden -> hidden\nW_hy = np.random.randn(num_chars, hidden_size)*0.01 # weight hidden -> output\n\nb_h = np.zeros((hidden_size, 1)) # hidden bias\nb_y = np.zeros((num_chars, 1)) # output bias\n\nh_prev = np.zeros((hidden_size,1)) # h_(t-1)",
"_____no_output_____"
],
[
"def forwardprop(inputs, targets, h_prev):\n \n # Since the RNN receives the sequence, the weights are not updated during one sequence.\n xs, hs, ys, ps = {}, {}, {}, {} # dictionary\n hs[-1] = np.copy(h_prev) # Copy previous hidden state vector to -1 key value.\n loss = 0 # loss initialization\n \n for t in range(len(inputs)): # t is a \"time step\" and is used as a key(dic). \n \n xs[t] = np.zeros((num_chars,1)) \n xs[t][inputs[t]] = 1\n hs[t] = np.tanh(np.dot(W_xh, xs[t]) + np.dot(W_hh, hs[t-1]) + b_h) # hidden state. \n ys[t] = np.dot(W_hy, hs[t]) + b_y # unnormalized log probabilities for next chars\n ps[t] = np.exp(ys[t]) / np.sum(np.exp(ys[t])) # probabilities for next chars. \n \n # Softmax. -> The sum of probabilities is 1 even without the exp() function, but all of the elements are positive through the exp() function.\n loss += -np.log(ps[t][targets[t],0]) # softmax (cross-entropy loss). Efficient and simple code\n\n# y_class = np.zeros((num_chars, 1)) \n# y_class[targets[t]] =1\n# loss += np.sum(y_class*(-np.log(ps[t]))) # softmax (cross-entropy loss) \n\n return loss, ps, hs, xs",
"_____no_output_____"
],
[
"def backprop(ps, inputs, hs, xs, targets):\n\n dWxh, dWhh, dWhy = np.zeros_like(W_xh), np.zeros_like(W_hh), np.zeros_like(W_hy) # make all zero matrices.\n dbh, dby = np.zeros_like(b_h), np.zeros_like(b_y)\n dhnext = np.zeros_like(hs[0]) # (hidden_size,1) \n\n # reversed\n for t in reversed(range(len(inputs))):\n dy = np.copy(ps[t]) # shape (num_chars,1). \"dy\" means \"dloss/dy\"\n dy[targets[t]] -= 1 # backprop into y. After taking the soft max in the input vector, subtract 1 from the value of the element corresponding to the correct label.\n dWhy += np.dot(dy, hs[t].T)\n dby += dy \n dh = np.dot(W_hy.T, dy) + dhnext # backprop into h. \n dhraw = (1 - hs[t] * hs[t]) * dh # backprop through tanh nonlinearity #tanh'(x) = 1-tanh^2(x)\n dbh += dhraw\n dWxh += np.dot(dhraw, xs[t].T)\n dWhh += np.dot(dhraw, hs[t-1].T)\n dhnext = np.dot(W_hh.T, dhraw)\n for dparam in [dWxh, dWhh, dWhy, dbh, dby]: \n np.clip(dparam, -5, 5, out=dparam) # clip to mitigate exploding gradients. \n \n return dWxh, dWhh, dWhy, dbh, dby",
"_____no_output_____"
],
[
"%%time\n\ndata_pointer = 0\n\n# memory variables for Adagrad\nmWxh, mWhh, mWhy = np.zeros_like(W_xh), np.zeros_like(W_hh), np.zeros_like(W_hy)\nmbh, mby = np.zeros_like(b_h), np.zeros_like(b_y) \n\nfor i in range(iteration):\n h_prev = np.zeros((hidden_size,1)) # reset RNN memory\n data_pointer = 0 # go from start of data\n \n for b in range(batch_size):\n \n inputs = [char_to_int[ch] for ch in text[data_pointer:data_pointer+sequence_length]]\n targets = [char_to_int[ch] for ch in text[data_pointer+1:data_pointer+sequence_length+1]] # t+1 \n \n if (data_pointer+sequence_length+1 >= len(text) and b == batch_size-1): # processing of the last part of the input data. \n# targets.append(char_to_int[txt_data[0]]) # When the data doesn't fit, add the first char to the back.\n targets.append(char_to_int[\" \"]) # When the data doesn't fit, add space(\" \") to the back.\n\n\n # forward\n loss, ps, hs, xs = forwardprop(inputs, targets, h_prev)\n# print(loss)\n \n # backward\n dWxh, dWhh, dWhy, dbh, dby = backprop(ps, inputs, hs, xs, targets) \n \n \n # perform parameter update with Adagrad\n for param, dparam, mem in zip([W_xh, W_hh, W_hy, b_h, b_y], \n [dWxh, dWhh, dWhy, dbh, dby], \n [mWxh, mWhh, mWhy, mbh, mby]):\n mem += dparam * dparam # elementwise\n param += -learning_rate * dparam / np.sqrt(mem + 1e-8) # adagrad update \n \n data_pointer += sequence_length # move data pointer\n \n if i % 2 == 0:\n print ('iter %d, loss: %f' % (i, loss)) # print progress",
"iter 0, loss: 15.097292\niter 2, loss: 15.400502\n"
],
[
"def predict(test_char, length):\n x = np.zeros((num_chars, 1)) \n x[char_to_int[test_char]] = 1\n ixes = []\n h = np.zeros((hidden_size,1))\n\n for t in range(length):\n h = np.tanh(np.dot(W_xh, x) + np.dot(W_hh, h) + b_h) \n y = np.dot(W_hy, h) + b_y\n p = np.exp(y) / np.sum(np.exp(y)) \n ix = np.random.choice(range(num_chars), p=p.ravel()) # ravel -> rank0\n # \"ix\" is a list of indexes selected according to the soft max probability.\n x = np.zeros((num_chars, 1)) # init\n x[ix] = 1 \n ixes.append(ix) # list\n txt = test_char + ''.join(int_to_char[i] for i in ixes)\n print ('----\\n %s \\n----' % (txt, ))",
"_____no_output_____"
],
[
"predict('A', 1000)",
"_____no_output_____"
],
[
"",
"_____no_output_____"
]
],
[
[
"# Resources and Stretch Goals",
"_____no_output_____"
],
[
"## Stretch goals:\n- Refine the training and generation of text to be able to ask for different genres/styles of Shakespearean text (e.g. plays versus sonnets)\n- Train a classification model that takes text and returns which work of Shakespeare it is most likely to be from\n- Make it more performant! Many possible routes here - lean on Keras, optimize the code, and/or use more resources (AWS, etc.)\n- Revisit the news example from class, and improve it - use categories or tags to refine the model/generation, or train a news classifier\n- Run on bigger, better data\n\n## Resources:\n- [The Unreasonable Effectiveness of Recurrent Neural Networks](https://karpathy.github.io/2015/05/21/rnn-effectiveness/) - a seminal writeup demonstrating a simple but effective character-level NLP RNN\n- [Simple NumPy implementation of RNN](https://github.com/JY-Yoon/RNN-Implementation-using-NumPy/blob/master/RNN%20Implementation%20using%20NumPy.ipynb) - Python 3 version of the code from \"Unreasonable Effectiveness\"\n- [TensorFlow RNN Tutorial](https://github.com/tensorflow/models/tree/master/tutorials/rnn) - code for training a RNN on the Penn Tree Bank language dataset\n- [4 part tutorial on RNN](http://www.wildml.com/2015/09/recurrent-neural-networks-tutorial-part-1-introduction-to-rnns/) - relates RNN to the vanishing gradient problem, and provides example implementation\n- [RNN training tips and tricks](https://github.com/karpathy/char-rnn#tips-and-tricks) - some rules of thumb for parameterizing and training your RNN",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown"
] |
[
[
"markdown",
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown",
"markdown"
]
] |
cbd82c61c6595f7385c03ad648ad052fbde372dd
| 705,309 |
ipynb
|
Jupyter Notebook
|
Data Description and Analysis.ipynb
|
ShaoniMukherjee/Instacart-Market-Basket-Analysis
|
4691cda5038247780e04ed822682c46079946e16
|
[
"MIT"
] | 18 |
2020-04-15T02:03:21.000Z
|
2022-03-29T10:51:57.000Z
|
Data Description and Analysis.ipynb
|
ShaoniMukherjee/Instacart-Market-Basket-Analysis
|
4691cda5038247780e04ed822682c46079946e16
|
[
"MIT"
] | 2 |
2020-10-08T14:51:24.000Z
|
2021-07-15T13:50:47.000Z
|
Data Description and Analysis.ipynb
|
ShaoniMukherjee/Instacart-Market-Basket-Analysis
|
4691cda5038247780e04ed822682c46079946e16
|
[
"MIT"
] | 12 |
2020-06-18T04:30:45.000Z
|
2022-02-07T19:27:08.000Z
| 258.165813 | 68,476 | 0.901103 |
[
[
[
"## Data Description and Analysis",
"_____no_output_____"
]
],
[
[
"import numpy as np\nimport pandas as pd\npd.set_option('max_columns', 150)\n\nimport gc\nimport os\n\n# matplotlib and seaborn for plotting\nimport matplotlib\nmatplotlib.rcParams['figure.dpi'] = 120 #resolution\nmatplotlib.rcParams['figure.figsize'] = (8,6) #figure size\n\nimport matplotlib.pyplot as plt\nsns.set_style('darkgrid')\nimport seaborn as sns\ncolor = sns.color_palette()\n\nroot = 'C:/Data/instacart-market-basket-analysis/'",
"_____no_output_____"
]
],
[
[
"The dataset contains relational set of files describing customers' orders over time. For each user, 4 to 100 orders are provided with the sequence of products purchased in each order. The data of the order's week and hour of the day as well as a relative measure of time between orders is provided.\n\n**Files in the Dataset:** ",
"_____no_output_____"
]
],
[
[
"os.listdir(root)",
"_____no_output_____"
],
[
"aisles = pd.read_csv(root + 'aisles.csv')\ndepartments = pd.read_csv(root + 'departments.csv')\norders = pd.read_csv(root + 'orders.csv')\norder_products_prior = pd.read_csv(root + 'order_products__prior.csv')\norder_products_train = pd.read_csv(root + 'order_products__train.csv')\nproducts = pd.read_csv(root + 'products.csv')",
"_____no_output_____"
]
],
[
[
"### aisles: \nThis file contains different aisles and there are total 134 unique aisles.",
"_____no_output_____"
]
],
[
[
"aisles.head()",
"_____no_output_____"
],
[
"aisles.tail()",
"_____no_output_____"
],
[
"len(aisles.aisle.unique())",
"_____no_output_____"
],
[
"aisles.aisle.unique()",
"_____no_output_____"
]
],
[
[
"### departments: \nThis file contains different departments and there are total 21 unique departments.",
"_____no_output_____"
]
],
[
[
"departments.head()",
"_____no_output_____"
],
[
"departments.tail()",
"_____no_output_____"
],
[
"len(departments.department.unique())",
"_____no_output_____"
],
[
"departments.department.unique()",
"_____no_output_____"
]
],
[
[
"### orders: \nThis file contains all the orders made by different users. From below analysis, we can conclude following:\n\n- There are total 3421083 orders made by total 206209 users.\n- There are three sets of orders: Prior, Train and Test. The distributions of orders in Train and Test sets are similar whereas the distribution of orders in Prior set is different.\n- The total orders per customer ranges from 0 to 100. \n- Based on the plot of 'Orders VS Day of Week' we can map 0 and 1 as Saturday and Sunday respectively based on the assumption that most of the people buy groceries on weekends.\n- Majority of the orders are made during the day time.\n- Customers order once in a week which is supported by peaks at 7, 14, 21 and 30 in 'Orders VS Days since prior order' graph.\n- Based on the heatmap between 'Day of Week' and 'Hour of Day,' we can say that Saturday afternoons and Sunday mornings are prime time for orders.",
"_____no_output_____"
]
],
[
[
"orders.head(12)",
"_____no_output_____"
],
[
"orders.tail()",
"_____no_output_____"
],
[
"orders.info()",
"<class 'pandas.core.frame.DataFrame'>\nRangeIndex: 3421083 entries, 0 to 3421082\nData columns (total 7 columns):\norder_id int64\nuser_id int64\neval_set object\norder_number int64\norder_dow int64\norder_hour_of_day int64\ndays_since_prior_order float64\ndtypes: float64(1), int64(5), object(1)\nmemory usage: 182.7+ MB\n"
],
[
"len(orders.order_id.unique())",
"_____no_output_____"
],
[
"len(orders.user_id.unique())",
"_____no_output_____"
],
[
"orders.eval_set.value_counts()",
"_____no_output_____"
],
[
"orders.order_number.describe().apply(lambda x: format(x, '.2f'))",
"_____no_output_____"
],
[
"order_number = orders.groupby('user_id')['order_number'].max()\norder_number = order_number.value_counts()",
"_____no_output_____"
],
[
"fig, ax = plt.subplots(figsize=(15,8))\nax = sns.barplot(x = order_number.index, y = order_number.values, color = color[3])\nax.set_xlabel('Orders per customer')\nax.set_ylabel('Count')\nax.xaxis.set_tick_params(rotation=90, labelsize=10)\nax.set_title('Frequency of Total Orders by Customers')\nfig.savefig('Frequency of Total Orders by Customers.png')",
"_____no_output_____"
],
[
"fig, ax = plt.subplots(figsize = (8,4))\nax = sns.kdeplot(orders.order_number[orders.eval_set == 'prior'], label = \"Prior set\", lw = 1)\nax = sns.kdeplot(orders.order_number[orders.eval_set == 'train'], label = \"Train set\", lw = 1)\nax = sns.kdeplot(orders.order_number[orders.eval_set == 'test'], label = \"Test set\", lw = 1)\nax.set_xlabel('Order Number')\nax.set_ylabel('Count')\nax.tick_params(axis = 'both', labelsize = 10)\nax.set_title('Distribution of Orders in Various Sets')\nfig.savefig('Distribution of Orders in Various Sets.png')\nplt.show()",
"_____no_output_____"
],
[
"fig, ax = plt.subplots(figsize = (5,3))\nax = sns.countplot(orders.order_dow)\nax.set_xlabel('Day of Week', size = 10)\nax.set_ylabel('Orders', size = 10)\nax.tick_params(axis = 'both', labelsize = 8)\nax.set_title('Total Orders per Day of Week')\nfig.savefig('Total Orders per Day of Week.png')\nplt.show()",
"_____no_output_____"
],
[
"temp_df = orders.groupby('order_dow')['user_id'].nunique()\n\nfig, ax = plt.subplots(figsize = (5,3))\nax = sns.barplot(x = temp_df.index, y = temp_df.values)\nax.set_xlabel('Day of Week', size = 10)\nax.set_ylabel('Total Unique Users', size = 10)\nax.tick_params(axis = 'both', labelsize = 8)\nax.set_title('Total Unique Users per Day of Week')\nfig.savefig('Total Unique Users per Day of Week.png')\nplt.show()",
"_____no_output_____"
],
[
"fig, ax = plt.subplots(figsize = (10,5))\nax = sns.countplot(orders.order_hour_of_day, color = color[2])\nax.set_xlabel('Hour of Day', size = 10 )\nax.set_ylabel('Orders', size = 10)\nax.tick_params(axis = 'both', labelsize = 8)\nax.set_title('Total Orders per Hour of Day')\nfig.savefig('Total Orders per Hour of Day.png')\nplt.show()",
"_____no_output_____"
],
[
"fig, ax = plt.subplots(figsize = (10,5))\nax = sns.countplot(orders.days_since_prior_order, color = color[2])\nax.set_xlabel('Days since prior order', size = 10)\nax.set_ylabel('Orders', size = 10)\nax.tick_params(axis = 'both', labelsize = 8)\nax.set_title('Orders VS Days since prior order')\nfig.savefig('Orders VS Days since prior order.png')\nplt.show()",
"_____no_output_____"
],
[
"temp_df = orders.groupby([\"order_dow\", \"order_hour_of_day\"])[\"order_number\"].aggregate(\"count\").reset_index()\ntemp_df = temp_df.pivot('order_dow', 'order_hour_of_day', 'order_number')\ntemp_df.head()",
"_____no_output_____"
],
[
"ax = plt.subplots(figsize=(7,3))\nax = sns.heatmap(temp_df, cmap=\"YlGnBu\", linewidths=.5)\nax.set_title(\"Frequency of Day of week Vs Hour of day\", size = 12)\nax.set_xlabel(\"Hour of Day\", size = 10)\nax.set_ylabel(\"Day of Week\", size = 10)\nax.tick_params(axis = 'both', labelsize = 8)\ncbar = ax.collections[0].colorbar\ncbar.ax.tick_params(labelsize=10)\nfig = ax.get_figure()\nfig.savefig(\"Frequency of Day of week Vs Hour of day.png\")\nplt.show()",
"_____no_output_____"
]
],
[
[
"### order_products_prior:\n\nThis file gives information about which products were ordered and in which order they were added in the cart. It also tells us that if the product was reordered or not. \n\n- In this file there is an information of total 3214874 orders through which total 49677 products were ordered. \n- From the 'Count VS Items in cart' plot, we can say that most of the people buy 1-15 items in an order and there were a maximum of 145 items in an order.\n- The percentage of reorder items in this set is 58.97%.",
"_____no_output_____"
]
],
[
[
"order_products_prior.head(10)",
"_____no_output_____"
],
[
"order_products_prior.tail()",
"_____no_output_____"
],
[
"len(order_products_prior.order_id.unique())",
"_____no_output_____"
],
[
"len(order_products_prior.product_id.unique())",
"_____no_output_____"
],
[
"add_to_cart_order_prior = order_products_prior.groupby('order_id')['add_to_cart_order'].count()\nadd_to_cart_order_prior = add_to_cart_order_prior.value_counts()",
"_____no_output_____"
],
[
"add_to_cart_order_prior.head()",
"_____no_output_____"
],
[
"add_to_cart_order_prior.tail()",
"_____no_output_____"
],
[
"add_to_cart_order_prior.index.max()",
"_____no_output_____"
],
[
"fig, ax = plt.subplots(figsize = (15,8))\nax = sns.barplot(x = add_to_cart_order_prior.index, y = add_to_cart_order_prior.values, color = color[3])\nax.set_xlabel('Items in cart')\nax.set_ylabel('Count')\nax.xaxis.set_tick_params(rotation=90, labelsize = 9)\nax.set_title('Frequency of Items in Cart in Prior set', size = 15)\nfig.savefig('Frequency of Items in Cart in Prior set.png')",
"_____no_output_____"
],
[
"fig, ax = plt.subplots(figsize=(3,3))\nax = sns.barplot(x = order_products_prior.reordered.value_counts().index, \n y = order_products_prior.reordered.value_counts().values, color = color[3])\nax.set_xlabel('Reorder', size = 10)\nax.set_ylabel('Count', size = 10)\nax.tick_params(axis = 'both', labelsize = 8)\nax.ticklabel_format(style='plain', axis='y')\nax.set_title('Reorder Frequency in Prior Set')\nfig.savefig('Reorder Frequency in Prior Set')\nplt.show()",
"_____no_output_____"
],
[
"print('Percentage of reorder in prior set:',\n format(order_products_prior[order_products_prior.reordered == 1].shape[0]*100/order_products_prior.shape[0], '.2f'))",
"Percentage of reorder in prior set: 58.97\n"
]
],
[
[
"### order_products_train:\n\nThis file gives information about which products were ordered and in which order they were added in the cart. It also tells us that if the product was reordered or not. \n\n- In this file there is an information of total 131209 orders through which total 39123 products were ordered. \n- From the 'Count VS Items in cart' plot, we can say that most of the people buy 1-15 items in an order and there were a maximum of 145 items in an order.\n- The percentage of reorder items in this set is 59.86%.",
"_____no_output_____"
]
],
[
[
"order_products_train.head(10)",
"_____no_output_____"
],
[
"order_products_train.tail()",
"_____no_output_____"
],
[
"len(order_products_train.order_id.unique())",
"_____no_output_____"
],
[
"len(order_products_train.product_id.unique())",
"_____no_output_____"
],
[
"add_to_cart_order_train = order_products_prior.groupby('order_id')['add_to_cart_order'].count()\nadd_to_cart_order_train = add_to_cart_order_train.value_counts()",
"_____no_output_____"
],
[
"add_to_cart_order_train.head()",
"_____no_output_____"
],
[
"add_to_cart_order_train.tail()",
"_____no_output_____"
],
[
"add_to_cart_order_train.index.max()",
"_____no_output_____"
],
[
"fig, ax = plt.subplots(figsize = (15,8))\nax = sns.barplot(x = add_to_cart_order_train.index, y = add_to_cart_order_train.values, color = color[2])\nax.set_xlabel('Items in cart')\nax.set_ylabel('Count')\nax.xaxis.set_tick_params(rotation=90, labelsize = 8)\nax.set_title('Frequency of Items in Cart in Train set', size = 15)\nfig.savefig('Frequency of Items in Cart in Train set.png')",
"_____no_output_____"
],
[
"fig, ax = plt.subplots(figsize=(3,3))\nax = sns.barplot(x = order_products_train.reordered.value_counts().index, \n y = order_products_train.reordered.value_counts().values, color = color[2])\nax.set_xlabel('Reorder', size = 10)\nax.set_ylabel('Count', size = 10)\nax.tick_params(axis = 'both', labelsize = 8)\nax.set_title('Reorder Frequency in Train Set')\nfig.savefig('Reorder Frequency in Train Set')\nplt.show()",
"_____no_output_____"
],
[
"print('Percentage of reorder in train set:',\n format(order_products_train[order_products_train.reordered == 1].shape[0]*100/order_products_train.shape[0], '.2f'))",
"Percentage of reorder in train set: 59.86\n"
]
],
[
[
"### products:\n\nThis file contains the list of total 49688 products and their aisle as well as department. The number of products in different aisles and different departments are different.",
"_____no_output_____"
]
],
[
[
"products.head(10)",
"_____no_output_____"
],
[
"products.tail()",
"_____no_output_____"
],
[
"len(products.product_name.unique())",
"_____no_output_____"
],
[
"len(products.aisle_id.unique())",
"_____no_output_____"
],
[
"len(products.department_id.unique())",
"_____no_output_____"
],
[
"temp_df = products.groupby('aisle_id')['product_id'].count()\n\nfig, ax = plt.subplots(figsize = (15,6))\nax = sns.barplot(x = temp_df.index, y = temp_df.values, color = color[3])\nax.set_xlabel('Aisle Id')\nax.set_ylabel('Total products in aisle')\nax.xaxis.set_tick_params(rotation=90, labelsize = 7)\nax.set_title('Total Products in Aisle VS Aisle ID', size = 12)\nfig.savefig('Total Products in Aisle VS Aisle ID.png')",
"_____no_output_____"
],
[
"temp_df = products.groupby('department_id')['product_id'].count()\n\nfig, ax = plt.subplots(figsize = (8,5))\nax = sns.barplot(x = temp_df.index, y = temp_df.values, color = color[2])\nax.set_xlabel('Department Id')\nax.set_ylabel('Total products in department')\nax.xaxis.set_tick_params(rotation=90, labelsize = 9)\nax.set_title('Total Products in Department VS Department ID', size = 10)\nfig.savefig('Total Products in Department VS Department ID.png')",
"_____no_output_____"
],
[
"temp_df = products.groupby('department_id')['aisle_id'].nunique()\n\nfig, ax = plt.subplots(figsize = (8,5))\nax = sns.barplot(x = temp_df.index, y = temp_df.values)\nax.set_xlabel('Department Id')\nax.set_ylabel('Total Aisles in department')\nax.xaxis.set_tick_params(rotation=90, labelsize = 9)\nax.set_title('Total Aisles in Department VS Department ID', size = 10)\nfig.savefig('Total Aisles in Department VS Department ID.png')",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
]
] |
cbd83928714ce8de68448f23288915782e6b2891
| 194,099 |
ipynb
|
Jupyter Notebook
|
notebooks/0.2.2-saw-data-preparation.ipynb
|
sward8080/housing-comp
|
c91962bfde8ae229ec37f6eeebfda14c7a45f275
|
[
"MIT"
] | null | null | null |
notebooks/0.2.2-saw-data-preparation.ipynb
|
sward8080/housing-comp
|
c91962bfde8ae229ec37f6eeebfda14c7a45f275
|
[
"MIT"
] | null | null | null |
notebooks/0.2.2-saw-data-preparation.ipynb
|
sward8080/housing-comp
|
c91962bfde8ae229ec37f6eeebfda14c7a45f275
|
[
"MIT"
] | null | null | null | 204.099895 | 60,768 | 0.884126 |
[
[
[
"---\n## Data Prep",
"_____no_output_____"
],
[
"### Dataset Cleaning",
"_____no_output_____"
]
],
[
[
"import numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\nimport seaborn as sns\n\n%matplotlib inline\n\nfrom time import time\nfrom src.features import build_features as bf\nfrom sklearn.preprocessing import MinMaxScaler\nfrom sklearn.model_selection import GridSearchCV\nfrom sklearn.metrics import make_scorer\nfrom sklearn.ensemble import AdaBoostRegressor\nfrom sklearn.ensemble import RandomForestRegressor\nfrom sklearn.metrics import fbeta_score, accuracy_score\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.preprocessing import StandardScaler\nfrom sklearn.preprocessing import OneHotEncoder\n\nfrom sklearn.experimental import enable_iterative_imputer\nfrom sklearn.impute import IterativeImputer, MissingIndicator\nfrom sklearn.pipeline import FeatureUnion, make_pipeline\n\nsns.set()",
"_____no_output_____"
]
],
[
[
"---\n## Data Preprocessing",
"_____no_output_____"
]
],
[
[
"features_raw = pd.read_csv('../data/interim/features_raw.csv', index_col='Id')\ntest_raw = pd.read_csv('../data/raw/test.csv', index_col='Id')\ntarget = pd.read_csv('../data/interim/target_raw.csv', index_col='Id', squeeze=True)\n\ntest = test_raw.copy()",
"_____no_output_____"
],
[
"features_raw.head()",
"_____no_output_____"
],
[
"features_raw.shape",
"_____no_output_____"
],
[
"df_zscore = pd.read_csv('../data/interim/df_zscore.csv', index_col='Id')\noutlier_idx = df_zscore[(df_zscore >= 5).any(1)].index\noutlier_idx",
"_____no_output_____"
]
],
[
[
"### Handle Outliers",
"_____no_output_____"
]
],
[
[
"df = features_raw.drop(index=outlier_idx)\ntarget = target.drop(index=outlier_idx)\nindex = df.index\n\n# Uncomment this line to save the DataFrames\n# target.to_csv('../data/interim/target_no_outliers.csv')",
"_____no_output_____"
],
[
"df.shape",
"_____no_output_____"
]
],
[
[
"### Assess Missing Data",
"_____no_output_____"
],
[
"#### Assess Missing Data in Each Column",
"_____no_output_____"
]
],
[
[
"df.isna().sum().sort_values(ascending=False).head(10)",
"_____no_output_____"
]
],
[
[
"#### Assess Missing Data in Each Column",
"_____no_output_____"
]
],
[
[
"nan_count = df.isna().sum()\nnan_count = nan_count[nan_count > 0]\nnan_cols = df[nan_count.index].columns\n(nan_count / df.shape[0]).sort_values(ascending=False)",
"_____no_output_____"
],
[
"# Investigate patterns in the amount of missing data in each column.\n\n# plt.rcParams.update({'figure.dpi':100})\n\nplt.figure(figsize=(9, 8))\nax = sns.histplot(nan_count, kde=False)\nax.set_title('Histogram of Missing Data by Column')\nax.set(xlabel='Total Missing or Unknown', ylabel='Total Occurrences')\nplt.show()\n\n# Uncomment this line to save the figure.\n# plt.savefig('../reports/figures/MissingDatabyCol_Histogram.svg')",
"_____no_output_____"
],
[
"nan_rows = df.isna().sum(axis=1)\nplt.figure(figsize=(9, 8))\nax = sns.histplot(nan_rows, kde=False)\nax.set_title('Histogram of Missing Data by Row')\nax.set(xlabel='Total Missing or Unknown', ylabel='Total Occurrences')\nplt.show()\n\n# Uncomment this line to save the figure.\n# plt.savefig('../reports/figures/MissingDatabyRow_Histogram.svg')",
"_____no_output_____"
]
],
[
[
"#### Assessment Summary\n\nThere is a fair amount of missing data in this dataset. Four features in particular (PoolQC, MiscFeature, Alley, Fence) contain >50% missing or unknown values. For PoolQC, we may be able to imply whether or not the home has a pool performing some feature engineering on the PoolArea feature. In addition, several features such as 'GarageArea' indicate the total square footage of a garage (if any) but our dataset does not seem to indicate whether a not a particular home has a garage or not. We'll create features for these. Let's investigate these in turn.",
"_____no_output_____"
]
],
[
[
"def fill_categorical(val):\n return 1 if val != 'NA' else 0\n\ndef fill_numerical(val):\n return 1 if val > 0 else 0\n\nna_cols = ['PoolArea', 'Fence', 'Alley', 'BsmtQual', 'BsmtCond', 'BsmtExposure',\n 'BsmtFinType1', 'BsmtFinType2', 'FireplaceQu']\n\ndf[na_cols] = df[na_cols].fillna('NA')\n\n# Most homes don't have pools. Let's use the value of 'PoolArea' to assign 'NA' to 'PoolQC' if the area == 0\n\ndf.loc[df['PoolArea'] == 0, 'PoolQC'] = 'NA' # If no 'PoolArea', (0), then there is no pool \n\n# Similarly, let's apply the same logic to other features where necessary:\n\ndf['HasFence'] = df['Fence'].apply(lambda x: fill_categorical(x))\ndf['HasAlley'] = df['Alley'].apply(lambda x: fill_categorical(x))\ndf['HasFireplace'] = df['FireplaceQu'].apply(lambda x: fill_categorical(x))\ndf['HasPool'] = df['PoolArea'].apply(lambda x: fill_numerical(x))\ndf['HasGarage'] = df['GarageArea'].apply(lambda x: fill_numerical(x))\ndf['HasBasement'] = df['TotalBsmtSF'].apply(lambda x: fill_numerical(x))\n\n# -------------------------------------------------\n# Apply above feature engineering steps on test set\n# -------------------------------------------------\n\ntest[na_cols] = test[na_cols].fillna('NA')\n\ntest.loc[test['PoolArea'] == 0, 'PoolQC'] = 'NA' # If no 'PoolArea', (0), then there is no pool \n\ntest['HasFence'] = test['Fence'].apply(lambda x: fill_categorical(x))\ntest['HasAlley'] = test['Alley'].apply(lambda x: fill_categorical(x))\ntest['HasFireplace'] = test['FireplaceQu'].apply(lambda x: fill_categorical(x))\ntest['HasPool'] = test['PoolArea'].apply(lambda x: fill_numerical(x))\ntest['HasGarage'] = test['GarageArea'].apply(lambda x: fill_numerical(x))\ntest['HasBasement'] = test['TotalBsmtSF'].apply(lambda x: fill_numerical(x))",
"_____no_output_____"
],
[
"categorical_cols = df.select_dtypes(include=object).columns\nnumerical_cols = df.select_dtypes(include=np.number).columns",
"_____no_output_____"
],
[
"# Perform One-Hot Encoding on our Categorical Data\n\nfeatures_enc = df.copy()\nfeatures_onehot_enc = pd.get_dummies(data=features_enc, columns=categorical_cols, dummy_na=True)\n\n# Uncomment this line to export DataFrame\n# features_onehot_enc.to_csv('../data/interim/features_onehot_enc.csv')\n\n# Print the number of features after one-hot encoding\nencoded = list(features_onehot_enc.columns)\nprint(f'{len(encoded)} total features after one-hot encoding.')\n\n# Uncomment the following line to see the encoded feature names\n# print(encoded)\n\n# -------------------------------------------------\n# Apply above feature engineering steps on test set\n# -------------------------------------------------\n\ntest_enc = test.copy()\ntest_onehot_enc = pd.get_dummies(data=test_enc, columns=categorical_cols, dummy_na=True)\n\n# Uncomment this line to export DataFrame\n# test_onehot_enc.to_csv('../data/interim/test_onehot_enc.csv')",
"345 total features after one-hot encoding.\n['MSSubClass', 'LotFrontage', 'LotArea', 'OverallQual', 'OverallCond', 'YearBuilt', 'YearRemodAdd', 'MasVnrArea', 'BsmtFinSF1', 'BsmtFinSF2', 'BsmtUnfSF', 'TotalBsmtSF', '1stFlrSF', '2ndFlrSF', 'LowQualFinSF', 'GrLivArea', 'BsmtFullBath', 'BsmtHalfBath', 'FullBath', 'HalfBath', 'BedroomAbvGr', 'KitchenAbvGr', 'TotRmsAbvGrd', 'Fireplaces', 'GarageYrBlt', 'GarageCars', 'GarageArea', 'WoodDeckSF', 'OpenPorchSF', 'EnclosedPorch', '3SsnPorch', 'ScreenPorch', 'PoolArea', 'MiscVal', 'MoSold', 'YrSold', 'HasFence', 'HasAlley', 'HasFireplace', 'HasPool', 'HasGarage', 'HasBasement', 'MSZoning_C (all)', 'MSZoning_FV', 'MSZoning_RH', 'MSZoning_RL', 'MSZoning_RM', 'MSZoning_nan', 'Street_Grvl', 'Street_Pave', 'Street_nan', 'Alley_Grvl', 'Alley_NA', 'Alley_Pave', 'Alley_nan', 'LotShape_IR1', 'LotShape_IR2', 'LotShape_IR3', 'LotShape_Reg', 'LotShape_nan', 'LandContour_Bnk', 'LandContour_HLS', 'LandContour_Low', 'LandContour_Lvl', 'LandContour_nan', 'Utilities_AllPub', 'Utilities_NoSeWa', 'Utilities_nan', 'LotConfig_Corner', 'LotConfig_CulDSac', 'LotConfig_FR2', 'LotConfig_FR3', 'LotConfig_Inside', 'LotConfig_nan', 'LandSlope_Gtl', 'LandSlope_Mod', 'LandSlope_Sev', 'LandSlope_nan', 'Neighborhood_Blmngtn', 'Neighborhood_Blueste', 'Neighborhood_BrDale', 'Neighborhood_BrkSide', 'Neighborhood_ClearCr', 'Neighborhood_CollgCr', 'Neighborhood_Crawfor', 'Neighborhood_Edwards', 'Neighborhood_Gilbert', 'Neighborhood_IDOTRR', 'Neighborhood_MeadowV', 'Neighborhood_Mitchel', 'Neighborhood_NAmes', 'Neighborhood_NPkVill', 'Neighborhood_NWAmes', 'Neighborhood_NoRidge', 'Neighborhood_NridgHt', 'Neighborhood_OldTown', 'Neighborhood_SWISU', 'Neighborhood_Sawyer', 'Neighborhood_SawyerW', 'Neighborhood_Somerst', 'Neighborhood_StoneBr', 'Neighborhood_Timber', 'Neighborhood_Veenker', 'Neighborhood_nan', 'Condition1_Artery', 'Condition1_Feedr', 'Condition1_Norm', 'Condition1_PosA', 'Condition1_PosN', 'Condition1_RRAe', 'Condition1_RRAn', 'Condition1_RRNe', 'Condition1_RRNn', 'Condition1_nan', 'Condition2_Artery', 'Condition2_Feedr', 'Condition2_Norm', 'Condition2_PosA', 'Condition2_PosN', 'Condition2_RRAe', 'Condition2_RRAn', 'Condition2_RRNn', 'Condition2_nan', 'BldgType_1Fam', 'BldgType_2fmCon', 'BldgType_Duplex', 'BldgType_Twnhs', 'BldgType_TwnhsE', 'BldgType_nan', 'HouseStyle_1.5Fin', 'HouseStyle_1.5Unf', 'HouseStyle_1Story', 'HouseStyle_2.5Fin', 'HouseStyle_2.5Unf', 'HouseStyle_2Story', 'HouseStyle_SFoyer', 'HouseStyle_SLvl', 'HouseStyle_nan', 'RoofStyle_Flat', 'RoofStyle_Gable', 'RoofStyle_Gambrel', 'RoofStyle_Hip', 'RoofStyle_Mansard', 'RoofStyle_Shed', 'RoofStyle_nan', 'RoofMatl_CompShg', 'RoofMatl_Membran', 'RoofMatl_Metal', 'RoofMatl_Roll', 'RoofMatl_Tar&Grv', 'RoofMatl_WdShake', 'RoofMatl_WdShngl', 'RoofMatl_nan', 'Exterior1st_AsbShng', 'Exterior1st_AsphShn', 'Exterior1st_BrkComm', 'Exterior1st_BrkFace', 'Exterior1st_CBlock', 'Exterior1st_CemntBd', 'Exterior1st_HdBoard', 'Exterior1st_ImStucc', 'Exterior1st_MetalSd', 'Exterior1st_Plywood', 'Exterior1st_Stone', 'Exterior1st_Stucco', 'Exterior1st_VinylSd', 'Exterior1st_Wd Sdng', 'Exterior1st_WdShing', 'Exterior1st_nan', 'Exterior2nd_AsbShng', 'Exterior2nd_AsphShn', 'Exterior2nd_Brk Cmn', 'Exterior2nd_BrkFace', 'Exterior2nd_CBlock', 'Exterior2nd_CmentBd', 'Exterior2nd_HdBoard', 'Exterior2nd_ImStucc', 'Exterior2nd_MetalSd', 'Exterior2nd_Other', 'Exterior2nd_Plywood', 'Exterior2nd_Stone', 'Exterior2nd_Stucco', 'Exterior2nd_VinylSd', 'Exterior2nd_Wd Sdng', 'Exterior2nd_Wd Shng', 'Exterior2nd_nan', 'MasVnrType_BrkCmn', 'MasVnrType_BrkFace', 'MasVnrType_None', 'MasVnrType_Stone', 'MasVnrType_nan', 'ExterQual_Ex', 'ExterQual_Fa', 'ExterQual_Gd', 'ExterQual_TA', 'ExterQual_nan', 'ExterCond_Ex', 'ExterCond_Fa', 'ExterCond_Gd', 'ExterCond_Po', 'ExterCond_TA', 'ExterCond_nan', 'Foundation_BrkTil', 'Foundation_CBlock', 'Foundation_PConc', 'Foundation_Slab', 'Foundation_Stone', 'Foundation_Wood', 'Foundation_nan', 'BsmtQual_Ex', 'BsmtQual_Fa', 'BsmtQual_Gd', 'BsmtQual_NA', 'BsmtQual_TA', 'BsmtQual_nan', 'BsmtCond_Fa', 'BsmtCond_Gd', 'BsmtCond_NA', 'BsmtCond_Po', 'BsmtCond_TA', 'BsmtCond_nan', 'BsmtExposure_Av', 'BsmtExposure_Gd', 'BsmtExposure_Mn', 'BsmtExposure_NA', 'BsmtExposure_No', 'BsmtExposure_nan', 'BsmtFinType1_ALQ', 'BsmtFinType1_BLQ', 'BsmtFinType1_GLQ', 'BsmtFinType1_LwQ', 'BsmtFinType1_NA', 'BsmtFinType1_Rec', 'BsmtFinType1_Unf', 'BsmtFinType1_nan', 'BsmtFinType2_ALQ', 'BsmtFinType2_BLQ', 'BsmtFinType2_GLQ', 'BsmtFinType2_LwQ', 'BsmtFinType2_NA', 'BsmtFinType2_Rec', 'BsmtFinType2_Unf', 'BsmtFinType2_nan', 'Heating_Floor', 'Heating_GasA', 'Heating_GasW', 'Heating_Grav', 'Heating_OthW', 'Heating_Wall', 'Heating_nan', 'HeatingQC_Ex', 'HeatingQC_Fa', 'HeatingQC_Gd', 'HeatingQC_Po', 'HeatingQC_TA', 'HeatingQC_nan', 'CentralAir_N', 'CentralAir_Y', 'CentralAir_nan', 'Electrical_FuseA', 'Electrical_FuseF', 'Electrical_FuseP', 'Electrical_Mix', 'Electrical_SBrkr', 'Electrical_nan', 'KitchenQual_Ex', 'KitchenQual_Fa', 'KitchenQual_Gd', 'KitchenQual_TA', 'KitchenQual_nan', 'Functional_Maj1', 'Functional_Maj2', 'Functional_Min1', 'Functional_Min2', 'Functional_Mod', 'Functional_Sev', 'Functional_Typ', 'Functional_nan', 'FireplaceQu_Ex', 'FireplaceQu_Fa', 'FireplaceQu_Gd', 'FireplaceQu_NA', 'FireplaceQu_Po', 'FireplaceQu_TA', 'FireplaceQu_nan', 'GarageType_2Types', 'GarageType_Attchd', 'GarageType_Basment', 'GarageType_BuiltIn', 'GarageType_CarPort', 'GarageType_Detchd', 'GarageType_nan', 'GarageFinish_Fin', 'GarageFinish_RFn', 'GarageFinish_Unf', 'GarageFinish_nan', 'GarageQual_Ex', 'GarageQual_Fa', 'GarageQual_Gd', 'GarageQual_Po', 'GarageQual_TA', 'GarageQual_nan', 'GarageCond_Ex', 'GarageCond_Fa', 'GarageCond_Gd', 'GarageCond_Po', 'GarageCond_TA', 'GarageCond_nan', 'PavedDrive_N', 'PavedDrive_P', 'PavedDrive_Y', 'PavedDrive_nan', 'PoolQC_Ex', 'PoolQC_Fa', 'PoolQC_Gd', 'PoolQC_NA', 'PoolQC_nan', 'Fence_GdPrv', 'Fence_GdWo', 'Fence_MnPrv', 'Fence_MnWw', 'Fence_NA', 'Fence_nan', 'MiscFeature_Gar2', 'MiscFeature_Othr', 'MiscFeature_Shed', 'MiscFeature_TenC', 'MiscFeature_nan', 'SaleType_COD', 'SaleType_CWD', 'SaleType_Con', 'SaleType_ConLD', 'SaleType_ConLI', 'SaleType_ConLw', 'SaleType_New', 'SaleType_Oth', 'SaleType_WD', 'SaleType_nan', 'SaleCondition_Abnorml', 'SaleCondition_AdjLand', 'SaleCondition_Alloca', 'SaleCondition_Family', 'SaleCondition_Normal', 'SaleCondition_Partial', 'SaleCondition_nan']\n"
],
[
"imp = IterativeImputer(missing_values=np.nan, random_state=5, max_iter=20)\n\nimputed_arr = imp.fit_transform(features_onehot_enc)\nfeatures_imputed = pd.DataFrame(imputed_arr, columns=features_onehot_enc.columns)\nfeatures_imputed.index = features_onehot_enc.index\n\n# Uncomment this line to export DataFrame\n# features_imputed.to_csv('../data/interim/features_imputed.csv')\n\n# -------------------------------------------------\n# Apply above imputation steps on test set\n# -------------------------------------------------\n\nimputed_arr_test = imp.fit_transform(test_onehot_enc)\ntest_imputed = pd.DataFrame(imputed_arr_test, columns=test_onehot_enc.columns)\ntest_imputed.index = test_onehot_enc.index",
"_____no_output_____"
],
[
"def align_dataframes(train_set, test_set):\n if train_set.shape[1] > test_set.shape[1]:\n cols = train_set.columns.difference(test_set.columns)\n df = pd.DataFrame(0, index=train_set.index, columns=cols)\n test_set[df.columns] = df\n elif train_set.shape[1] < test_set.shape[1]:\n cols = test_set.columns.difference(train_set.columns)\n df = pd.DataFrame(0, index=test_set.index, columns=cols)\n train_set[df.columns] = df\n \nalign_dataframes(features_imputed, test_imputed)",
"_____no_output_____"
],
[
"test_imputed = test_imputed.fillna(value=0)\n\n# Uncomment this line to export DataFrame\n# test_imputed.to_csv('../data/interim/test_imputed.csv')",
"_____no_output_____"
]
],
[
[
"### Feature Transformation\n#### Transforming Skewed Continuous Features\n\nA dataset may sometimes contain at least one feature whose values tend to lie near a single number, but will also have a non-trivial number of vastly larger or smaller values than that single number. Algorithms can be sensitive to such distributions of values and can underperform if the range is not properly normalized. We'll need to check the following continuous data features for 'skew'.\n\n- LotFrontage\n- LotArea\n- MasVnrArea\n- BsmtFinSF1\n- BsmtFinSF2\n- TotalBsmtSF\n- 1stFlrSF\n- 2ndFlrSF\n- LowQualFinSF\n- GrLivArea\n- GarageArea\n- WoodDeckSF\n- OpenPorchSF\n- EnclosedPorch\n- 3SsnPorch\n- ScreenPorch\n- PoolArea\n- MiscVal",
"_____no_output_____"
]
],
[
[
"continuous_cols = ['LotFrontage', 'LotArea', 'MasVnrArea', 'BsmtFinSF1', 'BsmtFinSF2', 'TotalBsmtSF', '1stFlrSF', \n '2ndFlrSF', 'LowQualFinSF', 'GrLivArea', 'GarageArea', 'WoodDeckSF', 'OpenPorchSF',\n 'EnclosedPorch', '3SsnPorch', 'ScreenPorch', 'PoolArea', 'MiscVal']",
"_____no_output_____"
],
[
"skewed = ['ScreenPorch', 'PoolArea', 'LotFrontage', '3SsnPorch', 'LowQualFinSF']\n\nfig = plt.figure(figsize = (16,10));\n\n# Skewed feature plotting\nfor i, feature in enumerate(skewed):\n ax = fig.add_subplot(2, 3, i+1)\n sns.histplot(features_imputed[feature], bins=20, color='#00A0A0')\n ax.set_title(\"'%s' Feature Distribution\"%(feature), fontsize = 14)\n ax.set_xlabel(\"Value\")\n ax.set_ylabel(\"Number of Records\")\n ax.set_ylim((0, 1600))\n ax.set_yticks([0, 400, 800, 1200, 1600])\n ax.set_yticklabels([0, 400, 800, 1200, \">1600\"])\n\n# Plot aesthetics\nfig.suptitle(\"Skewed Distributions of Continuous Data Features\", \\\n fontsize = 16, y = 1.03)\n\nfig.tight_layout()\n\n# Uncomment this line to save the figure.\n# plt.savefig('../reports/figures/Skewed_Distributions.svg')",
"_____no_output_____"
],
[
"features_log_xformed = pd.DataFrame(data = features_imputed)\n\n# Since the logarithm of 0 is undefined, translate values a small amount to apply the logarithm successfully\nfeatures_log_xformed[continuous_cols] = features_imputed[continuous_cols].apply(lambda x: np.log(x + 1))\n\nfig = plt.figure(figsize = (15,10));\n\n# Skewed feature plotting\nfor i, feature in enumerate(skewed):\n ax = fig.add_subplot(2, 3, i+1)\n sns.histplot(features_log_xformed[feature], bins=20, color='#00A0A0')\n ax.set_title(\"'%s' Feature Distribution\"%(feature), fontsize = 14)\n ax.set_xlabel(\"Value\")\n ax.set_ylabel(\"Number of Records\")\n ax.set_ylim((0, 1600))\n ax.set_yticks([0, 400, 800, 1200, 1600])\n ax.set_yticklabels([0, 400, 800, 1200, \">1600\"])\n\n# Plot aesthetics\nfig.suptitle(\"Log-Transformed Distributions of Continuous Data Features\", \\\n fontsize = 16, y = 1.03)\n\nfig.tight_layout()\n\n# Uncomment this line to save the figure.\n# plt.savefig('../reports/figures/Log_Xformed_Distributions.svg')\n\n# -------------------------------------------------\n# Apply above log transformation steps on test set\n# -------------------------------------------------\n\ntest_log_xformed = pd.DataFrame(data = test_imputed)\n\n# Since the logarithm of 0 is undefined, translate values a small amount to apply the logarithm successfully\ntest_log_xformed[continuous_cols] = test_imputed[continuous_cols].apply(lambda x: np.log(x + 1))",
"_____no_output_____"
]
],
[
[
"We also need to perform a log-transformation on our target variable 'SalePrice' to remove skew.",
"_____no_output_____"
]
],
[
[
"target_log_xformed = target.transform(np.log)\n\n# Uncomment this line to export DataFrame\n# target_log_xformed.to_csv('../data/interim/target_log_xformed.csv')",
"_____no_output_____"
]
],
[
[
"##### Feature Scaling\n#### Normalizing Numerical Features\n\nIn addition to performing transformations on features that are highly skewed, it is often good practice to perform some type of scaling on numerical features. Applying a scaling to the data does not change the shape of each feature's distribution. Normalization does, however, ensure that each feature is treated equally when applying supervised learners.",
"_____no_output_____"
]
],
[
[
"# Initialize scaler, then apply it to the features\nscaler = StandardScaler()\nnumerical = features_log_xformed.select_dtypes(include=np.number).columns\n\nfeatures_scaled = pd.DataFrame(data = features_log_xformed)\nfeatures_scaled[numerical] = scaler.fit_transform(features_log_xformed[numerical])\n\nfeatures_final = features_scaled.copy()\n\n# Uncomment this line to export DataFrame\n# features_scaled.to_csv('../data/interim/features_scaled.csv')\n\n# Uncomment this line to export DataFrame\n# features_final.to_csv('../data/processed/features_final.csv')\n\n# Show an example of a record with scaling applied\nfeatures_final.head()\n\n# -------------------------------------------------\n# Apply above feature scaling steps on test set\n# -------------------------------------------------\n\nnumerical = test_log_xformed.select_dtypes(include=np.number).columns\n\ntest_scaled = pd.DataFrame(data = test_log_xformed)\ntest_scaled[numerical] = scaler.fit_transform(test_log_xformed[numerical])\n\ntest_final = test_scaled.copy()\n\n# Uncomment this line to export DataFrame\n# test_scaled.to_csv('../data/interim/test_scaled.csv')\n\n# Uncomment this line to export DataFrame\n# test_final.to_csv('../data/processed/test_final.csv')",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
]
] |
cbd84513e05192d95aabe904122dc99f46c816f1
| 349,180 |
ipynb
|
Jupyter Notebook
|
samples/opencv/load_and_save.ipynb
|
RyoWakabayashi/jupyter-image-processing-samples
|
ae4cb47814ceb240b3bb2d6cd6eed46670382bf8
|
[
"MIT"
] | 2 |
2020-04-22T15:06:57.000Z
|
2020-04-23T08:23:27.000Z
|
samples/opencv/load_and_save.ipynb
|
RyoWakabayashi/jupyter-image-processing-samples
|
ae4cb47814ceb240b3bb2d6cd6eed46670382bf8
|
[
"MIT"
] | null | null | null |
samples/opencv/load_and_save.ipynb
|
RyoWakabayashi/jupyter-image-processing-samples
|
ae4cb47814ceb240b3bb2d6cd6eed46670382bf8
|
[
"MIT"
] | null | null | null | 1,995.314286 | 173,488 | 0.962051 |
[
[
[
"# OpenCV による画像の読込・書込",
"_____no_output_____"
]
],
[
[
"import cv2\nfrom matplotlib import pyplot as plt",
"_____no_output_____"
]
],
[
[
"## 画像の読込",
"_____no_output_____"
]
],
[
[
"img = cv2.imread('../lena.jpg')",
"_____no_output_____"
]
],
[
[
"## matplotlib で表示するため、BGRからRGBに変換する",
"_____no_output_____"
]
],
[
[
"img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)",
"_____no_output_____"
]
],
[
[
"## 画像の表示",
"_____no_output_____"
]
],
[
[
"plt.imshow(img)\nplt.show()",
"_____no_output_____"
]
],
[
[
"## 画像の反転",
"_____no_output_____"
]
],
[
[
"img = cv2.flip(img, 1)\nplt.imshow(img)\nplt.show()",
"_____no_output_____"
]
],
[
[
"## 保存前にRGBからGBRに戻す",
"_____no_output_____"
]
],
[
[
"img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)",
"_____no_output_____"
]
],
[
[
"## 画像の保存",
"_____no_output_____"
]
],
[
[
"cv2.imwrite('../lena_flip.jpg', img)",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
]
] |
cbd8523dab0c697ac3d6a7be3173091e83fbc389
| 542,129 |
ipynb
|
Jupyter Notebook
|
figures/MOC_z_wBasinCalc.ipynb
|
zouberou-sayibou/DNN4Cli
|
c5e6d94b2516d168f76655e199a0856cd5666ead
|
[
"MIT"
] | null | null | null |
figures/MOC_z_wBasinCalc.ipynb
|
zouberou-sayibou/DNN4Cli
|
c5e6d94b2516d168f76655e199a0856cd5666ead
|
[
"MIT"
] | null | null | null |
figures/MOC_z_wBasinCalc.ipynb
|
zouberou-sayibou/DNN4Cli
|
c5e6d94b2516d168f76655e199a0856cd5666ead
|
[
"MIT"
] | null | null | null | 652.381468 | 342,748 | 0.941518 |
[
[
[
"import scipy.io, os\nimport numpy as np\nimport matplotlib.pyplot as plt\nfrom netCDF4 import Dataset\nfrom fastjmd95 import rho\nfrom matplotlib.colors import ListedColormap\nimport seaborn as sns; sns.set()\nsns.set()\nimport seawater as sw\nfrom mpl_toolkits.axes_grid1.inset_locator import inset_axes \nimport matp\n\nlotlib as mpl\n\ncolours=sns.color_palette('colorblind', 10)\nmy_cmap = ListedColormap(colours)\n\ncolor_list=colours",
"_____no_output_____"
]
],
[
[
"## Code to plot the meridional overturning and density structure from the North Atlantic from Sonnewald and Lguensat (2021).\n\nData used are from the ECCOv4 State Estimate available: https://ecco-group.org/products-ECCO-V4r4.html\n\nNote: Data is generated for the North Atlantic, also including the Southern Ocean and Artcic basin. Data fro the Paciic and Indian ocean are also generated, and the bleow code can be adjusted to plot this also.\n",
"_____no_output_____"
]
],
[
[
"gridInfo=np.load('latLonDepthLevelECCOv4.npz')\nzLev=gridInfo['depthLevel'][:]\ndepthPlot=zLev.cumsum()\nlat=gridInfo['lat'][:]\nlon=gridInfo['lon'][:]\n\nzMat=np.repeat(zLev,720*360).reshape((50,360,720))\n\ndvx=np.rot90(0.5*111000*np.cos(lat*(np.pi/180)),1)\n\nmasks=np.load('regimeMasks.npz')\nmaskMD=masks['maskMD']\nmaskSSV=masks['maskSSV']\nmaskNSV=masks['maskNSV']\nmaskTR=masks['maskTR']\nmaskSO=masks['maskSO']\nmaskNL=masks['maskNL']",
"_____no_output_____"
],
[
"def getData(NR):\n arr = os.listdir('/home/maike/Documents/ECCO_BV/NVELSTAR/.')\n f =Dataset('/home/maike/Documents/ECCO_BV/NVELSTAR/'+arr[NR])\n nvelS =f.variables['NVELSTAR'][:]\n\n arr = os.listdir('/home/maike/Documents/ECCO_BV/NVELMASS/.')\n f =Dataset('/home/maike/Documents/ECCO_BV/NVELMASS/'+arr[NR])\n nvelM =f.variables['NVELMASS'][:]\n\n return(nvelS+nvelM)",
"_____no_output_____"
]
],
[
[
"## Creating the basin masks",
"_____no_output_____"
]
],
[
[
"nvel= getData(1) #To get the shape",
"_____no_output_____"
],
[
"globalMask=np.ones(nvel[0].shape)\n\nmaskArea=np.zeros(nvel[0].shape)*np.nan\nmaskArea[:,65:360,0:222]=1\nmaskArea[:,65:360,500:720]=1\nmaskArea[:,310:360,:]=np.nan\nmaskArea[:,210:350,160:250]=np.nan\nmaskArea[:,0:140,500:650]=np.nan\nmaskArea[:,0:165,500:620]=np.nan\nmaskArea[:,0:255,500:560]=np.nan\nmaskArea[:,0:210,500:570]=np.nan\nmaskArea[:,0:185,500:590]=np.nan\npacificMask=maskArea\n\nmaskArea=np.zeros(nvel[0].shape)*np.nan\nmaskArea[:,:,221:400]=1\nmaskArea[:,200:360,160:400]=1\nmaskArea[:,0:65,:]=1\nmaskArea[:,310:360,:]=1\nmaskArea[:,199:215,160:180]=np.nan\nmaskArea[:,199:210,160:190]=np.nan\natlanticMask=maskArea\n\nmaskArea=np.ones(nvel[0].shape)\nindA=np.where(atlanticMask==1)\nindP=np.where(pacificMask==1)\nmaskArea[indA]=np.nan\nmaskArea[indP]=np.nan\nmaskArea[:,100:250,100:250]=np.nan\nindianMask=maskArea\n\nplt.figure()\nplt.imshow(np.flipud(globalMask[0]*nvel[0,0]))\nplt.figure()\nplt.imshow(np.flipud(atlanticMask[0]*nvel[0,0]))\nplt.figure()\nplt.imshow(np.flipud(pacificMask[0]*nvel[0,0]))\nplt.figure()\nplt.imshow(np.flipud(indianMask[0]*nvel[0,0]))",
"_____no_output_____"
]
],
[
[
"## Calculating the streamfunction\n\nThe overall meridional overturning ($\\Psi_{z\\theta}$) from Fig. 3 in Sonnewald and Lguensat (2021) is defined as:\n\n$$\\Psi_{z\\theta}(\\theta,z)=- \\int^z_{-H} \\int_{\\phi_2}^{\\phi_1} v(\\phi,\\theta,z')d\\phi dz',$$\n\n\\noindent where $z$ is the relative level depth and $v$ is the meridional (north-south) component of velocity. For the regimes, the relevant velocity fields were then used. A positive $\\Psi_{z\\theta}$ signifies a clockwise circulation, while a negative $\\Psi_{z\\theta}$ signifies an anticlockwise circulation.",
"_____no_output_____"
]
],
[
[
"def psiZ(NVEL_IN, mask):\n '''Function to calculate overturning in depth space as described in Sonnewald and Lguensat (2021).'''\n ntrans=np.zeros(NVEL_IN[:,:,0].shape);\n gmoc=np.zeros(NVEL_IN[:,:,0].shape);\n NVEL=NVEL_IN*mask\n\n # zonal transport integral\n for zz in np.arange(0,50):\n ntrans[zz,:]=np.nansum(NVEL[zz,:,:]*dvx,axis=1); \n\n for zz in np.flipud(np.arange(0,49)):\n gmoc[zz,:]=gmoc[zz+1,:]+ntrans[zz+1,:]*zLev[zz+1];\n\n gmoc=gmoc/1e6; \n return(gmoc)",
"_____no_output_____"
],
[
"def psiMaskedCalc(mask):\n '''Calculating the overturning in depth space for the different regimes, as plotted in Fig. 3 in Sonnewald and Lguensat (2021).'''\n yrs, months=20,12\n\n PSI_all = np.zeros((yrs*months, 50, 360))*np.nan\n PSI_NL = np.zeros((yrs*months, 50, 360))*np.nan\n PSI_SO = np.zeros((yrs*months, 50, 360))*np.nan\n PSI_SSV = np.zeros((yrs*months, 50, 360))*np.nan\n PSI_NSV = np.zeros((yrs*months, 50, 360))*np.nan\n PSI_MD = np.zeros((yrs*months, 50, 360))*np.nan\n PSI_TR = np.zeros((yrs*months, 50, 360))*np.nan\n\n ITTER=0\n for NR in np.arange(0,yrs):\n nvel= getData(NR)\n # print('Got data')\n for MM in np.arange(0,months):\n PSI_all[ITTER]=psiZ(nvel[MM], np.ones(maskSO.shape)*mask)\n PSI_NL[ITTER]=psiZ(nvel[MM], maskNL*mask)\n PSI_SO[ITTER]=psiZ(nvel[MM], maskSO*mask)\n PSI_SSV[ITTER]=psiZ(nvel[MM], maskSSV*mask)\n PSI_NSV[ITTER]=psiZ(nvel[MM], maskNSV*mask)\n PSI_MD[ITTER]=psiZ(nvel[MM], maskMD*mask)\n PSI_TR[ITTER]=psiZ(nvel[MM], maskTR*mask)\n ITTER+=1 \n return PSI_all, PSI_NL, PSI_SO, PSI_SSV, PSI_NSV, PSI_MD, PSI_TR",
"_____no_output_____"
],
[
"PSI_all_A, PSI_NL_A, PSI_SO_A, PSI_SSV_A, PSI_NSV_A, PSI_MD_A, PSI_TR_A = psiMaskedCalc(atlanticMask)\n\nPSI_all_P, PSI_NL_P, PSI_SO_P, PSI_SSV_P, PSI_NSV_P, PSI_MD_P, PSI_TR_P = psiMaskedCalc(pacificMask)\n\nPSI_all_I, PSI_NL_I, PSI_SO_I, PSI_SSV_I, PSI_NSV_I, PSI_MD_I, PSI_TR_I = psiMaskedCalc(indianMask)\n\nPSI_all_G, PSI_NL_G, PSI_SO_G, PSI_SSV_G, PSI_NSV_G, PSI_MD_G, PSI_TR_G = psiMaskedCalc(globalMask)\n\n",
"_____no_output_____"
],
[
"#Save the data\n\nnp.savez('PSI_global', PSI_all_G=PSI_all_G, PSI_NL_G=PSI_NL_G, PSI_SO_G=PSI_SO_G, PSI_SSV_G=PSI_SSV_G, PSI_NSV_G=PSI_NSV_G, PSI_MD_G=PSI_MD_G, PSI_TR_G=PSI_TR_G)\n\nnp.savez('PSI_atlantic', PSI_all_A=PSI_all_A, PSI_NL_A=PSI_NL_A, PSI_SO_A=PSI_SO_A, PSI_SSV_A=PSI_SSV_A, PSI_NSV_A=PSI_NSV_A, PSI_MD_A=PSI_MD_A, PSI_TR_A=PSI_TR_A)\n\nnp.savez('PSI_pacific', PSI_all_P=PSI_all_P, PSI_NL_P=PSI_NL_P, PSI_SO_P=PSI_SO_P, PSI_SSV_P=PSI_SSV_P, PSI_NSV_P=PSI_NSV_P, PSI_MD_P=PSI_MD_P, PSI_TR_P=PSI_TR_P)\n\nnp.savez('PSI_indian', PSI_all_I=PSI_all_I, PSI_NL_I=PSI_NL_I, PSI_SO_I=PSI_SO_I, PSI_SSV_I=PSI_SSV_I, PSI_NSV_I=PSI_NSV_I, PSI_MD_I=PSI_MD_I, PSI_TR_I=PSI_TR_I)",
"_____no_output_____"
]
],
[
[
"## Calculate the density in $\\sigma_2$",
"_____no_output_____"
]
],
[
[
"def getDataTS(NR):\n '''Retrieve the T and S data. Data from the ECCOv4 state estimate.'''\n arr = os.listdir('/home/maike/Documents/ECCO_BV/THETA/.')\n f =Dataset('/home/maike/Documents/ECCO_BV/THETA/'+arr[NR])\n T =f.variables['THETA'][:]\n\n arr = os.listdir('/home/maike/Documents/ECCO_BV/SALT/.')\n f =Dataset('/home/maike/Documents/ECCO_BV/SALT/'+arr[NR])\n S =f.variables['SALT'][:]\n\n return(T, S)",
"_____no_output_____"
],
[
"dens=np.zeros((50,360,720))\nITTER=1\nyrs=20\nmonths=12\nfor NR in np.arange(0,yrs):\n T,S = getDataTS(NR)\n print('Got data', NR)\n #Tin=sw.eos80.temp(S, T, -np.cumsum(zMat, axis=0), pr=np.cumsum(zMat, axis=0))\n for MM in np.arange(0,months):\n dens = dens+rho(S[MM], T[MM], 2000) - 1000\n ITTER+=1\ndens=dens/ITTER",
"Got data 0\nGot data 1\nGot data 2\nGot data 3\nGot data 4\nGot data 5\nGot data 6\nGot data 7\nGot data 8\nGot data 9\nGot data 10\nGot data 11\nGot data 12\nGot data 13\nGot data 14\nGot data 15\nGot data 16\nGot data 17\nGot data 18\nGot data 19\n"
],
[
"#Save the density data.\nnp.save('density20yr', np.array(dens))",
"_____no_output_____"
]
],
[
[
"# Finally, we plot the data.\n\nThe plot is a composite of different subplots.\n",
"_____no_output_____"
]
],
[
[
"\nlevs=[32,33,34, 34.5, 35, 35.5,36,36.5,37,37.25,37.5,37.75,38]\ncols=plt.cm.viridis([300,250, 200,150, 125, 100, 50,30, 10,15,10,9,1])\nLand=np.ones(np.nansum(PSI_all_A, axis=0).shape)*np.nan\nLand[np.nansum(PSI_all_A, axis=0)==0.0]=0\nland3D=np.ones(dens.shape)\nland3D[dens==0]=np.nan\n\ndef zPlotSurf(ax, data,zMin, zMax,label,mm,latMin,latMax,RGB,Ticks,saveName='test'):\n land=np.ones(np.nanmean(data, axis=0).shape)*np.nan\n land[np.nansum(data, axis=0)==0.0]=0\n n=50\n levels = np.linspace(-20, 20, n+1)\n ax.contourf(lat[0,latMin:latMax],-depthPlot[zMin:zMax],-np.nanmean(data, axis=0)[zMin:zMax,latMin:latMax], levels=np.linspace(-20, 20, n+1),cmap=plt.cm.seismic, extend='both')\n\n n2=30 \n densityPlot=np.nanmean((dens*land3D*mm), axis=2)\n assert(len(levs)==len(cols))\n CS=ax.contour(lat[0,latMin:latMax],-depthPlot[zMin:zMax],densityPlot[zMin:zMax,latMin:latMax],\n levels=levs,\n linewidths=3,colors=cols, extend='both')\n\n ax.tick_params(axis='y', labelsize=20)\n if Ticks == 0:\n ax.set_xticklabels( () )\n elif Ticks == 1:\n ax.set_xticklabels( () )\n ax.set_yticklabels( () )\n ax.contourf(lat[0,latMin:latMax],-depthPlot[zMin:zMax],land[zMin:zMax,latMin:latMax], 1,cmap=plt.cm.Set2)\n ax.contourf(lat[0,latMin:latMax],-depthPlot[zMin:zMax],Land[zMin:zMax,latMin:latMax], 50,cmap=plt.cm.bone)\n\n yL=ax.get_ylim()\n xL=ax.get_xlim()\n plt.text(xL[0]+0.02*np.ptp(xL), yL[0]+0.4*np.ptp(yL), label, fontsize=20, size=30, \n weight='bold', bbox={'facecolor':'white', 'alpha':0.7}, va='bottom')\n \n \ndef zPlotDepth(ax, data,zMin, zMax,label,mm,latMin,latMax,RGB,Ticks,saveName='test'):\n land=np.ones(np.nanmean(data, axis=0).shape)*np.nan\n land[np.nansum(data, axis=0)==0.0]=0\n n=50\n levels = np.linspace(-20, 20, n+1)\n ax.contourf(lat[0,latMin:latMax],-depthPlot[zMin:zMax],-np.nanmean(data, axis=0)[zMin:zMax,latMin:latMax], levels=np.linspace(-20, 20, n+1),cmap=plt.cm.seismic, extend='both')\n n2=30\n densityPlot=np.nanmean((dens*land3D*mm), axis=2)\n ax.contour(lat[0,latMin:latMax],-depthPlot[zMin:zMax],densityPlot[zMin:zMax,latMin:latMax], colors=cols,\n levels=levs,\n linewidths=3, extend='both')\n\n if Ticks == 0:\n ax.tick_params(axis='y', labelsize=20)\n #ax.set_xticklabels( () )\n elif Ticks== 1:\n #ax.set_xticklabels( () )\n ax.set_yticklabels( () )\n plt.tick_params(axis='both', labelsize=20)\n #plt.clim(cmin, cmax)\n \n ax.contourf(lat[0,latMin:latMax],-depthPlot[zMin:zMax],land[zMin:zMax,latMin:latMax], 1,cmap=plt.cm.Set2)\n ax.contourf(lat[0,latMin:latMax],-depthPlot[zMin:zMax],Land[zMin:zMax,latMin:latMax], 50,cmap=plt.cm.bone)\n yL=ax.get_ylim()\n xL=ax.get_xlim()\n plt.text(xL[0]+0.03*np.ptp(xL), yL[0]+0.03*np.ptp(yL), label, fontsize=20, size=30, \n weight='bold', bbox={'facecolor':RGB, 'alpha':1}, va='bottom')\n \n",
"_____no_output_____"
],
[
"# Set general figure options\n\n# figure layout\nxs = 15.5 # figure width in inches\nnx = 2 # number of axes in x dimension\nny = 3 # number of sub-figures in y dimension (each sub-figure has two axes)\nnya = 2 # number of axes per sub-figure\nidy = [2.0, 1.0] # size of the figures in the y dimension\nxm = [0.07, 0.07,0.9, 0.07] # x margins of the figure (left to right)\nym = [1.5] + ny*[0.07, 0.1] + [0.3] # y margins of the figure (bottom to top)\n\n# pre-calculate some things\nxcm = np.cumsum(xm) # cumulative margins\nycm = np.cumsum(ym) # cumulative margins\nidx = (xs - np.sum(xm))/nx\nidy_off = [0] + idy\nys = np.sum(idy)*ny + np.sum(ym) # size of figure in y dimension\n\n# make the figure!\nfig = plt.figure(figsize=(xs, ys))\n\n# loop through sub-figures\n\n\nix,iy=0,0\n\n\n#print(loc[0], loc[1], loc[0] + loc[2], loc[1] + loc[3])\n# create the axis\n#ax = plt.axes(loc)\nfor iys in range(nya):\n # (bottom left corner x, bottom left corner y, width, height)\n loc = ((xcm[ix] + (ix*idx))/xs,\n (ycm[nya*iy + iys] + np.sum(idy)*iy+ idy_off[iys])/ys,\n idx/xs,\n idy[iys]/ys)\n #print(loc[0], loc[1], loc[0] + loc[2], loc[1] + loc[3])\n # create the axis\n ax = plt.axes(loc)\n\n # split between your two figure types\n if iys == 0:\n zPlotDepth(ax, PSI_TR_A,1,50,'TR', maskTR,200, 310, color_list[1],'')\n # if not the bottom figure remove x ticks\n if iy > 0:\n ax.set_xticks([])\n else:\n xticks = ax.get_xticks()\n ax.set_xticklabels(['{:0.0f}$^\\circ$N'.format(xtick) for xtick in xticks])\n elif iys == 1:\n zPlotSurf(ax, PSI_TR_A,0,10,'', maskTR,200, 310, color_list[1],'')\n # remove x ticks\n ax.set_xticks([]) \n\n\n\n\nix,iy=0,1\n\n#print(loc[0], loc[1], loc[0] + loc[2], loc[1] + loc[3])\n# create the axis\n#ax = plt.axes(loc)\nfor iys in range(nya):\n # (bottom left corner x, bottom left corner y, width, height)\n loc = ((xcm[ix] + (ix*idx))/xs,\n (ycm[nya*iy + iys] + np.sum(idy)*iy+ idy_off[iys])/ys,\n idx/xs,\n idy[iys]/ys)\n #print(loc[0], loc[1], loc[0] + loc[2], loc[1] + loc[3])\n # create the axis\n ax = plt.axes(loc)\n\n # split between your two figure types\n if iys == 0:\n zPlotDepth(ax, PSI_NL_A,1,50,'NL', maskNL,200, 310, color_list[-1],'')\n # if not the bottom figure remove x ticks\n if iy > 0:\n ax.set_xticks([])\n elif iys == 1:\n zPlotSurf(ax, PSI_NL_A,0,10,'', maskNL,200, 310, color_list[4],'')\n # remove x ticks\n ax.set_xticks([]) \n \n\n \n############### n-SV\n\nix,iy=0,2\n\nloc = ((xcm[ix] + (ix*idx))/xs,\n (ycm[nya*iy + iys] + np.sum(idy)*iy+ idy_off[iys])/ys,\n idx/xs,\n idy[iys]/ys)\n\n#print(loc[0], loc[1], loc[0] + loc[2], loc[1] + loc[3])\n# create the axis\n#ax = plt.axes(loc)\nfor iys in range(nya):\n # (bottom left corner x, bottom left corner y, width, height)\n loc = ((xcm[ix] + (ix*idx))/xs,\n (ycm[nya*iy + iys] + np.sum(idy)*iy+ idy_off[iys])/ys,\n idx/xs,\n idy[iys]/ys)\n #print(loc[0], loc[1], loc[0] + loc[2], loc[1] + loc[3])\n # create the axis\n ax = plt.axes(loc)\n\n # split between your two figure types\n if iys == 0:\n zPlotDepth(ax, PSI_NSV_A,1,50,'N-SV', maskNSV,200, 310, color_list[4],'')\n # if not the bottom figure remove x ticks\n if iy > 0:\n ax.set_xticks([])\n elif iys == 1:\n zPlotSurf(ax, PSI_NSV_A,0,10,'', maskNSV,200, 310, color_list[-1],'')\n # remove x ticks\n ax.set_xticks([]) \n \n# \n#_______________________________________________________________________\n# S-SV \n\nix,iy=1,2\n\nloc = ((xcm[ix] + (ix*idx))/xs,\n (ycm[nya*iy + iys] + np.sum(idy)*iy+ idy_off[iys])/ys,\n idx/xs,\n idy[iys]/ys)\n\n#print(loc[0], loc[1], loc[0] + loc[2], loc[1] + loc[3])\n# create the axis\n# ax = plt.axes(loc)\nfor iys in range(nya):\n # (bottom left corner x, bottom left corner y, width, height)\n loc = ((xcm[ix] + (ix*idx))/xs,\n (ycm[nya*iy + iys] + np.sum(idy)*iy+ idy_off[iys])/ys,\n idx/xs,\n idy[iys]/ys)\n #print(loc[0], loc[1], loc[0] + loc[2], loc[1] + loc[3])\n # create the axis\n ax = plt.axes(loc)\n\n # split between your two figure types\n if iys == 0:\n zPlotDepth(ax, PSI_SSV_A,1,50,'S-SV', maskSSV,200, 310, color_list[2],1,'')\n # if not the bottom figure remove x ticks\n if iy > 0:\n ax.set_xticks([])\n elif iys == 1:\n zPlotSurf(ax, PSI_SSV_A,0,10,'', maskSSV,200, 310, color_list[-3],1,'')\n # remove x ticks\n ax.set_xticks([]) \n\n \n\n\n#%%%%%%%%%%%%%%%%%%%%%%%%% SO\n \nix,iy=1,1\n\nloc = ((xcm[ix] + (ix*idx))/xs,\n (ycm[nya*iy + iys] + np.sum(idy)*iy+ idy_off[iys])/ys,\n idx/xs,\n idy[iys]/ys)\n\n#print(loc[0], loc[1], loc[0] + loc[2], loc[1] + loc[3])\n# create the axis\n#ax = plt.axes(loc)\nfor iys in range(nya):\n # (bottom left corner x, bottom left corner y, width, height)\n loc = ((xcm[ix] + (ix*idx))/xs,\n (ycm[nya*iy + iys] + np.sum(idy)*iy+ idy_off[iys])/ys,\n idx/xs,\n idy[iys]/ys)\n #print(loc[0], loc[1], loc[0] + loc[2], loc[1] + loc[3])\n # create the axis\n ax = plt.axes(loc)\n\n # split between your two figure types\n if iys == 0:\n zPlotDepth(ax, PSI_SO_A,1,50,'SO', maskSO,200, 310, color_list[-3],1,'')\n # if not the bottom figure remove x ticks\n if iy > 0:\n ax.set_xticks([])\n elif iys == 1:\n zPlotSurf(ax, PSI_SO_A,0,10,'', maskSO,200, 310, color_list[-3],1,'')\n # remove x ticks\n ax.set_xticks([]) \n \n\n#%%%%%%%MD\n \nix,iy=1,0\n\nloc = ((xcm[ix] + (ix*idx))/xs,\n (ycm[nya*iy + iys] + np.sum(idy)*iy+ idy_off[iys])/ys,\n idx/xs,\n idy[iys]/ys)\n\n#print(loc[0], loc[1], loc[0] + loc[2], loc[1] + loc[3])\n# create the axis\n#ax = plt.axes(loc)\nfor iys in range(nya):\n # (bottom left corner x, bottom left corner y, width, height)\n loc = ((xcm[ix] + (ix*idx))/xs,\n (ycm[nya*iy + iys] + np.sum(idy)*iy+ idy_off[iys])/ys,\n idx/xs,\n idy[iys]/ys)\n #print(loc[0], loc[1], loc[0] + loc[2], loc[1] + loc[3])\n # create the axis\n ax = plt.axes(loc)\n\n # split between your two figure types\n if iys == 0:\n zPlotDepth(ax, PSI_MD_A,1,50,'MD', maskMD,200, 310, color_list[0],1,'')\n # if not the bottom figure remove x ticks\n if iy > 0:\n ax.set_xticks([])\n else:\n xticks = ax.get_xticks()\n ax.set_xticklabels(['{:0.0f}$^\\circ$N'.format(xtick) for xtick in xticks])\n elif iys == 1:\n zPlotSurf(ax, PSI_MD_A,0,10,'', maskMD,200, 310, color_list[-3],1,'')\n # remove x ticks\n ax.set_xticks([]) \n \n\n\n\n\ncmap = plt.get_cmap('viridis')\ncmap = mpl.colors.ListedColormap(cols)\nncol = len(levs)\naxes = plt.axes([(xcm[0])/(xs), (ym[0]-0.6)/ys, (2*idx + xm[1])/(xs*2), (0.2)/ys])\ncb = fig.colorbar(plt.cm.ScalarMappable(norm=mpl.colors.Normalize(-0.5, ncol - 0.5), cmap=cmap), \n cax=axes, orientation='horizontal')\ncb.ax.set_xticks(np.arange(ncol))\ncb.ax.set_xticklabels(['{:0.2f}'.format(lev) for lev in levs])\ncb.ax.tick_params(labelsize=20) \ncb.set_label(label=r'Density, $\\sigma_2$',weight='bold', fontsize=20)\n\ncmap = plt.get_cmap('seismic')\nncol = len(cols)\naxes = plt.axes([(xcm[2]+2*idx)/(xs*2), (ym[0]-0.6)/ys, (2*idx+xm[3])/(xs*2), (0.2)/ys])\ncb = fig.colorbar(plt.cm.ScalarMappable(norm=mpl.colors.Normalize(-20,20), cmap=cmap), \n cax=axes, label='title', orientation='horizontal', extend='both',format='%.0f',\n boundaries=np.linspace(-20, 20, 41))\ncb.ax.tick_params(labelsize=20) \ncb.set_label(label=r'SV ($10^{6}m^{2}s^{-2}$)',weight='bold', fontsize=20)\n# save as a png\n#fig.savefig('psiRho_NAtl_sigma2.png', dpi=200, bbox_inches='tight')\n\n",
"/home/maike/Downloads/anaconda3/lib/python3.7/site-packages/ipykernel_launcher.py:44: RuntimeWarning: Mean of empty slice\n/home/maike/Downloads/anaconda3/lib/python3.7/site-packages/ipykernel_launcher.py:16: RuntimeWarning: Mean of empty slice\n app.launch_new_instance()\n/home/maike/Downloads/anaconda3/lib/python3.7/site-packages/matplotlib/contour.py:1520: UserWarning: Warning: converting a masked element to nan.\n self.zmax = float(z.max())\n/home/maike/Downloads/anaconda3/lib/python3.7/site-packages/matplotlib/contour.py:1521: UserWarning: Warning: converting a masked element to nan.\n self.zmin = float(z.min())\n/home/maike/Downloads/anaconda3/lib/python3.7/site-packages/matplotlib/contour.py:1169: RuntimeWarning: invalid value encountered in less\n under = np.nonzero(lev < self.zmin)[0]\n/home/maike/Downloads/anaconda3/lib/python3.7/site-packages/matplotlib/contour.py:1171: RuntimeWarning: invalid value encountered in greater\n over = np.nonzero(lev > self.zmax)[0]\n/home/maike/Downloads/anaconda3/lib/python3.7/site-packages/ipykernel_launcher.py:237: UserWarning: Use the colorbar set_ticks() method instead.\n"
]
]
] |
[
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
]
] |
cbd86e7d3b87e5be36bf0ef4e492f49ca4bf97d1
| 30,207 |
ipynb
|
Jupyter Notebook
|
Tryout.ipynb
|
prajitdhar/Compositionality
|
d0920ca1ab866b1859621eb10437b01723a5582d
|
[
"Apache-2.0"
] | 1 |
2021-07-31T21:53:29.000Z
|
2021-07-31T21:53:29.000Z
|
Tryout.ipynb
|
prajitdhar/Compositionality
|
d0920ca1ab866b1859621eb10437b01723a5582d
|
[
"Apache-2.0"
] | null | null | null |
Tryout.ipynb
|
prajitdhar/Compositionality
|
d0920ca1ab866b1859621eb10437b01723a5582d
|
[
"Apache-2.0"
] | 3 |
2019-03-20T19:58:41.000Z
|
2020-06-13T10:48:08.000Z
| 40.875507 | 206 | 0.466051 |
[
[
[
"import pandas as pd\nimport numpy as np\nfrom sklearn.decomposition import PCA,TruncatedSVD,NMF\nfrom sklearn.preprocessing import Normalizer\nimport argparse\nimport time\nimport pickle as pkl",
"_____no_output_____"
],
[
"def year_binner(year,val=10):\n return year - year%val",
"_____no_output_____"
],
[
"def dim_reduction(df,rows):\n df_svd = TruncatedSVD(n_components=300, n_iter=10, random_state=args.seed)\n print(f'Explained variance ratio {(df_svd.fit(df).explained_variance_ratio_.sum()):2.3f}')\n #df_list=df_svd.fit(df).explained_variance_ratio_\n df_reduced = df_svd.fit_transform(df)\n df_reduced = Normalizer(copy=False).fit_transform(df_reduced)\n df_reduced=pd.DataFrame(df_reduced,index=rows)\n #df_reduced.reset_index(inplace=True)\n if args.temporal!=0:\n df_reduced.index = pd.MultiIndex.from_tuples(df_reduced.index, names=['common', 'time'])\n return df_reduced",
"_____no_output_____"
],
[
"parser = argparse.ArgumentParser(description='Gather data necessary for performing Regression')\n\nparser.add_argument('--inputdir',type=str,\n help='Provide directory that has the files with the fivegram counts')\nparser.add_argument('--outputdir',type=str,\n help='Provide directory in that the output files should be stored')\nparser.add_argument('--temporal', type=int, default=0,\n help='Value to bin the temporal information: 0 (remove temporal information), 1 (no binning), 10 (binning to decades), 20 (binning each 20 years) or 50 (binning each 50 years)')\n\nparser.add_argument('--contextual', action='store_true',\n help='Is the model contextual')\n\nparser.add_argument('--cutoff', type=int, default=50,\n help='Cut-off frequency for each compound per time period : none (0), 20, 50 and 100')\n\n\nparser.add_argument('--seed', type=int, default=1991,\n help='random seed')\n\nparser.add_argument('--storedf', action='store_true',\n help='Should the embeddings be saved')\n\nparser.add_argument('--dims', type=int, default=300,\n help='Desired number of reduced dimensions')\n\nparser.add_argument('--input_format',type=str,default='csv',choices=['csv','pkl'],\n help='In what format are the input files : csv or pkl')\nparser.add_argument('--save_format', type=str,default='pkl',choices=['pkl','csv'],\n help='In what format should the reduced datasets be saved : csv or pkl')\n\n\nargs = parser.parse_args('--inputdir ../Compounding/coha_compounds/ --outputdir ../Compounding/coha_compounds/ --cutoff 10 --storedf --input_format csv --save_format csv'.split())",
"_____no_output_____"
],
[
"print(f'Cutoff: {args.cutoff}')\nprint(f'Time span: {args.temporal}')\nprint(f'Dimensionality: {args.dims}')",
"Cutoff: 10\nTime span: 0\nDimensionality: 300\n"
],
[
"print(\"Creating dense embeddings\")\nif args.contextual:\n print(\"CompoundCentric Model\")\n print(\"Loading the constituent and compound vector datasets\")\n\n if args.input_format==\"csv\":\n compounds=pd.read_csv(args.inputdir+\"/compounds.csv\",sep=\"\\t\")\n elif args.input==\"pkl\":\n compounds=pd.read_pickle(args.inputdir+\"/compounds.pkl\")\n compounds.reset_index(inplace=True)\n compounds.year=compounds.year.astype(\"int32\")\n compounds=compounds.query('1800 <= year <= 2010').copy()\n compounds['common']=compounds['modifier']+\" \"+compounds['head']\n\n #head_list_reduced=compounds['head'].unique().tolist()\n #modifier_list_reduced=compounds['modifier'].unique().tolist()\n\n if args.temporal==0:\n print('No temporal information is stored')\n compounds=compounds.groupby(['common','context'])['count'].sum().to_frame()\n compounds.reset_index(inplace=True)\n compounds=compounds.loc[compounds.groupby(['common'])['count'].transform('sum').gt(args.cutoff)]\n compounds=compounds.groupby(['common','context'])['count'].sum()\n\n else:\n compounds['time']=year_binner(compounds['year'].values,args.temporal)\n compounds=compounds.groupby(['common','context','time'])['count'].sum().to_frame()\n compounds.reset_index(inplace=True)\n compounds=compounds.loc[compounds.groupby(['common','time'])['count'].transform('sum').gt(args.cutoff)]\n compounds=compounds.groupby(['common','time','context'])['count'].sum()\n\n\n\n\n if args.input_format==\"csv\":\n modifiers=pd.read_csv(args.inputdir+\"/modifiers.csv\",sep=\"\\t\")\n elif args.input==\"pkl\":\n modifiers=pd.read_pickle(args.inputdir+\"/modifiers.pkl\")\n modifiers.reset_index(inplace=True)\n modifiers.year=modifiers.year.astype(\"int32\")\n modifiers=modifiers.query('1800 <= year <= 2010').copy()\n modifiers.columns=['common','context','year','count']\n modifiers['common']=modifiers['common'].str.replace(r'_noun$', r'_m', regex=True)\n \n if args.temporal==0:\n print('No temporal information is stored')\n modifiers=modifiers.groupby(['common','context'])['count'].sum().to_frame()\n modifiers.reset_index(inplace=True)\n modifiers=modifiers.loc[modifiers.groupby(['common'])['count'].transform('sum').gt(args.cutoff)]\n modifiers=modifiers.groupby(['common','context'])['count'].sum()\n else:\n modifiers['time']=year_binner(modifiers['year'].values,args.temporal)\n modifiers=modifiers.groupby(['common','context','time'])['count'].sum().to_frame()\n modifiers=modifiers.loc[modifiers.groupby(['common','time'])['count'].transform('sum').gt(args.cutoff)]\n modifiers=modifiers.groupby(['common','time','context'])['count'].sum()\n\n if args.input_format==\"csv\":\n heads=pd.read_csv(args.inputdir+\"/heads.csv\",sep=\"\\t\")\n elif args.input_format==\"pkl\":\n heads=pd.read_pickle(args.inputdir+\"/heads.pkl\")\n heads.reset_index(inplace=True)\n heads.year=heads.year.astype(\"int32\")\n heads=heads.query('1800 <= year <= 2010').copy()\n heads.columns=['common','context','year','count']\n heads['common']=heads['common'].str.replace(r'_noun$', r'_h', regex=True)\n if args.temporal==0:\n print('No temporal information is stored')\n heads=heads.groupby(['common','context'])['count'].sum().to_frame()\n heads.reset_index(inplace=True)\n heads=heads.loc[heads.groupby(['common'])['count'].transform('sum').gt(args.cutoff)]\n heads=heads.groupby(['common','context'])['count'].sum()\n else:\n heads['time']=year_binner(heads['year'].values,args.temporal)\n heads=heads.groupby(['common','context','time'])['count'].sum().to_frame()\n heads=heads.loc[heads.groupby(['common','time'])['count'].transform('sum').gt(args.cutoff)]\n heads=heads.groupby(['common','time','context'])['count'].sum()\n\n print('Concatenating all the datasets together')\n df=pd.concat([heads,modifiers,compounds], sort=True)\n\nelse:\n print(\"CompoundAgnostic Model\")\n wordlist = pkl.load( open( \"data/coha_wordlist.pkl\", \"rb\" ) )\n \n if args.input_format==\"csv\":\n compounds=pd.read_csv(args.inputdir+\"/phrases.csv\",sep=\"\\t\")\n elif args.input_format==\"pkl\":\n compounds=pd.read_pickle(args.inputdir+\"/phrases.pkl\")\n compounds.reset_index(inplace=True)\n compounds.year=compounds.year.astype(\"int32\")\n compounds=compounds.query('1800 <= year <= 2010').copy()\n compounds['common']=compounds['modifier']+\" \"+compounds['head']\n\n\n if args.temporal==0:\n print('No temporal information is stored')\n compounds=compounds.groupby(['common','context'])['count'].sum().to_frame()\n compounds.reset_index(inplace=True)\n compounds=compounds.loc[compounds.groupby(['common'])['count'].transform('sum').gt(args.cutoff)]\n compounds=compounds.groupby(['common','context'])['count'].sum()\n else:\n compounds['time']=year_binner(compounds['year'].values,args.temporal)\n #compounds = dd.from_pandas(compounds, npartitions=100)\n compounds=compounds.groupby(['common','context','time'])['count'].sum().to_frame()\n compounds=compounds.loc[compounds.groupby(['common','time'])['count'].transform('sum').gt(args.cutoff)]\n compounds=compounds.groupby(['common','time','context'])['count'].sum()\n \n if args.input_format==\"csv\":\n constituents=pd.read_csv(args.outputdir+\"/words.csv\",sep=\"\\t\")\n elif args.input_format==\"pkl\":\n constituents=pd.read_pickle(args.outputdir+\"/words.pkl\")\n constituents.reset_index(inplace=True)\n constituents.year=constituents.year.astype(\"int32\")\n constituents=constituents.query('1800 <= year <= 2010').copy()\n constituents.columns=['common','context','year','count']\n constituents.query('common in @wordlist',inplace=True)\n if args.temporal==0:\n print('No temporal information is stored')\n constituents=constituents.groupby(['common','context'])['count'].sum().to_frame()\n constituents.reset_index(inplace=True)\n constituents=constituents.loc[constituents.groupby(['common'])['count'].transform('sum').gt(args.cutoff)]\n constituents=constituents.groupby(['common','context'])['count'].sum() \n else:\n constituents['time']=year_binner(constituents['year'].values,args.temporal)\n constituents=constituents.groupby(['common','context','time'])['count'].sum().to_frame()\n constituents.reset_index(inplace=True)\n constituents=constituents.loc[constituents.groupby(['common','time'])['count'].transform('sum').gt(args.cutoff)]\n constituents=constituents.groupby(['common','time','context'])['count'].sum()\n\n print('Concatenating all the datasets together')\n df=pd.concat([constituents,compounds], sort=True)",
"Creating dense embeddings\nCompoundAgnostic Model\nNo temporal information is stored\nNo temporal information is stored\nConcatenating all the datasets together\n"
],
[
"dtype = pd.SparseDtype(np.float, fill_value=0)\ndf=df.astype(dtype)\nif args.temporal!=0: \n df, rows, _ = df.sparse.to_coo(row_levels=['common','time'],column_levels=['context'],sort_labels=False)\n\nelse:\n df, rows, _ = df.sparse.to_coo(row_levels=['common'],column_levels=['context'],sort_labels=False)",
"_____no_output_____"
],
[
"print('Running SVD') \ndf_reduced=dim_reduction(df,rows)\n\nprint('Splitting back into individual datasets are saving them')\nif args.temporal!=0:\n df_reduced.index.names = ['common','time']\nelse:\n df_reduced.index.names = ['common']",
"Running SVD\nExplained variance ratio 0.971\nSplitting back into individual datasets are saving them\n"
],
[
"compounds_reduced=df_reduced.loc[df_reduced.index.get_level_values(0).str.contains(r'\\w \\w')]\ncompounds_reduced.reset_index(inplace=True)\n #print(compounds_reduced.head())\n#compounds_reduced['modifier'],compounds_reduced['head']=compounds_reduced['common'].str.split(' ', 1).str",
"_____no_output_____"
],
[
"compounds_reduced[['modifier','head']]=compounds_reduced['common'].str.split(' ', n=1,expand=True).copy()\ncompounds_reduced",
"/home/users0/pageljs/dh/build/anaconda3/envs/pytorch/lib/python3.7/site-packages/pandas/core/frame.py:3062: SettingWithCopyWarning: \nA value is trying to be set on a copy of a slice from a DataFrame.\nTry using .loc[row_indexer,col_indexer] = value instead\n\nSee the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n self[k1] = value[k2]\n"
]
]
] |
[
"code"
] |
[
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
]
] |
cbd8735fc725ad50ac467ef7e62a75234cd8b4ed
| 13,589 |
ipynb
|
Jupyter Notebook
|
FourrierNeuralNet2.ipynb
|
szat/FourierNeuralNet-TF2
|
2cce2f4f0a82f06a6f597881bb0442a368691948
|
[
"MIT"
] | null | null | null |
FourrierNeuralNet2.ipynb
|
szat/FourierNeuralNet-TF2
|
2cce2f4f0a82f06a6f597881bb0442a368691948
|
[
"MIT"
] | null | null | null |
FourrierNeuralNet2.ipynb
|
szat/FourierNeuralNet-TF2
|
2cce2f4f0a82f06a6f597881bb0442a368691948
|
[
"MIT"
] | null | null | null | 37.747222 | 178 | 0.487306 |
[
[
[
"Load libs and utilities. ",
"_____no_output_____"
]
],
[
[
"!pip install -U -q PyDrive\n\nfrom pydrive.auth import GoogleAuth\nfrom pydrive.drive import GoogleDrive\nfrom google.colab import auth\nfrom oauth2client.client import GoogleCredentials\n\nfrom google.colab import drive\ndrive.mount('/content/drive')\n%cd \"drive/MyDrive/Projects/Fourier\"\n!pip install import-ipynb\nimport import_ipynb",
"Mounted at /content/drive\n/content/drive/MyDrive/Projects/Fourier\nCollecting import-ipynb\n Downloading https://files.pythonhosted.org/packages/63/35/495e0021bfdcc924c7cdec4e9fbb87c88dd03b9b9b22419444dc370c8a45/import-ipynb-0.1.3.tar.gz\nBuilding wheels for collected packages: import-ipynb\n Building wheel for import-ipynb (setup.py) ... \u001b[?25l\u001b[?25hdone\n Created wheel for import-ipynb: filename=import_ipynb-0.1.3-cp37-none-any.whl size=2976 sha256=be3f9d46cb95832a7fcb579be3d210152d056eb7f8c24e67b0c0e8b1aab4b407\n Stored in directory: /root/.cache/pip/wheels/b4/7b/e9/a3a6e496115dffdb4e3085d0ae39ffe8a814eacc44bbf494b5\nSuccessfully built import-ipynb\nInstalling collected packages: import-ipynb\nSuccessfully installed import-ipynb-0.1.3\n"
],
[
"import os\nimport tensorflow as tf\nprint(\"Tensorflow version: \" + tf.__version__)\n\nfrom tensorflow import keras\nfrom tensorflow.keras import layers\nfrom tensorflow.keras.utils import plot_model\nfrom pathlib import Path\n\nfrom utils import *",
"Tensorflow version: 2.4.1\n"
],
[
"class UnitGaussianNormalizer:\n def __init__(self, x, eps=0.00001):\n super(UnitGaussianNormalizer, self).__init__()\n self.mean = tf.math.reduce_mean(x, 0)\n self.std = tf.math.reduce_std(x, 0)\n self.eps = eps\n\n def encode(self, x):\n x = (x - self.mean) / (self.std + self.eps)\n return x\n\n def decode(self, x):\n std = self.std + self.eps\n mean = self.mean\n x = (x * std) + mean\n return x",
"_____no_output_____"
],
[
"PROJECT_PATH = Path(os.path.abspath('')).parent.parent.resolve().__str__()\nTRAIN_PATH = PROJECT_PATH + '/Datasets/Fourier/piececonst_r241_N1024_smooth1.mat'\nTEST_PATH = PROJECT_PATH + '/Datasets/Fourier/piececonst_r241_N1024_smooth2.mat'\n\nN_TRAIN = 1000\nW = 49 #width\nFTS = 32 #features\nR = 5 #refinement\nMODES = 12\n\n# ...\ntry:\n if DATA_IS_LOADED:\n print(\"Not reloading data!\")\nexcept:\n reader = MatReader()\n if reader.is_not_loaded():\n reader.load_file(TRAIN_PATH)\n\nDATA_IS_LOADED = True\n# ...\n\nx_train = reader.read_field('coeff')[:N_TRAIN,::R,::R]\ny_train = reader.read_field('sol')[:N_TRAIN,::R,::R]\n\nS_ = x_train.shape[1]\ngrids = []\ngrids.append(np.linspace(0, 1, S_))\ngrids.append(np.linspace(0, 1, S_))\ngrid = np.vstack([xx.ravel() for xx in np.meshgrid(*grids)]).T\ngrid = grid.reshape(1,S_,S_,2)\n\nprint(x_train.shape)\n\nx_train = tf.convert_to_tensor(x_train, dtype=tf.float32)\ny_train = tf.convert_to_tensor(y_train, dtype=tf.float32)\ngrid = tf.convert_to_tensor(grid, dtype=tf.float32)\nx_train = tf.expand_dims(x_train, axis=3)\ngrid = tf.repeat(grid, repeats = N_TRAIN, axis = 0)\nx_train = tf.concat([x_train, grid], axis=3)\ny_train = tf.expand_dims(y_train, axis=3)\n\nx_normalizer = UnitGaussianNormalizer(x_train)\nx_train = x_normalizer.encode(x_train)\n\ny_normalizer = UnitGaussianNormalizer(y_train)\ny_train = y_normalizer.encode(y_train)\n\nprint(\"x_train dims: \" + str(x_train.shape))\nprint(\"y_train dims: \" + str(y_train.shape))",
"(1000, 49, 49)\nx_train dims: (1000, 49, 49, 3)\ny_train dims: (1000, 49, 49, 1)\n"
],
[
"class FourierLayer(layers.Layer):\n def __init__(self):\n super(FourierLayer, self).__init__() \n self.weight_fft1 = tf.Variable(tf.random.uniform([FTS, FTS, MODES, MODES], minval=0, maxval=1),name=\"Wfft1\", trainable=True)\n self.weight_fft2 = tf.Variable(tf.random.uniform([FTS, FTS, MODES, MODES], minval=0, maxval=1),name=\"Wfft2\", trainable=True)\n\n def call(self, input, training=True):\n weight_fft_complex = tf.complex(self.weight_fft1, self.weight_fft2)\n x = input\n x = keras.layers.Lambda(lambda v: tf.signal.rfft2d(v, tf.constant([49, 49])))(x)\n x = x[:,:,:MODES, :MODES]\n x = keras.layers.Lambda(lambda v: tf.einsum('ioxy,bixy->boxy', weight_fft_complex, v))(x)\n x = keras.layers.Lambda(lambda v: tf.signal.irfft2d(v, tf.constant([49, 49])))(x)\n return x",
"_____no_output_____"
],
[
"class FourierUnit(layers.Layer):\n def __init__(self):\n super(FourierUnit, self).__init__()\n self.W = tf.keras.layers.Conv1D(W, 1)\n self.fourier = FourierLayer() \n self.add = tf.keras.layers.Add()\n self.bn = tf.keras.layers.BatchNormalization()\n\n def call(self, input, training=True):\n x = input\n x1 = self.fourier(x)\n x2 = self.W(x)\n x = self.add([x1, x2])\n x = self.bn(x)\n return x",
"_____no_output_____"
],
[
"class MyModel(keras.Model):\n def __init__(self):\n super(MyModel, self).__init__()\n self.fc0 = tf.keras.layers.Dense(FTS)\n self.perm_pre = tf.keras.layers.Permute((3, 1, 2))\n \n self.fourier_unit_1 = FourierUnit()\n self.relu_1 = tf.keras.layers.ReLU()\n\n self.fourier_unit_2 = FourierUnit()\n\n self.relu = tf.keras.layers.ReLU()\n self.perm_post = tf.keras.layers.Permute((2, 3, 1))\n self.fc1 = tf.keras.layers.Dense(128)\n self.relu2 = tf.keras.layers.ReLU()\n self.fc2 = tf.keras.layers.Dense(1)\n\n def call(self, input):\n x = self.fc0(input)\n x = self.perm_pre(x)\n \n x = self.fourier_unit_1(x)\n x = self.relu_1(x)\n\n x = self.fourier_unit_2(x)\n\n x = self.perm_post(x)\n x = self.fc1(x)\n x = self.relu2(x)\n x = self.fc2(x)\n return x\n\n def model(self):\n x = keras.Input(shape=(W, W, 3))\n return keras.Model(inputs=[x], outputs=self.call(x))",
"_____no_output_____"
],
[
"model = MyModel()\nmse = tf.keras.losses.MeanSquaredError()\nmodel.compile(\n loss=mse,\n optimizer=keras.optimizers.Adam(lr=3e-4),\n metrics=[tf.keras.metrics.RootMeanSquaredError()],\n)\nmodel.fit(x_train, y_train, batch_size=64, epochs=2, verbose=2)\nmodel.model().summary()",
"Epoch 1/2\n16/16 - 36s - loss: 1.0500 - root_mean_squared_error: 1.0247\nEpoch 2/2\n16/16 - 1s - loss: 0.7540 - root_mean_squared_error: 0.8683\nModel: \"model\"\n_________________________________________________________________\nLayer (type) Output Shape Param # \n=================================================================\ninput_1 (InputLayer) [(None, 49, 49, 3)] 0 \n_________________________________________________________________\ndense (Dense) (None, 49, 49, 32) 128 \n_________________________________________________________________\npermute (Permute) (None, 32, 49, 49) 0 \n_________________________________________________________________\nfourier_unit (FourierUnit) (None, 32, 49, 49) 297558 \n_________________________________________________________________\nre_lu (ReLU) (None, 32, 49, 49) 0 \n_________________________________________________________________\nfourier_unit_1 (FourierUnit) (None, 32, 49, 49) 297558 \n_________________________________________________________________\npermute_1 (Permute) (None, 49, 49, 32) 0 \n_________________________________________________________________\ndense_1 (Dense) (None, 49, 49, 128) 4224 \n_________________________________________________________________\nre_lu_2 (ReLU) (None, 49, 49, 128) 0 \n_________________________________________________________________\ndense_2 (Dense) (None, 49, 49, 1) 129 \n=================================================================\nTotal params: 599,597\nTrainable params: 599,401\nNon-trainable params: 196\n_________________________________________________________________\n"
]
]
] |
[
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
]
] |
cbd882551c32ced7bf2163e80eccf7957812a750
| 18,410 |
ipynb
|
Jupyter Notebook
|
charis/dehra_dun/04_Temperature_Index_Melt_Model.ipynb
|
mjbrodzik/ipython_notebooks
|
65ff0e0eb3fa9545987d009e51087fd0fd2979b7
|
[
"Apache-2.0"
] | null | null | null |
charis/dehra_dun/04_Temperature_Index_Melt_Model.ipynb
|
mjbrodzik/ipython_notebooks
|
65ff0e0eb3fa9545987d009e51087fd0fd2979b7
|
[
"Apache-2.0"
] | null | null | null |
charis/dehra_dun/04_Temperature_Index_Melt_Model.ipynb
|
mjbrodzik/ipython_notebooks
|
65ff0e0eb3fa9545987d009e51087fd0fd2979b7
|
[
"Apache-2.0"
] | 1 |
2020-09-12T14:51:10.000Z
|
2020-09-12T14:51:10.000Z
| 29.26868 | 416 | 0.513905 |
[
[
[
"empty"
]
]
] |
[
"empty"
] |
[
[
"empty"
]
] |
cbd8861d32d3f51697645da86ad11e5c0ea19d3a
| 442,346 |
ipynb
|
Jupyter Notebook
|
_notebooks/2022-02-11-Neural-Networks-Regression-with-TensorFlow.ipynb
|
sandeshkatakam/My-Machine_learning-Blog
|
2d71f3bcac3662617b54d6b90a46c85a6ebc6830
|
[
"Apache-2.0"
] | 1 |
2022-02-01T11:58:52.000Z
|
2022-02-01T11:58:52.000Z
|
_notebooks/2022-02-11-Neural-Networks-Regression-with-TensorFlow.ipynb
|
sandeshkatakam/My-Machine_learning-Blog
|
2d71f3bcac3662617b54d6b90a46c85a6ebc6830
|
[
"Apache-2.0"
] | 5 |
2022-02-01T12:00:39.000Z
|
2022-02-18T03:44:00.000Z
|
_notebooks/2022-02-11-Neural-Networks-Regression-with-TensorFlow.ipynb
|
sandeshkatakam/My-Machine_learning-Blog
|
2d71f3bcac3662617b54d6b90a46c85a6ebc6830
|
[
"Apache-2.0"
] | null | null | null | 73.589419 | 25,422 | 0.704297 |
[
[
[
"# Neural Networks for Regression with TensorFlow\n> Notebook demonstrates Neural Networks for Regression Problems with TensorFlow\n\n- toc: true \n- badges: true\n- comments: true\n- categories: [DeepLearning, NeuralNetworks, TensorFlow, Python, LinearRegression]\n- image: images/nntensorflow.png",
"_____no_output_____"
],
[
"## Neural Network Regression Model with TensorFlow",
"_____no_output_____"
],
[
"This notebook is continuation of the Blog post [TensorFlow Fundamentals](https://sandeshkatakam.github.io/My-Machine_learning-Blog/tensorflow/machinelearning/2022/02/09/TensorFlow-Fundamentals.html). **The notebook is an account of my working for the Tensorflow tutorial by Daniel Bourke on Youtube**. \n**The Notebook will cover the following concepts:**\n* Architecture of a neural network regression model.\n* Input shapes and output shapes of a regression model(features and labels).\n* Creating custom data to view and fit.\n* Steps in modelling\n * Creating a model, compiling a model, fitting a model, evaluating a model.\n* Different evaluation methods.\n* Saving and loading models.",
"_____no_output_____"
],
[
"**Regression Problems**: \nA regression problem is when the output variable is a real or continuous value, such as “salary” or “weight”. Many different models can be used, the simplest is the linear regression. It tries to fit data with the best hyper-plane which goes through the points.\nExamples:\n* How much will this house sell for?\n* How many people will buy this app?\n* How much will my health insurace be?\n* How much should I save each week for fuel?",
"_____no_output_____"
],
[
"We can also use the regression model to try and predict where the bounding boxes should be in object detection problem. Object detection thus involves both regression and then classifying the image in the box(classification problem).\n",
"_____no_output_____"
],
[
" ### Regression Inputs and outputs\n",
"_____no_output_____"
],
[
"Architecture of a regression model:\n* Hyperparameters:\n * Input Layer Shape : same as shape of number of features.\n * Hidden Layrer(s): Problem specific\n * Neurons per hidden layer : Problem specific.\n * Output layer shape: same as hape of desired prediction shape.\n * Hidden activation : Usually ReLU(rectified linear unit) sometimes sigmoid.\n * Output acitvation: None, ReLU, logistic/tanh.\n * Loss function : MSE(Mean squared error) or MAE(Mean absolute error) or combination of both.\n * Optimizer: SGD(Stochastic Gradient Descent), Adam optimizer.\n\n**Source:** Adapted from page 239 of [Hands-On Machine learning with Scikit-Learn, Keras & TensorFlow](https://www.oreilly.com/library/view/hands-on-machine-learning/9781492032632/)\n",
"_____no_output_____"
],
[
"Example of creating a sample regression model in TensorFlow:\n\n```\n# 1. Create a model(specific to your problem)\n\nmodel = tf.keras.Sequential([\n tf.keras.Input(shape = (3,)),\n tf.keras.layers.Dense(100, activation = \"relu\"),\n tf.keras.layers.Dense(100, activation = \"relu\"),\n tf.keras.layers.Dense(100, activation = \"relu\"),\n tf.keras.layers.Dense(1, activation = None)\n])\n\n# 2. Compile the model\n\nmodel.compile(loss = tf.keras.losses.mae, optimizer = tf.keras.optimizers.Adam(lr = 0.0001), metrics = [\"mae\"])\n\n# 3. Fit the model\n\nmodel.fit(X_train, Y_train, epochs = 100)\n\n```",
"_____no_output_____"
],
[
"### Introduction to Regression with Neural Networks in TensorFlow",
"_____no_output_____"
]
],
[
[
"# Import TensorFlow\nimport tensorflow as tf\nprint(tf.__version__)",
"2.7.0\n"
],
[
"## Creating data to view and fit\nimport numpy as np\nimport matplotlib.pyplot as plt\nfrom matplotlib import style\nstyle.use('dark_background')\n# create features\nX = np.array([-7.0,-4.0,-1.0,2.0,5.0,8.0,11.0,14.0])\n\n# Create labels\ny = np.array([3.0,6.0,9.0,12.0,15.0,18.0,21.0,24.0])\n\n# Visualize it\nplt.scatter(X,y)",
"_____no_output_____"
],
[
"y == X + 10",
"_____no_output_____"
]
],
[
[
"Yayy.. we got the relation by just seeing the data. Since the data is small and the relation ship is just linear, it was easy to guess the relation.",
"_____no_output_____"
],
[
"### Input and Output shapes",
"_____no_output_____"
]
],
[
[
"# Create a demo tensor for the housing price prediction problem\nhouse_info = tf.constant([\"bedroom\",\"bathroom\", \"garage\"])\nhouse_price = tf.constant([939700])\nhouse_info, house_price",
"_____no_output_____"
],
[
"X[0], y[0]",
"_____no_output_____"
],
[
"X[1], y[1]",
"_____no_output_____"
],
[
"input_shape = X[0].shape\noutput_shape = y[0].shape\ninput_shape, output_shape",
"_____no_output_____"
],
[
"X[0].ndim",
"_____no_output_____"
]
],
[
[
"we are specifically looking at scalars here. Scalars have 0 dimension",
"_____no_output_____"
]
],
[
[
"# Turn our numpy arrays into tensors\nX = tf.cast(tf.constant(X), dtype = tf.float32)\ny = tf.cast(tf.constant(y), dtype = tf.float32)\nX.shape, y.shape",
"_____no_output_____"
],
[
"input_shape = X[0].shape\noutput_shape = y[0].shape\ninput_shape, output_shape",
"_____no_output_____"
],
[
"plt.scatter(X,y)",
"_____no_output_____"
]
],
[
[
"### Steps in modelling with Tensorflow\n\n1. **Creating a model** - define the input and output layers, as well as the hidden layers of a deep learning model.\n\n2. **Compiling a model** - define the loss function(how wrong the prediction of our model is) and the optimizer (tells our model how to improve the partterns its learning) and evaluation metrics(what we can use to interpret the performance of our model).\n\n3. Fitting a model - letting the model try to find the patterns between X & y (features and labels).",
"_____no_output_____"
]
],
[
[
"X,y",
"_____no_output_____"
],
[
"X.shape",
"_____no_output_____"
],
[
"# Set random seed\ntf.random.set_seed(42)\n\n# Create a model using the Sequential API\nmodel = tf.keras.Sequential([\n tf.keras.layers.Dense(1)\n])\n\n# Compile the model\nmodel.compile(loss=tf.keras.losses.mae, # mae is short for mean absolute error\n optimizer=tf.keras.optimizers.SGD(), # SGD is short for stochastic gradient descent\n metrics=[\"mae\"])\n\n# Fit the model\n# model.fit(X, y, epochs=5) # this will break with TensorFlow 2.7.0+\nmodel.fit(tf.expand_dims(X, axis=-1), y, epochs=5)",
"Epoch 1/5\n1/1 [==============================] - 0s 295ms/step - loss: 11.5048 - mae: 11.5048\nEpoch 2/5\n1/1 [==============================] - 0s 5ms/step - loss: 11.3723 - mae: 11.3723\nEpoch 3/5\n1/1 [==============================] - 0s 5ms/step - loss: 11.2398 - mae: 11.2398\nEpoch 4/5\n1/1 [==============================] - 0s 5ms/step - loss: 11.1073 - mae: 11.1073\nEpoch 5/5\n1/1 [==============================] - 0s 5ms/step - loss: 10.9748 - mae: 10.9748\n"
],
[
"# Check out X and y\nX, y",
"_____no_output_____"
],
[
"# Try and make a prediction using our model\ny_pred = model.predict([17.0])\ny_pred",
"_____no_output_____"
]
],
[
[
"The output is very far off from the actual value. So, Our model is not working correctly. Let's go and improve our model in the next section.",
"_____no_output_____"
],
[
"### Improving our Model\n\nLet's take a look about the three steps when we created the above model.\n\nWe can improve the model by altering the steps we took to create a model.\n\n1. **Creating a model** - here we might add more layers, increase the number of hidden units(all called neurons) within each of the hidden layers, change the activation function of each layer.\n\n2. **Compiling a model** - here we might change the optimization function or perhaps the learning rate of the optimization function.\n\n3. **Fitting a model** - here we might fit a model for more **epochs** (leave it for training longer) or on more data (give the model more examples to learn from)\n",
"_____no_output_____"
]
],
[
[
"# Let's rebuild our model with change in the epoch number\n\n# 1. Create the model\nmodel = tf.keras.Sequential([\n tf.keras.layers.Dense(1)\n \n])\n\n# 2. Compile the model\nmodel.compile(loss = tf.keras.losses.mae,\n optimizer = tf.keras.optimizers.SGD(),\n metrics = [\"mae\"])\n\n# 3. Fit the model to our dataset\n\nmodel.fit(tf.expand_dims(X, axis=-1), y, epochs=100, verbose = 0)",
"_____no_output_____"
],
[
"# Our data \nX , y",
"_____no_output_____"
],
[
"# Let's see if our model's prediction has improved\nmodel.predict([17.0])",
"_____no_output_____"
]
],
[
[
"We got so close the actual value is 27 we performed a better prediction than the last model we trained. But we need to improve much better. \nLet's see what more we change and how close can we get to our actual output",
"_____no_output_____"
]
],
[
[
"# Let's rebuild our model with changing the optimization function to Adam\n\n# 1. Create the model\nmodel = tf.keras.Sequential([\n tf.keras.layers.Dense(1)\n \n])\n\n# 2. Compile the model\nmodel.compile(loss = tf.keras.losses.mae,\n optimizer = tf.keras.optimizers.Adam(lr = 0.0001), # lr stands for learning rate \n metrics = [\"mae\"])\n\n# 3. Fit the model to our dataset\n\nmodel.fit(tf.expand_dims(X, axis=-1), y, epochs=100, verbose = 0)",
"/usr/local/lib/python3.7/dist-packages/keras/optimizer_v2/adam.py:105: UserWarning: The `lr` argument is deprecated, use `learning_rate` instead.\n super(Adam, self).__init__(name, **kwargs)\n"
],
[
"# Prediction of our newly trained model:\nmodel.predict([17.0]) # we are going to predict for the same input value 17 \n",
"_____no_output_____"
]
],
[
[
"Oh..god!! This result went really bad for us. ",
"_____no_output_____"
]
],
[
[
"# Let's rebuild our model by adding one extra hidden layer with 100 units\n\n# 1. Create the model\nmodel = tf.keras.Sequential([\n tf.keras.layers.Dense(100, activation = \"relu\"), # only difference we made\n tf.keras.layers.Dense(1)\n])\n\n# 2. Compile the model\nmodel.compile(loss = \"mae\",\n optimizer = tf.keras.optimizers.SGD(),\n metrics = [\"mae\"])\n\n# 3. Fit the model to our dataset\n\nmodel.fit(tf.expand_dims(X, axis=-1), y, epochs=100, verbose = 0) # verbose will hide the output from epochs",
"_____no_output_____"
],
[
"X , y",
"_____no_output_____"
],
[
"# It's prediction time!\nmodel.predict([17.0])",
"_____no_output_____"
]
],
[
[
"Oh, this should be 27 but this prediction is very far off from our previous prediction. \nIt seems that our previous model did better than this. \n\nEven though we find the values of our loss function are very low than that of our previous model. We still are far away from our label value. \n**Why is that so??** \nThe explanation is our model is overfitting the dataset. That means it is trying to map a function that just fits the already provided examples correctly but it cannot fit the new examples that we are giving. \nSo, the `mae` and `loss value` if not the ultimate metric to check for improving the model. because we need to get less error for new examples that the model has not seen before.\n",
"_____no_output_____"
]
],
[
[
"# Let's rebuild our model by using Adam optimizer\n\n# 1. Create the model\nmodel = tf.keras.Sequential([\n tf.keras.layers.Dense(100, activation = \"relu\"), # only difference we made\n tf.keras.layers.Dense(1)\n])\n\n# 2. Compile the model\nmodel.compile(loss = \"mae\",\n optimizer = tf.keras.optimizers.Adam(), \n metrics = [\"mae\"])\n\n# 3. Fit the model to our dataset\n\nmodel.fit(tf.expand_dims(X, axis=-1), y, epochs=100, verbose = 0)# verbose will hide the epochs output",
"_____no_output_____"
],
[
"model.predict([17.0])",
"_____no_output_____"
]
],
[
[
"Still not better!!",
"_____no_output_____"
]
],
[
[
"# Let's rebuild our model by adding more layers\n\n# 1. Create the model\nmodel = tf.keras.Sequential([\n tf.keras.layers.Dense(100, activation = \"relu\"), \n tf.keras.layers.Dense(100, activation = \"relu\"),\n tf.keras.layers.Dense(100, activation = \"relu\"),# only difference we made\n tf.keras.layers.Dense(1)\n])\n# default value of lr is 0.001\n# 2. Compile the model\nmodel.compile(loss = \"mae\",\n optimizer = tf.keras.optimizers.Adam(lr = 0.01), # lr stands for learning rate \n metrics = [\"mae\"])\n\n# 3. Fit the model to our dataset\n\nmodel.fit(tf.expand_dims(X, axis=-1), y, epochs=100, verbose = 0) # verbose will hide the epochs output",
"/usr/local/lib/python3.7/dist-packages/keras/optimizer_v2/adam.py:105: UserWarning: The `lr` argument is deprecated, use `learning_rate` instead.\n super(Adam, self).__init__(name, **kwargs)\n"
]
],
[
[
"The learning rate is the most important hyperparameter for all the Neural Networks",
"_____no_output_____"
],
[
"### Evaluating our model\n\nIn practice, a typical workflow you'll go through when building a neural network is: \n\n```\nBuild a model -> fit it -> evaluate it -> tweak a model -> fit it -> evaluate it -> tweak it -> fit it \n``` \nCommon ways to improve a deep model: \n* Adding Layers\n* Increase the number of hidden units\n* Change the activation functions\n* Change the optimization function\n* Change the learning rate\n* Fitting on more data\n* Train for longer (more epochs) \n\n**Because we can alter each of these they are called hyperparameters**",
"_____no_output_____"
],
[
"When it comes to evaluation.. there are 3 words you should memorize:\n\n> \"Visualize, Visualize, Visualize\"\n\nIt's a good idea to visualize:\n* The data - what data are working with? What does it look like\n* The model itself - What does our model look like?\n* The training of a model - how does a model perform while it learns?\n* The predictions of the model - how does the prediction of the model line up against the labels(original value)\n",
"_____no_output_____"
]
],
[
[
"# Make a bigger dataset \nX_large = tf.range(-100,100,4)\nX_large",
"_____no_output_____"
],
[
"y_large = X_large + 10\ny_large",
"_____no_output_____"
],
[
"import matplotlib.pyplot as plt\n\nplt.scatter(X_large,y_large)",
"_____no_output_____"
]
],
[
[
"### The 3 sets ...\n\n* **Training set** - The model learns from this data, which is typically 70-80% of the total data you have available.\n\n* **validation set** - The model gets tuned on this data, which is typically 10-15% of the data avaialable.\n\n* **Test set** - The model gets evaluated on this data to test what it has learned. This set is typically 10-15%.\n",
"_____no_output_____"
]
],
[
[
"# Check the length of how many samples we have \nlen(X_large)",
"_____no_output_____"
],
[
"# split the data into train and test sets\n# since the dataset is small we can skip the valdation set \nX_train = X_large[:40]\n\nX_test = X_large[40:]\n\ny_train = y_large[:40]\n\ny_test = y_large[40:]\n\nlen(X_train), len(X_test), len(y_train), len(y_test)",
"_____no_output_____"
]
],
[
[
"### Visualizing the data\n\nNow we've got our data in training and test sets. Let's visualize it.",
"_____no_output_____"
]
],
[
[
"plt.figure(figsize = (10,7))\n# Plot the training data in blue\nplt.scatter(X_train, y_train, c= 'b', label = \"Training data\")\n# Plot the test data in green\nplt.scatter(X_test, y_test, c = \"g\", label = \"Training data\")\nplt.legend();",
"_____no_output_____"
],
[
"# Let's have a look at how to build neural network for our data\n\n# 1. Create the model\nmodel = tf.keras.Sequential([\n tf.keras.layers.Dense(1)\n])\n# default value of lr is 0.001\n# 2. Compile the model\nmodel.compile(loss = \"mae\",\n optimizer = tf.keras.optimizers.SGD(), # lr stands for learning rate \n metrics = [\"mae\"])\n\n# 3. Fit the model to our dataset\n\n#model.fit(tf.expand_dims(X_train, axis=-1), y_train, epochs=100)",
"_____no_output_____"
]
],
[
[
"Let's visualize it before fitting the model ",
"_____no_output_____"
]
],
[
[
"model.summary()",
"_____no_output_____"
]
],
[
[
"model.summary() doesn't work without building the model or fitting the model",
"_____no_output_____"
]
],
[
[
"X[0], y[0]",
"_____no_output_____"
],
[
"# Let's create a model which builds automatically by defining the input_shape arguments\ntf.random.set_seed(42)\n\n# Create a model(same as above)\nmodel = tf.keras.Sequential([\n tf.keras.layers.Dense(1, input_shape = [1]) # input_shape is 1 refer above code cell\n])\n\n# Compile the model\nmodel.compile(loss= \"mae\",\n optimizer = tf.keras.optimizers.SGD(),\n metrics = [\"mae\"])\n",
"_____no_output_____"
],
[
"model.summary()",
"Model: \"sequential_31\"\n_________________________________________________________________\n Layer (type) Output Shape Param # \n=================================================================\n dense_86 (Dense) (None, 1) 2 \n \n=================================================================\nTotal params: 2\nTrainable params: 2\nNon-trainable params: 0\n_________________________________________________________________\n"
]
],
[
[
"* **Total params** - total number of parameters in the model.\n* **Trainable parameters**- these are the parameters (patterns) the model can update as it trains.\n* **Non-Trainable parameters** - these parameters aren't updated during training(this is typical when you have paramters from other models during **transfer learning**)",
"_____no_output_____"
]
],
[
[
"# Let's have a look at how to build neural network for our data\n\n# 1. Create the model\nmodel = tf.keras.Sequential([\n tf.keras.layers.Dense(10, input_shape = [1], name= \"input_layer\"),\n tf.keras.layers.Dense(1, name = \"output_layer\")\n], name = \"model_1\")\n\n# 2. Compile the model\nmodel.compile(loss = \"mae\",\n optimizer = tf.keras.optimizers.SGD(), # lr stands for learning rate \n metrics = [\"mae\"])",
"_____no_output_____"
],
[
"model.summary()",
"Model: \"model_1\"\n_________________________________________________________________\n Layer (type) Output Shape Param # \n=================================================================\n input_layer (Dense) (None, 10) 20 \n \n output_layer (Dense) (None, 1) 11 \n \n=================================================================\nTotal params: 31\nTrainable params: 31\nNon-trainable params: 0\n_________________________________________________________________\n"
]
],
[
[
"We have changed the layer names and added our custom model name.",
"_____no_output_____"
]
],
[
[
"from tensorflow.keras.utils import plot_model\nplot_model(model = model, to_file = 'model1.png', show_shapes = True)",
"_____no_output_____"
],
[
"# Let's have a look at how to build neural network for our data\n\n# 1. Create the model\nmodel = tf.keras.Sequential([\n tf.keras.layers.Dense(100, activation = \"relu\"), \n tf.keras.layers.Dense(100, activation = \"relu\"),\n tf.keras.layers.Dense(100, activation = \"relu\"),# only difference we made\n tf.keras.layers.Dense(1)\n], name)\n# default value of lr is 0.001\n# 2. Compile the model\nmodel.compile(loss = \"mae\",\n optimizer = tf.keras.optimizers.Adam(lr = 0.01), # lr stands for learning rate \n metrics = [\"mae\"])\n\n# 3. Fit the model to our dataset\n\nmodel.fit(tf.expand_dims(X_train, axis=-1), y_train, epochs=100, verbose = 0)",
"_____no_output_____"
],
[
"model.predict(X_test)",
"_____no_output_____"
]
],
[
[
"wow, we are so close!!!",
"_____no_output_____"
]
],
[
[
"model.summary()",
"Model: \"model_1\"\n_________________________________________________________________\n Layer (type) Output Shape Param # \n=================================================================\n input_layer (Dense) (None, 10) 20 \n \n output_layer (Dense) (None, 1) 11 \n \n=================================================================\nTotal params: 31\nTrainable params: 31\nNon-trainable params: 0\n_________________________________________________________________\n"
],
[
"from tensorflow.keras.utils import plot_model\n\nplot_model(model = model, to_file = 'model.png', show_shapes = True)",
"_____no_output_____"
]
],
[
[
"### Visualizing our model's predictions\n\nTo visualize predictions, it's a good idea to plot them against the ground truth labels.\n\nOften you'll see this in the form of `y_test` or `y_true` versus `y_pred`",
"_____no_output_____"
]
],
[
[
"# Set random seed\ntf.random.set_seed(42)\n\n# Create a model (same as above)\nmodel = tf.keras.Sequential([\n tf.keras.layers.Dense(10, input_shape = [1], name = \"input_layer\"),\n tf.keras.layers.Dense(1, name = \"output_layer\") # define the input_shape to our model\n], name = \"revised_model_1\")\n\n# Compile model (same as above)\nmodel.compile(loss=tf.keras.losses.mae,\n optimizer=tf.keras.optimizers.SGD(),\n metrics=[\"mae\"])",
"_____no_output_____"
],
[
"model.summary()",
"Model: \"revised_model_1\"\n_________________________________________________________________\n Layer (type) Output Shape Param # \n=================================================================\n input_layer (Dense) (None, 10) 20 \n \n output_layer (Dense) (None, 1) 11 \n \n=================================================================\nTotal params: 31\nTrainable params: 31\nNon-trainable params: 0\n_________________________________________________________________\n"
],
[
"model.fit(X_train, y_train, epochs=100, verbose=0)",
"_____no_output_____"
],
[
"model.summary()",
"Model: \"revised_model_1\"\n_________________________________________________________________\n Layer (type) Output Shape Param # \n=================================================================\n input_layer (Dense) (None, 10) 20 \n \n output_layer (Dense) (None, 1) 11 \n \n=================================================================\nTotal params: 31\nTrainable params: 31\nNon-trainable params: 0\n_________________________________________________________________\n"
],
[
"# Make some predictions\ny_pred = model.predict(X_test)\ntf.constant(y_pred)",
"_____no_output_____"
]
],
[
[
"These are our predictions! ",
"_____no_output_____"
]
],
[
[
"y_test",
"_____no_output_____"
]
],
[
[
"These are the ground truth labels!",
"_____no_output_____"
]
],
[
[
"plot_model(model, show_shapes=True)",
"_____no_output_____"
]
],
[
[
"**Note:** IF you feel like you're going to reuse some kind of functionality in future,\nit's a good idea to define a function so that we can reuse it whenever we need.",
"_____no_output_____"
]
],
[
[
"#Let's create a plotting function\ndef plot_predictions(train_data= X_train,\n train_labels = y_train,\n test_data = X_test,\n test_labels =y_test,\n predictions = y_pred):\n \"\"\"\n\n Plots training data, test data and compares predictions to ground truth labels\n\n \"\"\"\n plt.figure(figsize = (10,7))\n # Plot training data in blue\n plt.scatter(train_data, train_labels, c= \"b\", label = \"Training data\")\n # Plot testing data in green\n plt.scatter(test_data, test_labels, c= \"g\", label = \"Testing data\")\n # Plot model's predictions in red\n plt.scatter(test_data, predictions, c= \"r\", label = \"Predictions\")\n # Show legends\n plt.legend();\n",
"_____no_output_____"
],
[
"plot_predictions(train_data=X_train,\n train_labels=y_train,\n test_data=X_test,\n test_labels=y_test,\n predictions=y_pred)",
"_____no_output_____"
]
],
[
[
"We tuned our model very well this time. The predictions are really close to the actual values.",
"_____no_output_____"
],
[
"### Evaluating our model's predictions with regression evaluation metrics\n\nDepending on the problem you're working on, there will be different evaluation metrics to evaluate your model's performance.\n\nSince, we're working on a regression, two of the main metrics:\n\n* **MAE** - mean absolute error, \"on average, how wrong id each of my model's predictions\"\n * TensorFlow code: `tf.keras.losses.MAE()`\n * or `tf.metrics.mean_absolute_error()`\n $$ MAE = \\frac{Σ_{i=1}^{n} |y_i - x_i| }{n} $$\n* **MSE** - mean square error, \"square of the average errors\"\n * `tf.keras.losses.MSE()`\n * `tf.metrics.mean_square_error()`\n $$ MSE = \\frac{1}{n} Σ_{i=1}^{n}(Y_i - \\hat{Y_i})^2$$\n\n $\\hat{Y_i}$ is the prediction our model makes.\n $Y_i$ is the label value.\n\n* **Huber** - Combination of MSE and MAE, Less sensitive to outliers than MSE.\n * `tf.keras.losses.Huber()`\n\n",
"_____no_output_____"
]
],
[
[
"# Evaluate the model on test set\nmodel.evaluate(X_test, y_test)",
"1/1 [==============================] - 0s 118ms/step - loss: 3.1969 - mae: 3.1969\n"
],
[
"# calculate the mean absolute error\nmae = tf.metrics.mean_absolute_error(y_true = y_test,\n y_pred = tf.constant(y_pred))\nmae",
"_____no_output_____"
]
],
[
[
"We got the metric values wrong..why did this happen??",
"_____no_output_____"
]
],
[
[
"tf.constant(y_pred)",
"_____no_output_____"
],
[
"y_test",
"_____no_output_____"
]
],
[
[
"Notice that the shape of `y_pred` is (10,1) and the shape of `y_test` is (10,) \nThey might seem the same but they are not of the same shape.\nLet's reshape the tensor to make the shapes equal.",
"_____no_output_____"
]
],
[
[
"tf.squeeze(y_pred)",
"_____no_output_____"
],
[
"# Calculate the mean absolute error\nmae = tf.metrics.mean_absolute_error(y_true = y_test,\n y_pred = tf.squeeze(y_pred))\n\nmae",
"_____no_output_____"
]
],
[
[
"Now,we got our metric value. The mean absolute error of our model is 3.1969407. \nNow, let's calculate the mean squared error and see how that goes.",
"_____no_output_____"
]
],
[
[
"# Calculate the mean squared error\nmse = tf.metrics.mean_squared_error(y_true = y_test,\n y_pred = tf.squeeze(y_pred))\n\nmse",
"_____no_output_____"
]
],
[
[
"Our mean squared error is 13.070143. Remember, the mean squared error squares the error for every example in the test set and averages the values. So, generally, the mse is largeer than mae. \nWhen larger errors are more significant than smaller errors, then it is best to use mse. \nMAE can be used as a great starter metric for any regression problem. \nWe can also try Huber and see how that goes.",
"_____no_output_____"
]
],
[
[
"# Calculate the Huber metric for our model\nhuber_metric = tf.losses.huber(y_true = y_test,\n y_pred = tf.squeeze(y_pred))\n\nhuber_metric",
"_____no_output_____"
],
[
"# Make some functions to reuse MAE and MSE and also Huber\ndef mae(y_true, y_pred):\n return tf.metrics.mean_absolute_error(y_true = y_test,\n y_pred = tf.squeeze(y_pred))\n \ndef mse(y_true, y_pred):\n return tf.metrics.mean_squared_error(y_true = y_test,\n y_pred = tf.squeeze(y_pred))\n \ndef huber(y_true, y_pred):\n return tf.losses.huber(y_true = y_test,\n y_pred = tf.squeeze(y_pred))",
"_____no_output_____"
]
],
[
[
"### Running experiments to improve our model \n\n```\nBuild a model -> fit it -> evaluate it -> tweak a model -> fit it -> evaluate it -> tweak it -> fit it \n``` \n\n1. Get more data - get more examples for your model to train on(more oppurtunities to learn patterns or relationships between features and labels). \n2. Make your mode larger(using a more complex model) - this might come in the form of more layeres or more hidden unites in each layer. \n3. Train for longer - give your model more of a chance to find patterns in the data. \n\nLet's do a few modelling experiments: \n1. `model_1` - same as original model, 1 layer, trained for 100 epochs. \n2. `model_2` - 2 layers, trained for 100 epochs \n3. `model_3` - 2 layers, trained for 500 epochs. \n\nYou can design more experiments too to make the model more better\n\n\n\n**Build `Model_1`**",
"_____no_output_____"
]
],
[
[
"X_train, y_train",
"_____no_output_____"
],
[
"# Set random seed\ntf.random.set_seed(42)\n\n# 1. Create the model\n\nmodel_1 = tf.keras.Sequential([\n tf.keras.layers.Dense(1, input_shape = [1])\n], name = \"Model_1\")\n\n# 2. Compile the model\nmodel_1.compile(loss = tf.keras.losses.mae,\n optimizer = tf.keras.optimizers.SGD(),\n metrics = [\"mae\"])\n\n# 3. Fit the model\nmodel_1.fit(X_train, y_train ,epochs = 100, verbose = 0)",
"_____no_output_____"
],
[
"model_1.summary()",
"Model: \"Model_1\"\n_________________________________________________________________\n Layer (type) Output Shape Param # \n=================================================================\n dense_91 (Dense) (None, 1) 2 \n \n=================================================================\nTotal params: 2\nTrainable params: 2\nNon-trainable params: 0\n_________________________________________________________________\n"
],
[
"# Make and plot the predictions for model_1\ny_preds_1 = model_1.predict(X_test)\nplot_predictions(predictions = y_preds_1)",
"_____no_output_____"
],
[
"# Calculate model_1 evaluation metrics\nmae_1 = mae(y_test, y_preds_1)\nmse_1 = mse(y_test, y_preds_1)\nmae_1, mse_1",
"_____no_output_____"
]
],
[
[
"**Build `Model_2`** \n\n* 2 dense layers, trained for 100 epochs",
"_____no_output_____"
]
],
[
[
"# Set random seed\ntf.random.set_seed(42)\n\n# 1. Create the model\n\nmodel_2 = tf.keras.Sequential([\n tf.keras.layers.Dense(10, input_shape =[1]),\n tf.keras.layers.Dense(1)\n], name = \"model_2\")\n\n# 2. Compile the model\nmodel_2.compile(loss = tf.keras.losses.mae,\n optimizer = tf.keras.optimizers.SGD(),\n metrics = [\"mse\"]) # Let's build this model with mse as eval metric.\n\n# 3. Fit the model\nmodel_2.fit(X_train, y_train ,epochs = 100, verbose = 0)",
"_____no_output_____"
],
[
"model_2.summary()",
"Model: \"model_2\"\n_________________________________________________________________\n Layer (type) Output Shape Param # \n=================================================================\n dense_92 (Dense) (None, 10) 20 \n \n dense_93 (Dense) (None, 1) 11 \n \n=================================================================\nTotal params: 31\nTrainable params: 31\nNon-trainable params: 0\n_________________________________________________________________\n"
],
[
"# Make and plot predictions of model_2\ny_preds_2 = model_2.predict(X_test)\n\nplot_predictions(predictions = y_preds_2)",
"_____no_output_____"
]
],
[
[
"Yeah,we improved this model very much than the previous one. \nIf you want to compare with previous one..scroll up and see the plot_predictions of\nprevious one and compare it with this one.",
"_____no_output_____"
]
],
[
[
"# Calculate the model_2 evaluation metrics\nmae_2 = mae(y_test, y_preds_2)\nmse_2 = mse(y_test, y_preds_2)\nmae_2, mse_2",
"_____no_output_____"
]
],
[
[
"**Build `Model_3`**\n\n* 2 layers, trained for 500 epochs",
"_____no_output_____"
]
],
[
[
"# Set random seed\ntf.random.set_seed(42)\n\n# 1. Create the model\n\nmodel_3 = tf.keras.Sequential([\n tf.keras.layers.Dense(10, input_shape =[1]),\n tf.keras.layers.Dense(1)\n], name = \"model_3\")\n\n# 2. Compile the model\nmodel_3.compile(loss = tf.keras.losses.mae,\n optimizer = tf.keras.optimizers.SGD(),\n metrics = [\"mae\"]) # Let's build this model with mse as eval metric.\n\n# 3. Fit the model\nmodel_2.fit(X_train, y_train ,epochs = 500, verbose = 0)",
"_____no_output_____"
],
[
"# Make and plot some predictions\ny_preds_3 = model_3.predict(X_test)\nplot_predictions(predictions = y_preds_3)",
"_____no_output_____"
]
],
[
[
"This is even terrible performance than the first model. we have actually made the model worse. WHY?? \n\nWe, overfitted the model too much because we trained it for much longer than we are supposed to.",
"_____no_output_____"
]
],
[
[
"# Calculate the model_3 evaluation metrics\nmae_3 = mae(y_test, y_preds_3)\nmse_3 = mse(y_test, y_preds_3)\nmae_3, mse_3",
"_____no_output_____"
]
],
[
[
"whoaa, the error is extremely high. I think the best of our models is `model_2`",
"_____no_output_____"
],
[
"The Machine Learning practitioner's motto:\n\n`Experiment, experiment, experiment`",
"_____no_output_____"
],
[
"**Note:** You want to start with small experiments(small models) and make sure they work and then increase their scale when neccessary.",
"_____no_output_____"
],
[
"### Comparing the results of our experiments \nWe've run a few experiments, let's compare the results now.",
"_____no_output_____"
]
],
[
[
"# Let's compare our models'c results using pandas dataframe:\nimport pandas as pd\n\nmodel_results = [[\"model_1\", mae_1.numpy(), mse_1.numpy()],\n [\"model_2\", mae_2.numpy(), mse_2.numpy()],\n [\"model_3\", mae_3.numpy(), mse_3.numpy()]]\n\nall_results = pd.DataFrame(model_results, columns =[\"model\", \"mae\", \"mse\"])\nall_results",
"_____no_output_____"
]
],
[
[
"It looks like model_2 performed done the best. Let's look at what is model_2",
"_____no_output_____"
]
],
[
[
"model_2.summary()",
"Model: \"model_2\"\n_________________________________________________________________\n Layer (type) Output Shape Param # \n=================================================================\n dense_92 (Dense) (None, 10) 20 \n \n dense_93 (Dense) (None, 1) 11 \n \n=================================================================\nTotal params: 31\nTrainable params: 31\nNon-trainable params: 0\n_________________________________________________________________\n"
]
],
[
[
"This is the model that has done the best on our dataset.",
"_____no_output_____"
],
[
"**Note:** One of your main goals should be to minimize the time between your experiments. The more experiments you do, the more things you will figure out which don't work and in turn, get closer to figuring out what does work. Remeber, the machine learning pracitioner's motto : \"experiment, experiment, experiment\".",
"_____no_output_____"
],
[
"## Tracking your experiments:\n\nOne really good habit of machine learning modelling is to track the results of your experiments. \n\nAnd when doing so, it can be tedious if you are running lots of experiments. \n\nLuckily, there are tools to help us! \n\n**Resources:** As you build more models, you'll want to look into using: \n\n* TensorBoard - a component of TensorFlow library to help track modelling experiments. It is integrated into the TensorFlow library.\n\n* Weights & Biases - A tool for tracking all kinds of machine learning experiments (it plugs straight into tensorboard). ",
"_____no_output_____"
],
[
"## Saving our models \n\nSaving our models allows us to use them outside of Google Colab(or wherever they were trained) such as in a web application or a mobile app. \n\nThere are two main formats we can save our model: \n\n1. The SavedModel format\n2. The HDF5 format",
"_____no_output_____"
],
[
"`model.save()` allows us to save the model and we can use it again to do add things to the model after reloading it. ",
"_____no_output_____"
]
],
[
[
"# Save model using savedmodel format\nmodel_2.save(\"best_model_SavedModel_format\")",
"INFO:tensorflow:Assets written to: best_model_SavedModel_format/assets\n"
]
],
[
[
"If we are planning to use this model inside the tensorflow framework. we will be better off using the `SavedModel` format. But if we are planning to export the model else where and use it outside the tensorflow framework use the HDF5 format.",
"_____no_output_____"
]
],
[
[
"# Save model using HDF5 format\nmodel_2.save(\"best_model_HDF5_format.h5\")",
"_____no_output_____"
]
],
[
[
"Saving a model with SavedModel format will give us a folder with some files regarding our model. \nSaving a model with HDF5 format will give us just one file with our model.",
"_____no_output_____"
],
[
"### Loading in a saved model",
"_____no_output_____"
]
],
[
[
"# Load in the SavedModel format model\nloaded_SavedModel_format = tf.keras.models.load_model(\"/content/best_model_SavedModel_format\")\nloaded_SavedModel_format.summary()",
"Model: \"model_2\"\n_________________________________________________________________\n Layer (type) Output Shape Param # \n=================================================================\n dense_92 (Dense) (None, 10) 20 \n \n dense_93 (Dense) (None, 1) 11 \n \n=================================================================\nTotal params: 31\nTrainable params: 31\nNon-trainable params: 0\n_________________________________________________________________\n"
],
[
"# Let's check is that the same thing as model_2\nmodel_2.summary()",
"Model: \"model_2\"\n_________________________________________________________________\n Layer (type) Output Shape Param # \n=================================================================\n dense_92 (Dense) (None, 10) 20 \n \n dense_93 (Dense) (None, 1) 11 \n \n=================================================================\nTotal params: 31\nTrainable params: 31\nNon-trainable params: 0\n_________________________________________________________________\n"
],
[
"# Compare the model_2 predictions with SavedModel format model predictions\nmodel_2_preds = model_2.predict(X_test)\nloaded_SavedModel_format_preds = loaded_SavedModel_format.predict(X_test)\nmodel_2_preds == loaded_SavedModel_format_preds",
"_____no_output_____"
],
[
"mae(y_true = y_test, y_pred = model_2_preds) == mae(y_true = y_test, y_pred = loaded_SavedModel_format_preds)",
"_____no_output_____"
],
[
"# Load in a model using the .hf format\nloaded_h5_model = tf.keras.models.load_model(\"/content/best_model_HDF5_format.h5\")\nloaded_h5_model.summary()",
"Model: \"model_2\"\n_________________________________________________________________\n Layer (type) Output Shape Param # \n=================================================================\n dense_92 (Dense) (None, 10) 20 \n \n dense_93 (Dense) (None, 1) 11 \n \n=================================================================\nTotal params: 31\nTrainable params: 31\nNon-trainable params: 0\n_________________________________________________________________\n"
],
[
"model_2.summary()",
"Model: \"model_2\"\n_________________________________________________________________\n Layer (type) Output Shape Param # \n=================================================================\n dense_92 (Dense) (None, 10) 20 \n \n dense_93 (Dense) (None, 1) 11 \n \n=================================================================\nTotal params: 31\nTrainable params: 31\nNon-trainable params: 0\n_________________________________________________________________\n"
]
],
[
[
"Yeah the loading of .hf format model matched with our original mode_2 format. \nSo, our model loading worked correctly.",
"_____no_output_____"
]
],
[
[
"# Check to see if loaded .hf model predictions match model_2\nmodel_2_preds = model_2.predict(X_test)\nloaded_h5_model_preds = loaded_h5_model.predict(X_test)\nmodel_2_preds == loaded_h5_model_preds",
"_____no_output_____"
]
],
[
[
"### Download a model(or any other file) from google colab\n\nIf you want to download your files from Google Colab: \n\n1. you can go to the files tab and right click on the file you're after and click download. \n\n2. Use code(see the cell below).\n\n3. You can save it to google drive by connecting to google drive and copying it there.\n\n\n",
"_____no_output_____"
]
],
[
[
"# Download a file from Google Colab\nfrom google.colab import files\nfiles.download(\"/content/best_model_HDF5_format.h5\")",
"_____no_output_____"
],
[
"# Save a file from Google Colab to Google Drive(requires mounting google drive)\n!cp /content/best_model_HDF5_format.h5 /content/drive/MyDrive/tensor-flow-deep-learning\n",
"_____no_output_____"
],
[
"!ls /content/drive/MyDrive/tensor-flow-deep-learning",
"best_model_HDF5_format.h5\n"
]
],
[
[
"We have saved our model to our google drive !!!",
"_____no_output_____"
],
[
"## A larger example \n\nWe take a larger dataset to do create a regression model. The model we do is insurance forecast by using linear regression available from kaggle [Medical Cost Personal Datasets](https://www.kaggle.com/mirichoi0218/insurance)",
"_____no_output_____"
]
],
[
[
"# Import required libraries\nimport tensorflow as tf\nimport pandas as pd\nimport matplotlib.pyplot as plt",
"_____no_output_____"
],
[
"# Read in the insurance data set\ninsurance = pd.read_csv(\"https://raw.githubusercontent.com/stedy/Machine-Learning-with-R-datasets/master/insurance.csv\")\ninsurance",
"_____no_output_____"
]
],
[
[
"This is a quite bigger dataset than the one we have previously worked with.",
"_____no_output_____"
]
],
[
[
"# one hot encoding on a pandas dataframe\ninsurance_one_hot = pd.get_dummies(insurance) \ninsurance_one_hot.head()",
"_____no_output_____"
],
[
"# Create X & y values (features and labels)\nX = insurance_one_hot.drop(\"charges\", axis =1)\ny = insurance_one_hot[\"charges\"]",
"_____no_output_____"
],
[
"# View X\nX.head()",
"_____no_output_____"
],
[
"# View y\ny.head()",
"_____no_output_____"
],
[
"# Create training and test sets\nfrom sklearn.model_selection import train_test_split\nX_train, X_test, y_train, y_test = train_test_split(X, y,test_size = 0.2, random_state = 42)\nlen(X), len(X_train), len(X_test)",
"_____no_output_____"
],
[
"X_train",
"_____no_output_____"
],
[
"insurance[\"smoker\"] , insurance[\"sex\"]",
"_____no_output_____"
],
[
"# Build a neural network (sort of like model_2 above)\ntf.random.set_seed(42)\n\n# 1. Create a model\ninsurance_model = tf.keras.Sequential([\n tf.keras.layers.Dense(10),\n tf.keras.layers.Dense(1) \n])\n\n# 2. Compile the model\ninsurance_model.compile(loss = tf.keras.losses.mae,\n optimizer = tf.keras.optimizers.SGD(),\n metrics = [\"mae\"])\n\n#3. Fit the model\ninsurance_model.fit(X_train, y_train,epochs = 100, verbose = 0)",
"_____no_output_____"
],
[
"# Check the results of the insurance model on the test data\ninsurance_model.evaluate(X_test,y_test)",
"9/9 [==============================] - 0s 2ms/step - loss: 7023.3291 - mae: 7023.3291\n"
],
[
"y_train.median(), y_train.mean()",
"_____no_output_____"
]
],
[
[
"Right now it looks like our model is not performing well, lets try and improve it. \n\nTo try and improve our model, we'll run 2 experiments: \n1. Add an extra layer with more hidden units and use the Adam optimizer\n2. Train for longer (like 200 epochs)\n3. We can also do our custom experiments to improve it.",
"_____no_output_____"
]
],
[
[
"# Set random seed\ntf.random.set_seed(42)\n\n# 1. Create the model \ninsurance_model_2 = tf.keras.Sequential([\n tf.keras.layers.Dense(100),\n tf.keras.layers.Dense(10),\n tf.keras.layers.Dense(1) \n],name = \"insurace_model_2\")\n\n# 2. Compile the model\ninsurance_model_2.compile(loss = tf.keras.losses.mae,\n optimizer = tf.keras.optimizers.Adam(),\n metrics = [\"mae\"])\n\n# 3. Fit the model\ninsurance_model_2.fit(X_train, y_train, epochs = 100, verbose = 0)",
"_____no_output_____"
],
[
"insurance_model_2.evaluate(X_test, y_test)",
"9/9 [==============================] - 0s 2ms/step - loss: 4924.3477 - mae: 4924.3477\n"
],
[
"# Set random seed\ntf.random.set_seed(42)\n\n# 1. Create the model \ninsurance_model_3 = tf.keras.Sequential([\n tf.keras.layers.Dense(100),\n tf.keras.layers.Dense(10),\n tf.keras.layers.Dense(1) \n],name = \"insurace_model_2\")\n\n# 2. Compile the model\ninsurance_model_3.compile(loss = tf.keras.losses.mae,\n optimizer = tf.keras.optimizers.Adam(),\n metrics = [\"mae\"])\n\n# 3. Fit the model\nhistory = insurance_model_3.fit(X_train, y_train, epochs = 200, verbose = 0)",
"_____no_output_____"
],
[
"# Evaluate our third model\ninsurance_model_3.evaluate(X_test, y_test)",
"9/9 [==============================] - 0s 2ms/step - loss: 3491.2961 - mae: 3491.2961\n"
],
[
"# Plot history (also known as a loss curve or a training curve)\npd.DataFrame(history.history).plot()\nplt.ylabel(\"loss\")\nplt.xlabel(\"epochs\")\nplt.title(\"Training curve of our model\")",
"_____no_output_____"
]
],
[
[
"**Question:** How long should you train for? \n\nIt depends, It really depends on problem you are working on. However, many people have asked this question before, so TensorFlow has a solution!, It is called [EarlyStopping callback](https://www.tensorflow.org/api_docs/python/tf/keras/callbacks/EarlyStopping), which is a TensorFlow component you can add to your model to stop training once it stops improving a certain metric.\n\n",
"_____no_output_____"
],
[
"## Preprocessing data (normalization and standardization)",
"_____no_output_____"
],
[
"Short review of our modelling steps in TensorFlow: \n1. Get data ready(turn into tensors)\n2. Build or pick a pretrained model (to suit your problem)\n3. Fit the model to the data and make a prediction.\n4. Evaluate the model.\n5. Imporve through experimentation.\n6. Save and reload your trained models.\n\nwe are going to focus on the step 1 to make our data set more rich for training.\nsome steps involved in getting data ready: \n1. Turn all data into numbers(neural networks can't handle strings).\n2. Make sure all of your tensors are the right shape.\n3. Scale features(normalize or standardize, neural networks tend to prefer normalization) -- this is the one thing we haven't done while preparing our data. \n\n**If you are not sure on which to use for scaling, you could try both and see which perform better**",
"_____no_output_____"
]
],
[
[
"# Import required libraries\nimport tensorflow as tf\nimport pandas as pd\nimport matplotlib.pyplot as plt",
"_____no_output_____"
],
[
"# Read in the insurance dataframe\ninsurance = pd.read_csv(\"https://raw.githubusercontent.com/stedy/Machine-Learning-with-R-datasets/master/insurance.csv\")\ninsurance",
"_____no_output_____"
]
],
[
[
"To prepare our data, we can borrow few classes from Scikit-Learn",
"_____no_output_____"
]
],
[
[
"from sklearn.compose import make_column_transformer\nfrom sklearn.preprocessing import MinMaxScaler, OneHotEncoder\nfrom sklearn.model_selection import train_test_split",
"_____no_output_____"
]
],
[
[
"**Feature Scaling**:\n\n\n| **Scaling type** | **what it does** | **Scikit-Learn Function** | **when to use** | \n| --- | --- | --- | --- | \n| scale(refers to as normalization) | converts all values to between 0 and 1 whilst preserving the original distribution | `MinMaxScaler` | Use as default scaler with neural networks | \n| Standarization | Removes the mean and divides each value by the standard deviation | `StandardScaler` | Transform a feature to have close to normal distribution |\n ",
"_____no_output_____"
]
],
[
[
"#Create a column transformer\nct = make_column_transformer(\n (MinMaxScaler(), [\"age\", \"bmi\", \"children\"]), # Turn all values in these columns between 0 and 1\n (OneHotEncoder(handle_unknown = \"ignore\"), [\"sex\", \"smoker\", \"region\"])\n)\n\n# Create our X and Y values\n# because we reimported our dataframe\nX = insurance.drop(\"charges\", axis = 1)\ny = insurance[\"charges\"]\n\n# Build our train and test set\nX_train, X_test, y_train, y_test = train_test_split(X,y, test_size = 0.2, random_state = 42)\n\n# Fit the column transformer to our training data (only training data)\nct.fit(X_train)\n\n# Transform training and test data with normalization(MinMaxScaler) and OneHotEncoder\nX_train_normal = ct.transform(X_train)\nX_test_normal = ct.transform(X_test)",
"_____no_output_____"
],
[
"# What does our data look like now??\nX_train.loc[0]",
"_____no_output_____"
],
[
"X_train_normal[0], X_train_normal[12], X_train_normal[78] \n# we have turned all our data into numerical encoding and aso normalized the data\n",
"_____no_output_____"
],
[
"X_train.shape, X_train_normal.shape",
"_____no_output_____"
]
],
[
[
"Beautiful! our data has been normalized and One hot encoded. Let's build Neural Network on it and see how it goes.",
"_____no_output_____"
]
],
[
[
"# Build a neural network model to fit on our normalized data\ntf.random.set_seed(42)\n\n# 1. Create the model\ninsurance_model_4 = tf.keras.Sequential([\n tf.keras.layers.Dense(100),\n tf.keras.layers.Dense(10),\n tf.keras.layers.Dense(1) \n])\n\n# 2. Compile the model\ninsurance_model_4.compile(loss = tf.keras.losses.mae,\n optimizer = tf.keras.optimizers.Adam(),\n metrics = [\"mae\"])\n\n# 3. Fit the model\nhistory = insurance_model_4.fit(X_train_normal, y_train, epochs= 100, verbose = 0)",
"_____no_output_____"
],
[
"# Evaluate our insurance model trained on normalized data\ninsurance_model_4.evaluate(X_test_normal, y_test)",
"9/9 [==============================] - 0s 2ms/step - loss: 3438.7844 - mae: 3438.7844\n"
],
[
"insurance_model_4.summary()",
"Model: \"sequential_33\"\n_________________________________________________________________\n Layer (type) Output Shape Param # \n=================================================================\n dense_110 (Dense) (None, 100) 1200 \n \n dense_111 (Dense) (None, 10) 1010 \n \n dense_112 (Dense) (None, 1) 11 \n \n=================================================================\nTotal params: 2,221\nTrainable params: 2,221\nNon-trainable params: 0\n_________________________________________________________________\n"
],
[
"pd.DataFrame(history.history).plot()\nplt.ylabel(\"loss\")\nplt.xlabel(\"epochs\")\nplt.title(\"Training curve of insurance_model_4\")",
"_____no_output_____"
]
],
[
[
"Let's just plot some graphs. Since we have use them the least in this notebook.",
"_____no_output_____"
]
],
[
[
"X[\"age\"].plot(kind = \"hist\")",
"_____no_output_____"
],
[
"X[\"bmi\"].plot(kind = \"hist\")",
"_____no_output_____"
],
[
"X[\"children\"].value_counts()",
"_____no_output_____"
]
],
[
[
"## **External Resources:** \n* [MIT introduction deep learning lecture 1](https://youtu.be/njKP3FqW3Sk)\n* [Kaggle's datasets](https://www.kaggle.com/data)\n* [Lion Bridge's collection of datasets](https://lionbridge.ai/datasets/)",
"_____no_output_____"
],
[
"## Bibliography:\n\n* [Learn TensorFlow and Deep Learning fundamentals with Python (code-first introduction) Part 1/2](https://www.youtube.com/watch?v=tpCFfeUEGs8&list=RDCMUCr8O8l5cCX85Oem1d18EezQ&start_radio=1&rv=tpCFfeUEGs8&t=3)\n\n* [Medical cost personal dataset](https://www.kaggle.com/mirichoi0218/insurance)\n\n* [TensorFlow documentation](https://www.tensorflow.org/api_docs/python/tf)\n\n* [TensorFlow and Deep learning Daniel Bourke GitHub Repo](https://github.com/mrdbourke/tensorflow-deep-learning)",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown"
] |
[
[
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code"
],
[
"markdown",
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown",
"markdown"
]
] |
cbd889e2c9c5141f1fb684af2cc38d26ab86b1b2
| 276,536 |
ipynb
|
Jupyter Notebook
|
.ipynb_checkpoints/pandas-basico-analise-carros-Copy1-checkpoint.ipynb
|
andersoncordeiro/Learn-Pandas
|
a4e43dba9e90deba994d590c912ff44f9455ad3e
|
[
"MIT"
] | 2 |
2021-11-05T02:41:56.000Z
|
2021-12-09T14:47:03.000Z
|
.ipynb_checkpoints/pandas-basico-analise-carros-Copy1-checkpoint.ipynb
|
andersoncordeiro/Learn-Pandas
|
a4e43dba9e90deba994d590c912ff44f9455ad3e
|
[
"MIT"
] | null | null | null |
.ipynb_checkpoints/pandas-basico-analise-carros-Copy1-checkpoint.ipynb
|
andersoncordeiro/Learn-Pandas
|
a4e43dba9e90deba994d590c912ff44f9455ad3e
|
[
"MIT"
] | 4 |
2021-03-04T15:11:11.000Z
|
2021-03-18T23:30:56.000Z
| 64.565958 | 23,896 | 0.664008 |
[
[
[
"# Análise de Dados com Python\n\n\nNeste notebook, utilizaremos dados de automóveis para analisar a influência das características de um carro em seu preço, tentando posteriormente prever qual será o preço de venda de um carro. Utilizaremos como fonte de dados um arquivo .csv com dados já tratados em outro notebook. Caso você tenha dúvidas quanto a como realizar o tratamento dos dados, dê uma olhada no meu repositório Learn-Pandas",
"_____no_output_____"
]
],
[
[
"import pandas as pd\nimport numpy as np",
"_____no_output_____"
],
[
"df = pd.read_csv('clean_auto_df.csv')\ndf.head()",
"_____no_output_____"
]
],
[
[
"<h4> Utilizando visualização de dados para verificar padrões de características individuais</h4>",
"_____no_output_____"
]
],
[
[
"# Importando as bibliotecas \"Matplotlib\" e \"Seaborn\n# utilizando \"%matplotlib inline\" para plotar o gráfico dentro do notebook.\nimport matplotlib.pyplot as plt\nimport seaborn as sns\n%matplotlib inline ",
"\nBad key savefig.frameon in file /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle, line 421 ('savefig.frameon : True')\nYou probably need to get an updated matplotlibrc file from\nhttps://github.com/matplotlib/matplotlib/blob/v3.3.4/matplotlibrc.template\nor from the matplotlib source distribution\n\nBad key verbose.level in file /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle, line 472 ('verbose.level : silent # one of silent, helpful, debug, debug-annoying')\nYou probably need to get an updated matplotlibrc file from\nhttps://github.com/matplotlib/matplotlib/blob/v3.3.4/matplotlibrc.template\nor from the matplotlib source distribution\n\nBad key verbose.fileo in file /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle, line 473 ('verbose.fileo : sys.stdout # a log filename, sys.stdout or sys.stderr')\nYou probably need to get an updated matplotlibrc file from\nhttps://github.com/matplotlib/matplotlib/blob/v3.3.4/matplotlibrc.template\nor from the matplotlib source distribution\nIn /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle: \nThe text.latex.preview rcparam was deprecated in Matplotlib 3.3 and will be removed two minor releases later.\nIn /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle: \nThe mathtext.fallback_to_cm rcparam was deprecated in Matplotlib 3.3 and will be removed two minor releases later.\nIn /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle: Support for setting the 'mathtext.fallback_to_cm' rcParam is deprecated since 3.3 and will be removed two minor releases later; use 'mathtext.fallback : 'cm' instead.\nIn /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle: \nThe validate_bool_maybe_none function was deprecated in Matplotlib 3.3 and will be removed two minor releases later.\nIn /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle: \nThe savefig.jpeg_quality rcparam was deprecated in Matplotlib 3.3 and will be removed two minor releases later.\nIn /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle: \nThe keymap.all_axes rcparam was deprecated in Matplotlib 3.3 and will be removed two minor releases later.\nIn /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle: \nThe animation.avconv_path rcparam was deprecated in Matplotlib 3.3 and will be removed two minor releases later.\nIn /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle: \nThe animation.avconv_args rcparam was deprecated in Matplotlib 3.3 and will be removed two minor releases later.\n"
]
],
[
[
"<h4> Como escolher o método de visualização correto? </h4>\n<p> Ao visualizar variáveis individuais, é importante primeiro entender com que tipo de variável você está lidando. Isso nos ajudará a encontrar o método de visualização correto para essa variável. Por exemplo, podemos calcular a correlação entre variáveis do tipo \"int64\" ou \"float64\" usando o método \"corr\":</p>",
"_____no_output_____"
]
],
[
[
"df.corr()",
"_____no_output_____"
]
],
[
[
"Os elementos diagonais são sempre um; (estudaremos isso, mais precisamente a correlação de Pearson no final do notebook)\n",
"_____no_output_____"
]
],
[
[
"# se quisermos verificar a correlação de apenas algumas colunas\n\ndf[['bore', 'stroke', 'compression-ratio', 'horsepower']].corr()",
"_____no_output_____"
]
],
[
[
"<h2> Variáveis numéricas contínuas: </h2>\n\n<p> Variáveis numéricas contínuas são variáveis que podem conter qualquer valor dentro de algum intervalo. Variáveis numéricas contínuas podem ter o tipo \"int64\" ou \"float64\". Uma ótima maneira de visualizar essas variáveis é usando gráficos de dispersão com linhas ajustadas. </p>\n\n<p> Para começar a compreender a relação (linear) entre uma variável individual e o preço. Podemos fazer isso usando \"regplot\", que plota o gráfico de dispersão mais a linha de regressão ajustada para os dados. </p>",
"_____no_output_____"
],
[
"<h4> Relação linear positiva </h4>\nVamos encontrar o gráfico de dispersão de \"engine-size\" e \"price\"",
"_____no_output_____"
]
],
[
[
"# Engine size as potential predictor variable of price\nsns.regplot(x=\"engine-size\", y=\"price\", data=df)\nplt.ylim(0,)",
"_____no_output_____"
]
],
[
[
"<p> Note que conforme o tamanho do motor aumenta, o preço sobe: isso indica uma correlação direta positiva entre essas duas variáveis. O tamanho do motor parece um bom preditor de preço, já que a linha de regressão é quase uma linha diagonal perfeita. </p>",
"_____no_output_____"
]
],
[
[
"# Podemos examinar a correlação entre 'engine-size' e 'price' e ver que é aproximadamente 0,87\ndf[[\"engine-size\", \"price\"]].corr()",
"_____no_output_____"
]
],
[
[
"<h4> Relação linear Negativa </h4>\n",
"_____no_output_____"
]
],
[
[
"# city-mpg também pode ser um bom preditor para a variável price: \nsns.regplot(x=\"city-mpg\", y=\"price\", data=df)",
"_____no_output_____"
]
],
[
[
"<p> À medida que o city-mpg sobe, o preço desce: isso indica uma relação inversa / negativa entre essas duas variáveis, podendo ser um indicador de preço. </p>",
"_____no_output_____"
]
],
[
[
"df[['city-mpg', 'price']].corr()",
"_____no_output_____"
]
],
[
[
"<h4> Relação linear neutra (ou fraca) </h4>\n",
"_____no_output_____"
]
],
[
[
"sns.regplot(x=\"peak-rpm\", y=\"price\", data=df)",
"_____no_output_____"
]
],
[
[
"<p> A variável peak-rpm não parece ser um bom preditor do preço, pois a linha de regressão está próxima da horizontal. Além disso, os pontos de dados estão muito dispersos e distantes da linha ajustada, apresentando grande variabilidade. Portanto, não é uma variável confiável. </p>",
"_____no_output_____"
]
],
[
[
"df[['peak-rpm','price']].corr()",
"_____no_output_____"
]
],
[
[
"<h2> Variáveis categóricas: </h2>\n\n\n<p> Essas são variáveis que descrevem uma 'característica' de uma unidade de dados e são selecionadas a partir de um pequeno grupo de categorias. As variáveis categóricas podem ser do tipo \"objeto\" ou \"int64\". Uma boa maneira de visualizar variáveis categóricas é usar boxplots. </p>",
"_____no_output_____"
]
],
[
[
"sns.boxplot(x=\"body-style\", y=\"price\", data=df)",
"_____no_output_____"
]
],
[
[
"Vemos que as distribuições de preço entre as diferentes categorias de body-style têm uma sobreposição significativa e, portanto, body-style não seria um bom preditor de preço. Vamos examinar a \"engine-location\" e o \"price\" do motor:",
"_____no_output_____"
]
],
[
[
"sns.boxplot(x=\"engine-location\", y=\"price\", data=df)",
"_____no_output_____"
]
],
[
[
"<p> Aqui, vemos que a distribuição de preço entre essas duas categorias de localização do motor, dianteira e traseira, são distintas o suficiente para considerar a localização do motor como um bom indicador de preço em potencial. </p>",
"_____no_output_____"
]
],
[
[
"# drive-wheels\nsns.boxplot(x=\"drive-wheels\", y=\"price\", data=df)",
"_____no_output_____"
]
],
[
[
"<p> Aqui vemos que a distribuição de preço entre as diferentes categorias de drive-wheels difere e podem ser um indicador de preço. </p>",
"_____no_output_____"
],
[
"<h2> Estatística Descritiva </h2>\n",
"_____no_output_____"
],
[
"<p> Vamos primeiro dar uma olhada nas variáveis usando um método de descrição. </p>\n\n<p> A função <b> describe </b> calcula automaticamente estatísticas básicas para todas as variáveis contínuas. Quaisquer valores NaN são automaticamente ignorados nessas estatísticas. </p>\n\nIsso mostrará:\n\n<ul>\n <li> a contagem dessa variável </li>\n <li> a média </li>\n <li> o desvio padrão (std) </li>\n <li> o valor mínimo </li>\n <li> o IQR (intervalo interquartil: 25%, 50% e 75%) </li>\n <li> o valor máximo </li>\n<ul>",
"_____no_output_____"
]
],
[
[
"df.describe()",
"_____no_output_____"
],
[
"# A configuração padrão de \"describe\" ignora variáveis do tipo de objeto. \n# Podemos aplicar o método \"describe\" nas variáveis do tipo 'objeto' da seguinte forma:\ndf.describe(include=['object'])",
"_____no_output_____"
]
],
[
[
"<h3>Value Counts</h3>\n\nA contagem de valores é uma boa maneira de entender quantas unidades de cada característica / variável temos. \nPodemos aplicar o método \"value_counts\" na coluna 'drive-wheels'. \nNão se esqueça que o método \"value_counts\" só funciona na série Pandas, não nos Dataframes Pandas. \nPor isso, incluímos apenas um colchete \"df ['drive-wheels']\" e não dois colchetes \"df [['drive-wheels']]\".\n",
"_____no_output_____"
]
],
[
[
"df['drive-wheels'].value_counts()",
"_____no_output_____"
],
[
"# nós podemos converter a série para um dataframe:\n\ndf['drive-wheels'].value_counts().to_frame()",
"_____no_output_____"
],
[
"drive_wheels_counts = df['drive-wheels'].value_counts().to_frame()\ndrive_wheels_counts.rename(columns={'drive-wheels': 'value_counts'}, inplace=True)\ndrive_wheels_counts",
"_____no_output_____"
],
[
"# vamos renomear o index para 'drive-wheels':\n\ndrive_wheels_counts.index.name = 'drive-wheels'\ndrive_wheels_counts",
"_____no_output_____"
],
[
"# repetindo o processo para engine-location\nengine_loc_counts = df['engine-location'].value_counts().to_frame()\nengine_loc_counts.rename(columns={'engine-location': 'value_counts'}, inplace=True)\nengine_loc_counts.index.name = 'engine-location'\nengine_loc_counts.head()",
"_____no_output_____"
]
],
[
[
"<h2>Agrupando</h2>\n\n\n<p> O método \"groupby\" agrupa os dados por categorias diferentes. Os dados são agrupados com base em uma ou várias variáveis e a análise é realizada nos grupos individuais. </p>\n\n<p> Por exemplo, vamos agrupar pela variável \"drive-wheels\". Vemos que existem 3 categorias diferentes de rodas motrizes. </p>",
"_____no_output_____"
]
],
[
[
"df['drive-wheels'].unique()",
"_____no_output_____"
]
],
[
[
"<p> Se quisermos saber, em média, qual tipo de drive-wheels é mais valiosa, podemos agrupar \"drive-wheels\" e depois fazer a média delas. </p>\n\n<p> Podemos selecionar as colunas 'drive-wheels', 'body-style' e 'price' e, em seguida, atribuí-las à variável \"df_group_one\". </p>",
"_____no_output_____"
]
],
[
[
"df_group_one = df[['drive-wheels','body-style','price']]",
"_____no_output_____"
],
[
"# Podemos então calcular o preço médio para cada uma das diferentes categorias de dados\ndf_group_one = df_group_one.groupby(['drive-wheels'],as_index=False).mean()\ndf_group_one",
"_____no_output_____"
]
],
[
[
"<p> Pelos nossos dados, parece que os veículos com tração traseira são, em média, os mais caros, enquanto as 4 rodas e as rodas dianteiras têm preços aproximadamente iguais. </p>\n\n<p> Você também pode agrupar com várias variáveis. Por exemplo, vamos agrupar por 'drive-wheels' e 'body-style'. Isso agrupa o dataframe pelas combinações exclusivas 'drive-wheels' e 'body-style'. Podemos armazenar os resultados na variável 'grouped_test1'. </p>",
"_____no_output_____"
]
],
[
[
"df_gptest = df[['drive-wheels','body-style','price']]\ngrouped_test1 = df_gptest.groupby(['drive-wheels','body-style'],as_index=False).mean()\ngrouped_test1",
"_____no_output_____"
]
],
[
[
"Esses dados agrupados são muito mais fáceis de visualizar quando transformados em uma tabela dinâmica. Uma tabela dinâmica é como uma planilha do Excel, com uma variável ao longo da coluna e outra ao longo da linha. Podemos converter o dataframe em uma tabela dinâmica usando o método \"pivô\" para criar uma tabela dinâmica a partir dos grupos.\n\nNesse caso, deixaremos a variável da \tdrive-wheels como as linhas da tabela e giraremos no estilo do corpo para se tornar as colunas da tabela:",
"_____no_output_____"
]
],
[
[
"grouped_pivot = grouped_test1.pivot(index='drive-wheels',columns='body-style')\ngrouped_pivot",
"_____no_output_____"
]
],
[
[
"As vezes não teremos dados para algumas das células pivô. Podemos preencher essas células ausentes com o valor 0, mas qualquer outro valor também pode ser usado. Deve ser mencionado que a falta de dados é um assunto bastante complexo...",
"_____no_output_____"
]
],
[
[
"grouped_pivot = grouped_pivot.fillna(0) #fill missing values with 0\ngrouped_pivot",
"_____no_output_____"
],
[
"df_gptest2 = df[['body-style','price']]\ngrouped_test_bodystyle = df_gptest2.groupby(['body-style'],as_index= False).mean()\ngrouped_test_bodystyle",
"_____no_output_____"
]
],
[
[
"<h2>Visualização dos dados</h2>\n\nVamos usar um mapa de calor para visualizar a relação entre body-style e price.",
"_____no_output_____"
]
],
[
[
"import matplotlib.pyplot as plt\n%matplotlib inline ",
"\nBad key savefig.frameon in file /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle, line 421 ('savefig.frameon : True')\nYou probably need to get an updated matplotlibrc file from\nhttps://github.com/matplotlib/matplotlib/blob/v3.3.4/matplotlibrc.template\nor from the matplotlib source distribution\n\nBad key verbose.level in file /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle, line 472 ('verbose.level : silent # one of silent, helpful, debug, debug-annoying')\nYou probably need to get an updated matplotlibrc file from\nhttps://github.com/matplotlib/matplotlib/blob/v3.3.4/matplotlibrc.template\nor from the matplotlib source distribution\n\nBad key verbose.fileo in file /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle, line 473 ('verbose.fileo : sys.stdout # a log filename, sys.stdout or sys.stderr')\nYou probably need to get an updated matplotlibrc file from\nhttps://github.com/matplotlib/matplotlib/blob/v3.3.4/matplotlibrc.template\nor from the matplotlib source distribution\nIn /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle: \nThe text.latex.preview rcparam was deprecated in Matplotlib 3.3 and will be removed two minor releases later.\nIn /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle: \nThe mathtext.fallback_to_cm rcparam was deprecated in Matplotlib 3.3 and will be removed two minor releases later.\nIn /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle: Support for setting the 'mathtext.fallback_to_cm' rcParam is deprecated since 3.3 and will be removed two minor releases later; use 'mathtext.fallback : 'cm' instead.\nIn /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle: \nThe validate_bool_maybe_none function was deprecated in Matplotlib 3.3 and will be removed two minor releases later.\nIn /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle: \nThe savefig.jpeg_quality rcparam was deprecated in Matplotlib 3.3 and will be removed two minor releases later.\nIn /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle: \nThe keymap.all_axes rcparam was deprecated in Matplotlib 3.3 and will be removed two minor releases later.\nIn /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle: \nThe animation.avconv_path rcparam was deprecated in Matplotlib 3.3 and will be removed two minor releases later.\nIn /home/anderson/.local/lib/python3.6/site-packages/matplotlib/mpl-data/stylelib/_classic_test.mplstyle: \nThe animation.avconv_args rcparam was deprecated in Matplotlib 3.3 and will be removed two minor releases later.\n"
],
[
"plt.pcolor(grouped_pivot, cmap='RdBu')\nplt.colorbar()\nplt.show()",
"_____no_output_____"
]
],
[
[
"<p> O mapa de calor representa a variável alvo (price) proporcional à cor em relação às variáveis 'drive-wheels' e 'body-style' nos eixos vertical e horizontal, respectivamente. Isso nos permite visualizar como o preço está relacionado a 'drive-wheels' e 'body-style'. </p>\n\n<p> Os rótulos padrão não transmitem informações úteis para nós. Vamos mudar isso: </p>",
"_____no_output_____"
]
],
[
[
"fig, ax = plt.subplots()\nim = ax.pcolor(grouped_pivot, cmap='RdBu')\n\n#label names\nrow_labels = grouped_pivot.columns.levels[1]\ncol_labels = grouped_pivot.index\n\n#move ticks and labels to the center\nax.set_xticks(np.arange(grouped_pivot.shape[1]) + 0.5, minor=False)\nax.set_yticks(np.arange(grouped_pivot.shape[0]) + 0.5, minor=False)\n\n#insert labels\nax.set_xticklabels(row_labels, minor=False)\nax.set_yticklabels(col_labels, minor=False)\n\n#rotate label if too long\nplt.xticks(rotation=90)\n\nfig.colorbar(im)\nplt.show()",
"_____no_output_____"
]
],
[
[
"<p> A visualização é muito importante na ciência de dados e os pacotes de visualização oferecem grande liberdade</p>\n\n<p> A principal questão que queremos responder neste notebook é \"Quais são as principais características que têm mais impacto no preço do carro?\". </p>\n\n<p> Para obter uma melhor medida das características importantes, olhamos para a correlação dessas variáveis com o preço do carro, em outras palavras: como o preço do carro depende dessa variável? </p>",
"_____no_output_____"
],
[
"<h2>Correlação e Causalidade</h2>",
"_____no_output_____"
],
[
"<p> <b> Correlação </b>: uma medida da extensão da interdependência entre as variáveis. </p>\n\n<p> <b> Causalidade </b>: a relação entre causa e efeito entre duas variáveis. </p>\n\n<p> É importante saber a diferença entre os dois e que a correlação não implica causalidade. Determinar a correlação é muito mais simples do que determinar a causalidade, pois a causalidade pode exigir experimentação independente. </p>",
"_____no_output_____"
],
[
"<p3> Correlação de Pearson </p>\n\n<p> A Correlação de Pearson mede a dependência linear entre duas variáveis X e Y. </p>\n<p> O coeficiente resultante é um valor entre -1 e 1 inclusive, onde: </p>\n<ul>\n <li> <b> 1 </b>: Correlação linear positiva total. </li>\n <li> <b> 0 </b>: Sem correlação linear, as duas variáveis provavelmente não se afetam. </li>\n <li> <b> -1 </b>: Correlação linear negativa total. </li>\n</ul>\n\n<p> Correlação de Pearson é o método padrão da função \"corr\". Como antes, podemos calcular a Correlação de Pearson das variáveis 'int64' ou 'float64'. </p>",
"_____no_output_____"
]
],
[
[
"df.corr()",
"_____no_output_____"
]
],
[
[
"<b> P-value </b>:\n\n<p>P-value é o valor da probabilidade de que a correlação entre essas duas variáveis seja estatisticamente significativa. Normalmente, escolhemos um nível de significância de 0.05, o que significa que temos 95% de confiança de que a correlação entre as variáveis é significativa. </p>\n\nPor convenção, quando o\n\n<ul>\n <li> o valor de p é $ <$ 0.001: afirmamos que há fortes evidências de que a correlação é significativa. </li>\n <li> o valor p é $ <$ 0.05: há evidências moderadas de que a correlação é significativa. </li>\n <li> o valor p é $ <$ 0.1: há evidências fracas de que a correlação é significativa. </li>\n <li> o valor p é $> $ 0.1: não há evidências de que a correlação seja significativa. </li>\n</ul>",
"_____no_output_____"
]
],
[
[
"# Podemos obter essas informações usando o módulo \"stats\" da biblioteca \"scipy\"\n\nfrom scipy import stats",
"_____no_output_____"
]
],
[
[
"<h3>Wheel-base vs Price</h3>\n\nVamos calcular o coeficiente de correlação de Pearson e o P-value entre 'wheel-base' e 'price'.",
"_____no_output_____"
]
],
[
[
"pearson_coef, p_value = stats.pearsonr(df['wheel-base'], df['price'])\nprint('Coeficiente de Pearson', pearson_coef)\nprint('P-value', p_value)",
"Coeficiente de Pearson 0.584641822265508\nP-value 8.076488270733218e-20\n"
]
],
[
[
"A notacão científica do resultado indica que o valor é muito maior ou muito pequeno.\n\nNo caso de 8.076488270733218e-20 significa:\n\n\n8.076488270733218 vezes 10 elevado a menos 20 (o que faz andar a casa decimal 20 vezes para esquerda):\n0,0000000000000000008076488270733218",
"_____no_output_____"
],
[
"<h5> Conclusão: </h5>\n<p> Como o P-value é $ <$ 0.001, a correlação entre wheel-base e price é estatisticamente significativa, embora a relação linear não seja extremamente forte (~ 0,585) </p>",
"_____no_output_____"
],
[
"<h3>Horsepower vs Price</h3>\n",
"_____no_output_____"
]
],
[
[
"pearson_coef, p_value = stats.pearsonr(df['horsepower'], df['price'])\nprint('Coeficiente de Pearson', pearson_coef)\nprint('P-value', p_value)",
"Coeficiente de Pearson 0.8096068016571054\nP-value 6.273536270650504e-48\n"
]
],
[
[
"<h5> Conclusão: </h5>\n\n<p> Como o P-value é $ <$ 0,001, a correlação entre a horsepower e price é estatisticamente significativa, e a relação linear é bastante forte (~ 0,809, próximo de 1) </p>",
"_____no_output_____"
],
[
"<h3>Length vs Price</h3>\n",
"_____no_output_____"
]
],
[
[
"pearson_coef, p_value = stats.pearsonr(df['length'], df['price'])\nprint('Coeficiente de Pearson', pearson_coef)\nprint('P-value', p_value)",
"Coeficiente de Pearson 0.6906283804483638\nP-value 8.016477466159556e-30\n"
]
],
[
[
"<h5> Conclusão: </h5>\n<p> Como o valor p é $ <$ 0,001, a correlação entre length e price é estatisticamente significativa, e a relação linear é moderadamente forte (~ 0,691). </p>",
"_____no_output_____"
],
[
"<h3>Width vs Price</h3>\n",
"_____no_output_____"
]
],
[
[
"pearson_coef, p_value = stats.pearsonr(df['width'], df['price'])\nprint('Coeficiente de Pearson', pearson_coef)\nprint('P-value', p_value)",
"Coeficiente de Pearson 0.7512653440522674\nP-value 9.200335510481516e-38\n"
]
],
[
[
"##### Conclusão:\n\nComo o valor p é <0,001, a correlação entre largura e preço é estatisticamente significativa e a relação linear é bastante forte (~ 0,751).",
"_____no_output_____"
],
[
"<h2>ANOVA</h2>\n",
"_____no_output_____"
],
[
"<p> A Análise de Variância (ANOVA) é um método estatístico usado para testar se existem diferenças significativas entre as médias de dois ou mais grupos. ANOVA retorna dois parâmetros: </p>\n\n<p> <b> F-test score </b>: ANOVA assume que as médias de todos os grupos são iguais, calcula o quanto as médias reais se desviam da suposição e relata como a pontuação do F-test. Uma pontuação maior significa que há uma diferença maior entre as médias. </p>\n\n<p> <b> P-value </b>: P-value diz o quão estatisticamente significativo é nosso valor de pontuação calculado. </p>\n\n<p> Se nossa variável de preço estiver fortemente correlacionada com a variável que estamos analisando, espere que a ANOVA retorne uma pontuação considerável no F-test e um pequeno P-value. </p>",
"_____no_output_____"
],
[
"<h3>Drive Wheels</h3>\n\n<p> Uma vez que ANOVA analisa a diferença entre diferentes grupos da mesma variável, a função groupby será útil. Como o algoritmo ANOVA calcula a média dos dados automaticamente, não precisamos tirar a média antes. </p>\n\n<p> Vamos ver se diferentes tipos de 'drive wheels' afetam o 'price', agrupamos os dados. </ p>\n",
"_____no_output_____"
]
],
[
[
"grouped_test2=df_gptest[['drive-wheels', 'price']].groupby(['drive-wheels'])\ngrouped_test2.head(2)",
"_____no_output_____"
],
[
"# Podemos obter os valores do grupo de métodos usando o método \"get_group\".\ngrouped_test2.get_group('4wd')['price']",
"_____no_output_____"
],
[
"# podemos usar a função 'f_oneway' no módulo 'stats' para obter pontuação do test-F e o P-value\nf_val, p_val = stats.f_oneway(grouped_test2.get_group('fwd')['price'], grouped_test2.get_group('rwd')['price'], grouped_test2.get_group('4wd')['price']) \n \nprint( \"ANOVA: F=\", f_val, \", P =\", p_val) ",
"ANOVA: F= 67.95406500780399 , P = 3.3945443577151245e-23\n"
]
],
[
[
"Este é um ótimo resultado, com uma grande pontuação no test-F mostrando uma forte correlação e um P-value de quase 0 implicando em significância estatística quase certa. Mas isso significa que todos os três grupos testados são altamente correlacionados?",
"_____no_output_____"
]
],
[
[
"#### fwd e rwd\nf_val, p_val = stats.f_oneway(grouped_test2.get_group('fwd')['price'], grouped_test2.get_group('rwd')['price']) \nprint( \"ANOVA: F=\", f_val, \", P =\", p_val )\n",
"ANOVA: F= 130.5533160959111 , P = 2.2355306355677845e-23\n"
],
[
"#### 4wd and rwd\nf_val, p_val = stats.f_oneway(grouped_test2.get_group('4wd')['price'], grouped_test2.get_group('rwd')['price']) \nprint( \"ANOVA: F=\", f_val, \", P =\", p_val) ",
"ANOVA: F= 8.580681368924756 , P = 0.004411492211225333\n"
],
[
"#### 4wd and fwd\nf_val, p_val = stats.f_oneway(grouped_test2.get_group('4wd')['price'], grouped_test2.get_group('fwd')['price']) \nprint(\"ANOVA: F=\", f_val, \", P =\", p_val) ",
"ANOVA: F= 0.665465750252303 , P = 0.41620116697845666\n"
]
],
[
[
"<h3>Conclusão</h3>\n\n<p> Agora temos uma ideia melhor de como são os nossos dados e quais variáveis são importantes levar em consideração ao prever o preço do carro.</p>\n\n \n<p> À medida que avançamos na construção de modelos de aprendizado de máquina para automatizar nossa análise, alimentar o modelo com variáveis que afetam significativamente nossa variável de destino melhorará o desempenho de previsão do nosso modelo. </p>",
"_____no_output_____"
],
[
"# É isso!\n\n### Este é apenas um exemplo de análise de dados com Python\nEste notebook faz parte de uma série de notebooks com conteúdos extraídos de cursos dos quais participei como aluno, ouvinte, professor, monitor... Reunidos para consulta futura e compartilhamento de idéias, soluções e conhecimento!\n\n### Muito obrigado pela sua leitura!\n\n\n<h4>Anderson Cordeiro</h4>\n\nVocê pode encontrar mais conteúdo no meu Medium<br> ou então entrar em contato comigo :D\n\n<a href=\"https://www.linkedin.com/in/andercordeiro/\" target=\"_blank\">[LinkedIn]</a>\n<a href=\"https://medium.com/@andcordeiro\" target=\"_blank\">[Medium]</a>\n",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown"
] |
[
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown",
"markdown"
]
] |
cbd8b1b0f97c88c13c7505e81759a16e00b9ab55
| 61,468 |
ipynb
|
Jupyter Notebook
|
HURDAT_JRDISCHG-Fran 1996.ipynb
|
williampc8985/VT-JamesRiver
|
6bacd10f4fd6158db74973ddc1abd89b650efc9f
|
[
"MIT"
] | null | null | null |
HURDAT_JRDISCHG-Fran 1996.ipynb
|
williampc8985/VT-JamesRiver
|
6bacd10f4fd6158db74973ddc1abd89b650efc9f
|
[
"MIT"
] | null | null | null |
HURDAT_JRDISCHG-Fran 1996.ipynb
|
williampc8985/VT-JamesRiver
|
6bacd10f4fd6158db74973ddc1abd89b650efc9f
|
[
"MIT"
] | null | null | null | 141.631336 | 49,656 | 0.87465 |
[
[
[
"import pandas as pd",
"_____no_output_____"
],
[
"#This is the Richmond USGS Data gage\nriver_richmnd = pd.read_csv('JR_Richmond02037500.csv')",
"/Users/williampc/opt/anaconda3/envs/geop/lib/python3.9/site-packages/IPython/core/interactiveshell.py:3165: DtypeWarning: Columns (7) have mixed types.Specify dtype option on import or set low_memory=False.\n has_raised = await self.run_ast_nodes(code_ast.body, cell_name,\n"
],
[
"river_richmnd.dropna();",
"_____no_output_____"
],
[
"#Hurricane data for the basin - Names of Relevant Storms - This will be used for getting the storms from the larger set\nJR_stormnames = pd.read_csv('gis_match.csv')\n",
"_____no_output_____"
],
[
"# Bring in the Big HURDAT data, from 1950 forward (satellites and data quality, etc.)\nHURDAT = pd.read_csv('hurdatcleanva_1950_present.csv')\n",
"_____no_output_____"
],
[
"VA_JR_stormmatch = JR_stormnames.merge(HURDAT)\n",
"_____no_output_____"
],
[
"# Now the common storms for the James Basin have been created. We now have time and storms together for the basin\n#checking some things about the data",
"_____no_output_____"
],
[
"# How many unique storms within the basin since 1950? 62 here and 53 in the Data on the Coast.NOAA.gov's website. \n#I think we are close enough here, digging may show some other storms, but I think we have at least captured the ones \n#from NOAA\nlen(VA_JR_stormmatch['Storm Number'].unique());",
"_____no_output_____"
],
[
"#double ck the lat and long parameters\nprint(VA_JR_stormmatch['Lat'].min(),\nVA_JR_stormmatch['Lon'].min(),\nVA_JR_stormmatch['Lat'].max(),\nVA_JR_stormmatch['Lon'].max())",
"36.1 -83.7 39.9 -75.1\n"
],
[
"#Make a csv of this data\nVA_JR_stormmatch.to_csv('storms_in_basin.csv', sep=',',encoding = 'utf-8')",
"_____no_output_____"
],
[
"#names of storms \nlen(VA_JR_stormmatch['Storm Number'].unique())\nVA_JR_stormmatch['Storm Number'].unique()\nnumbers = VA_JR_stormmatch['Storm Number']",
"_____no_output_____"
],
[
"#grab a storm from this list and lok at the times\n#Bill = pd.DataFrame(VA_JR_stormmatch['Storm Number'=='AL032003'])\n\nstorm = VA_JR_stormmatch[(VA_JR_stormmatch[\"Storm Number\"] == 'AL061996')]\nstorm\n#so this is the data for a storm named Bill that had a pth through the basin * BILL WAS A BACKDOOR Storm\n\n",
"_____no_output_____"
],
[
"# plotting for the USGS river Gage data \nimport matplotlib\nimport matplotlib.pyplot as plt\nfrom climata.usgs import DailyValueIO\nfrom datetime import datetime\nfrom pandas.plotting import register_matplotlib_converters\nimport numpy as np\n\nregister_matplotlib_converters()\nplt.style.use('ggplot')\nplt.rcParams['figure.figsize'] = (20.0, 10.0)\n# set parameters\nnyears = 1\nndays = 365 * nyears\nstation_id = \"02037500\"\nparam_id = \"00060\"\n\ndatelist = pd.date_range(end=datetime.today(), periods=ndays).tolist()\n#take an annual average for the river\nannual_data = DailyValueIO(\n start_date=\"1996-01-01\",\n end_date=\"1997-01-01\",\n station=station_id,\n parameter=param_id,)\nfor series in annual_data:\n flow = [r[1] for r in series.data]\n si_flow_annual = np.asarray(flow) * 0.0283168\n flow_mean = np.mean(si_flow_annual)\n\n#now for the storm \ndischg = DailyValueIO(\n start_date=\"1996-09-03\",\n end_date=\"1996-09-17\",\n station=station_id,\n parameter=param_id,)\n#create lists of date-flow values\nfor series in dischg:\n flow = [r[1] for r in series.data]\n si_flow = np.asarray(flow) * 0.0283168\n dates = [r[0] for r in series.data]\nplt.plot(dates, si_flow)\nplt.axhline(y=flow_mean, color='r', linestyle='-')\nplt.xlabel('Date')\nplt.ylabel('Discharge (m^3/s)')\nplt.title(\"TS Fran - 1996 (Atlantic)\")\nplt.xticks(rotation='vertical')\nplt.show()",
"_____no_output_____"
],
[
"max(si_flow)",
"_____no_output_____"
],
[
"percent_incr= (abs(max(si_flow)-flow_mean)/abs(flow_mean))*100\npercent_incr",
"_____no_output_____"
],
[
"#take an annual average for the river\nannual_data = DailyValueIO(\n start_date=\"1996-03-01\",\n end_date=\"1996-10-01\",\n station=station_id,\n parameter=param_id,)\nfor series in annual_data:\n flow = [r[1] for r in series.data]\n si_flow_annual = np.asarray(flow) * 0.0283168\n flow_mean_season = np.mean(si_flow_annual)\nprint(abs(flow_mean-flow_mean_season))",
"60.149465996223284\n"
]
]
] |
[
"code"
] |
[
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
]
] |
cbd8b7bf61b5f4787e1b7f6ccd2fbf9b67b56340
| 15,899 |
ipynb
|
Jupyter Notebook
|
NAIP/ndwi.ipynb
|
pberezina/earthengine-py-notebooks
|
4cbe3c52bcc9ed3f1337bf097aa5799442991a5e
|
[
"MIT"
] | 1 |
2020-03-20T19:39:34.000Z
|
2020-03-20T19:39:34.000Z
|
NAIP/ndwi.ipynb
|
pberezina/earthengine-py-notebooks
|
4cbe3c52bcc9ed3f1337bf097aa5799442991a5e
|
[
"MIT"
] | null | null | null |
NAIP/ndwi.ipynb
|
pberezina/earthengine-py-notebooks
|
4cbe3c52bcc9ed3f1337bf097aa5799442991a5e
|
[
"MIT"
] | null | null | null | 81.117347 | 9,700 | 0.835587 |
[
[
[
"<table class=\"ee-notebook-buttons\" align=\"left\">\n <td><a target=\"_blank\" href=\"https://github.com/giswqs/earthengine-py-notebooks/tree/master/NAIP/ndwi.ipynb\"><img width=32px src=\"https://www.tensorflow.org/images/GitHub-Mark-32px.png\" /> View source on GitHub</a></td>\n <td><a target=\"_blank\" href=\"https://nbviewer.jupyter.org/github/giswqs/earthengine-py-notebooks/blob/master/NAIP/ndwi.ipynb\"><img width=26px src=\"https://upload.wikimedia.org/wikipedia/commons/thumb/3/38/Jupyter_logo.svg/883px-Jupyter_logo.svg.png\" />Notebook Viewer</a></td>\n <td><a target=\"_blank\" href=\"https://mybinder.org/v2/gh/giswqs/earthengine-py-notebooks/master?filepath=NAIP/ndwi.ipynb\"><img width=58px src=\"https://mybinder.org/static/images/logo_social.png\" />Run in binder</a></td>\n <td><a target=\"_blank\" href=\"https://colab.research.google.com/github/giswqs/earthengine-py-notebooks/blob/master/NAIP/ndwi.ipynb\"><img src=\"https://www.tensorflow.org/images/colab_logo_32px.png\" /> Run in Google Colab</a></td>\n</table>",
"_____no_output_____"
],
[
"## Install Earth Engine API\nInstall the [Earth Engine Python API](https://developers.google.com/earth-engine/python_install) and [geehydro](https://github.com/giswqs/geehydro). The **geehydro** Python package builds on the [folium](https://github.com/python-visualization/folium) package and implements several methods for displaying Earth Engine data layers, such as `Map.addLayer()`, `Map.setCenter()`, `Map.centerObject()`, and `Map.setOptions()`.\nThe following script checks if the geehydro package has been installed. If not, it will install geehydro, which automatically install its dependencies, including earthengine-api and folium.",
"_____no_output_____"
]
],
[
[
"import subprocess\n\ntry:\n import geehydro\nexcept ImportError:\n print('geehydro package not installed. Installing ...')\n subprocess.check_call([\"python\", '-m', 'pip', 'install', 'geehydro'])",
"_____no_output_____"
]
],
[
[
"Import libraries",
"_____no_output_____"
]
],
[
[
"import ee\nimport folium\nimport geehydro",
"_____no_output_____"
]
],
[
[
"Authenticate and initialize Earth Engine API. You only need to authenticate the Earth Engine API once. ",
"_____no_output_____"
]
],
[
[
"try:\n ee.Initialize()\nexcept Exception as e:\n ee.Authenticate()\n ee.Initialize()",
"_____no_output_____"
]
],
[
[
"## Create an interactive map \nThis step creates an interactive map using [folium](https://github.com/python-visualization/folium). The default basemap is the OpenStreetMap. Additional basemaps can be added using the `Map.setOptions()` function. \nThe optional basemaps can be `ROADMAP`, `SATELLITE`, `HYBRID`, `TERRAIN`, or `ESRI`.",
"_____no_output_____"
]
],
[
[
"Map = folium.Map(location=[40, -100], zoom_start=4)\nMap.setOptions('HYBRID')",
"_____no_output_____"
]
],
[
[
"## Add Earth Engine Python script ",
"_____no_output_____"
]
],
[
[
"collection = ee.ImageCollection('USDA/NAIP/DOQQ')\nfromFT = ee.FeatureCollection('ft:1CLldB-ULPyULBT2mxoRNv7enckVF0gCQoD2oH7XP')\npolys = fromFT.geometry()\ncentroid = polys.centroid()\nlng, lat = centroid.getInfo()['coordinates']\n# print(\"lng = {}, lat = {}\".format(lng, lat))\n\n# lng_lat = ee.Geometry.Point(lng, lat)\nnaip = collection.filterBounds(polys)\nnaip_2015 = naip.filterDate('2015-01-01', '2015-12-31')\nppr = naip_2015.mosaic().clip(polys)\n\n# print(naip_2015.size().getInfo()) # count = 120\nvis = {'bands': ['N', 'R', 'G']}\nMap.setCenter(lng, lat, 10)\n# Map.addLayer(naip_2015,vis)\nMap.addLayer(ppr,vis)\n# Map.addLayer(fromFT)\n\nndwi = ppr.normalizedDifference(['G', 'N'])\nndwiViz = {'min': 0, 'max': 1, 'palette': ['00FFFF', '0000FF']}\nndwiMasked = ndwi.updateMask(ndwi.gte(0.05))\nndwi_bin = ndwiMasked.gt(0)\nMap.addLayer(ndwiMasked, ndwiViz)\n\npatch_size = ndwi_bin.connectedPixelCount(256, True)\n# Map.addLayer(patch_size)\n\npatch_id = ndwi_bin.connectedComponents(ee.Kernel.plus(1), 256)\nMap.addLayer(patch_id)",
"_____no_output_____"
]
],
[
[
"## Display Earth Engine data layers ",
"_____no_output_____"
]
],
[
[
"Map.setControlVisibility(layerControl=True, fullscreenControl=True, latLngPopup=True)\nMap",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
]
] |
cbd8c0c7a9857942bfc8be7c06dc93aae0cc5190
| 52,347 |
ipynb
|
Jupyter Notebook
|
utf-8''module_3.ipynb
|
Thisislegit/Python-Project-pillow-tesseract-and-opencv-
|
bf0d44e485676a17807ba4baceb5144dfd80c2f3
|
[
"MIT"
] | 1 |
2019-11-25T02:13:47.000Z
|
2019-11-25T02:13:47.000Z
|
utf-8''module_3.ipynb
|
Thisislegit/Python-Project-pillow-tesseract-and-opencv-
|
bf0d44e485676a17807ba4baceb5144dfd80c2f3
|
[
"MIT"
] | 1 |
2022-03-12T00:53:46.000Z
|
2022-03-12T00:53:46.000Z
|
utf-8''module_3.ipynb
|
Thisislegit/Python-Project-pillow-tesseract-and-opencv-
|
bf0d44e485676a17807ba4baceb5144dfd80c2f3
|
[
"MIT"
] | null | null | null | 44.741026 | 113 | 0.637782 |
[
[
[
"## Release the Kraken!",
"_____no_output_____"
]
],
[
[
"# The next library we're going to look at is called Kraken, which was developed by Université \n# PSL in Paris. It's actually based on a slightly older code base, OCRopus. You can see how the\n# flexible open-source licenses allow new ideas to grow by building upon older ideas. And, in\n# this case, I fully support the idea that the Kraken - a mythical massive sea creature - is the\n# natural progression of an octopus!\n#\n# What we are going to use Kraken for is to detect lines of text as bounding boxes in a given\n# image. The biggest limitation of tesseract is the lack of a layout engine inside of it. Tesseract\n# expects to be using fairly clean text, and gets confused if we don't crop out other artifacts.\n# It's not bad, but Kraken can help us out be segmenting pages. Lets take a look.",
"_____no_output_____"
],
[
"# First, we'll take a look at the kraken module itself\nimport kraken\nhelp(kraken)",
"_____no_output_____"
],
[
"# There isn't much of a discussion here, but there are a number of sub-modules that look\n# interesting. I spend a bit of time on their website, and I think the pageseg module, which\n# handles all of the page segmentation, is the one we want to use. Lets look at it\nfrom kraken import pageseg\nhelp(pageseg)",
"_____no_output_____"
],
[
"# So it looks like there are a few different functions we can call, and the segment\n# function looks particularly appropriate. I love how expressive this library is on the\n# documentation front -- I can see immediately that we are working with PIL.Image files,\n# and the author has even indicated that we need to pass in either a binarized (e.g. '1')\n# or grayscale (e.g. 'L') image. We can also see that the return value is a dictionary\n# object with two keys, \"text_direction\" which will return to us a string of the\n# direction of the text, and \"boxes\" which appears to be a list of tuples, where each\n# tuple is a box in the original image.\n#\n# Lets try this on the image of text. I have a simple bit of text in a file called\n# two_col.png which is from a newspaper on campus here\nfrom PIL import Image\nim=Image.open(\"readonly/two_col.png\")\n# Lets display the image inline\ndisplay(im)\n# Lets now convert it to black and white and segment it up into lines with kraken\nbounding_boxes=pageseg.segment(im.convert('1'))['boxes']\n# And lets print those lines to the screen\nprint(bounding_boxes)",
"_____no_output_____"
],
[
"# Ok, pretty simple two column text and then a list of lists which are the bounding boxes of \n# lines of that text. Lets write a little routine to try and see the effects a bit more\n# clearly. I'm going to clean up my act a bit and write real documentation too, it's a good\n# practice\ndef show_boxes(img):\n '''Modifies the passed image to show a series of bounding boxes on an image as run by kraken\n \n :param img: A PIL.Image object\n :return img: The modified PIL.Image object\n '''\n # Lets bring in our ImageDraw object\n from PIL import ImageDraw\n # And grab a drawing object to annotate that image\n drawing_object=ImageDraw.Draw(img)\n # We can create a set of boxes using pageseg.segment\n bounding_boxes=pageseg.segment(img.convert('1'))['boxes']\n # Now lets go through the list of bounding boxes\n for box in bounding_boxes:\n # An just draw a nice rectangle\n drawing_object.rectangle(box, fill = None, outline ='red')\n # And to make it easy, lets return the image object\n return img\n\n# To test this, lets use display\ndisplay(show_boxes(Image.open(\"readonly/two_col.png\")))",
"_____no_output_____"
],
[
"# Not bad at all! It's interesting to see that kraken isn't completely sure what to do with this\n# two column format. In some cases, kraken has identified a line in just a single column, while\n# in other cases kraken has spanned the line marker all the way across the page. Does this matter?\n# Well, it really depends on our goal. In this case, I want to see if we can improve a bit on this.\n#\n# So we're going to go a bit off script here. While this week of lectures is about libraries, the\n# goal of this last course is to give you confidence that you can apply your knowledge to actual\n# programming tasks, even if the library you are using doesn't quite do what you want. \n#\n# I'd like to pause the video for the moment and collect your thoughts. Looking at the image above,\n# with the two column example and red boxes, how do you think we might modify this image to improve\n# kraken's ability to text lines?",
"_____no_output_____"
],
[
"# Thanks for sharing your thoughts, I'm looking forward to seeing the breadth of ideas that everyone\n# in the course comes up with. Here's my partial solution -- while looking through the kraken docs on \n# the pageseg() function I saw that there are a few parameters we can supply in order to improve \n# segmentation. One of these is the black_colseps parameter. If set to True, kraken will assume that \n# columns will be separated by black lines. This isn't our case here, but, I think we have all of the\n# tools to go through and actually change the source image to have a black separator between columns.\n#\n# The first step is that I want to update the show_boxes() function. I'm just going to do a quick\n# copy and paste from the above but add in the black_colseps=True parameter\ndef show_boxes(img):\n '''Modifies the passed image to show a series of bounding boxes on an image as run by kraken\n \n :param img: A PIL.Image object\n :return img: The modified PIL.Image object\n '''\n # Lets bring in our ImageDraw object\n from PIL import ImageDraw\n # And grab a drawing object to annotate that image\n drawing_object=ImageDraw.Draw(img)\n # We can create a set of boxes using pageseg.segment\n bounding_boxes=pageseg.segment(img.convert('1'), black_colseps=True)['boxes']\n # Now lets go through the list of bounding boxes\n for box in bounding_boxes:\n # An just draw a nice rectangle\n drawing_object.rectangle(box, fill = None, outline ='red')\n # And to make it easy, lets return the image object\n return img",
"_____no_output_____"
],
[
"# The next step is to think of the algorithm we want to apply to detect a white column separator.\n# In experimenting a bit I decided that I only wanted to add the separator if the space of was\n# at least 25 pixels wide, which is roughly the width of a character, and six lines high. The\n# width is easy, lets just make a variable\nchar_width=25\n# The height is harder, since it depends on the height of the text. I'm going to write a routine\n# to calculate the average height of a line\ndef calculate_line_height(img):\n '''Calculates the average height of a line from a given image\n :param img: A PIL.Image object\n :return: The average line height in pixels\n '''\n # Lets get a list of bounding boxes for this image\n bounding_boxes=pageseg.segment(img.convert('1'))['boxes']\n # Each box is a tuple of (top, left, bottom, right) so the height is just top - bottom\n # So lets just calculate this over the set of all boxes\n height_accumulator=0\n for box in bounding_boxes:\n height_accumulator=height_accumulator+box[3]-box[1]\n # this is a bit tricky, remember that we start counting at the upper left corner in PIL!\n # now lets just return the average height\n # lets change it to the nearest full pixel by making it an integer\n return int(height_accumulator/len(bounding_boxes))\n\n# And lets test this with the image with have been using\nline_height=calculate_line_height(Image.open(\"readonly/two_col.png\"))\nprint(line_height)",
"_____no_output_____"
],
[
"# Ok, so the average height of a line is 31.\n# Now, we want to scan through the image - looking at each pixel in turn - to determine if there\n# is a block of whitespace. How bit of a block should we look for? That's a bit more of an art\n# than a science. Looking at our sample image, I'm going to say an appropriate block should be\n# one char_width wide, and six line_heights tall. But, I honestly just made this up by eyeballing\n# the image, so I would encourage you to play with values as you explore.\n# Lets create a new box called gap box that represents this area\ngap_box=(0,0,char_width,line_height*6)\ngap_box",
"_____no_output_____"
],
[
"# It seems we will want to have a function which, given a pixel in an image, can check to see\n# if that pixel has whitespace to the right and below it. Essentially, we want to test to see\n# if the pixel is the upper left corner of something that looks like the gap_box. If so, then\n# we should insert a line to \"break up\" this box before sending to kraken\n#\n# Lets call this new function gap_check\ndef gap_check(img, location):\n '''Checks the img in a given (x,y) location to see if it fits the description\n of a gap_box\n :param img: A PIL.Image file\n :param location: A tuple (x,y) which is a pixel location in that image\n :return: True if that fits the definition of a gap_box, otherwise False\n '''\n # Recall that we can get a pixel using the img.getpixel() function. It returns this value\n # as a tuple of integers, one for each color channel. Our tools all work with binarized\n # images (black and white), so we should just get one value. If the value is 0 it's a black\n # pixel, if it's white then the value should be 255\n #\n # We're going to assume that the image is in the correct mode already, e.g. it has been\n # binarized. The algorithm to check our bounding box is fairly easy: we have a single location \n # which is our start and then we want to check all the pixels to the right of that location \n # up to gap_box[2]\n for x in range(location[0], location[0]+gap_box[2]):\n # the height is similar, so lets iterate a y variable to gap_box[3]\n for y in range(location[1], location[1]+gap_box[3]):\n # we want to check if the pixel is white, but only if we are still within the image\n if x < img.width and y < img.height:\n # if the pixel is white we don't do anything, if it's black, we just want to\n # finish and return False\n if img.getpixel((x,y)) != 255:\n return False\n # If we have managed to walk all through the gap_box without finding any non-white pixels\n # then we can return true -- this is a gap!\n return True",
"_____no_output_____"
],
[
"# Alright, we have a function to check for a gap, called gap_check. What should we do once\n# we find a gap? For this, lets just draw a line in the middle of it. Lets create a new function\ndef draw_sep(img,location):\n '''Draws a line in img in the middle of the gap discovered at location. Note that\n this doesn't draw the line in location, but draws it at the middle of a gap_box\n starting at location.\n :param img: A PIL.Image file\n :param location: A tuple(x,y) which is a pixel location in the image\n '''\n # First lets bring in all of our drawing code\n from PIL import ImageDraw\n drawing_object=ImageDraw.Draw(img)\n # next, lets decide what the middle means in terms of coordinates in the image\n x1=location[0]+int(gap_box[2]/2)\n # and our x2 is just the same thing, since this is a one pixel vertical line\n x2=x1\n # our starting y coordinate is just the y coordinate which was passed in, the top of the box\n y1=location[1]\n # but we want our final y coordinate to be the bottom of the box\n y2=y1+gap_box[3]\n drawing_object.rectangle((x1,y1,x2,y2), fill = 'black', outline ='black')\n # and we don't have anything we need to return from this, because we modified the image",
"_____no_output_____"
],
[
"# Now, lets try it all out. This is pretty easy, we can just iterate through each pixel \n# in the image, check if there is a gap, then insert a line if there is.\ndef process_image(img):\n '''Takes in an image of text and adds black vertical bars to break up columns\n :param img: A PIL.Image file\n :return: A modified PIL.Image file\n '''\n # we'll start with a familiar iteration process\n for x in range(img.width):\n for y in range(img.height):\n # check if there is a gap at this point\n if (gap_check(img, (x,y))):\n # then update image to one which has a separator drawn on it\n draw_sep(img, (x,y))\n # and for good measure we'll return the image we modified\n return img\n\n# Lets read in our test image and convert it through binarization\ni=Image.open(\"readonly/two_col.png\").convert(\"L\")\ni=process_image(i)\ndisplay(i)\n\n#Note: This will take some time to run! Be patient!",
"_____no_output_____"
],
[
"# Not bad at all! The effect at the bottom of the image is a bit unexpected to me, but it makes\n# sense. You can imagine that there are several ways we might try and control this. Lets see how \n# this new image works when run through the kraken layout engine\ndisplay(show_boxes(i))",
"_____no_output_____"
],
[
"# Looks like that is pretty accurate, and fixes the problem we faced. Feel free to experiment\n# with different settings for the gap heights and width and share in the forums. You'll notice though \n# method we created is really quite slow, which is a bit of a problem if we wanted to use\n# this on larger text. But I wanted to show you how you can mix your own logic and work with\n# libraries you're using. Just because Kraken didn't work perfectly, doesn't mean we can't\n# build something more specific to our use case on top of it.\n#\n# I want to end this lecture with a pause and to ask you to reflect on the code we've written\n# here. We started this course with some pretty simple use of libraries, but now we're\n# digging in deeper and solving problems ourselves with the help of these libraries. Before we\n# go on to our last library, how well prepared do you think you are to take your python\n# skills out into the wild?",
"_____no_output_____"
]
],
[
[
"## Comparing Image Data Structures",
"_____no_output_____"
]
],
[
[
"# OpenCV supports reading of images in most file formats, such as JPEG, PNG, and TIFF. Most image and \n# video analysis requires converting images into grayscale first. This simplifies the image and reduces \n# noise allowing for improved analysis. Let's write some code that reads an image of as person, Floyd \n# Mayweather and converts it into greyscale.\n\n# First we will import the open cv package cv2 \nimport cv2 as cv\n# We'll load the floyd.jpg image \nimg = cv.imread('readonly/floyd.jpg')\n# And we'll convert it to grayscale using the cvtColor image\ngray = cv.cvtColor(img, cv.COLOR_BGR2GRAY)\n\n# Now, before we get to the result, lets talk about docs. Just like tesseract, opencv is an external\n# package written in C++, and the docs for python are really poor. This is unfortunatly quite common\n# when python is being used as a wrapper. Thankfully, the web docs for opencv are actually pretty good,\n# so hit the website docs.opencv.org when you want to learn more about a particular function. In this\n# case cvtColor converts from one color space to another, and we are convering our image to grayscale.\n# Of course, we already know at least two different ways of doing this, using binarization and PIL\n# color spaces conversions\n\n# Lets instpec this object that has been returned.\nimport inspect\ninspect.getmro(type(gray))",
"_____no_output_____"
],
[
"# We see that it is of type ndarray, which is a fundamental list type coming from the numerical\n# python project. That's a bit surprising - up until this point we have been used to working with\n# PIL.Image objects. OpenCV, however, wants to represent an image as a two dimensional sequence \n# of bytes, and the ndarray, which stands for n dimensional array, is the ideal way to do this.\n# Lets look at the array contents.\ngray",
"_____no_output_____"
],
[
"# The array is shown here as a list of lists, where the inner lists are filled with integers.\n# The dtype=uint8 definition indicates that each of the items in an array is an 8 bit unsigned\n# integer, which is very common for black and white images. So this is a pixel by pixel definition\n# of the image.\n#\n# The display package, however, doesn't know what to do with this image. So lets convert it\n# into a PIL object to render it in the browser.\nfrom PIL import Image\n\n# PIL can take an array of data with a given color format and convert this into a PIL object.\n# This is perfect for our situation, as the PIL color mode, \"L\" is just an array of luminance\n# values in unsigned integers\nimage = Image.fromarray(gray, \"L\")\ndisplay(image)",
"_____no_output_____"
],
[
"# Lets talk a bit more about images for a moment. Numpy arrays are multidimensional. For \n# instance, we can define an array in a single dimension:\nimport numpy as np\nsingle_dim = np.array([25, 50 , 25, 10, 10])\n\n# In an image, this is analagous to a single row of 5 pixels each in grayscale. But actually,\n# all imaging libraries tend to expect at least two dimensions, a width and a height, and to\n# show a matrix. So if we put the single_dim inside of another array, this would be a two\n# dimensional array with element in the height direction, and five in the width direction\ndouble_dim = np.array([single_dim])\n\ndouble_dim",
"_____no_output_____"
],
[
"# This should look pretty familiar, it's a lot like a list of lists! Lets see what this new\n# two dimensional array looks like if we display it\ndisplay(Image.fromarray(double_dim, \"L\"))",
"_____no_output_____"
],
[
"# Pretty unexciting - it's just a little line. Five pixels in a row to be exact, of different\n# levels of black. The numpy library has a nice attribute called shape that allows us to see how\n# many dimensions big an array is. The shape attribute returns a tuple that shows the height of\n# the image, by the width of the image\ndouble_dim.shape",
"_____no_output_____"
],
[
"# Lets take a look at the shape of our initial image which we loaded into the img variable\nimg.shape",
"_____no_output_____"
],
[
"# This image has three dimensions! That's because it has a width, a height, and what's called\n# a color depth. In this case, the color is represented as an array of three values. Lets take a \n# look at the color of the first pixel\nfirst_pixel=img[0][0]\nfirst_pixel",
"_____no_output_____"
],
[
"# Here we see that the color value is provided in full RGB using an unsigned integer. This\n# means that each color can have one of 256 values, and the total number of unique colors\n# that can be represented by this data is 256 * 256 *256 which is roughly 16 million colors.\n# We call this 24 bit color, which is 8+8+8.\n#\n# If you find yourself shopping for a television, you might notice that some expensive models\n# are advertised as having 10 bit or even 12 bit panels. These are televisions where each of\n# the red, green, and blue color channels are represented by 10 or 12 bits instead of 8. For\n# ten bit panels this means that there are 1 billion colors capable, and 12 bit panels are\n# capable of over 68 billion colors!",
"_____no_output_____"
],
[
"# We're not going to talk much more about color in this course, but it's a fun subject. Instead,\n# lets go back to this array representation of images, because we can do some interesting things\n# with this.\n#\n# One of the most common things to do with an ndarray is to reshape it -- to change the number\n# of rows and columns that are represented so that we can do different kinds of operations.\n# Here is our original two dimensional image\nprint(\"Original image\")\nprint(gray)\n# If we wanted to represent that as a one dimensional image, we just call reshape\nprint(\"New image\")\n# And reshape takes the image as the first parameter, and a new shape as the second\nimage1d=np.reshape(gray,(1,gray.shape[0]*gray.shape[1]))\nprint(image1d)",
"_____no_output_____"
],
[
"# So, why are we talking about these nested arrays of bytes, we were supposed to be talking\n# about OpenCV as a library. Well, I wanted to show you that often libraries working on the\n# same kind of principles, in this case images stored as arrays of bytes, are not representing\n# data in the same way in their APIs. But, by exploring a bit you can learn how the internal\n# representation of data is stored, and build routines to convert between formats.\n#\n# For instance, remember in the last lecture when we wanted to look for gaps in an image so\n# that we could draw lines to feed into kraken? Well, we use PIL to do this, using getpixel()\n# to look at individual pixels and see what the luminosity was, then ImageDraw.rectangle to\n# actually fill in a black bar separator. This was a nice high level API, and let us write\n# routines to do the work we wanted without having to understand too much about how the images\n# were being stored. But it was computationally very slow.\n#\n# Instead, we could write the code to do this using matrix features within numpy. Lets take\n# a look.\nimport cv2 as cv\n# We'll load the 2 column image\nimg = cv.imread('readonly/two_col.png')\n# And we'll convert it to grayscale using the cvtColor image\ngray = cv.cvtColor(img, cv.COLOR_BGR2GRAY)",
"_____no_output_____"
],
[
"# Now, remember how slicing on a list works, if you have a list of number such as \n# a=[0,1,2,3,4,5] then a[2:4] will return the sublist of numbers at position 2 through 4 \n# inclusive - don't forget that lists start indexing at 0!\n# If we have a two dimensional array, we can slice out a smaller piece of that using the\n# format a[2:4,1:3]. You can think of this as first slicing along the rows dimension, then\n# in the columns dimension. So in this example, that would be a matrix of rows 2, and 3,\n# and columns 1, and 2. Here's a look at our image.\ngray[2:4,1:3]",
"_____no_output_____"
],
[
"# So we see that it is all white. We can use this as a \"window\" and move it around our\n# our big image.\n#\n# Finally, the ndarray library has lots of matrix functions which are generally very fast\n# to run. One that we want to consider in this case is count_nonzero(), which just returns\n# the number of entries in the matrix which are not zero.\nnp.count_nonzero(gray[2:4,1:3])",
"_____no_output_____"
],
[
"# Ok, the last benefit of going to this low level approach to images is that we can change\n# pixels very fast as well. Previously we were drawing rectangles and setting a fill and line\n# width. This is nice if you want to do something like change the color of the fill from the\n# line, or draw complex shapes. But we really just want a line here. That's really easy to\n# do - we just want to change a number of luminosity values from 255 to 0.\n#\n# As an example, lets create a big white matrix\nwhite_matrix=np.full((12,12),255,dtype=np.uint8)\ndisplay(Image.fromarray(white_matrix,\"L\"))\nwhite_matrix",
"_____no_output_____"
],
[
"# looks pretty boring, it's just a giant white square we can't see. But if we want, we can\n# easily color a column to be black\nwhite_matrix[:,6]=np.full((1,12),0,dtype=np.uint8)\ndisplay(Image.fromarray(white_matrix,\"L\"))\nwhite_matrix",
"_____no_output_____"
],
[
"# And that's exactly what we wanted to do. So, why do it this way, when it seems so much\n# more low level? Really, the answer is speed. This paradigm of using matricies to store\n# and manipulate bytes of data for images is much closer to how low level API and hardware\n# developers think about storing files and bytes in memory.\n#\n# How much faster is it? Well, that's up to you to discover; there's an optional assignment\n# for this week to convert our old code over into this new format, to compare both the\n# readability and speed of the two different approaches.",
"_____no_output_____"
]
],
[
[
"## OpenCV",
"_____no_output_____"
]
],
[
[
"# Ok, we're just about at the project for this course. If you reflect on the specialization \n# as a whole you'll realize that you started with probably little or no understanding of python,\n# progressed through the basic control structures and libraries included with the language\n# with the help of a digital textbook, moved on to more high level representations of data\n# and functions with objects, and now started to explore third party libraries that exist for\n# python which allow you to manipulate and display images. This is quite an achievement!\n#\n# You have also no doubt found that as you have progressed the demands on you to engage in self-\n# discovery have also increased. Where the first assignments were maybe straight forward, the\n# ones in this week require you to struggle a bit more with planning and debugging code as\n# you develop.\n#\n# But, you've persisted, and I'd like to share with you just one more set of features before\n# we head over to a project. The OpenCV library contains mechanisms to do face detection on\n# images. The technique used is based on Haar cascades, which is a machine learning approach.\n# Now, we're not going to go into the machine learning bits, we have another specialization on\n# Applied Data Science with Python which you can take after this if you're interested in that topic.\n# But here we'll treat OpenCV like a black box.\n#\n# OpenCV comes with trained models for detecting faces, eyes, and smiles which we'll be using.\n# You can train models for detecting other things - like hot dogs or flutes - and if you're\n# interested in that I'd recommend you check out the Open CV docs on how to train a cascade\n# classifier: https://docs.opencv.org/3.4/dc/d88/tutorial_traincascade.html\n# However, in this lecture we just want to use the current classifiers and see if we can detect\n# portions of an image which are interesting.\n#\n# First step is to load opencv and the XML-based classifiers\nimport cv2 as cv\nface_cascade = cv.CascadeClassifier('readonly/haarcascade_frontalface_default.xml')\neye_cascade = cv.CascadeClassifier('readonly/haarcascade_eye.xml')",
"_____no_output_____"
],
[
"# Ok, with the classifiers loaded, we now want to try and detect a face. Lets pull in the\n# picture we played with last time\nimg = cv.imread('readonly/floyd.jpg')\n# And we'll convert it to grayscale using the cvtColor image\ngray = cv.cvtColor(img, cv.COLOR_BGR2GRAY)\n# The next step is to use the face_cascade classifier. I'll let you go explore the docs if you\n# would like to, but the norm is to use the detectMultiScale() function. This function returns\n# a list of objects as rectangles. The first parameter is an ndarray of the image.\nfaces = face_cascade.detectMultiScale(gray)\n# And lets just print those faces out to the screen\nfaces",
"_____no_output_____"
],
[
"faces.tolist()[0]",
"_____no_output_____"
],
[
"# The resulting rectangles are in the format of (x,y,w,h) where x and y denote the upper\n# left hand point for the image and the width and height represent the bounding box. We know\n# how to handle this in PIL\nfrom PIL import Image\n\n# Lets create a PIL image object\npil_img=Image.fromarray(gray,mode=\"L\")\n\n# Now lets bring in our drawing object\nfrom PIL import ImageDraw\n# And lets create our drawing context\ndrawing=ImageDraw.Draw(pil_img)\n\n# Now lets pull the rectangle out of the faces object\nrec=faces.tolist()[0]\n\n# Now we just draw a rectangle around the bounds\ndrawing.rectangle(rec, outline=\"white\")\n\n# And display\ndisplay(pil_img)",
"_____no_output_____"
],
[
"# So, not quite what we were looking for. What do you think went wrong?\n# Well, a quick double check of the docs and it is apparent that OpenCV is return the coordinates\n# as (x,y,w,h), while PIL.ImageDraw is looking for (x1,y1,x2,y2). Looks like an easy fix\n# Wipe our old image\npil_img=Image.fromarray(gray,mode=\"L\")\n# Setup our drawing context\ndrawing=ImageDraw.Draw(pil_img)\n# And draw the new box\ndrawing.rectangle((rec[0],rec[1],rec[0]+rec[2],rec[1]+rec[3]), outline=\"white\")\n# And display\ndisplay(pil_img)",
"_____no_output_____"
],
[
"# We see the face detection works pretty good on this image! Note that it's apparent that this is\n# not head detection, but that the haarcascades file we used is looking for eyes and a mouth.\n# Lets try this on something a bit more complex, lets read in our MSI recruitment image\nimg = cv.imread('readonly/msi_recruitment.gif')\n# And lets take a look at that image\ndisplay(Image.fromarray(img))",
"_____no_output_____"
],
[
"# Whoa, what's that error about? It looks like there is an error on a line deep within the PIL\n# Image.py file, and it is trying to call an internal private member called __array_interface__\n# on the img object, but this object is None\n#\n# It turns out that the root of this error is that OpenCV can't work with Gif images. This is\n# kind of a pain and unfortunate. But we know how to fix that right? One was is that we could\n# just open this in PIL and then save it as a png, then open that in open cv.\n#\n# Lets use PIL to open our image\npil_img=Image.open('readonly/msi_recruitment.gif')\n# now lets convert it to greyscale for opencv, and get the bytestream\nopen_cv_version=pil_img.convert(\"L\")\n# now lets just write that to a file\nopen_cv_version.save(\"msi_recruitment.png\")",
"_____no_output_____"
],
[
"# Ok, now that the conversion of format is done, lets try reading this back into opencv\ncv_img=cv.imread('msi_recruitment.png')\n# We don't need to color convert this, because we saved it as grayscale\n# lets try and detect faces in that image\nfaces = face_cascade.detectMultiScale(cv_img)\n\n# Now, we still have our PIL color version in a gif\npil_img=Image.open('readonly/msi_recruitment.gif')\n# Set our drawing context\ndrawing=ImageDraw.Draw(pil_img)\n\n# For each item in faces, lets surround it with a red box\nfor x,y,w,h in faces:\n # That might be new syntax for you! Recall that faces is a list of rectangles in (x,y,w,h)\n # format, that is, a list of lists. Instead of having to do an iteration and then manually\n # pull out each item, we can use tuple unpacking to pull out individual items in the sublist\n # directly to variables. A really nice python feature\n #\n # Now we just need to draw our box\n drawing.rectangle((x,y,x+w,y+h), outline=\"white\")\ndisplay(pil_img)",
"_____no_output_____"
],
[
"# What happened here!? We see that we have detected faces, and that we have drawn boxes\n# around those faces on the image, but that the colors have gone all weird! This, it turns\n# out, has to do with color limitations for gif images. In short, a gif image has a very\n# limited number of colors. This is called a color pallette after the pallette artists\n# use to mix paints. For gifs the pallette can only be 256 colors -- but they can be *any*\n# 256 colors. When a new color is introduced, is has to take the space of an old color.\n# In this case, PIL adds white to the pallette but doesn't know which color to replace and\n# thus messes up the image.\n#\n# Who knew there was so much to learn about image formats? We can see what mode the image\n# is in with the .mode attribute\npil_img.mode",
"_____no_output_____"
],
[
"# We can see a list of modes in the PILLOW documentation, and they correspond with the\n# color spaces we have been using. For the moment though, lets change back to RGB, which\n# represents color as a three byte tuple instead of in a pallette.\n# Lets read in the image\npil_img=Image.open('readonly/msi_recruitment.gif')\n# Lets convert it to RGB mode\npil_img = pil_img.convert(\"RGB\")\n# And lets print out the mode\npil_img.mode",
"_____no_output_____"
],
[
"# Ok, now lets go back to drawing rectangles. Lets get our drawing object\ndrawing=ImageDraw.Draw(pil_img)\n# And iterate through the faces sequence, tuple unpacking as we go\nfor x,y,w,h in faces:\n # And remember this is width and height so we have to add those appropriately.\n drawing.rectangle((x,y,x+w,y+h), outline=\"white\")\ndisplay(pil_img)",
"_____no_output_____"
],
[
"# Awesome! We managed to detect a bunch of faces in that image. Looks like we have missed \n# four faces. In the machine learning world we would call these false negatives - something\n# which the machine thought was not a face (so a negative), but that it was incorrect on.\n# Consequently, we would call the actual faces that were detected as true positives -\n# something that the machine thought was a face and it was correct on. This leaves us with\n# false positives - something the machine thought was a face but it wasn't. We see there are\n# two of these in the image, picking up shadow patterns or textures in shirts and matching\n# them with the haarcascades. Finally, we have true negatives, or the set of all possible\n# rectangles the machine learning classifier could consider where it correctly indicated that\n# the result was not a face. In this case there are many many true negatives.",
"_____no_output_____"
],
[
"# There are a few ways we could try and improve this, and really, it requires a lot of \n# experimentation to find good values for a given image. First, lets create a function\n# which will plot rectanges for us over the image\ndef show_rects(faces):\n #Lets read in our gif and convert it\n pil_img=Image.open('readonly/msi_recruitment.gif').convert(\"RGB\")\n # Set our drawing context\n drawing=ImageDraw.Draw(pil_img)\n # And plot all of the rectangles in faces\n for x,y,w,h in faces:\n drawing.rectangle((x,y,x+w,y+h), outline=\"white\")\n #Finally lets display this\n display(pil_img)",
"_____no_output_____"
],
[
"# Ok, first up, we could try and binarize this image. It turns out that opencv has a built in\n# binarization function called threshold(). You simply pass in the image, the midpoint, and\n# the maximum value, as well as a flag which indicates whether the threshold should be\n# binary or something else. Lets try this.\ncv_img_bin=cv.threshold(img,120,255,cv.THRESH_BINARY)[1] # returns a list, we want the second value\n# Now do the actual face detection\nfaces = face_cascade.detectMultiScale(cv_img_bin)\n# Now lets see the results\nshow_rects(faces)",
"_____no_output_____"
],
[
"# That's kind of interesting. Not better, but we do see that there is one false positive\n# towards the bottom, where the classifier detected the sunglasses as eyes and the dark shadow\n# line below as a mouth.\n#\n# If you're following in the notebook with this video, why don't you pause things and try a\n# few different parameters for the thresholding value?",
"_____no_output_____"
],
[
"# The detectMultiScale() function from OpenCV also has a couple of parameters. The first of\n# these is the scale factor. The scale factor changes the size of rectangles which are\n# considered against the model, that is, the haarcascades XML file. You can think of it as if\n# it were changing the size of the rectangles which are on the screen.\n#\n# Lets experiment with the scale factor. Usually it's a small value, lets try 1.05\nfaces = face_cascade.detectMultiScale(cv_img,1.05)\n# Show those results\nshow_rects(faces)\n# Now lets also try 1.15\nfaces = face_cascade.detectMultiScale(cv_img,1.15)\n# Show those results\nshow_rects(faces)\n# Finally lets also try 1.25\nfaces = face_cascade.detectMultiScale(cv_img,1.25)\n# Show those results\nshow_rects(faces)",
"_____no_output_____"
],
[
"# We can see that as we change the scale factor we change the number of true and \n# false positives and negatives. With the scale set to 1.05, we have 7 true positives,\n# which are correctly identified faces, and 3 false negatives, which are faces which\n# are there but not detected, and 3 false positives, where are non-faces which\n# opencv thinks are faces. When we change this to 1.15 we lose the false positives but\n# also lose one of the true positives, the person to the right wearing a hat. And\n# when we change this to 1.25 we lost more true positives as well.\n#\n# This is actually a really interesting phenomena in machine learning and artificial\n# intelligence. There is a trade off between not only how accurate a model is, but how\n# the inaccuracy actually happens. Which of these three models do you think is best?",
"_____no_output_____"
],
[
"# Well, the answer to that question is really, \"it depends\". It depends why you are trying\n# to detect faces, and what you are going to do with them. If you think these issues\n# are interesting, you might want to check out the Applied Data Science with Python\n# specialization Michigan offers on Coursera.\n#\n# Ok, beyond an opportunity to advertise, did you notice anything else that happened when\n# we changed the scale factor? It's subtle, but the speed at which the processing ran\n# took longer at smaller scale factors. This is because more subimages are being considered\n# for these scales. This could also affect which method we might use.\n#\n# Jupyter has nice support for timing commands. You might have seen this before, a line\n# that starts with a percentage sign in jupyter is called a \"magic function\". This isn't\n# normal python - it's actually a shorthand way of writing a function which Jupyter\n# has predefined. It looks a lot like the decorators we talked about in a previous\n# lecture, but the magic functions were around long before decorators were part of the\n# python language. One of the built-in magic functions in juptyer is called timeit, and this\n# repeats a piece of python ten times (by default) and tells you the average speed it\n# took to complete.\n#\n# Lets time the speed of detectmultiscale when using a scale of 1.05\n%timeit face_cascade.detectMultiScale(cv_img,1.05)",
"_____no_output_____"
],
[
"# Ok, now lets compare that to the speed at scale = 1.15\n%timeit face_cascade.detectMultiScale(cv_img,1.15)",
"_____no_output_____"
],
[
"# You can see that this is a dramatic difference, roughly two and a half times slower\n# when using the smaller scale!\n#\n# This wraps up our discussion of detecting faces in opencv. You'll see that, like OCR, this\n# is not a foolproof process. But we can build on the work others have done in machine learning\n# and leverage powerful libraries to bring us closer to building a turn key python-based\n# solution. Remember that the detection mechanism isn't specific to faces, that's just the\n# haarcascades training data we used. On the web you'll be able to find other training data\n# to detect other objects, including eyes, animals, and so forth.",
"_____no_output_____"
]
],
[
[
"## More Jupyter Widgets",
"_____no_output_____"
]
],
[
[
"# One of the nice things about using the Jupyter notebook systems is that there is a\n# rich set of contributed plugins that seek to extend this system. In this lecture I\n# want to introduce you to one such plugin, call ipy web rtc. Webrtc is a fairly new\n# protocol for real time communication on the web. Yup, I'm talking about chatting.\n# The widget brings this to the Jupyter notebook system. Lets take a look.\n#\n# First, lets import from this library two different classes which we'll use in a\n# demo, one for the camera and one for images.\nfrom ipywebrtc import CameraStream, ImageRecorder\n# Then lets take a look at the camera stream object\nhelp(CameraStream)",
"_____no_output_____"
],
[
"# We see from the docs that it's east to get a camera facing the user, and we can have\n# the audio on or off. We don't need audio for this demo, so lets create a new camera\n# instance\ncamera = CameraStream.facing_user(audio=False)\n# The next object we want to look at is the ImageRecorder\nhelp(ImageRecorder)",
"_____no_output_____"
],
[
"# The image recorder lets us actually grab images from the camera stream. There are features\n# for downloading and using the image as well. We see that the default format is a png file.\n# Lets hook up the ImageRecorder to our stream\nimage_recorder = ImageRecorder(stream=camera)\n# Now, the docs are a little unclear how to use this within Jupyter, but if we call the\n# download() function it will actually store the results of the camera which is hooked up\n# in image_recorder.image. Lets try it out\n# First, lets tell the recorder to start capturing data\nimage_recorder.recording=True\n# Now lets download the image\nimage_recorder.download()\n# Then lets inspect the type of the image\ntype(image_recorder.image)",
"_____no_output_____"
],
[
"# Ok, the object that it stores is an ipywidgets.widgets.widget_media.Image. How do we do\n# something useful with this? Well, an inspection of the object shows that there is a handy\n# value field which actually holds the bytes behind the image. And we know how to display\n# those.\n# Lets import PIL Image\nimport PIL.Image\n# And lets import io\nimport io\n# And now lets create a PIL image from the bytes\nimg = PIL.Image.open(io.BytesIO(image_recorder.image.value))\n# And render it to the screen\ndisplay(img)",
"_____no_output_____"
],
[
"# Great, you see a picture! Hopefully you are following along in one of the notebooks\n# and have been able to try this out for yourself!\n#\n# What can you do with this? This is a great way to get started with a bit of computer vision.\n# You already know how to identify a face in the webcam picture, or try and capture text\n# from within the picture. With OpenCV there are any number of other things you can do, simply\n# with a webcam, the Jupyter notebooks, and python!",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code"
]
] |
cbd8c0df012599d3bdd9cfd6105472924764c1d3
| 26,233 |
ipynb
|
Jupyter Notebook
|
Ch4/07_BERT_Sentiment_Classification_IMDB_ktrain.ipynb
|
quicksilverTrx/practical-nlp
|
bb372ec32885a7b25da7857434e1c9f7a100a94e
|
[
"MIT"
] | 5 |
2020-12-02T23:17:57.000Z
|
2021-04-14T01:17:09.000Z
|
Ch4/07_BERT_Sentiment_Classification_IMDB_ktrain.ipynb
|
quicksilverTrx/practical-nlp
|
bb372ec32885a7b25da7857434e1c9f7a100a94e
|
[
"MIT"
] | 5 |
2021-08-23T20:56:47.000Z
|
2022-02-10T04:38:21.000Z
|
Ch4/07_BERT_Sentiment_Classification_IMDB_ktrain.ipynb
|
quicksilverTrx/practical-nlp
|
bb372ec32885a7b25da7857434e1c9f7a100a94e
|
[
"MIT"
] | 3 |
2020-12-02T23:42:01.000Z
|
2021-03-03T23:04:00.000Z
| 52.049603 | 453 | 0.582549 |
[
[
[
"#### We need to install the ktrain library. Its a light weight wrapper for keras to help train neural networks. With only a few lines of code it allows you to build models, estimate optimal learning rate, loading and preprocessing text and image data from various sources and much more. More about our approach can be found at [this](https://towardsdatascience.com/bert-text-classification-in-3-lines-of-code-using-keras-264db7e7a358) article.",
"_____no_output_____"
]
],
[
[
"!pip3 install ktrain==0.2.2",
"Collecting ktrain==0.2.2\n\u001b[?25l Downloading https://files.pythonhosted.org/packages/83/ce/f8dd172bec1486c02f20cc5099055fb2e8850fc414eb7bc922f29e4e13ec/ktrain-0.2.2.tar.gz (43kB)\n\r\u001b[K |███████▋ | 10kB 26.3MB/s eta 0:00:01\r\u001b[K |███████████████▎ | 20kB 1.7MB/s eta 0:00:01\r\u001b[K |██████████████████████▉ | 30kB 2.5MB/s eta 0:00:01\r\u001b[K |██████████████████████████████▌ | 40kB 1.7MB/s eta 0:00:01\r\u001b[K |████████████████████████████████| 51kB 1.8MB/s \n\u001b[?25hRequirement already satisfied: keras>=2.2.4 in /usr/local/lib/python3.6/dist-packages (from ktrain==0.2.2) (2.2.5)\nRequirement already satisfied: scikit-learn>=0.20.0 in /usr/local/lib/python3.6/dist-packages (from ktrain==0.2.2) (0.22.1)\nRequirement already satisfied: matplotlib>=3.0.0 in /usr/local/lib/python3.6/dist-packages (from ktrain==0.2.2) (3.1.2)\nRequirement already satisfied: pandas>=0.24.2 in /usr/local/lib/python3.6/dist-packages (from ktrain==0.2.2) (0.25.3)\nRequirement already satisfied: fastprogress>=0.1.21 in /usr/local/lib/python3.6/dist-packages (from ktrain==0.2.2) (0.2.2)\nCollecting keras_bert\n Downloading https://files.pythonhosted.org/packages/2c/0f/cdc886c1018943ea62d3209bc964413d5aa9d0eb7e493abd8545be679294/keras-bert-0.81.0.tar.gz\nRequirement already satisfied: requests in /usr/local/lib/python3.6/dist-packages (from ktrain==0.2.2) (2.21.0)\nRequirement already satisfied: six>=1.9.0 in /usr/local/lib/python3.6/dist-packages (from keras>=2.2.4->ktrain==0.2.2) (1.12.0)\nRequirement already satisfied: h5py in /usr/local/lib/python3.6/dist-packages (from keras>=2.2.4->ktrain==0.2.2) (2.8.0)\nRequirement already satisfied: numpy>=1.9.1 in /usr/local/lib/python3.6/dist-packages (from keras>=2.2.4->ktrain==0.2.2) (1.17.5)\nRequirement already satisfied: keras-applications>=1.0.8 in /usr/local/lib/python3.6/dist-packages (from keras>=2.2.4->ktrain==0.2.2) (1.0.8)\nRequirement already satisfied: keras-preprocessing>=1.1.0 in /usr/local/lib/python3.6/dist-packages (from keras>=2.2.4->ktrain==0.2.2) (1.1.0)\nRequirement already satisfied: pyyaml in /usr/local/lib/python3.6/dist-packages (from keras>=2.2.4->ktrain==0.2.2) (3.13)\nRequirement already satisfied: scipy>=0.14 in /usr/local/lib/python3.6/dist-packages (from keras>=2.2.4->ktrain==0.2.2) (1.4.1)\nRequirement already satisfied: joblib>=0.11 in /usr/local/lib/python3.6/dist-packages (from scikit-learn>=0.20.0->ktrain==0.2.2) (0.14.1)\nRequirement already satisfied: kiwisolver>=1.0.1 in /usr/local/lib/python3.6/dist-packages (from matplotlib>=3.0.0->ktrain==0.2.2) (1.1.0)\nRequirement already satisfied: python-dateutil>=2.1 in /usr/local/lib/python3.6/dist-packages (from matplotlib>=3.0.0->ktrain==0.2.2) (2.6.1)\nRequirement already satisfied: pyparsing!=2.0.4,!=2.1.2,!=2.1.6,>=2.0.1 in /usr/local/lib/python3.6/dist-packages (from matplotlib>=3.0.0->ktrain==0.2.2) (2.4.6)\nRequirement already satisfied: cycler>=0.10 in /usr/local/lib/python3.6/dist-packages (from matplotlib>=3.0.0->ktrain==0.2.2) (0.10.0)\nRequirement already satisfied: pytz>=2017.2 in /usr/local/lib/python3.6/dist-packages (from pandas>=0.24.2->ktrain==0.2.2) (2018.9)\nCollecting keras-transformer>=0.30.0\n Downloading https://files.pythonhosted.org/packages/54/0c/fede535ac576c03863c44bf2e0bf051fe21f5e10103631b6b6236ae446f3/keras-transformer-0.32.0.tar.gz\nRequirement already satisfied: idna<2.9,>=2.5 in /usr/local/lib/python3.6/dist-packages (from requests->ktrain==0.2.2) (2.8)\nRequirement already satisfied: urllib3<1.25,>=1.21.1 in /usr/local/lib/python3.6/dist-packages (from requests->ktrain==0.2.2) (1.24.3)\nRequirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.6/dist-packages (from requests->ktrain==0.2.2) (2019.11.28)\nRequirement already satisfied: chardet<3.1.0,>=3.0.2 in /usr/local/lib/python3.6/dist-packages (from requests->ktrain==0.2.2) (3.0.4)\nRequirement already satisfied: setuptools in /usr/local/lib/python3.6/dist-packages (from kiwisolver>=1.0.1->matplotlib>=3.0.0->ktrain==0.2.2) (45.1.0)\nCollecting keras-pos-embd>=0.10.0\n Downloading https://files.pythonhosted.org/packages/09/70/b63ed8fc660da2bb6ae29b9895401c628da5740c048c190b5d7107cadd02/keras-pos-embd-0.11.0.tar.gz\nCollecting keras-multi-head>=0.22.0\n Downloading https://files.pythonhosted.org/packages/40/3e/d0a64bb2ac5217928effe4507c26bbd19b86145d16a1948bc2d4f4c6338a/keras-multi-head-0.22.0.tar.gz\nCollecting keras-layer-normalization>=0.12.0\n Downloading https://files.pythonhosted.org/packages/a4/0e/d1078df0494bac9ce1a67954e5380b6e7569668f0f3b50a9531c62c1fc4a/keras-layer-normalization-0.14.0.tar.gz\nCollecting keras-position-wise-feed-forward>=0.5.0\n Downloading https://files.pythonhosted.org/packages/e3/59/f0faa1037c033059e7e9e7758e6c23b4d1c0772cd48de14c4b6fd4033ad5/keras-position-wise-feed-forward-0.6.0.tar.gz\nCollecting keras-embed-sim>=0.7.0\n Downloading https://files.pythonhosted.org/packages/bc/20/735fd53f6896e2af63af47e212601c1b8a7a80d00b6126c388c9d1233892/keras-embed-sim-0.7.0.tar.gz\nCollecting keras-self-attention==0.41.0\n Downloading https://files.pythonhosted.org/packages/1b/1c/01599219bef7266fa43b3316e4f55bcb487734d3bafdc60ffd564f3cfe29/keras-self-attention-0.41.0.tar.gz\nBuilding wheels for collected packages: ktrain, keras-bert, keras-transformer, keras-pos-embd, keras-multi-head, keras-layer-normalization, keras-position-wise-feed-forward, keras-embed-sim, keras-self-attention\n Building wheel for ktrain (setup.py) ... \u001b[?25l\u001b[?25hdone\n Created wheel for ktrain: filename=ktrain-0.2.2-cp36-none-any.whl size=50418 sha256=30ac3da49ca22b1ec10c57ae08d3bab9756222409cd3463f85ef190fe66dc8d0\n Stored in directory: /root/.cache/pip/wheels/fa/21/6f/e0452e1564ccab53339b601dbd1f37ed81e34868826200e8f8\n Building wheel for keras-bert (setup.py) ... \u001b[?25l\u001b[?25hdone\n Created wheel for keras-bert: filename=keras_bert-0.81.0-cp36-none-any.whl size=37913 sha256=180438bed7964563b41411bcd633f83794ceb0ac2817898291612600cfb39f3a\n Stored in directory: /root/.cache/pip/wheels/bd/27/da/ffc2d573aa48b87440ec4f98bc7c992e3a2d899edb2d22ef9e\n Building wheel for keras-transformer (setup.py) ... \u001b[?25l\u001b[?25hdone\n Created wheel for keras-transformer: filename=keras_transformer-0.32.0-cp36-none-any.whl size=13266 sha256=4f32a5e7c46dec4f5dbeb040523f6c94b46048bb5eab2f85a4ee1c71b9b71206\n Stored in directory: /root/.cache/pip/wheels/62/f0/ce/82fa5d024d5ef8e263f26a50dcee23820efe245680ce9c922a\n Building wheel for keras-pos-embd (setup.py) ... \u001b[?25l\u001b[?25hdone\n Created wheel for keras-pos-embd: filename=keras_pos_embd-0.11.0-cp36-none-any.whl size=7554 sha256=c6c8607ddbc95fc5c3c2412f94d41d2d6780d3b30004543791e9313492b2de49\n Stored in directory: /root/.cache/pip/wheels/5b/a1/a0/ce6b1d49ba1a9a76f592e70cf297b05c96bc9f418146761032\n Building wheel for keras-multi-head (setup.py) ... \u001b[?25l\u001b[?25hdone\n Created wheel for keras-multi-head: filename=keras_multi_head-0.22.0-cp36-none-any.whl size=15371 sha256=05c6dc4ab2d005d29f310419143e25714eb6a7ffb169e008d1c18666019fcf5d\n Stored in directory: /root/.cache/pip/wheels/bb/df/3f/81b36f41b66e6a9cd69224c70a737de2bb6b2f7feb3272c25e\n Building wheel for keras-layer-normalization (setup.py) ... \u001b[?25l\u001b[?25hdone\n Created wheel for keras-layer-normalization: filename=keras_layer_normalization-0.14.0-cp36-none-any.whl size=5268 sha256=affb377c258c1f72bd3be271eb6a578a1be567626b082c7d51c96af3c685707f\n Stored in directory: /root/.cache/pip/wheels/54/80/22/a638a7d406fd155e507aa33d703e3fa2612b9eb7bb4f4fe667\n Building wheel for keras-position-wise-feed-forward (setup.py) ... \u001b[?25l\u001b[?25hdone\n Created wheel for keras-position-wise-feed-forward: filename=keras_position_wise_feed_forward-0.6.0-cp36-none-any.whl size=5623 sha256=794c37b5c74aa2d635f3581d8f1e82e8e4e08f9709c8b08beba747cc84d786a2\n Stored in directory: /root/.cache/pip/wheels/39/e2/e2/3514fef126a00574b13bc0b9e23891800158df3a3c19c96e3b\n Building wheel for keras-embed-sim (setup.py) ... \u001b[?25l\u001b[?25hdone\n Created wheel for keras-embed-sim: filename=keras_embed_sim-0.7.0-cp36-none-any.whl size=4676 sha256=922770c4ee921ab176dd5e405a9a0c30a2abd4bbb9b02b313f81308d1c00cd3f\n Stored in directory: /root/.cache/pip/wheels/d1/bc/b1/b0c45cee4ca2e6c86586b0218ffafe7f0703c6d07fdf049866\n Building wheel for keras-self-attention (setup.py) ... \u001b[?25l\u001b[?25hdone\n Created wheel for keras-self-attention: filename=keras_self_attention-0.41.0-cp36-none-any.whl size=17288 sha256=1ad03d750cc19684d453b79f17fcc64565051904d5c87d91d0c17fb6a9bbced4\n Stored in directory: /root/.cache/pip/wheels/cc/dc/17/84258b27a04cd38ac91998abe148203720ca696186635db694\nSuccessfully built ktrain keras-bert keras-transformer keras-pos-embd keras-multi-head keras-layer-normalization keras-position-wise-feed-forward keras-embed-sim keras-self-attention\nInstalling collected packages: keras-pos-embd, keras-self-attention, keras-multi-head, keras-layer-normalization, keras-position-wise-feed-forward, keras-embed-sim, keras-transformer, keras-bert, ktrain\nSuccessfully installed keras-bert-0.81.0 keras-embed-sim-0.7.0 keras-layer-normalization-0.14.0 keras-multi-head-0.22.0 keras-pos-embd-0.11.0 keras-position-wise-feed-forward-0.6.0 keras-self-attention-0.41.0 keras-transformer-0.32.0 ktrain-0.2.2\n"
],
[
"#Importing\nimport ktrain\nfrom ktrain import text",
"Using TensorFlow backend.\n"
],
[
"#obtain the dataset\nimport tensorflow as tf\ndataset = tf.keras.utils.get_file(\n fname=\"aclImdb.tar.gz\", \n origin=\"http://ai.stanford.edu/~amaas/data/sentiment/aclImdb_v1.tar.gz\", \n extract=True,\n)",
"Downloading data from http://ai.stanford.edu/~amaas/data/sentiment/aclImdb_v1.tar.gz\n84131840/84125825 [==============================] - 4s 0us/step\n"
],
[
"%cd /root/.keras/datasets/aclImdb\n!ls",
"/root/.keras/datasets/aclImdb\nimdbEr.txt imdb.vocab\tREADME\ttest train\n"
],
[
"# set path to dataset\nimport os.path\ndataset = '/root/.keras/datasets/aclImdb'\nIMDB_DATADIR = os.path.join(os.path.dirname(dataset), 'aclImdb')\nprint(IMDB_DATADIR)",
"/root/.keras/datasets/aclImdb\n"
]
],
[
[
"## STEP 1: Preprocessing\n####The texts_from_folder function will load the training and validation data from the specified folder and automatically preprocess it according to BERT's requirements. In doing so, the BERT model and vocabulary will be automatically downloaded.",
"_____no_output_____"
]
],
[
[
"\n\n(x_train, y_train), (x_test, y_test), preproc = text.texts_from_folder(IMDB_DATADIR, \n maxlen=500, \n preprocess_mode='bert',\n train_test_names=['train', \n 'test'],\n classes=['pos', 'neg'])",
"downloading pretrained BERT model and vocabulary...\n[██████████████████████████████████████████████████]\nextracting pretrained BERT model and vocabulary...\ndone.\n\ncleanup downloaded zip...\ndone.\n\npreprocessing train...\n"
]
],
[
[
"### STEP 2: Loading a pre trained BERT and wrapping it in a ktrain.learner object",
"_____no_output_____"
]
],
[
[
"\nmodel = text.text_classifier('bert', (x_train, y_train), preproc=preproc)\nlearner = ktrain.get_learner(model,train_data=(x_train, y_train), val_data=(x_test, y_test), batch_size=6)",
"Is Multi-Label? False\nmaxlen is 500\nWARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:541: The name tf.placeholder is deprecated. Please use tf.compat.v1.placeholder instead.\n\nWARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:4432: The name tf.random_uniform is deprecated. Please use tf.random.uniform instead.\n\nWARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:66: The name tf.get_default_graph is deprecated. Please use tf.compat.v1.get_default_graph instead.\n\nWARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:148: The name tf.placeholder_with_default is deprecated. Please use tf.compat.v1.placeholder_with_default instead.\n\nWARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:3733: calling dropout (from tensorflow.python.ops.nn_ops) with keep_prob is deprecated and will be removed in a future version.\nInstructions for updating:\nPlease use `rate` instead of `keep_prob`. Rate should be set to `rate = 1 - keep_prob`.\nWARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:4479: The name tf.truncated_normal is deprecated. Please use tf.random.truncated_normal instead.\n\nWARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:190: The name tf.get_default_session is deprecated. Please use tf.compat.v1.get_default_session instead.\n\nWARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:197: The name tf.ConfigProto is deprecated. Please use tf.compat.v1.ConfigProto instead.\n\nWARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:203: The name tf.Session is deprecated. Please use tf.compat.v1.Session instead.\n\nWARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:207: The name tf.global_variables is deprecated. Please use tf.compat.v1.global_variables instead.\n\nWARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:216: The name tf.is_variable_initialized is deprecated. Please use tf.compat.v1.is_variable_initialized instead.\n\nWARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:223: The name tf.variables_initializer is deprecated. Please use tf.compat.v1.variables_initializer instead.\n\nWARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/optimizers.py:793: The name tf.train.Optimizer is deprecated. Please use tf.compat.v1.train.Optimizer instead.\n\nWARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:3576: The name tf.log is deprecated. Please use tf.math.log instead.\n\ndone.\n"
]
],
[
[
"### STEP 3: Training and Tuning the model's parameters",
"_____no_output_____"
]
],
[
[
"learner.fit_onecycle(2e-5, 4)",
"\n\nbegin training using onecycle policy with max lr of 2e-05...\nWARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/tensorflow_core/python/ops/math_grad.py:1424: where (from tensorflow.python.ops.array_ops) is deprecated and will be removed in a future version.\nInstructions for updating:\nUse tf.where in 2.0, which has the same broadcast rule as np.where\nWARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:1033: The name tf.assign_add is deprecated. Please use tf.compat.v1.assign_add instead.\n\nWARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:1020: The name tf.assign is deprecated. Please use tf.compat.v1.assign instead.\n\nTrain on 25000 samples, validate on 25000 samples\nEpoch 1/4\n25000/25000 [==============================] - 4557s 182ms/step - loss: 0.2551 - acc: 0.8944 - val_loss: 0.1694 - val_acc: 0.9342\nEpoch 2/4\n25000/25000 [==============================] - 4545s 182ms/step - loss: 0.1607 - acc: 0.9419 - val_loss: 0.1692 - val_acc: 0.9362\nEpoch 3/4\n25000/25000 [==============================] - 4545s 182ms/step - loss: 0.0875 - acc: 0.9706 - val_loss: 0.1860 - val_acc: 0.9378\nEpoch 4/4\n25000/25000 [==============================] - 4543s 182ms/step - loss: 0.0242 - acc: 0.9932 - val_loss: 0.2394 - val_acc: 0.9385\n"
],
[
"",
"_____no_output_____"
],
[
"",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
]
] |
cbd8c83cc594a689aa89382d90c235639b7b03d8
| 44,887 |
ipynb
|
Jupyter Notebook
|
model_monitoring.ipynb
|
giranntu/EDA_to_pipeline
|
ff2f30724616b400b681502baf6ec1767f82bbf4
|
[
"Apache-2.0"
] | 2 |
2021-09-17T04:17:13.000Z
|
2021-10-08T04:33:55.000Z
|
model_monitoring.ipynb
|
giranntu/EDA_to_pipeline
|
ff2f30724616b400b681502baf6ec1767f82bbf4
|
[
"Apache-2.0"
] | null | null | null |
model_monitoring.ipynb
|
giranntu/EDA_to_pipeline
|
ff2f30724616b400b681502baf6ec1767f82bbf4
|
[
"Apache-2.0"
] | 1 |
2021-09-14T00:52:09.000Z
|
2021-09-14T00:52:09.000Z
| 41.142988 | 642 | 0.623031 |
[
[
[
"# @title Copyright & License (click to expand)\n# Copyright 2021 Google LLC\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# https://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.",
"_____no_output_____"
]
],
[
[
"# Vertex Model Monitoring\n\n<table align=\"left\">\n <td>\n <a href=\"https://console.cloud.google.com/ai-platform/notebooks/deploy-notebook?name=Model%20Monitoring&download_url=https%3A%2F%2Fraw.githubusercontent.com%2FGoogleCloudPlatform%2Fai-platform-samples%2Fmaster%2Fai-platform-unified%2Fnotebooks%2Fofficial%2Fmodel_monitoring%2Fmodel_monitoring.ipynb\">\n <img src=\"https://cloud.google.com/images/products/ai/ai-solutions-icon.svg\" alt=\"Google Cloud Notebooks\"> Open in GCP Notebooks\n </a>\n </td> \n <td>\n <a href=\"https://colab.research.google.com/github/GoogleCloudPlatform/ai-platform-samples/blob/master/ai-platform-unified/notebooks/official/model_monitoring/model_monitoring.ipynb\">\n <img src=\"https://cloud.google.com/ml-engine/images/colab-logo-32px.png\" alt=\"Colab logo\"> Open in Colab\n </a>\n </td>\n <td>\n <a href=\"https://github.com/GoogleCloudPlatform/ai-platform-samples/blob/master/ai-platform-unified/notebooks/official/model_monitoring/model_monitoring.ipynb\">\n <img src=\"https://cloud.google.com/ml-engine/images/github-logo-32px.png\" alt=\"GitHub logo\">\n View on GitHub\n </a>\n </td>\n</table>",
"_____no_output_____"
],
[
"## Overview",
"_____no_output_____"
],
[
"### What is Model Monitoring?\n\nModern applications rely on a well established set of capabilities to monitor the health of their services. Examples include:\n\n* software versioning\n* rigorous deployment processes\n* event logging\n* alerting/notication of situations requiring intervention\n* on-demand and automated diagnostic tracing\n* automated performance and functional testing\n\nYou should be able to manage your ML services with the same degree of power and flexibility with which you can manage your applications. That's what MLOps is all about - managing ML services with the best practices Google and the broader computing industry have learned from generations of experience deploying well engineered, reliable, and scalable services.\n\nModel monitoring is only one piece of the ML Ops puzzle - it helps answer the following questions:\n\n* How well do recent service requests match the training data used to build your model? This is called **training-serving skew**.\n* How significantly are service requests evolving over time? This is called **drift detection**.\n\nIf production traffic differs from training data, or varies substantially over time, that's likely to impact the quality of the answers your model produces. When that happens, you'd like to be alerted automatically and responsively, so that **you can anticipate problems before they affect your customer experiences or your revenue streams**.",
"_____no_output_____"
],
[
"### Objective\n\nIn this notebook, you will learn how to... \n\n* deploy a pre-trained model\n* configure model monitoring\n* generate some artificial traffic\n* understand how to interpret the statistics, visualizations, other data reported by the model monitoring feature",
"_____no_output_____"
],
[
"### Costs \n\nThis tutorial uses billable components of Google Cloud:\n\n* Vertex AI\n* BigQuery\n\nLearn about [Vertex AI\npricing](https://cloud.google.com/vertex-ai/pricing) and [Cloud Storage\npricing](https://cloud.google.com/storage/pricing), and use the [Pricing\nCalculator](https://cloud.google.com/products/calculator/)\nto generate a cost estimate based on your projected usage.",
"_____no_output_____"
],
[
"### The example model\n\nThe model you'll use in this notebook is based on [this blog post](https://cloud.google.com/blog/topics/developers-practitioners/churn-prediction-game-developers-using-google-analytics-4-ga4-and-bigquery-ml). The idea behind this model is that your company has extensive log data describing how your game users have interacted with the site. The raw data contains the following categories of information:\n\n- identity - unique player identitity numbers\n- demographic features - information about the player, such as the geographic region in which a player is located\n- behavioral features - counts of the number of times a player has triggered certain game events, such as reaching a new level\n- churn propensity - this is the label or target feature, it provides an estimated probability that this player will churn, i.e. stop being an active player.\n\nThe blog article referenced above explains how to use BigQuery to store the raw data, pre-process it for use in machine learning, and train a model. Because this notebook focuses on model monitoring, rather than training models, you're going to reuse a pre-trained version of this model, which has been exported to Google Cloud Storage. In the next section, you will setup your environment and import this model into your own project.",
"_____no_output_____"
],
[
"## Before you begin",
"_____no_output_____"
],
[
"### Setup your dependencies",
"_____no_output_____"
]
],
[
[
"import os\nimport sys\n\nassert sys.version_info.major == 3, \"This notebook requires Python 3.\"\n\n# Google Cloud Notebook requires dependencies to be installed with '--user'\nIS_GOOGLE_CLOUD_NOTEBOOK = os.path.exists(\"/opt/deeplearning/metadata/env_version\")\nUSER_FLAG = \"\"\nif IS_GOOGLE_CLOUD_NOTEBOOK:\n USER_FLAG = \"--user\"\n\nif 'google.colab' in sys.modules: \n from google.colab import auth\n auth.authenticate_user()\n\n# Install Python package dependencies.\n! pip3 install {USER_FLAG} --quiet --upgrade google-api-python-client google-auth-oauthlib \\\n google-auth-httplib2 oauth2client requests \\\n google-cloud-aiplatform google-cloud-storage==1.32.0\n\nif not os.getenv(\"IS_TESTING\"): \n # Automatically restart kernel after installs\n import IPython\n app = IPython.Application.instance() \n app.kernel.do_shutdown(True)",
"_____no_output_____"
]
],
[
[
"### Set up your Google Cloud project\n\n**The following steps are required, regardless of your notebook environment.**\n\n1. [Select or create a Google Cloud project](https://console.cloud.google.com/cloud-resource-manager). When you first create an account, you get a $300 free credit towards your compute/storage costs.\n\n1. Enter your project id in the first line of the cell below.\n\n1. [Make sure that billing is enabled for your project](https://cloud.google.com/billing/docs/how-to/modify-project).\n\n1. If you are running this notebook locally, you will need to install the [Cloud SDK](https://cloud.google.com/sdk).\n\n1. You'll use the *gcloud* command throughout this notebook. In the following cell, enter your project name and run the cell to authenticate yourself with the Google Cloud and initialize your *gcloud* configuration settings.\n\n**Model monitoring is currently supported in regions us-central1, europe-west4, asia-east1, and asia-southeast1. To keep things simple for this lab, we're going to use region us-central1 for all our resources (BigQuery training data, Cloud Storage bucket, model and endpoint locations, etc.). You can use any supported region, so long as all resources are co-located.**",
"_____no_output_____"
]
],
[
[
"# Import globally needed dependencies here, after kernel restart.\nimport copy\nimport numpy as np\nimport os\nimport pprint as pp\nimport random\nimport sys\nimport time\n\nPROJECT_ID = \"[your-project-id]\" # @param {type:\"string\"}\nREGION = \"us-central1\" # @param {type:\"string\"}\nSUFFIX = \"aiplatform.googleapis.com\"\nAPI_ENDPOINT = f\"{REGION}-{SUFFIX}\"\nPREDICT_API_ENDPOINT = f\"{REGION}-prediction-{SUFFIX}\"\nif os.getenv(\"IS_TESTING\"):\n !gcloud --quiet components install beta\n !gcloud --quiet components update\n!gcloud config set project $PROJECT_ID\n!gcloud config set ai/region $REGION",
"_____no_output_____"
]
],
[
[
"### Login to your Google Cloud account and enable AI services",
"_____no_output_____"
]
],
[
[
"# The Google Cloud Notebook product has specific requirements\nIS_GOOGLE_CLOUD_NOTEBOOK = os.path.exists(\"/opt/deeplearning/metadata/env_version\")\n\n# If on Google Cloud Notebooks, then don't execute this code\nif not IS_GOOGLE_CLOUD_NOTEBOOK:\n if \"google.colab\" in sys.modules:\n from google.colab import auth as google_auth\n\n google_auth.authenticate_user()\n\n # If you are running this notebook locally, replace the string below with the\n # path to your service account key and run this cell to authenticate your GCP\n # account.\n elif not os.getenv(\"IS_TESTING\"):\n %env GOOGLE_APPLICATION_CREDENTIALS ''\n\n!gcloud services enable aiplatform.googleapis.com",
"_____no_output_____"
]
],
[
[
"### Define utilities\n\nRun the following cells to define some utility functions and distributions used later in this notebook. Although these utilities are not critical to understand the main concepts, feel free to expand the cells\nin this section if you're curious or want to dive deeper into how some of your API requests are made.",
"_____no_output_____"
]
],
[
[
"# @title Utility imports and constants\n\nfrom google.cloud.aiplatform_v1beta1.services.endpoint_service import \\\n EndpointServiceClient\nfrom google.cloud.aiplatform_v1beta1.services.job_service import \\\n JobServiceClient\nfrom google.cloud.aiplatform_v1beta1.services.prediction_service import \\\n PredictionServiceClient\nfrom google.cloud.aiplatform_v1beta1.types.io import BigQuerySource\nfrom google.cloud.aiplatform_v1beta1.types.model_deployment_monitoring_job import (\n ModelDeploymentMonitoringJob, ModelDeploymentMonitoringObjectiveConfig,\n ModelDeploymentMonitoringScheduleConfig)\nfrom google.cloud.aiplatform_v1beta1.types.model_monitoring import (\n ModelMonitoringAlertConfig, ModelMonitoringObjectiveConfig,\n SamplingStrategy, ThresholdConfig)\nfrom google.cloud.aiplatform_v1beta1.types.prediction_service import \\\n PredictRequest\nfrom google.protobuf import json_format\nfrom google.protobuf.duration_pb2 import Duration\nfrom google.protobuf.struct_pb2 import Value\n\n# This is the default value at which you would like the monitoring function to trigger an alert.\n# In other words, this value fine tunes the alerting sensitivity. This threshold can be customized\n# on a per feature basis but this is the global default setting.\nDEFAULT_THRESHOLD_VALUE = 0.001",
"_____no_output_____"
],
[
"# @title Utility functions\n\ndef create_monitoring_job(objective_configs):\n # Create sampling configuration.\n random_sampling = SamplingStrategy.RandomSampleConfig(sample_rate=LOG_SAMPLE_RATE)\n sampling_config = SamplingStrategy(random_sample_config=random_sampling)\n\n # Create schedule configuration.\n duration = Duration(seconds=MONITOR_INTERVAL)\n schedule_config = ModelDeploymentMonitoringScheduleConfig(monitor_interval=duration)\n\n # Create alerting configuration.\n emails = [USER_EMAIL]\n email_config = ModelMonitoringAlertConfig.EmailAlertConfig(user_emails=emails)\n alerting_config = ModelMonitoringAlertConfig(email_alert_config=email_config)\n\n # Create the monitoring job.\n endpoint = f\"projects/{PROJECT_ID}/locations/{REGION}/endpoints/{ENDPOINT_ID}\"\n predict_schema = \"\"\n analysis_schema = \"\"\n job = ModelDeploymentMonitoringJob(\n display_name=JOB_NAME,\n endpoint=endpoint,\n model_deployment_monitoring_objective_configs=objective_configs,\n logging_sampling_strategy=sampling_config,\n model_deployment_monitoring_schedule_config=schedule_config,\n model_monitoring_alert_config=alerting_config,\n predict_instance_schema_uri=predict_schema,\n analysis_instance_schema_uri=analysis_schema,\n )\n options = dict(api_endpoint=API_ENDPOINT)\n client = JobServiceClient(client_options=options)\n parent = f\"projects/{PROJECT_ID}/locations/{REGION}\"\n response = client.create_model_deployment_monitoring_job(\n parent=parent, model_deployment_monitoring_job=job\n )\n print(\"Created monitoring job:\")\n print(response)\n return response\n\n\ndef get_thresholds(default_thresholds, custom_thresholds):\n thresholds = {}\n default_threshold = ThresholdConfig(value=DEFAULT_THRESHOLD_VALUE)\n for feature in default_thresholds.split(\",\"):\n feature = feature.strip()\n thresholds[feature] = default_threshold\n for custom_threshold in custom_thresholds.split(\",\"):\n pair = custom_threshold.split(\":\")\n if len(pair) != 2:\n print(f\"Invalid custom skew threshold: {custom_threshold}\")\n return\n feature, value = pair\n thresholds[feature] = ThresholdConfig(value=float(value))\n return thresholds\n\n\ndef get_deployed_model_ids(endpoint_id):\n client_options = dict(api_endpoint=API_ENDPOINT)\n client = EndpointServiceClient(client_options=client_options)\n parent = f\"projects/{PROJECT_ID}/locations/{REGION}\"\n response = client.get_endpoint(name=f\"{parent}/endpoints/{endpoint_id}\")\n model_ids = []\n for model in response.deployed_models:\n model_ids.append(model.id)\n return model_ids\n\n\ndef set_objectives(model_ids, objective_template):\n # Use the same objective config for all models.\n objective_configs = []\n for model_id in model_ids:\n objective_config = copy.deepcopy(objective_template)\n objective_config.deployed_model_id = model_id\n objective_configs.append(objective_config)\n return objective_configs\n\n\ndef send_predict_request(endpoint, input):\n client_options = {\"api_endpoint\": PREDICT_API_ENDPOINT}\n client = PredictionServiceClient(client_options=client_options)\n params = {}\n params = json_format.ParseDict(params, Value())\n request = PredictRequest(endpoint=endpoint, parameters=params)\n inputs = [json_format.ParseDict(input, Value())]\n request.instances.extend(inputs)\n response = client.predict(request)\n return response\n\n\ndef list_monitoring_jobs():\n client_options = dict(api_endpoint=API_ENDPOINT)\n parent = f\"projects/{PROJECT_ID}/locations/us-central1\"\n client = JobServiceClient(client_options=client_options)\n response = client.list_model_deployment_monitoring_jobs(parent=parent)\n print(response)\n\n\ndef pause_monitoring_job(job):\n client_options = dict(api_endpoint=API_ENDPOINT)\n client = JobServiceClient(client_options=client_options)\n response = client.pause_model_deployment_monitoring_job(name=job)\n print(response)\n\n\ndef delete_monitoring_job(job):\n client_options = dict(api_endpoint=API_ENDPOINT)\n client = JobServiceClient(client_options=client_options)\n response = client.delete_model_deployment_monitoring_job(name=job)\n print(response)",
"_____no_output_____"
],
[
"# @title Utility distributions\n\n# This cell containers parameters enabling us to generate realistic test data that closely \n# models the feature distributions found in the training data.\nDAYOFWEEK = {1: 1040, 2: 1223, 3: 1352, 4: 1217, 5: 1078, 6: 1011, 7: 1110}\nLANGUAGE = {\n \"en-us\": 4807,\n \"en-gb\": 678,\n \"ja-jp\": 419,\n \"en-au\": 310,\n \"en-ca\": 299,\n \"de-de\": 147,\n \"en-in\": 130,\n \"en\": 127,\n \"fr-fr\": 94,\n \"pt-br\": 81,\n \"es-us\": 65,\n \"zh-tw\": 64,\n \"zh-hans-cn\": 55,\n \"es-mx\": 53,\n \"nl-nl\": 37,\n \"fr-ca\": 34,\n \"en-za\": 29,\n \"vi-vn\": 29,\n \"en-nz\": 29,\n \"es-es\": 25,\n}\nOS = {\"IOS\": 3980, \"ANDROID\": 3798, \"null\": 253}\nMONTH = {6: 3125, 7: 1838, 8: 1276, 9: 1718, 10: 74}\nCOUNTRY = {\n \"United States\": 4395,\n \"India\": 486,\n \"Japan\": 450,\n \"Canada\": 354,\n \"Australia\": 327,\n \"United Kingdom\": 303,\n \"Germany\": 144,\n \"Mexico\": 102,\n \"France\": 97,\n \"Brazil\": 93,\n \"Taiwan\": 72,\n \"China\": 65,\n \"Saudi Arabia\": 49,\n \"Pakistan\": 48,\n \"Egypt\": 46,\n \"Netherlands\": 45,\n \"Vietnam\": 42,\n \"Philippines\": 39,\n \"South Africa\": 38,\n}\n\n# Means and standard deviations for numerical features...\nMEAN_SD = {\n \"julianday\": (204.6, 34.7),\n \"cnt_user_engagement\": (30.8, 53.2),\n \"cnt_level_start_quickplay\": (7.8, 28.9),\n \"cnt_level_end_quickplay\": (5.0, 16.4),\n \"cnt_level_complete_quickplay\": (2.1, 9.9),\n \"cnt_level_reset_quickplay\": (2.0, 19.6),\n \"cnt_post_score\": (4.9, 13.8),\n \"cnt_spend_virtual_currency\": (0.4, 1.8),\n \"cnt_ad_reward\": (0.1, 0.6),\n \"cnt_challenge_a_friend\": (0.0, 0.3),\n \"cnt_completed_5_levels\": (0.1, 0.4),\n \"cnt_use_extra_steps\": (0.4, 1.7),\n}\n\nDEFAULT_INPUT = {\n \"cnt_ad_reward\": 0,\n \"cnt_challenge_a_friend\": 0,\n \"cnt_completed_5_levels\": 1,\n \"cnt_level_complete_quickplay\": 3,\n \"cnt_level_end_quickplay\": 5,\n \"cnt_level_reset_quickplay\": 2,\n \"cnt_level_start_quickplay\": 6,\n \"cnt_post_score\": 34,\n \"cnt_spend_virtual_currency\": 0,\n \"cnt_use_extra_steps\": 0,\n \"cnt_user_engagement\": 120,\n \"country\": \"Denmark\",\n \"dayofweek\": 3,\n \"julianday\": 254,\n \"language\": \"da-dk\",\n \"month\": 9,\n \"operating_system\": \"IOS\",\n \"user_pseudo_id\": \"104B0770BAE16E8B53DF330C95881893\",\n}",
"_____no_output_____"
]
],
[
[
"## Import your model\n\nThe churn propensity model you'll be using in this notebook has been trained in BigQuery ML and exported to a Google Cloud Storage bucket. This illustrates how you can easily export a trained model and move a model from one cloud service to another. \n\nRun the next cell to import this model into your project. **If you've already imported your model, you can skip this step.**",
"_____no_output_____"
]
],
[
[
"MODEL_NAME = \"churn\"\nIMAGE = \"us-docker.pkg.dev/cloud-aiplatform/prediction/tf2-cpu.2-4:latest\"\nARTIFACT = \"gs://mco-mm/churn\"\noutput = !gcloud --quiet beta ai models upload --container-image-uri=$IMAGE --artifact-uri=$ARTIFACT --display-name=$MODEL_NAME --format=\"value(model)\"\nMODEL_ID = output[1].split(\"/\")[-1]\nif _exit_code == 0:\n print(f\"Model {MODEL_NAME}/{MODEL_ID} created.\")\nelse:\n print(f\"Error creating model: {output}\")",
"_____no_output_____"
]
],
[
[
"## Deploy your endpoint\n\nNow that you've imported your model into your project, you need to create an endpoint to serve your model. An endpoint can be thought of as a channel through which your model provides prediction services. Once established, you'll be able to make prediction requests on your model via the public internet. Your endpoint is also serverless, in the sense that Google ensures high availability by reducing single points of failure, and scalability by dynamically allocating resources to meet the demand for your service. In this way, you are able to focus on your model quality, and freed from adminstrative and infrastructure concerns.\n\nRun the next cell to deploy your model to an endpoint. **This will take about ten minutes to complete. If you've already deployed a model to an endpoint, you can reuse your endpoint by running the cell after the next one.**",
"_____no_output_____"
]
],
[
[
"ENDPOINT_NAME = \"churn\"\noutput = !gcloud --quiet beta ai endpoints create --display-name=$ENDPOINT_NAME --format=\"value(name)\"\nif _exit_code == 0:\n print(\"Endpoint created.\")\nelse:\n print(f\"Error creating endpoint: {output}\")\nENDPOINT = output[-1]\nENDPOINT_ID = ENDPOINT.split(\"/\")[-1]\noutput = !gcloud --quiet beta ai endpoints deploy-model $ENDPOINT_ID --display-name=$ENDPOINT_NAME --model=$MODEL_ID --traffic-split=\"0=100\"\nDEPLOYED_MODEL_ID = output[-1].split()[-1][:-1]\nif _exit_code == 0:\n print(\n f\"Model {MODEL_NAME}/{MODEL_ID} deployed to Endpoint {ENDPOINT_NAME}/{ENDPOINT_ID}.\"\n )\nelse:\n print(f\"Error deploying model to endpoint: {output}\")",
"_____no_output_____"
]
],
[
[
"### If you already have a deployed endpoint\n\nYou can reuse your existing endpoint by filling in the value of your endpoint ID in the next cell and running it. **If you've just deployed an endpoint in the previous cell, you should skip this step.**",
"_____no_output_____"
]
],
[
[
"# @title Run this cell only if you want to reuse an existing endpoint.\nif not os.getenv(\"IS_TESTING\"):\n ENDPOINT_ID = \"\" # @param {type:\"string\"}\n if ENDPOINT_ID:\n ENDPOINT = f\"projects/{PROJECT_ID}/locations/us-central1/endpoints/{ENDPOINT_ID}\"\n print(f\"Using endpoint {ENDPOINT}\")\n else:\n print(\"If you want to reuse an existing endpoint, you must specify the endpoint id above.\")",
"_____no_output_____"
]
],
[
[
"## Run a prediction test\n\nNow that you have imported a model and deployed that model to an endpoint, you are ready to verify that it's working. Run the next cell to send a test prediction request. If everything works as expected, you should receive a response encoded in a text representation called JSON.\n\n**Try this now by running the next cell and examine the results.**",
"_____no_output_____"
]
],
[
[
"print(ENDPOINT)\nprint(\"request:\")\npp.pprint(DEFAULT_INPUT)\ntry:\n resp = send_predict_request(ENDPOINT, DEFAULT_INPUT)\n print(\"response\")\n pp.pprint(resp)\nexcept Exception:\n print(\"prediction request failed\")",
"_____no_output_____"
]
],
[
[
"Taking a closer look at the results, we see the following elements:\n\n- **churned_values** - a set of possible values (0 and 1) for the target field\n- **churned_probs** - a corresponding set of probabilities for each possible target field value (5x10^-40 and 1.0, respectively)\n- **predicted_churn** - based on the probabilities, the predicted value of the target field (1)\n\nThis response encodes the model's prediction in a format that is readily digestible by software, which makes this service ideal for automated use by an application.",
"_____no_output_____"
],
[
"## Start your monitoring job\n\nNow that you've created an endpoint to serve prediction requests on your model, you're ready to start a monitoring job to keep an eye on model quality and to alert you if and when input begins to deviate in way that may impact your model's prediction quality.\n\nIn this section, you will configure and create a model monitoring job based on the churn propensity model you imported from BigQuery ML.",
"_____no_output_____"
],
[
"### Configure the following fields:\n\n1. User email - The email address to which you would like monitoring alerts sent.\n1. Log sample rate - Your prediction requests and responses are logged to BigQuery tables, which are automatically created when you create a monitoring job. This parameter specifies the desired logging frequency for those tables.\n1. Monitor interval - The time window over which to analyze your data and report anomalies. The minimum window is one hour (3600 seconds).\n1. Target field - The prediction target column name in training dataset.\n1. Skew detection threshold - The skew threshold for each feature you want to monitor.\n1. Prediction drift threshold - The drift threshold for each feature you want to monitor.",
"_____no_output_____"
]
],
[
[
"USER_EMAIL = \"[your-email-address]\" # @param {type:\"string\"}\nJOB_NAME = \"churn\"\n\n# Sampling rate (optional, default=.8)\nLOG_SAMPLE_RATE = 0.8 # @param {type:\"number\"}\n\n# Monitoring Interval in seconds (optional, default=3600).\nMONITOR_INTERVAL = 3600 # @param {type:\"number\"}\n\n# URI to training dataset.\nDATASET_BQ_URI = \"bq://mco-mm.bqmlga4.train\" # @param {type:\"string\"}\n# Prediction target column name in training dataset.\nTARGET = \"churned\"\n\n# Skew and drift thresholds.\nSKEW_DEFAULT_THRESHOLDS = \"country,language\" # @param {type:\"string\"}\nSKEW_CUSTOM_THRESHOLDS = \"cnt_user_engagement:.5\" # @param {type:\"string\"}\nDRIFT_DEFAULT_THRESHOLDS = \"country,language\" # @param {type:\"string\"}\nDRIFT_CUSTOM_THRESHOLDS = \"cnt_user_engagement:.5\" # @param {type:\"string\"}",
"_____no_output_____"
]
],
[
[
"### Create your monitoring job\n\nThe following code uses the Google Python client library to translate your configuration settings into a programmatic request to start a model monitoring job. To do this successfully, you need to specify your alerting thresholds (for both skew and drift), your training data source, and apply those settings to all deployed models on your new endpoint (of which there should only be one at this point).\n\nInstantiating a monitoring job can take some time. If everything looks good with your request, you'll get a successful API response. Then, you'll need to check your email to receive a notification that the job is running.",
"_____no_output_____"
]
],
[
[
"# Set thresholds specifying alerting criteria for training/serving skew and create config object.\nskew_thresholds = get_thresholds(SKEW_DEFAULT_THRESHOLDS, SKEW_CUSTOM_THRESHOLDS)\nskew_config = ModelMonitoringObjectiveConfig.TrainingPredictionSkewDetectionConfig(\n skew_thresholds=skew_thresholds\n)\n\n# Set thresholds specifying alerting criteria for serving drift and create config object.\ndrift_thresholds = get_thresholds(DRIFT_DEFAULT_THRESHOLDS, DRIFT_CUSTOM_THRESHOLDS)\ndrift_config = ModelMonitoringObjectiveConfig.PredictionDriftDetectionConfig(\n drift_thresholds=drift_thresholds\n)\n\n# Specify training dataset source location (used for schema generation).\ntraining_dataset = ModelMonitoringObjectiveConfig.TrainingDataset(target_field=TARGET)\ntraining_dataset.bigquery_source = BigQuerySource(input_uri=DATASET_BQ_URI)\n\n# Aggregate the above settings into a ModelMonitoringObjectiveConfig object and use\n# that object to adjust the ModelDeploymentMonitoringObjectiveConfig object.\nobjective_config = ModelMonitoringObjectiveConfig(\n training_dataset=training_dataset,\n training_prediction_skew_detection_config=skew_config,\n prediction_drift_detection_config=drift_config,\n)\nobjective_template = ModelDeploymentMonitoringObjectiveConfig(\n objective_config=objective_config\n)\n\n# Find all deployed model ids on the created endpoint and set objectives for each.\nmodel_ids = get_deployed_model_ids(ENDPOINT_ID)\nobjective_configs = set_objectives(model_ids, objective_template)\n\n# Create the monitoring job for all deployed models on this endpoint.\nmonitoring_job = create_monitoring_job(objective_configs)",
"_____no_output_____"
],
[
"# Run a prediction request to generate schema, if necessary.\ntry:\n _ = send_predict_request(ENDPOINT, DEFAULT_INPUT)\n print(\"prediction succeeded\")\nexcept Exception:\n print(\"prediction failed\")",
"_____no_output_____"
]
],
[
[
"After a minute or two, you should receive email at the address you configured above for USER_EMAIL. This email confirms successful deployment of your monitoring job. Here's a sample of what this email might look like:\n<br>\n<br>\n<img src=\"https://storage.googleapis.com/mco-general/img/mm6.png\" />\n<br>\nAs your monitoring job collects data, measurements are stored in Google Cloud Storage and you are free to examine your data at any time. The circled path in the image above specifies the location of your measurements in Google Cloud Storage. Run the following cell to take a look at your measurements in Cloud Storage.\n",
"_____no_output_____"
]
],
[
[
"!gsutil ls gs://cloud-ai-platform-fdfb4810-148b-4c86-903c-dbdff879f6e1/*/*",
"_____no_output_____"
]
],
[
[
"You will notice the following components in these Cloud Storage paths:\n\n- **cloud-ai-platform-..** - This is a bucket created for you and assigned to capture your service's prediction data. Each monitoring job you create will trigger creation of a new folder in this bucket.\n- **[model_monitoring|instance_schemas]/job-..** - This is your unique monitoring job number, which you can see above in both the response to your job creation requesst and the email notification. \n- **instance_schemas/job-../analysis** - This is the monitoring jobs understanding and encoding of your training data's schema (field names, types, etc.).\n- **instance_schemas/job-../predict** - This is the first prediction made to your model after the current monitoring job was enabled.\n- **model_monitoring/job-../serving** - This folder is used to record data relevant to drift calculations. It contains measurement summaries for every hour your model serves traffic.\n- **model_monitoring/job-../training** - This folder is used to record data relevant to training-serving skew calculations. It contains an ongoing summary of prediction data relative to training data.",
"_____no_output_____"
],
[
"### You can create monitoring jobs with other user interfaces\n\nIn the previous cells, you created a monitoring job using the Python client library. You can also use the *gcloud* command line tool to create a model monitoring job and, in the near future, you will be able to use the Cloud Console, as well for this function. \n",
"_____no_output_____"
],
[
"## Generate test data to trigger alerting\n\nNow you are ready to test the monitoring function. Run the following cell, which will generate fabricated test predictions designed to trigger the thresholds you specified above. It takes about five minutes to run this cell and at least an hour to assess and report anamolies in skew or drift so after running this cell, feel free to proceed with the notebook and you'll see how to examine the resulting alert later.",
"_____no_output_____"
]
],
[
[
"def random_uid():\n digits = [str(i) for i in range(10)] + [\"A\", \"B\", \"C\", \"D\", \"E\", \"F\"]\n return \"\".join(random.choices(digits, k=32))\n\n\ndef monitoring_test(count, sleep, perturb_num={}, perturb_cat={}):\n # Use random sampling and mean/sd with gaussian distribution to model\n # training data. Then modify sampling distros for two categorical features\n # and mean/sd for two numerical features.\n mean_sd = MEAN_SD.copy()\n country = COUNTRY.copy()\n for k, (mean_fn, sd_fn) in perturb_num.items():\n orig_mean, orig_sd = MEAN_SD[k]\n mean_sd[k] = (mean_fn(orig_mean), sd_fn(orig_sd))\n for k, v in perturb_cat.items():\n country[k] = v\n for i in range(0, count):\n input = DEFAULT_INPUT.copy()\n input[\"user_pseudo_id\"] = str(random_uid())\n input[\"country\"] = random.choices([*country], list(country.values()))[0]\n input[\"dayofweek\"] = random.choices([*DAYOFWEEK], list(DAYOFWEEK.values()))[0]\n input[\"language\"] = str(random.choices([*LANGUAGE], list(LANGUAGE.values()))[0])\n input[\"operating_system\"] = str(random.choices([*OS], list(OS.values()))[0])\n input[\"month\"] = random.choices([*MONTH], list(MONTH.values()))[0]\n for key, (mean, sd) in mean_sd.items():\n sample_val = round(float(np.random.normal(mean, sd, 1)))\n val = max(sample_val, 0)\n input[key] = val\n print(f\"Sending prediction {i}\")\n try:\n send_predict_request(ENDPOINT, input)\n except Exception:\n print(\"prediction request failed\")\n time.sleep(sleep)\n print(\"Test Completed.\")\n\n\ntest_time = 300\ntests_per_sec = 1\nsleep_time = 1 / tests_per_sec\niterations = test_time * tests_per_sec\nperturb_num = {\"cnt_user_engagement\": (lambda x: x * 3, lambda x: x / 3)}\nperturb_cat = {\"Japan\": max(COUNTRY.values()) * 2}\nmonitoring_test(iterations, sleep_time, perturb_num, perturb_cat)",
"_____no_output_____"
]
],
[
[
"## Interpret your results\n\nWhile waiting for your results, which, as noted, may take up to an hour, you can read ahead to get sense of the alerting experience.",
"_____no_output_____"
],
[
"### Here's what a sample email alert looks like...\n\n<img src=\"https://storage.googleapis.com/mco-general/img/mm7.png\" />\n",
"_____no_output_____"
],
[
"This email is warning you that the *cnt_user_engagement*, *country* and *language* feature values seen in production have skewed above your threshold between training and serving your model. It's also telling you that the *cnt_user_engagement* feature value is drifting significantly over time, again, as per your threshold specification.",
"_____no_output_____"
],
[
"### Monitoring results in the Cloud Console\n\nYou can examine your model monitoring data from the Cloud Console. Below is a screenshot of those capabilities.",
"_____no_output_____"
],
[
"#### Monitoring Status\n\n<img src=\"https://storage.googleapis.com/mco-general/img/mm1.png\" />\n\n#### Monitoring Alerts\n\n<img src=\"https://storage.googleapis.com/mco-general/img/mm2.png\" />",
"_____no_output_____"
],
[
"## Clean up\n\nTo clean up all Google Cloud resources used in this project, you can [delete the Google Cloud\nproject](https://cloud.google.com/resource-manager/docs/creating-managing-projects#shutting_down_projects) you used for the tutorial.\n\nOtherwise, you can delete the individual resources you created in this tutorial:",
"_____no_output_____"
]
],
[
[
"out = !gcloud ai endpoints undeploy-model $ENDPOINT_ID --deployed-model-id $DEPLOYED_MODEL_ID\nif _exit_code == 0:\n print(\"Model undeployed.\")\nelse:\n print(\"Error undeploying model:\", out)\n\nout = !gcloud ai endpoints delete $ENDPOINT_ID --quiet\nif _exit_code == 0:\n print(\"Endpoint deleted.\")\nelse:\n print(\"Error deleting endpoint:\", out)\n\nout = !gcloud ai models delete $MODEL_ID --quiet\nif _exit_code == 0:\n print(\"Model deleted.\")\nelse:\n print(\"Error deleting model:\", out)",
"_____no_output_____"
]
],
[
[
"## Learn more about model monitoring\n\n**Congratulations!** You've now learned what model monitoring is, how to configure and enable it, and how to find and interpret the results. Check out the following resources to learn more about model monitoring and ML Ops.\n\n- [TensorFlow Data Validation](https://www.tensorflow.org/tfx/guide/tfdv)\n- [Data Understanding, Validation, and Monitoring At Scale](https://blog.tensorflow.org/2018/09/introducing-tensorflow-data-validation.html)\n- [Vertex Product Documentation](https://cloud.google.com/vertex)\n- [Model Monitoring Reference Docs](https://cloud.google.com/vertex/docs/reference)\n- [Model Monitoring blog article]()",
"_____no_output_____"
]
]
] |
[
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown"
] |
[
[
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
]
] |
cbd8cb119020afb7951398c2bba662e6fa6a6650
| 10,568 |
ipynb
|
Jupyter Notebook
|
jupyter/jupyter_algo.ipynb
|
jfklorenz/Python-RMinimum
|
75ac91201774cff980c06cba5c2d7edcd271efbb
|
[
"MIT"
] | 2 |
2020-10-27T19:58:29.000Z
|
2020-10-27T21:40:09.000Z
|
jupyter/jupyter_algo.ipynb
|
jfklorenz/RMinimum-Algorithm
|
75ac91201774cff980c06cba5c2d7edcd271efbb
|
[
"MIT"
] | null | null | null |
jupyter/jupyter_algo.ipynb
|
jfklorenz/RMinimum-Algorithm
|
75ac91201774cff980c06cba5c2d7edcd271efbb
|
[
"MIT"
] | null | null | null | 30.72093 | 729 | 0.389099 |
[
[
[
"RMinimum : Full - Test",
"_____no_output_____"
]
],
[
[
"import math\nimport random\nimport queue",
"_____no_output_____"
]
],
[
[
"Testfall : $X = [0, \\cdots, n-1]$, $k$",
"_____no_output_____"
]
],
[
[
"# User input\nn = 2**10\nk = 2**5\n\n# Automatic\nX = [i for i in range(n)]\n\n# Show Testcase\nprint(' Testcase: ')\nprint('=============================')\nprint('X = [0, ..., ' + str(n - 1) + ']')\nprint('k =', k)",
" Testcase: \n=============================\nX = [0, ..., 1023]\nk = 32\n"
]
],
[
[
"Algorithmus : Full",
"_____no_output_____"
]
],
[
[
"def rminimum(X, k, cnt = [], rec = 0):\n\n # Generate empty cnt list if its not a recursive call\n if cnt == []:\n cnt = [0 for _ in range(max(X) + 1)]\n\n # Convert parameters if needed\n k = int(k)\n n = len(X)\n\n # Base case |X| = 3\n if len(X) == 3:\n if X[0] < X[1]:\n cnt[X[0]] += 2\n cnt[X[1]] += 1\n cnt[X[2]] += 1\n\n if X[0] < X[2]:\n mini = X[0]\n else:\n mini = X[2]\n else:\n cnt[X[0]] += 1\n cnt[X[1]] += 2\n cnt[X[2]] += 1\n\n if X[1] < X[2]:\n mini = X[1]\n else:\n mini = X[2]\n return mini, cnt, rec\n\n # Run phases\n W, L, cnt = phase1(X, cnt)\n M, cnt = phase2(L, k, cnt)\n Wnew, cnt = phase3(W, k, M, cnt)\n mini, cnt, rec = phase4(Wnew, k, n, cnt, rec)\n\n return mini, cnt, rec\n\n return mini, cnt, rec\n\n# --------------------------------------------------\ndef phase1(X, cnt):\n\n # Init W, L\n W = [0 for _ in range(len(X) // 2)]\n L = [0 for _ in range(len(X) // 2)]\n\n # Random pairs\n random.shuffle(X)\n for i in range(len(X) // 2):\n if X[2 * i] > X[2 * i + 1]:\n W[i] = X[2 * i + 1]\n L[i] = X[2 * i]\n else:\n W[i] = X[2 * i]\n L[i] = X[2 * i + 1]\n cnt[X[2 * i + 1]] += 1\n cnt[X[2 * i]] += 1\n\n return W, L, cnt\n\n# --------------------------------------------------\ndef phase2(L, k, cnt):\n \n # Generate subsets\n random.shuffle(L)\n subsets = [L[i * k:(i + 1) * k] for i in range((len(L) + k - 1) // k)]\n\n # Init M\n M = [0 for _ in range(len(subsets))]\n\n # Perfectly balanced tournament tree using a Queue\n for i in range(len(subsets)):\n q = queue.Queue()\n\n for ele in subsets[i]:\n q.put(ele)\n\n while q.qsize() > 1:\n a = q.get()\n b = q.get()\n\n if a < b:\n q.put(a)\n else:\n q.put(b)\n cnt[a] += 1\n cnt[b] += 1\n M[i] = q.get()\n\n return M, cnt\n\n# --------------------------------------------------\ndef phase3(W, k, M, cnt):\n\n # Generate subsets\n random.shuffle(W)\n W_i = [W[i * k:(i + 1) * k] for i in range((len(W) + k - 1) // k)]\n subsets_filtered = [0 for _ in range(len(subsets))]\n\n # Filter subsets\n for i in range(len(subsets_filtered)):\n subsets_filtered[i] = [elem for elem in subsets[i] if elem < M[i]]\n cnt[M[i]] += len(subsets[i])\n for elem in subsets[i]:\n cnt[elem] += 1\n\n # Merge subsets\n Wnew = [item for sublist in subsets_filtered for item in sublist]\n\n return Wnew, cnt\n\n# --------------------------------------------------\ndef phase4(Wnew, k, n0, cnt, rec):\n # Recursive call check\n if len(Wnew) <= math.log(n0, 2) ** 2:\n q = queue.Queue()\n\n for ele in Wnew:\n q.put(ele)\n while q.qsize() > 1:\n a = q.get()\n b = q.get()\n\n if a < b:\n q.put(a)\n else:\n q.put(b)\n\n cnt[a] += 1\n cnt[b] += 1\n mini = q.get()\n\n return mini, cnt, rec\n\n else:\n rec += 1\n rminimum(Wnew, k, cnt, rec)\n\n# ==================================================\n# Testcase\nmini, cnt, rec = rminimum(X, k)",
"_____no_output_____"
]
],
[
[
"Resultat :",
"_____no_output_____"
]
],
[
[
"def test(X, k, mini, cnt, rec):\n \n print('')\n print('Testfall n / k:', len(X), '/', k)\n print('====================================')\n print('Fragile Complexity:')\n print('-------------------')\n print('f_min :', cnt[0])\n print('f_rem :', max(cnt[1:]))\n print('f_n :', max(cnt))\n print('Work :', int(sum(cnt)/2))\n print('====================================')\n print('Process:')\n print('--------')\n print('Minimum :', mini)\n print('n :', len(X))\n print('log(n) :', round(math.log(len(X), 2), 2))\n print('log(k) :', round(math.log(k, 2), 2))\n print('lg / lglg :', round(math.log(len(X), 2) / math.log(math.log(len(X), 2), 2)))\n print('n / log(n) :', round(len(X) / math.log(len(X), 2)))\n print('====================================')\n return\n\n# Testfall\ntest(X, k, mini, cnt, rec)",
"\nTestfall n / k: 1024 / 32\n====================================\nFragile Complexity:\nf_min : 8\nf_rem : 40\nf_n : 40\nWork : 1747\n====================================\nProcess:\n---------\nMinimum : 0\nn : 1024\nlog(n) : 10.0\nlog(k) : 5.0\nlg / lglg : 3\nn / log(n) : 102\n====================================\n"
],
[
"",
"[2, 3]\n"
],
[
"",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
]
] |
cbd8cb7cb461fce9d44c1101a0473b4a1951574b
| 36,824 |
ipynb
|
Jupyter Notebook
|
docs/tutorials/rosetta_and_yangson_walkthrough.ipynb
|
marcushoff/ntc-rosetta
|
a4a34ae3fa01e8b087d49fcb709a97aefecb19f4
|
[
"Apache-2.0"
] | 95 |
2019-05-29T13:04:12.000Z
|
2022-03-11T20:38:13.000Z
|
docs/tutorials/rosetta_and_yangson_walkthrough.ipynb
|
marcushoff/ntc-rosetta
|
a4a34ae3fa01e8b087d49fcb709a97aefecb19f4
|
[
"Apache-2.0"
] | 35 |
2019-05-30T03:58:34.000Z
|
2021-03-03T15:30:05.000Z
|
docs/tutorials/rosetta_and_yangson_walkthrough.ipynb
|
marcushoff/ntc-rosetta
|
a4a34ae3fa01e8b087d49fcb709a97aefecb19f4
|
[
"Apache-2.0"
] | 28 |
2019-05-29T19:32:26.000Z
|
2021-04-05T09:49:30.000Z
| 34.223048 | 527 | 0.412476 |
[
[
[
"# Advanced topics\n\nThe following material is a deep-dive into Yangson, and is not necessarily representative of how one would perform manipulations in a production environment. Please refer to the other tutorials for a better picture of Rosetta's intended use. Keep in mind that the key feature of Yangson is to be able to manipulate YANG data models in a more human-readable format, ala JSON. What lies below digs beneath the higher-level abstractions and should paint a decent picture of the functional nature of Yangson. ",
"_____no_output_____"
],
[
"# Manipulating models with Rosetta and Yangson\n\nOne of the goals of many network operators is to provide abstractions in a multi-vendor environment. This can be done with YANG and OpenConfig data models, but as they say, the devil is in the details. It occurred to me that you should be able to parse configuration from one vendor and translate it to another. Unfortunately, as we all know, these configurations don't always translate well on a 1-to-1 basis. I will demonstrate this process below and show several features of the related libraries along the way.\n\nThe following example begins exactly the same as the Cisco parsing tutorial. Let's load up some Juniper config and parse it into a YANG data model. First, we'll read the file.",
"_____no_output_____"
]
],
[
[
"from ntc_rosetta import get_driver\nimport json\n\njunos = get_driver(\"junos\", \"openconfig\")\njunos_driver = junos()\n\n# Strip any rpc tags before and after `<configuration>...</configuration>`\nwith open(\"data/junos/dev_conf.xml\", \"r\") as fp:\n config = fp.read()\nprint(config)",
"<configuration>\n <system>\n <host-name>junos-host</host-name>\n <name-server>\n <name>8.8.8.8</name>\n </name-server>\n <name-server>\n <name>8.8.4.4</name>\n </name-server>\n <name-server>\n <name>2606:4700:4700::1111</name>\n </name-server>\n <name-server>\n <name>2606:4700:4700::1001</name>\n </name-server>\n <domain-name>packet.net</domain-name>\n <tacplus-server>\n <name>192.168.0.42</name>\n <secret>$9$kq5z3nCuBE/CLNb2GU/Ctp0IEcy</secret>\n </tacplus-server>\n <services>\n <ssh>\n <protocol-version>v2</protocol-version>\n </ssh>\n </services>\n <login>\n <user>\n <name>user</name>\n <class>super-user</class>\n <authentication>\n <ssh-rsa>\n <name>ssh-rsa some-ssh-key [email protected]</name>\n </ssh-rsa>\n </authentication>\n </user>\n <user>\n <name>other</name>\n <class>super-user</class>\n <authentication>\n <encrypted-password>$6$Ds.gV3Ce$V10YZ/FFypUTCcUtJ8C9WIgnD6wEwsXFivPH2RrVCBs/JzcS7WYgPL.EgVlsFwSVtK1F.HBENK2UbMMj8Dfc4/</encrypted-password>\n </authentication>\n </user>\n <announcement>foo\\nbar\\nbang</announcement>\n <message>A bunch of \\nmessage\\ntest</message>\n </login>\n <commit>\n <synchronize/>\n </commit>\n <ntp>\n <server>\n <name>129.6.15.28</name>\n </server>\n <server>\n <name>129.6.15.29</name>\n </server>\n <server>\n <name>129.6.15.30</name>\n </server>\n <server>\n <name>2610:20:6f15:15::27</name>\n </server>\n </ntp>\n <time-zone>America/Winnipeg</time-zone>\n </system>\n <interfaces>\n <interface>\n <name>xe-0/0/1</name>\n <unit>\n <name>0</name>\n <family>\n <ethernet-switching>\n <interface-mode>access</interface-mode>\n <vlan>\n <members>10</members>\n </vlan>\n </ethernet-switching>\n </family>\n </unit>\n </interface>\n <interface>\n <name>xe-0/0/3</name>\n <unit>\n <name>0</name>\n <family>\n <ethernet-switching>\n <interface-mode>trunk</interface-mode>\n <vlan>\n <members>10</members>\n <members>20</members>\n </vlan>\n </ethernet-switching>\n </family>\n </unit>\n </interface>\n <interface>\n <name>xe-0/0/4</name>\n <unit>\n <name>0</name>\n <family>\n <ethernet-switching>\n <interface-mode>trunk</interface-mode>\n <vlan>\n <members>VLAN-100</members>\n <members>VLAN-200</members>\n </vlan>\n </ethernet-switching>\n </family>\n </unit>\n </interface>\n <interface>\n <name>xe-0/0/5</name>\n <unit>\n <name>0</name>\n <family>\n <ethernet-switching>\n <interface-mode>access</interface-mode>\n <vlan>\n <members>VLAN-100</members>\n </vlan>\n </ethernet-switching>\n </family>\n </unit>\n </interface>\n </interfaces>\n <vlans>\n <vlan>\n <name>default</name>\n <vlan-id>1</vlan-id>\n </vlan>\n <vlan>\n <name>prod</name>\n <vlan-id>20</vlan-id>\n </vlan>\n <vlan inactive=\"inactive\">\n <vlan-id>10</vlan-id>\n </vlan>\n <vlan>\n <name>VLAN-100</name>\n <vlan-id>100</vlan-id>\n </vlan>\n <vlan>\n <name>VLAN-200</name>\n <vlan-id>200</vlan-id>\n </vlan>\n </vlans>\n</configuration>\n\n"
]
],
[
[
"## Junos parsing\n\nNow, we parse the config and take a look at the data model.",
"_____no_output_____"
]
],
[
[
"from sys import exc_info\nfrom yangson.exceptions import SemanticError\n\ntry:\n parsed = junos_driver.parse(\n native={\"dev_conf\": config},\n validate=False,\n include=[\n \"/openconfig-interfaces:interfaces\",\n \"/openconfig-network-instance:network-instances/network-instance/name\",\n \"/openconfig-network-instance:network-instances/network-instance/config\",\n \"/openconfig-network-instance:network-instances/network-instance/vlans\", \n ]\n )\nexcept SemanticError as e:\n print(f\"error: {e}\")\n\nprint(json.dumps(parsed.raw_value(), sort_keys=True, indent=2))",
"{\n \"openconfig-interfaces:interfaces\": {\n \"interface\": [\n {\n \"config\": {\n \"enabled\": true,\n \"name\": \"xe-0/0/1\",\n \"type\": \"iana-if-type:ethernetCsmacd\"\n },\n \"name\": \"xe-0/0/1\",\n \"openconfig-if-ethernet:ethernet\": {\n \"openconfig-vlan:switched-vlan\": {\n \"config\": {\n \"access-vlan\": 10,\n \"interface-mode\": \"ACCESS\"\n }\n }\n },\n \"subinterfaces\": {\n \"subinterface\": [\n {\n \"config\": {\n \"index\": 0\n },\n \"index\": 0\n }\n ]\n }\n },\n {\n \"config\": {\n \"enabled\": true,\n \"name\": \"xe-0/0/3\",\n \"type\": \"iana-if-type:ethernetCsmacd\"\n },\n \"name\": \"xe-0/0/3\",\n \"openconfig-if-ethernet:ethernet\": {\n \"openconfig-vlan:switched-vlan\": {\n \"config\": {\n \"interface-mode\": \"TRUNK\",\n \"trunk-vlans\": [\n 10,\n 20\n ]\n }\n }\n },\n \"subinterfaces\": {\n \"subinterface\": [\n {\n \"config\": {\n \"index\": 0\n },\n \"index\": 0\n }\n ]\n }\n },\n {\n \"config\": {\n \"enabled\": true,\n \"name\": \"xe-0/0/4\",\n \"type\": \"iana-if-type:ethernetCsmacd\"\n },\n \"name\": \"xe-0/0/4\",\n \"openconfig-if-ethernet:ethernet\": {\n \"openconfig-vlan:switched-vlan\": {\n \"config\": {\n \"interface-mode\": \"TRUNK\",\n \"trunk-vlans\": [\n 100,\n 200\n ]\n }\n }\n },\n \"subinterfaces\": {\n \"subinterface\": [\n {\n \"config\": {\n \"index\": 0\n },\n \"index\": 0\n }\n ]\n }\n },\n {\n \"config\": {\n \"enabled\": true,\n \"name\": \"xe-0/0/5\",\n \"type\": \"iana-if-type:ethernetCsmacd\"\n },\n \"name\": \"xe-0/0/5\",\n \"openconfig-if-ethernet:ethernet\": {\n \"openconfig-vlan:switched-vlan\": {\n \"config\": {\n \"access-vlan\": 100,\n \"interface-mode\": \"ACCESS\"\n }\n }\n },\n \"subinterfaces\": {\n \"subinterface\": [\n {\n \"config\": {\n \"index\": 0\n },\n \"index\": 0\n }\n ]\n }\n }\n ]\n },\n \"openconfig-network-instance:network-instances\": {\n \"network-instance\": [\n {\n \"config\": {\n \"name\": \"default\"\n },\n \"name\": \"default\",\n \"vlans\": {\n \"vlan\": [\n {\n \"config\": {\n \"name\": \"default\",\n \"status\": \"ACTIVE\",\n \"vlan-id\": 1\n },\n \"vlan-id\": 1\n },\n {\n \"config\": {\n \"name\": \"prod\",\n \"status\": \"ACTIVE\",\n \"vlan-id\": 20\n },\n \"vlan-id\": 20\n },\n {\n \"config\": {\n \"status\": \"SUSPENDED\",\n \"vlan-id\": 10\n },\n \"vlan-id\": 10\n },\n {\n \"config\": {\n \"name\": \"VLAN-100\",\n \"status\": \"ACTIVE\",\n \"vlan-id\": 100\n },\n \"vlan-id\": 100\n },\n {\n \"config\": {\n \"name\": \"VLAN-200\",\n \"status\": \"ACTIVE\",\n \"vlan-id\": 200\n },\n \"vlan-id\": 200\n }\n ]\n }\n }\n ]\n }\n}\n"
]
],
[
[
"## Naive translation\nSince we have a valid data model, let's see if Rosetta can translate it as-is.",
"_____no_output_____"
]
],
[
[
"ios = get_driver(\"ios\", \"openconfig\")\nios_driver = ios()\nnative = ios_driver.translate(candidate=parsed.raw_value())\n\nprint(native)",
"interface xe-0/0/1\n no shutdown\n switchport mode access\n switchport access vlan 10\n exit\n!\ninterface xe-0/0/3\n no shutdown\n switchport mode trunk\n switchport trunk allowed vlan 10,20\n exit\n!\ninterface xe-0/0/4\n no shutdown\n switchport mode trunk\n switchport trunk allowed vlan 100,200\n exit\n!\ninterface xe-0/0/5\n no shutdown\n switchport mode access\n switchport access vlan 100\n exit\n!\nvlan 1\n name default\n no shutdown\n exit\n!\nvlan 20\n name prod\n no shutdown\n exit\n!\nvlan 10\n shutdown\n exit\n!\nvlan 100\n name VLAN-100\n no shutdown\n exit\n!\nvlan 200\n name VLAN-200\n no shutdown\n exit\n!\n\n"
]
],
[
[
"Pretty cool, right?! Rosetta does a great job of parsing and translating, but it is a case of \"monkey see, monkey do\". Rosetta doesn't have any mechanisms to translate interface names, for example. It is up to the operator to perform this sort of manipulation.\n\n## Down the Yangson rabbit hole\nYangson allows the developer to easily translate between YANG data models and JSON. Most all of these manipulations can be performed on dictionaries in Python and loaded into data models using [`from_raw`](https://yangson.labs.nic.cz/datamodel.html#yangson.datamodel.DataModel.from_raw). The following examples may appear to be a little obtuse, but the goal is to demonstrate the internals of Yangson.\n\n### And it's mostly functional\nIt is critical to read the short description of the [zipper](https://yangson.labs.nic.cz/instance.html?highlight=zipper#yangson.instance.InstanceNode) interface in the InstanceNode section of the docs. Yanson never manipulates an object, but returns a copy with the manipulated attributes.\n\n### Show me the code!\nLet's take a look at fixing up the interface names and how we can manipulate data model attributes. To do that, we need to locate the attribute in the tree using the [`parse_resource_id`](https://yangson.labs.nic.cz/datamodel.html#yangson.datamodel.DataModel.parse_resource_id) method. This method returns an [`instance route'](https://yangson.labs.nic.cz/instance.html?highlight=arrayentry#yangson.instance.InstanceRoute). The string passed to the method is an xpath. \n",
"_____no_output_____"
]
],
[
[
"# Locate the interfaces in the tree. We need to modify this one\n# Note that we have to URL-escape the forward slashes per https://tools.ietf.org/html/rfc8040#section-3.5.3\nirt = parsed.datamodel.parse_resource_id(\"openconfig-interfaces:interfaces/interface=xe-0%2F0%2F1\")\ncurrent_data = parsed.root.goto(irt)\nprint(\"Current node configuration: \", json.dumps(current_data.raw_value(), sort_keys=True, indent=2))\nmodify_data = current_data.raw_value()\nifname = 'Ethernet0/0/1'\nmodify_data['name'] = ifname\nmodify_data['config']['name'] = ifname\nstub = current_data.update(modify_data, raw=True)\nprint(\"\\n\\nCandidate node configuration: \", json.dumps(stub.raw_value(), sort_keys=True, indent=2))",
"Current node configuration: {\n \"config\": {\n \"enabled\": true,\n \"name\": \"xe-0/0/1\",\n \"type\": \"iana-if-type:ethernetCsmacd\"\n },\n \"name\": \"xe-0/0/1\",\n \"openconfig-if-ethernet:ethernet\": {\n \"openconfig-vlan:switched-vlan\": {\n \"config\": {\n \"access-vlan\": 10,\n \"interface-mode\": \"ACCESS\"\n }\n }\n },\n \"subinterfaces\": {\n \"subinterface\": [\n {\n \"config\": {\n \"index\": 0\n },\n \"index\": 0\n }\n ]\n }\n}\n\n\nCandidate node configuration: {\n \"config\": {\n \"enabled\": true,\n \"name\": \"Ethernet0/0/1\",\n \"type\": \"iana-if-type:ethernetCsmacd\"\n },\n \"name\": \"Ethernet0/0/1\",\n \"openconfig-if-ethernet:ethernet\": {\n \"openconfig-vlan:switched-vlan\": {\n \"config\": {\n \"access-vlan\": 10,\n \"interface-mode\": \"ACCESS\"\n }\n }\n },\n \"subinterfaces\": {\n \"subinterface\": [\n {\n \"config\": {\n \"index\": 0\n },\n \"index\": 0\n }\n ]\n }\n}\n"
]
],
[
[
"### Instance routes\nYou will notice a `goto` method on child nodes. You _can_ access successors with this method, but you have to build the path from the root `datamodel` attribute as seen in the following example. If you aren't sure where an object is in the tree, you can also rely on its `path` attribute.\n\nQuick tangent... what is the difference between `parse_instance_id` and `parse_resource_id`? The answer can be found in the [Yangson glossary](https://yangson.labs.nic.cz/glossary.html) and the respective RFC's.",
"_____no_output_____"
]
],
[
[
"# TL;DR\nirt = parsed.datamodel.parse_instance_id('/openconfig-network-instance:network-instances/network-instance[1]/vlans/vlan[3]')\nprint(parsed.root.goto(irt).raw_value())\n\nirt = parsed.datamodel.parse_resource_id('openconfig-network-instance:network-instances/network-instance=default/vlans/vlan=10')\nprint(parsed.root.goto(irt).raw_value())",
"{'vlan-id': 10, 'config': {'vlan-id': 10, 'status': 'SUSPENDED'}}\n{'vlan-id': 10, 'config': {'vlan-id': 10, 'status': 'SUSPENDED'}}\n"
]
],
[
[
"What about the rest of the interfaces in the list? Yangson provides an iterator for array nodes.",
"_____no_output_____"
]
],
[
[
"import re\n\nirt = parsed.datamodel.parse_resource_id(\"openconfig-interfaces:interfaces/interface\")\niface_objs = parsed.root.goto(irt)\n# Swap the name as required\np, sub = re.compile(r'xe-'), 'Ethernet'\n\n# There are a couple challenges here. First is that Yanson doesn't impliment __len__\n# The second problem is that you cannot modify a list in-place, so we're basically\n# hacking this to hijack the index of the current element and looking it up from a \"clean\"\n# instance. This is a pet example! It would be much easier using Python dicts.\nnew_ifaces = None\nfor iface in iface_objs:\n name_irt = parsed.datamodel.parse_instance_id('/name')\n cname_irt = parsed.datamodel.parse_instance_id('/config/name')\n if new_ifaces:\n name = new_ifaces[iface.index].goto(name_irt)\n else:\n name = iface.goto(name_irt)\n name = name.update(p.sub(sub, name.raw_value()), raw=True)\n cname = name.up().goto(cname_irt)\n cname = cname.update(p.sub(sub, cname.raw_value()), raw=True)\n iface = cname.up().up()\n new_ifaces = iface.up()\nprint(json.dumps(new_ifaces.raw_value(), sort_keys=True, indent=2))",
"[\n {\n \"config\": {\n \"enabled\": true,\n \"name\": \"Ethernet0/0/1\",\n \"type\": \"iana-if-type:ethernetCsmacd\"\n },\n \"name\": \"Ethernet0/0/1\",\n \"openconfig-if-ethernet:ethernet\": {\n \"openconfig-vlan:switched-vlan\": {\n \"config\": {\n \"access-vlan\": 10,\n \"interface-mode\": \"ACCESS\"\n }\n }\n },\n \"subinterfaces\": {\n \"subinterface\": [\n {\n \"config\": {\n \"index\": 0\n },\n \"index\": 0\n }\n ]\n }\n },\n {\n \"config\": {\n \"enabled\": true,\n \"name\": \"Ethernet0/0/3\",\n \"type\": \"iana-if-type:ethernetCsmacd\"\n },\n \"name\": \"Ethernet0/0/3\",\n \"openconfig-if-ethernet:ethernet\": {\n \"openconfig-vlan:switched-vlan\": {\n \"config\": {\n \"interface-mode\": \"TRUNK\",\n \"trunk-vlans\": [\n 10,\n 20\n ]\n }\n }\n },\n \"subinterfaces\": {\n \"subinterface\": [\n {\n \"config\": {\n \"index\": 0\n },\n \"index\": 0\n }\n ]\n }\n },\n {\n \"config\": {\n \"enabled\": true,\n \"name\": \"Ethernet0/0/4\",\n \"type\": \"iana-if-type:ethernetCsmacd\"\n },\n \"name\": \"Ethernet0/0/4\",\n \"openconfig-if-ethernet:ethernet\": {\n \"openconfig-vlan:switched-vlan\": {\n \"config\": {\n \"interface-mode\": \"TRUNK\",\n \"trunk-vlans\": [\n 100,\n 200\n ]\n }\n }\n },\n \"subinterfaces\": {\n \"subinterface\": [\n {\n \"config\": {\n \"index\": 0\n },\n \"index\": 0\n }\n ]\n }\n },\n {\n \"config\": {\n \"enabled\": true,\n \"name\": \"Ethernet0/0/5\",\n \"type\": \"iana-if-type:ethernetCsmacd\"\n },\n \"name\": \"Ethernet0/0/5\",\n \"openconfig-if-ethernet:ethernet\": {\n \"openconfig-vlan:switched-vlan\": {\n \"config\": {\n \"access-vlan\": 100,\n \"interface-mode\": \"ACCESS\"\n }\n }\n },\n \"subinterfaces\": {\n \"subinterface\": [\n {\n \"config\": {\n \"index\": 0\n },\n \"index\": 0\n }\n ]\n }\n }\n]\n"
],
[
"# Translate to Cisco-speak\nnative = ios_driver.translate(candidate=new_ifaces.top().raw_value())\n\nprint(native)",
"interface Ethernet0/0/1\n no shutdown\n switchport mode access\n switchport access vlan 10\n exit\n!\ninterface Ethernet0/0/3\n no shutdown\n switchport mode trunk\n switchport trunk allowed vlan 10,20\n exit\n!\ninterface Ethernet0/0/4\n no shutdown\n switchport mode trunk\n switchport trunk allowed vlan 100,200\n exit\n!\ninterface Ethernet0/0/5\n no shutdown\n switchport mode access\n switchport access vlan 100\n exit\n!\nvlan 1\n name default\n no shutdown\n exit\n!\nvlan 20\n name prod\n no shutdown\n exit\n!\nvlan 10\n shutdown\n exit\n!\nvlan 100\n name VLAN-100\n no shutdown\n exit\n!\nvlan 200\n name VLAN-200\n no shutdown\n exit\n!\n\n"
]
],
[
[
"Hooray! That should work. One final approach, just to show you different ways of doing things. This is another pet example to demonstrate Yangson methods.",
"_____no_output_____"
]
],
[
[
"import re\nfrom typing import Dict\n\nirt = parsed.datamodel.parse_resource_id(\"openconfig-interfaces:interfaces\")\niface_objs = parsed.root.goto(irt)\n# Nuke the whole branch!\niface_objs = iface_objs.delete_item(\"interface\")\n\ndef build_iface(data: str) -> Dict:\n # Example template, this could be anything you like that conforms to the schema\n return {\n \"name\": f\"{data['name']}\",\n \"config\": {\n \"name\": f\"{data['name']}\",\n \"description\": f\"{data['description']}\",\n \"type\": \"iana-if-type:ethernetCsmacd\",\n \"enabled\": True\n },\n }\n\niface_data = [\n build_iface({\n \"name\": f\"TenGigabitEthernet0/{idx}\",\n \"description\": f\"This is interface TenGigabitEthernet0/{idx}\"\n }) for idx in range(10, 0, -1)\n]\n\ninitial = iface_data.pop()\n# Start a new interface list\niface_objs = iface_objs.put_member(\"interface\", [initial], raw=True)\ncur_obj = iface_objs[0]\n\n# Yangson exposes `next`, `insert_after`, and `insert_before` methods.\n# There is no `append`.\nwhile iface_data:\n new_obj = cur_obj.insert_after(iface_data.pop(), raw=True)\n cur_obj = new_obj",
"_____no_output_____"
],
[
"# Translate to Cisco-speak\nnative = ios_driver.translate(candidate=cur_obj.top().raw_value())\n\nprint(native)",
"interface TenGigabitEthernet0/1\n description This is interface TenGigabitEthernet0/1\n no shutdown\n exit\n!\ninterface TenGigabitEthernet0/2\n description This is interface TenGigabitEthernet0/2\n no shutdown\n exit\n!\ninterface TenGigabitEthernet0/3\n description This is interface TenGigabitEthernet0/3\n no shutdown\n exit\n!\ninterface TenGigabitEthernet0/4\n description This is interface TenGigabitEthernet0/4\n no shutdown\n exit\n!\ninterface TenGigabitEthernet0/5\n description This is interface TenGigabitEthernet0/5\n no shutdown\n exit\n!\ninterface TenGigabitEthernet0/6\n description This is interface TenGigabitEthernet0/6\n no shutdown\n exit\n!\ninterface TenGigabitEthernet0/7\n description This is interface TenGigabitEthernet0/7\n no shutdown\n exit\n!\ninterface TenGigabitEthernet0/8\n description This is interface TenGigabitEthernet0/8\n no shutdown\n exit\n!\ninterface TenGigabitEthernet0/9\n description This is interface TenGigabitEthernet0/9\n no shutdown\n exit\n!\ninterface TenGigabitEthernet0/10\n description This is interface TenGigabitEthernet0/10\n no shutdown\n exit\n!\nvlan 1\n name default\n no shutdown\n exit\n!\nvlan 20\n name prod\n no shutdown\n exit\n!\nvlan 10\n shutdown\n exit\n!\nvlan 100\n name VLAN-100\n no shutdown\n exit\n!\nvlan 200\n name VLAN-200\n no shutdown\n exit\n!\n\n"
]
],
[
[
"### Deleting individual items\nHere is an example of deleting an individual item. Navigating the tree can be a bit tricky, but it's not too bad once you get the hang of it.",
"_____no_output_____"
]
],
[
[
"# Locate a vlan by ID and delete it\nirt = parsed.datamodel.parse_resource_id(\"openconfig-network-instance:network-instances/network-instance=default/vlans/vlan=10\")\nvlan10 = parsed.root.goto(irt)\nvlans = vlan10.up().delete_item(vlan10.index)\nprint(json.dumps(vlans.raw_value(), sort_keys=True, indent=2))",
"[\n {\n \"config\": {\n \"name\": \"default\",\n \"status\": \"ACTIVE\",\n \"vlan-id\": 1\n },\n \"vlan-id\": 1\n },\n {\n \"config\": {\n \"name\": \"prod\",\n \"status\": \"ACTIVE\",\n \"vlan-id\": 20\n },\n \"vlan-id\": 20\n },\n {\n \"config\": {\n \"name\": \"VLAN-100\",\n \"status\": \"ACTIVE\",\n \"vlan-id\": 100\n },\n \"vlan-id\": 100\n },\n {\n \"config\": {\n \"name\": \"VLAN-200\",\n \"status\": \"ACTIVE\",\n \"vlan-id\": 200\n },\n \"vlan-id\": 200\n }\n]\n"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
]
] |
cbd8ef8eaba25a1916a9130c9cff5d6811a37e40
| 51,980 |
ipynb
|
Jupyter Notebook
|
3_planning_agent.ipynb
|
angelmtenor/AIND-projects
|
bccf0920fab44d3bebf0af5d8fea8ef53e661e1c
|
[
"MIT"
] | 1 |
2020-04-01T08:52:46.000Z
|
2020-04-01T08:52:46.000Z
|
3_planning_agent.ipynb
|
angelmtenor/AIND-projects
|
bccf0920fab44d3bebf0af5d8fea8ef53e661e1c
|
[
"MIT"
] | null | null | null |
3_planning_agent.ipynb
|
angelmtenor/AIND-projects
|
bccf0920fab44d3bebf0af5d8fea8ef53e661e1c
|
[
"MIT"
] | null | null | null | 44.125637 | 243 | 0.500962 |
[
[
[
"# Planning Search Agent\n\nNotebook version of the project [Implement a Planning Search](https://github.com/udacity/AIND-Planning) from [Udacity's Artificial Intelligence Nanodegree](https://www.udacity.com/course/artificial-intelligence-nanodegree--nd889) <br>\n\n**Goal**: Solve deterministic logistics planning problems for an Air Cargo transport system using a planning search agent\n\n\nAll problems are in the Air Cargo domain. They have the same action schema defined, but different initial states and goals:\n\n```\nAction(Load(c, p, a),\n\tPRECOND: At(c, a) ∧ At(p, a) ∧ Cargo(c) ∧ Plane(p) ∧ Airport(a)\n\tEFFECT: ¬ At(c, a) ∧ In(c, p))\nAction(Unload(c, p, a),\n\tPRECOND: In(c, p) ∧ At(p, a) ∧ Cargo(c) ∧ Plane(p) ∧ Airport(a)\n\tEFFECT: At(c, a) ∧ ¬ In(c, p))\nAction(Fly(p, from, to),\n\tPRECOND: At(p, from) ∧ Plane(p) ∧ Airport(from) ∧ Airport(to)\n\tEFFECT: ¬ At(p, from) ∧ At(p, to))\n```",
"_____no_output_____"
],
[
"## Planning Graph nodes",
"_____no_output_____"
]
],
[
[
"from planning_agent.aimacode.planning import Action\nfrom planning_agent.aimacode.search import Problem\nfrom planning_agent.aimacode.utils import expr\nfrom planning_agent.lp_utils import decode_state\n\n\nclass PgNode():\n \"\"\" Base class for planning graph nodes.\n\n includes instance sets common to both types of nodes used in a planning graph\n parents: the set of nodes in the previous level\n children: the set of nodes in the subsequent level\n mutex: the set of sibling nodes that are mutually exclusive with this node\n \"\"\"\n\n def __init__(self):\n self.parents = set()\n self.children = set()\n self.mutex = set()\n\n def is_mutex(self, other) -> bool:\n \"\"\" Boolean test for mutual exclusion\n\n :param other: PgNode\n the other node to compare with\n :return: bool\n True if this node and the other are marked mutually exclusive (mutex)\n \"\"\"\n if other in self.mutex:\n return True\n return False\n\n def show(self):\n \"\"\" helper print for debugging shows counts of parents, children, siblings\n\n :return:\n print only\n \"\"\"\n print(\"{} parents\".format(len(self.parents)))\n print(\"{} children\".format(len(self.children)))\n print(\"{} mutex\".format(len(self.mutex)))\n\n\nclass PgNode_s(PgNode):\n \"\"\"\n A planning graph node representing a state (literal fluent) from a planning\n problem.\n\n Args:\n ----------\n symbol : str\n A string representing a literal expression from a planning problem\n domain.\n\n is_pos : bool\n Boolean flag indicating whether the literal expression is positive or\n negative.\n \"\"\"\n\n def __init__(self, symbol: str, is_pos: bool):\n \"\"\" S-level Planning Graph node constructor\n\n :param symbol: expr\n :param is_pos: bool\n Instance variables calculated:\n literal: expr\n fluent in its literal form including negative operator if applicable\n Instance variables inherited from PgNode:\n parents: set of nodes connected to this node in previous A level; initially empty\n children: set of nodes connected to this node in next A level; initially empty\n mutex: set of sibling S-nodes that this node has mutual exclusion with; initially empty\n \"\"\"\n PgNode.__init__(self)\n self.symbol = symbol\n self.is_pos = is_pos\n self.literal = expr(self.symbol)\n if not self.is_pos:\n self.literal = expr('~{}'.format(self.symbol))\n\n def show(self):\n \"\"\"helper print for debugging shows literal plus counts of parents, children, siblings\n\n :return:\n print only\n \"\"\"\n print(\"\\n*** {}\".format(self.literal))\n PgNode.show(self)\n\n def __eq__(self, other):\n \"\"\"equality test for nodes - compares only the literal for equality\n\n :param other: PgNode_s\n :return: bool\n \"\"\"\n if isinstance(other, self.__class__):\n return (self.symbol == other.symbol) \\\n and (self.is_pos == other.is_pos)\n\n def __hash__(self):\n return hash(self.symbol) ^ hash(self.is_pos)\n\n\nclass PgNode_a(PgNode):\n \"\"\"A-type (action) Planning Graph node - inherited from PgNode\n \"\"\"\n\n def __init__(self, action: Action):\n \"\"\"A-level Planning Graph node constructor\n\n :param action: Action\n a ground action, i.e. this action cannot contain any variables\n Instance variables calculated:\n An A-level will always have an S-level as its parent and an S-level as its child.\n The preconditions and effects will become the parents and children of the A-level node\n However, when this node is created, it is not yet connected to the graph\n prenodes: set of *possible* parent S-nodes\n effnodes: set of *possible* child S-nodes\n is_persistent: bool True if this is a persistence action, i.e. a no-op action\n Instance variables inherited from PgNode:\n parents: set of nodes connected to this node in previous S level; initially empty\n children: set of nodes connected to this node in next S level; initially empty\n mutex: set of sibling A-nodes that this node has mutual exclusion with; initially empty\n \"\"\"\n PgNode.__init__(self)\n self.action = action\n self.prenodes = self.precond_s_nodes()\n self.effnodes = self.effect_s_nodes()\n self.is_persistent = False\n if self.prenodes == self.effnodes:\n self.is_persistent = True\n\n def show(self):\n \"\"\"helper print for debugging shows action plus counts of parents, children, siblings\n\n :return:\n print only\n \"\"\"\n print(\"\\n*** {}{}\".format(self.action.name, self.action.args))\n PgNode.show(self)\n\n def precond_s_nodes(self):\n \"\"\"precondition literals as S-nodes (represents possible parents for this node).\n It is computationally expensive to call this function; it is only called by the\n class constructor to populate the `prenodes` attribute.\n\n :return: set of PgNode_s\n \"\"\"\n nodes = set()\n for p in self.action.precond_pos:\n n = PgNode_s(p, True)\n nodes.add(n)\n for p in self.action.precond_neg:\n n = PgNode_s(p, False)\n nodes.add(n)\n return nodes\n\n def effect_s_nodes(self):\n \"\"\"effect literals as S-nodes (represents possible children for this node).\n It is computationally expensive to call this function; it is only called by the\n class constructor to populate the `effnodes` attribute.\n\n :return: set of PgNode_s\n \"\"\"\n nodes = set()\n for e in self.action.effect_add:\n n = PgNode_s(e, True)\n nodes.add(n)\n for e in self.action.effect_rem:\n n = PgNode_s(e, False)\n nodes.add(n)\n return nodes\n\n def __eq__(self, other):\n \"\"\"equality test for nodes - compares only the action name for equality\n\n :param other: PgNode_a\n :return: bool\n \"\"\"\n if isinstance(other, self.__class__):\n return (self.action.name == other.action.name) \\\n and (self.action.args == other.action.args)\n\n def __hash__(self):\n return hash(self.action.name) ^ hash(self.action.args)\n",
"_____no_output_____"
]
],
[
[
"## Planning Graph",
"_____no_output_____"
]
],
[
[
"def mutexify(node1: PgNode, node2: PgNode):\n \"\"\" adds sibling nodes to each other's mutual exclusion (mutex) set. These should be sibling nodes!\n :param node1: PgNode (or inherited PgNode_a, PgNode_s types)\n :param node2: PgNode (or inherited PgNode_a, PgNode_s types)\n :return:\n node mutex sets modified\n \"\"\"\n if type(node1) != type(node2):\n raise TypeError('Attempted to mutex two nodes of different types')\n node1.mutex.add(node2)\n node2.mutex.add(node1)\n\n\nclass PlanningGraph():\n \"\"\"\n A planning graph as described in chapter 10 of the AIMA text. The planning\n graph can be used to reason about \n \"\"\"\n\n def __init__(self, problem: Problem, state: str, serial_planning=True):\n \"\"\"\n :param problem: PlanningProblem (or subclass such as AirCargoProblem or HaveCakeProblem)\n :param state: str (will be in form TFTTFF... representing fluent states)\n :param serial_planning: bool (whether or not to assume that only one action can occur at a time)\n Instance variable calculated:\n fs: FluentState\n the state represented as positive and negative fluent literal lists\n all_actions: list of the PlanningProblem valid ground actions combined with calculated no-op actions\n s_levels: list of sets of PgNode_s, where each set in the list represents an S-level in the planning graph\n a_levels: list of sets of PgNode_a, where each set in the list represents an A-level in the planning graph\n \"\"\"\n self.problem = problem\n self.fs = decode_state(state, problem.state_map)\n self.serial = serial_planning\n self.all_actions = self.problem.actions_list + self.noop_actions(self.problem.state_map)\n self.s_levels = []\n self.a_levels = []\n self.create_graph()\n\n def noop_actions(self, literal_list):\n \"\"\"create persistent action for each possible fluent\n\n \"No-Op\" actions are virtual actions (i.e., actions that only exist in\n the planning graph, not in the planning problem domain) that operate\n on each fluent (literal expression) from the problem domain. No op\n actions \"pass through\" the literal expressions from one level of the\n planning graph to the next.\n\n The no-op action list requires both a positive and a negative action\n for each literal expression. Positive no-op actions require the literal\n as a positive precondition and add the literal expression as an effect\n in the output, and negative no-op actions require the literal as a\n negative precondition and remove the literal expression as an effect in\n the output.\n\n This function should only be called by the class constructor.\n\n :param literal_list:\n :return: list of Action\n \"\"\"\n action_list = []\n for fluent in literal_list:\n act1 = Action(expr(\"Noop_pos({})\".format(fluent)), ([fluent], []), ([fluent], []))\n action_list.append(act1)\n act2 = Action(expr(\"Noop_neg({})\".format(fluent)), ([], [fluent]), ([], [fluent]))\n action_list.append(act2)\n return action_list\n\n def create_graph(self):\n \"\"\" build a Planning Graph as described in Russell-Norvig 3rd Ed 10.3 or 2nd Ed 11.4\n\n The S0 initial level has been implemented for you. It has no parents and includes all of\n the literal fluents that are part of the initial state passed to the constructor. At the start\n of a problem planning search, this will be the same as the initial state of the problem. However,\n the planning graph can be built from any state in the Planning Problem\n\n This function should only be called by the class constructor.\n\n :return:\n builds the graph by filling s_levels[] and a_levels[] lists with node sets for each level\n \"\"\"\n # the graph should only be built during class construction\n if (len(self.s_levels) != 0) or (len(self.a_levels) != 0):\n raise Exception(\n 'Planning Graph already created; construct a new planning graph for each new state in the planning sequence')\n\n # initialize S0 to literals in initial state provided.\n leveled = False\n level = 0\n self.s_levels.append(set()) # S0 set of s_nodes - empty to start\n # for each fluent in the initial state, add the correct literal PgNode_s\n for literal in self.fs.pos:\n self.s_levels[level].add(PgNode_s(literal, True))\n for literal in self.fs.neg:\n self.s_levels[level].add(PgNode_s(literal, False))\n # no mutexes at the first level\n\n # continue to build the graph alternating A, S levels until last two S levels contain the same literals,\n # i.e. until it is \"leveled\"\n while not leveled:\n self.add_action_level(level)\n self.update_a_mutex(self.a_levels[level])\n\n level += 1\n self.add_literal_level(level)\n self.update_s_mutex(self.s_levels[level])\n\n if self.s_levels[level] == self.s_levels[level - 1]:\n leveled = True\n\n def add_action_level(self, level):\n \"\"\" add an A (action) level to the Planning Graph\n\n :param level: int\n the level number alternates S0, A0, S1, A1, S2, .... etc the level number is also used as the\n index for the node set lists self.a_levels[] and self.s_levels[]\n :return:\n adds A nodes to the current level in self.a_levels[level]\n \"\"\"\n\n self.a_levels.append(set()) # set of a_nodes\n for a in self.all_actions:\n a_node = PgNode_a(a)\n if set(a_node.prenodes).issubset(set(self.s_levels[level])): # True: Valid A node\n for s_node in self.s_levels[level]:\n if s_node in a_node.prenodes: # search for the right parents\n a_node.parents.add(s_node)\n s_node.children.add(a_node)\n self.a_levels[level].add(a_node)\n\n def add_literal_level(self, level):\n \"\"\" add an S (literal) level to the Planning Graph\n\n :param level: int\n the level number alternates S0, A0, S1, A1, S2, .... etc the level number is also used as the\n index for the node set lists self.a_levels[] and self.s_levels[]\n :return:\n adds S nodes to the current level in self.s_levels[level]\n \"\"\"\n\n self.s_levels.append(set()) # set of s_nodes\n for a in self.a_levels[level-1]:\n for s_node in a.effnodes: # Valid S nodes\n a.children.add(s_node)\n s_node.parents.add(a)\n self.s_levels[level].add(s_node)\n\n def update_a_mutex(self, nodeset):\n \"\"\" Determine and update sibling mutual exclusion for A-level nodes\n\n Mutex action tests section from 3rd Ed. 10.3 or 2nd Ed. 11.4\n A mutex relation holds between two actions a given level\n if the planning graph is a serial planning graph and the pair are nonpersistence actions\n or if any of the three conditions hold between the pair:\n Inconsistent Effects\n Interference\n Competing needs\n\n :param nodeset: set of PgNode_a (siblings in the same level)\n :return:\n mutex set in each PgNode_a in the set is appropriately updated\n \"\"\"\n nodelist = list(nodeset)\n for i, n1 in enumerate(nodelist[:-1]):\n for n2 in nodelist[i + 1:]:\n if (self.serialize_actions(n1, n2) or\n self.inconsistent_effects_mutex(n1, n2) or\n self.interference_mutex(n1, n2) or\n self.competing_needs_mutex(n1, n2)):\n mutexify(n1, n2)\n\n def serialize_actions(self, node_a1: PgNode_a, node_a2: PgNode_a) -> bool:\n \"\"\"\n Test a pair of actions for mutual exclusion, returning True if the\n planning graph is serial, and if either action is persistent; otherwise\n return False. Two serial actions are mutually exclusive if they are\n both non-persistent.\n\n :param node_a1: PgNode_a\n :param node_a2: PgNode_a\n :return: bool\n \"\"\"\n #\n if not self.serial:\n return False\n if node_a1.is_persistent or node_a2.is_persistent:\n return False\n return True\n\n def inconsistent_effects_mutex(self, node_a1: PgNode_a, node_a2: PgNode_a) -> bool:\n \"\"\"\n Test a pair of actions for inconsistent effects, returning True if\n one action negates an effect of the other, and False otherwise.\n\n HINT: The Action instance associated with an action node is accessible\n through the PgNode_a.action attribute. See the Action class\n documentation for details on accessing the effects and preconditions of\n an action.\n\n :param node_a1: PgNode_a\n :param node_a2: PgNode_a\n :return: bool\n \"\"\"\n\n # Create 1 set with all the adding effects and 1 set with all the removing effects.\n # (a single action cannot result in inconsistent effects)\n # If the intersection (&) of the two sets is not empty, then at least one effect negates another\n effects_add = node_a1.action.effect_add + node_a2.action.effect_add\n effects_rem = node_a1.action.effect_rem + node_a2.action.effect_rem\n return bool(set(effects_add) & set(effects_rem))\n\n def interference_mutex(self, node_a1: PgNode_a, node_a2: PgNode_a) -> bool:\n \"\"\"\n Test a pair of actions for mutual exclusion, returning True if the\n effect of one action is the negation of a precondition of the other.\n\n HINT: The Action instance associated with an action node is accessible\n through the PgNode_a.action attribute. See the Action class\n documentation for details on accessing the effects and preconditions of\n an action.\n\n :param node_a1: PgNode_a\n :param node_a2: PgNode_a\n :return: bool\n \"\"\"\n\n # Similar implementation of inconsistent_effects_mutex but crossing the adding/removing effect of each action\n # with the negative/positive precondition of the other.\n # 4 sets are used for 2 separated intersections. The intersection of 2 large sets (pos_add and neg_rem) would\n # also result True for inconsistent_effects\n cross_pos = node_a1.action.effect_add + node_a2.action.precond_pos\n cross_neg = node_a1.action.precond_neg + node_a2.action.effect_rem\n cross_pos2 = node_a2.action.effect_add + node_a1.action.precond_pos\n cross_neg2 = node_a2.action.precond_neg + node_a1.action.effect_rem\n\n return bool(set(cross_pos) & set(cross_neg)) or bool(set(cross_pos2) & set(cross_neg2))\n\n def competing_needs_mutex(self, node_a1: PgNode_a, node_a2: PgNode_a) -> bool:\n \"\"\"\n Test a pair of actions for mutual exclusion, returning True if one of\n the precondition of one action is mutex with a precondition of the\n other action.\n\n :param node_a1: PgNode_a\n :param node_a2: PgNode_a\n :return: bool\n \"\"\"\n\n # Create a list with the parents of one action node that are mutually exclusive with the parents of the other\n # and return True if the list is not empty\n mutex = [i for i in node_a1.parents for j in node_a2.parents if i.is_mutex(j)]\n return bool(mutex)\n\n\n def update_s_mutex(self, nodeset: set):\n \"\"\" Determine and update sibling mutual exclusion for S-level nodes\n\n Mutex action tests section from 3rd Ed. 10.3 or 2nd Ed. 11.4\n A mutex relation holds between literals at a given level\n if either of the two conditions hold between the pair:\n Negation\n Inconsistent support\n\n :param nodeset: set of PgNode_a (siblings in the same level)\n :return:\n mutex set in each PgNode_a in the set is appropriately updated\n \"\"\"\n nodelist = list(nodeset)\n for i, n1 in enumerate(nodelist[:-1]):\n for n2 in nodelist[i + 1:]:\n if self.negation_mutex(n1, n2) or self.inconsistent_support_mutex(n1, n2):\n mutexify(n1, n2)\n\n def negation_mutex(self, node_s1: PgNode_s, node_s2: PgNode_s) -> bool:\n \"\"\"\n Test a pair of state literals for mutual exclusion, returning True if\n one node is the negation of the other, and False otherwise.\n\n HINT: Look at the PgNode_s.__eq__ defines the notion of equivalence for\n literal expression nodes, and the class tracks whether the literal is\n positive or negative.\n\n :param node_s1: PgNode_s\n :param node_s2: PgNode_s\n :return: bool\n \"\"\"\n\n # Mutual exclusive nodes have the same 'symbol' and different 'is_pos' attributes\n return (node_s1.symbol == node_s2.symbol) and (node_s1.is_pos != node_s2.is_pos)\n\n def inconsistent_support_mutex(self, node_s1: PgNode_s, node_s2: PgNode_s):\n \"\"\"\n Test a pair of state literals for mutual exclusion, returning True if\n there are no actions that could achieve the two literals at the same\n time, and False otherwise. In other words, the two literal nodes are\n mutex if all of the actions that could achieve the first literal node\n are pairwise mutually exclusive with all of the actions that could\n achieve the second literal node.\n\n HINT: The PgNode.is_mutex method can be used to test whether two nodes\n are mutually exclusive.\n\n :param node_s1: PgNode_s\n :param node_s2: PgNode_s\n :return: bool\n \"\"\"\n\n # Get a list with the parents of one node that are not mutually exclusive with at least one parent of the other\n # Here the inconsistent is detected if the list is empty (none of the actions can lead to these pair of nodes at\n # the same time)\n compatible_parents_s1 = [a for a in node_s1.parents for b in node_s2.parents if not a.is_mutex(b)]\n return not bool(compatible_parents_s1)\n\n\n def h_levelsum(self) -> int:\n \"\"\"The sum of the level costs of the individual goals (admissible if goals independent)\n\n :return: int\n \"\"\"\n level_sum = 0\n\n # for each goal in the problem, determine the level cost, then add them together\n remaining_goals = set(self.problem.goal) # remaining goals to find to determine the level cost\n # Search for all the goals simultaneously from level 0\n for level in range(len(self.s_levels)+1):\n literals = set([node.literal for node in self.s_levels[level]]) # literals found in the current level\n match = literals & remaining_goals # set of goals found in literals (empty set if none)\n level_sum += len(match)*level # add cost of the found goals (0 if none)\n remaining_goals -= match # remove found goals from the remaining goals\n if not remaining_goals: # return when all goals are found\n return level_sum\n raise Exception(\"Goal not found\")",
"_____no_output_____"
]
],
[
[
"## Air Cargo Problem",
"_____no_output_____"
]
],
[
[
"from planning_agent.aimacode.logic import PropKB\nfrom planning_agent.aimacode.planning import Action\nfrom planning_agent.aimacode.search import Node, Problem\nfrom planning_agent.aimacode.utils import expr\nfrom planning_agent.lp_utils import FluentState, encode_state, decode_state\n\n\nclass AirCargoProblem(Problem):\n def __init__(self, cargos, planes, airports, initial: FluentState, goal: list):\n \"\"\"\n :param cargos: list of str\n cargos in the problem\n :param planes: list of str\n planes in the problem\n :param airports: list of str\n airports in the problem\n :param initial: FluentState object\n positive and negative literal fluents (as expr) describing initial state\n :param goal: list of expr\n literal fluents required for goal test\n \"\"\"\n self.state_map = initial.pos + initial.neg\n self.initial_state_TF = encode_state(initial, self.state_map)\n Problem.__init__(self, self.initial_state_TF, goal=goal)\n self.cargos = cargos\n self.planes = planes\n self.airports = airports\n self.actions_list = self.get_actions()\n\n def get_actions(self):\n \"\"\"\n This method creates concrete actions (no variables) for all actions in the problem\n domain action schema and turns them into complete Action objects as defined in the\n aimacode.planning module. It is computationally expensive to call this method directly;\n however, it is called in the constructor and the results cached in the `actions_list` property.\n Returns:\n ----------\n list<Action>\n list of Action objects\n \"\"\"\n\n def load_actions():\n \"\"\"Create all concrete Load actions and return a list\n :return: list of Action objects\n \"\"\"\n loads = []\n\n for c in self.cargos:\n for p in self.planes:\n for a in self.airports:\n precond_pos = [expr(\"At({}, {})\".format(c, a)),\n expr(\"At({}, {})\".format(p, a))]\n precond_neg = []\n effect_add = [expr(\"In({}, {})\".format(c, p))]\n effect_rem = [expr(\"At({}, {})\".format(c, a))]\n load = Action(expr(\"Load({}, {}, {})\".format(c, p, a)),\n [precond_pos, precond_neg],\n [effect_add, effect_rem])\n loads.append(load)\n return loads\n\n def unload_actions():\n \"\"\"Create all concrete Unload actions and return a list\n :return: list of Action objects\n \"\"\"\n unloads = []\n\n for c in self.cargos:\n for p in self.planes:\n for a in self.airports:\n precond_pos = [expr(\"In({}, {})\".format(c, p)),\n expr(\"At({}, {})\".format(p, a))]\n precond_neg = []\n effect_add = [expr(\"At({}, {})\".format(c, a))]\n effect_rem = [expr(\"In({}, {})\".format(c, p))]\n unload = Action(expr(\"Unload({}, {}, {})\".format(c, p, a)),\n [precond_pos, precond_neg],\n [effect_add, effect_rem])\n unloads.append(unload)\n return unloads\n\n def fly_actions():\n \"\"\"Create all concrete Fly actions and return a list\n :return: list of Action objects\n \"\"\"\n flys = []\n for fr in self.airports:\n for to in self.airports:\n if fr != to:\n for p in self.planes:\n precond_pos = [expr(\"At({}, {})\".format(p, fr)),\n ]\n precond_neg = []\n effect_add = [expr(\"At({}, {})\".format(p, to))]\n effect_rem = [expr(\"At({}, {})\".format(p, fr))]\n fly = Action(expr(\"Fly({}, {}, {})\".format(p, fr, to)),\n [precond_pos, precond_neg],\n [effect_add, effect_rem])\n flys.append(fly)\n return flys\n\n return load_actions() + unload_actions() + fly_actions()\n\n def actions(self, state: str) -> list:\n \"\"\" Return the actions that can be executed in the given state.\n :param state: str\n state represented as T/F string of mapped fluents (state variables)\n e.g. 'FTTTFF'\n :return: list of Action objects\n \"\"\"\n\n possible_actions = []\n kb = PropKB()\n kb.tell(decode_state(state, self.state_map).pos_sentence())\n for action in self.actions_list:\n is_possible = True\n for clause in action.precond_pos:\n if clause not in kb.clauses:\n is_possible = False\n for clause in action.precond_neg:\n if clause in kb.clauses:\n is_possible = False\n if is_possible:\n possible_actions.append(action)\n return possible_actions\n\n def result(self, state: str, action: Action):\n \"\"\" Return the state that results from executing the given\n action in the given state. The action must be one of\n self.actions(state).\n\n :param state: state entering node\n :param action: Action applied\n :return: resulting state after action\n \"\"\"\n\n new_state = FluentState([], [])\n # Used the same implementation as cake example:\n old_state = decode_state(state, self.state_map)\n for fluent in old_state.pos:\n if fluent not in action.effect_rem:\n new_state.pos.append(fluent)\n for fluent in action.effect_add:\n if fluent not in new_state.pos:\n new_state.pos.append(fluent)\n for fluent in old_state.neg:\n if fluent not in action.effect_add:\n new_state.neg.append(fluent)\n for fluent in action.effect_rem:\n if fluent not in new_state.neg:\n new_state.neg.append(fluent)\n return encode_state(new_state, self.state_map)\n\n def goal_test(self, state: str) -> bool:\n \"\"\" Test the state to see if goal is reached\n :param state: str representing state\n :return: bool\n \"\"\"\n kb = PropKB()\n kb.tell(decode_state(state, self.state_map).pos_sentence())\n for clause in self.goal:\n if clause not in kb.clauses:\n return False\n return True\n\n def h_1(self, node: Node):\n # note that this is not a true heuristic\n h_const = 1\n return h_const\n\n def h_pg_levelsum(self, node: Node):\n \"\"\"\n This heuristic uses a planning graph representation of the problem\n state space to estimate the sum of all actions that must be carried\n out from the current state in order to satisfy each individual goal\n condition.\n \"\"\"\n # requires implemented PlanningGraph class\n pg = PlanningGraph(self, node.state)\n pg_levelsum = pg.h_levelsum()\n return pg_levelsum\n\n def h_ignore_preconditions(self, node: Node):\n \"\"\"\n This heuristic estimates the minimum number of actions that must be\n carried out from the current state in order to satisfy all of the goal\n conditions by ignoring the preconditions required for an action to be\n executed.\n \"\"\"\n\n # Note: We assume that the number of steps required to solve the relaxed ignore preconditions problem\n # is equal to the number of unsatisfied goals.\n # Thus no action results in multiple goals and no action undoes the effects of other actions\n kb = PropKB()\n kb.tell(decode_state(node.state, self.state_map).pos_sentence())\n # Unsatisfied goals are the ones not found in the clauses of PropKB() for the current state\n count = len(set(self.goal) - set(kb.clauses))\n # print(\"Current_state: \", kb.clauses, \" Goal state: \", self.goal)\n return count",
"_____no_output_____"
]
],
[
[
"## Scenarios",
"_____no_output_____"
]
],
[
[
"def air_cargo_p1() -> AirCargoProblem:\n cargos = ['C1', 'C2']\n planes = ['P1', 'P2']\n airports = ['JFK', 'SFO']\n pos = [expr('At(C1, SFO)'),\n expr('At(C2, JFK)'),\n expr('At(P1, SFO)'),\n expr('At(P2, JFK)'),\n ]\n neg = [expr('At(C2, SFO)'),\n expr('In(C2, P1)'),\n expr('In(C2, P2)'),\n expr('At(C1, JFK)'),\n expr('In(C1, P1)'),\n expr('In(C1, P2)'),\n expr('At(P1, JFK)'),\n expr('At(P2, SFO)'),\n ]\n init = FluentState(pos, neg)\n goal = [expr('At(C1, JFK)'),\n expr('At(C2, SFO)'),\n ]\n return AirCargoProblem(cargos, planes, airports, init, goal)\n\n\ndef air_cargo_p2() -> AirCargoProblem:\n\n cargos = ['C1', 'C2', 'C3']\n planes = ['P1', 'P2', 'P3']\n airports = ['SFO', 'JFK', 'ATL']\n pos = [expr('At(C1, SFO)'),\n expr('At(C2, JFK)'),\n expr('At(C3, ATL)'),\n expr('At(P1, SFO)'),\n expr('At(P2, JFK)'),\n expr('At(P3, ATL)'),\n ]\n neg = [expr('At(C1, JFK)'),\n expr('At(C1, ATL)'),\n expr('At(C2, SFO)'),\n expr('At(C2, ATL)'),\n expr('At(C3, SFO)'),\n expr('At(C3, JFK)'),\n expr('In(C1, P1)'),\n expr('In(C1, P2)'),\n expr('In(C1, P3)'),\n expr('In(C2, P1)'),\n expr('In(C2, P2)'),\n expr('In(C2, P3)'),\n expr('In(C3, P1)'),\n expr('In(C3, P2)'),\n expr('In(C3, P3)'),\n expr('At(P1, JFK)'),\n expr('At(P1, ATL)'),\n expr('At(P2, SFO)'),\n expr('At(P2, ATL)'),\n expr('At(P3, SFO)'),\n expr('At(P3, JFK)'),\n ]\n init = FluentState(pos, neg)\n goal = [expr('At(C1, JFK)'),\n expr('At(C2, SFO)'),\n expr('At(C3, SFO)'),\n ]\n return AirCargoProblem(cargos, planes, airports, init, goal)\n\n\ndef air_cargo_p3() -> AirCargoProblem:\n\n cargos = ['C1', 'C2', 'C3', 'C4']\n planes = ['P1', 'P2']\n airports = ['SFO', 'JFK', 'ATL', 'ORD']\n pos = [expr('At(C1, SFO)'),\n expr('At(C2, JFK)'),\n expr('At(C3, ATL)'),\n expr('At(C4, ORD)'),\n expr('At(P1, SFO)'),\n expr('At(P2, JFK)'),\n ]\n neg = [expr('At(C1, JFK)'),\n expr('At(C1, ATL)'),\n expr('At(C1, ORD)'),\n expr('At(C2, SFO)'),\n expr('At(C2, ATL)'),\n expr('At(C2, ORD)'),\n expr('At(C3, JFK)'),\n expr('At(C3, SFO)'),\n expr('At(C3, ORD)'),\n expr('At(C4, JFK)'),\n expr('At(C4, SFO)'),\n expr('At(C4, ATL)'),\n expr('In(C1, P1)'),\n expr('In(C1, P2)'),\n expr('In(C2, P1)'),\n expr('In(C2, P2)'),\n expr('In(C3, P1)'),\n expr('In(C3, P2)'),\n expr('In(C4, P1)'),\n expr('In(C4, P2)'),\n expr('At(P1, JFK)'),\n expr('At(P1, ATL)'),\n expr('At(P1, ORD)'),\n expr('At(P2, SFO)'),\n expr('At(P2, ATL)'),\n expr('At(P2, ORD)'),\n\n ]\n init = FluentState(pos, neg)\n goal = [expr('At(C1, JFK)'),\n expr('At(C2, SFO)'),\n expr('At(C3, JFK)'),\n expr('At(C4, SFO)'),\n ]\n return AirCargoProblem(cargos, planes, airports, init, goal)",
"_____no_output_____"
]
],
[
[
"- Problem 1 initial state and goal:\n```\nInit(At(C1, SFO) ∧ At(C2, JFK) \n\t∧ At(P1, SFO) ∧ At(P2, JFK) \n\t∧ Cargo(C1) ∧ Cargo(C2) \n\t∧ Plane(P1) ∧ Plane(P2)\n\t∧ Airport(JFK) ∧ Airport(SFO))\nGoal(At(C1, JFK) ∧ At(C2, SFO))\n```\n- Problem 2 initial state and goal:\n```\nInit(At(C1, SFO) ∧ At(C2, JFK) ∧ At(C3, ATL) \n\t∧ At(P1, SFO) ∧ At(P2, JFK) ∧ At(P3, ATL) \n\t∧ Cargo(C1) ∧ Cargo(C2) ∧ Cargo(C3)\n\t∧ Plane(P1) ∧ Plane(P2) ∧ Plane(P3)\n\t∧ Airport(JFK) ∧ Airport(SFO) ∧ Airport(ATL))\nGoal(At(C1, JFK) ∧ At(C2, SFO) ∧ At(C3, SFO))\n```\n- Problem 3 initial state and goal:\n```\nInit(At(C1, SFO) ∧ At(C2, JFK) ∧ At(C3, ATL) ∧ At(C4, ORD) \n\t∧ At(P1, SFO) ∧ At(P2, JFK) \n\t∧ Cargo(C1) ∧ Cargo(C2) ∧ Cargo(C3) ∧ Cargo(C4)\n\t∧ Plane(P1) ∧ Plane(P2)\n\t∧ Airport(JFK) ∧ Airport(SFO) ∧ Airport(ATL) ∧ Airport(ORD))\nGoal(At(C1, JFK) ∧ At(C3, JFK) ∧ At(C2, SFO) ∧ At(C4, SFO))\n```",
"_____no_output_____"
],
[
"## Solving the problem",
"_____no_output_____"
]
],
[
[
"import argparse\nfrom timeit import default_timer as timer\nfrom planning_agent.aimacode.search import InstrumentedProblem\nfrom planning_agent.aimacode.search import (breadth_first_search, astar_search,\n breadth_first_tree_search, depth_first_graph_search, uniform_cost_search,\n greedy_best_first_graph_search, depth_limited_search,\n recursive_best_first_search)\n\n\n\nPROBLEMS = [[\"Air Cargo Problem 1\", air_cargo_p1],\n [\"Air Cargo Problem 2\", air_cargo_p2],\n [\"Air Cargo Problem 3\", air_cargo_p3]]\nSEARCHES = [[\"breadth_first_search\", breadth_first_search, \"\"],\n ['breadth_first_tree_search', breadth_first_tree_search, \"\"],\n ['depth_first_graph_search', depth_first_graph_search, \"\"],\n ['depth_limited_search', depth_limited_search, \"\"],\n ['uniform_cost_search', uniform_cost_search, \"\"],\n ['recursive_best_first_search', recursive_best_first_search, 'h_1'],\n ['greedy_best_first_graph_search', greedy_best_first_graph_search, 'h_1'],\n ['astar_search', astar_search, 'h_1'],\n ['astar_search', astar_search, 'h_ignore_preconditions'],\n ['astar_search', astar_search, 'h_pg_levelsum'],\n ]\n\n\nclass PrintableProblem(InstrumentedProblem):\n \"\"\" InstrumentedProblem keeps track of stats during search, and this class modifies the print output of those\n statistics for air cargo problems \"\"\"\n\n def __repr__(self):\n return '{:^10d} {:^10d} {:^10d}'.format(self.succs, self.goal_tests, self.states)\n\n\ndef show_solution(node, elapsed_time):\n print(\"Plan length: {} Time elapsed in seconds: {}\".format(len(node.solution()), elapsed_time))\n for action in node.solution():\n print(\"{}{}\".format(action.name, action.args))\n\n\ndef run_search(problem, search_function, parameter=None):\n\n start = timer()\n ip = PrintableProblem(problem)\n if parameter is not None:\n node = search_function(ip, parameter)\n else:\n node = search_function(ip)\n end = timer()\n print(\"\\nExpansions Goal Tests New Nodes\")\n print(\"{}\\n\".format(ip))\n show_solution(node, end - start)\n print()\n\n\n\ndef main(p_choices, s_choices):\n\n problems = [PROBLEMS[i-1] for i in map(int, p_choices)]\n searches = [SEARCHES[i-1] for i in map(int, s_choices)]\n\n for pname, p in problems:\n\n for sname, s, h in searches:\n hstring = h if not h else \" with {}\".format(h)\n print(\"\\nSolving {} using {}{}...\".format(pname, sname, hstring))\n\n _p = p()\n _h = None if not h else getattr(_p, h)\n run_search(_p, s, _h)\n\n\n\n\nif __name__==\"__main__\":\n main([1,2,3],[1,9])",
"\nSolving Air Cargo Problem 1 using breadth_first_search...\n\nExpansions Goal Tests New Nodes\n 43 56 180 \n\nPlan length: 6 Time elapsed in seconds: 0.03141521600082342\nLoad(C1, P1, SFO)\nLoad(C2, P2, JFK)\nFly(P2, JFK, SFO)\nUnload(C2, P2, SFO)\nFly(P1, SFO, JFK)\nUnload(C1, P1, JFK)\n\n\nSolving Air Cargo Problem 1 using astar_search with h_ignore_preconditions...\n\nExpansions Goal Tests New Nodes\n 41 43 170 \n\nPlan length: 6 Time elapsed in seconds: 0.06203888300115068\nLoad(C1, P1, SFO)\nFly(P1, SFO, JFK)\nUnload(C1, P1, JFK)\nLoad(C2, P2, JFK)\nFly(P2, JFK, SFO)\nUnload(C2, P2, SFO)\n\n\nSolving Air Cargo Problem 2 using breadth_first_search...\n\nExpansions Goal Tests New Nodes\n 3346 4612 30534 \n\nPlan length: 9 Time elapsed in seconds: 14.21029247200022\nLoad(C1, P1, SFO)\nLoad(C2, P2, JFK)\nLoad(C3, P3, ATL)\nFly(P1, SFO, JFK)\nUnload(C1, P1, JFK)\nFly(P2, JFK, SFO)\nUnload(C2, P2, SFO)\nFly(P3, ATL, SFO)\nUnload(C3, P3, SFO)\n\n\nSolving Air Cargo Problem 2 using astar_search with h_ignore_preconditions...\n\nExpansions Goal Tests New Nodes\n 1506 1508 13820 \n\nPlan length: 9 Time elapsed in seconds: 13.803835380998862\nLoad(C3, P3, ATL)\nFly(P3, ATL, SFO)\nUnload(C3, P3, SFO)\nLoad(C2, P2, JFK)\nFly(P2, JFK, SFO)\nUnload(C2, P2, SFO)\nLoad(C1, P1, SFO)\nFly(P1, SFO, JFK)\nUnload(C1, P1, JFK)\n\n\nSolving Air Cargo Problem 3 using breadth_first_search...\n\nExpansions Goal Tests New Nodes\n 14120 17673 124926 \n\nPlan length: 12 Time elapsed in seconds: 99.77857781799867\nLoad(C1, P1, SFO)\nLoad(C2, P2, JFK)\nFly(P1, SFO, ATL)\nLoad(C3, P1, ATL)\nFly(P2, JFK, ORD)\nLoad(C4, P2, ORD)\nFly(P1, ATL, JFK)\nUnload(C1, P1, JFK)\nUnload(C3, P1, JFK)\nFly(P2, ORD, SFO)\nUnload(C2, P2, SFO)\nUnload(C4, P2, SFO)\n\n\nSolving Air Cargo Problem 3 using astar_search with h_ignore_preconditions...\n\nExpansions Goal Tests New Nodes\n 5118 5120 45650 \n\nPlan length: 12 Time elapsed in seconds: 87.09942936799962\nLoad(C2, P2, JFK)\nFly(P2, JFK, ORD)\nLoad(C4, P2, ORD)\nFly(P2, ORD, SFO)\nUnload(C4, P2, SFO)\nLoad(C1, P1, SFO)\nFly(P1, SFO, ATL)\nLoad(C3, P1, ATL)\nFly(P1, ATL, JFK)\nUnload(C3, P1, JFK)\nUnload(C2, P2, SFO)\nUnload(C1, P1, JFK)\n\n"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
]
] |
cbd8efeb091f75a7ae837e13d70456fdf108a6b2
| 17,284 |
ipynb
|
Jupyter Notebook
|
cs229_ml/lec10-DecisionTrees-EnsembleMethods.ipynb
|
chandrabsingh/learnings
|
a3f507bbbf46582ce5a64991983dfc0759db0af5
|
[
"MIT"
] | null | null | null |
cs229_ml/lec10-DecisionTrees-EnsembleMethods.ipynb
|
chandrabsingh/learnings
|
a3f507bbbf46582ce5a64991983dfc0759db0af5
|
[
"MIT"
] | null | null | null |
cs229_ml/lec10-DecisionTrees-EnsembleMethods.ipynb
|
chandrabsingh/learnings
|
a3f507bbbf46582ce5a64991983dfc0759db0af5
|
[
"MIT"
] | null | null | null | 40.86052 | 320 | 0.60339 |
[
[
[
">>> Work in Progress (Following are the lecture notes of Prof Andrew Ng/Head TA-Raphael Townshend - CS229 - Stanford. This is my interpretation of his excellent teaching and I take full responsibility of any misinterpretation/misinformation provided herein.)",
"_____no_output_____"
],
[
"## Lecture Notes\n\n#### Outline\n- Decision Trees\n- Ensemble Methods\n- Bagging\n- Random Forests\n- Boosting",
"_____no_output_____"
],
[
"### Decision Trees\n- Non-linear model\n- A model is called linear if the hypothesis function is of the form $h(x) = \\theta^{T}x$ \n- Ski example - months vs latitude - when you can ski\n - we cannot get a linear classifier or use SVM for this\n - with decision trees you will have a very natural way of classifying this\n - partition this into individual regions, isolating positive and negative examples\n\n#### Selecting Regions - Greedy, Top-Down, Recursive Partitioning\n- You ask question and partition the space and then iteratively keep asking new question, partitioning the space\n- Is latitude > 30\n - Yes\n - Is Month < 3\n - Yes\n - No\n - No\n \n- We are looking for a split function\n- Region $R_{p}$\n - Looking for a split $S_{p}$\n > $S_{p}(j,t) = (\\{ X|X_{j} \\lt t, X \\in R_{p}\\}, \\{ X|X_{j} \\ge t, X \\in R_{p}\\} ) = (R_{1}, R_{2})$\n - where j is the feature number and t is the threshold \n \n#### How to choose splits\n- isolate space of positives and negatives in this case\n- Define L(R): loss on R\n- Given C class, define $\\hat{p_{i}}$ to be the __porportion of examples__ in R that are of class C\n- Define misclassification loss of any region as \n> $L_{misclass}(R) = 1 - \\max\\limits_{C} \\hat{p}_{C}$\n - what we are saying here is for any region that we have subdivided, we want to predict the most common class there, which is the maximum of $\\hat{p}_{C}$. The remaining is the probability of misclassification errors.\n- We want to pick a split that maximizes the decrease of loss as much as possible over parent $R_{parent}$ and children regions $R_{1}, R_{2}$\n> $\\max\\limits_{j,t} L(R_{p}) - (L(R_{1}) + L(R_{2}))$\n\n#### Why is misclassification loss the right loss \n\n\n<img src=\"images/10_misclassificationLoss.png\" width=400 height=400> \n$\\tiny{\\text{YouTube-Stanford-CS229-Andrew Ng/Raphael Townshend}}$ \n\n- We might argue that the decision boundary on right scenario is better than left, because in the right we are isolating out more positives \n\n- Loss of R1 and R2 region = 100 on right scenario\n- Loss of R1' and R2' region = 100 on left scenario\n- The loss of both parent Rp is also 100\n\n- We can see that the misclassification loss is not sensitive enough\n - its not sensitive enough or the loss is not informative enough because the parent level loss is same as child level loss\n\n- Instead we can define __cross entropy loss__\n> $L_{cross}(R) = - \\sum\\limits_{c}\\hat{p}_{c} log_{2}\\hat{p}_{c}$\n - we are summing over the classes the proportion of elements in that class times the log of proportion in that class\n - if we know everything about one class, we dont need to communicate, as we know everything that it's a 100% chance that it is of one class\n - if we have a even split, then we need to communicate lot more information about the class\n \n- Cross entropy came from information theory where it is used for transmitting bits, where you can transmit bits of information, which is why it came up as log base 2\n ",
"_____no_output_____"
],
[
"#### Misclassification loss vs Cross-entropy loss\n- Let the plot be between $\\hat{p}$ - the proportion of positives in the set vs the loss \n- the cross-entropy loss is a strictly concave curve\n- Let $L(R_{1})$ and $L(R_{2})$ be the child loss plotted on the curve\n- Let there be equal number of examples in both $R_{1}$ and $R_{2}$, are equally weighted \n- the overall loss between the two is the average loss between the two, which is $\\frac{L(R_{1}) + L(R_{2})}{2}$ \n- the parent node loss is the projected loss on the curve $L(R_{p})$\n- the projection height is the change in loss \n\n- as we see below, \\hat{p} parent is the average of child proportions\n\n<img src=\"images/10_crossEntropyLoss.png\" width=400 height=400> \n$\\tiny{\\text{YouTube-Stanford-CS229-Andrew Ng/Raphael Townshend}}$ \n\n- the cross-entropy diagram\n\n<img src=\"images/10_crossEntropyDiagram.png\" width=400 height=400> \n$\\tiny{\\text{YouTube-Stanford-CS229-Andrew Ng/Raphael Townshend}}$ \n\n\n",
"_____no_output_____"
],
[
"- the misrepresenstation loss\n - if we end up with child node loss on the same side of the curve, there is no change in loss and hence no information gain based on this kind of representation\n - this is not strictly concave curve\n\n<img src=\"images/10_misrepresentationDiagram.png\" width=400 height=400> \n$\\tiny{\\text{YouTube-Stanford-CS229-Andrew Ng/Raphael Townshend}}$ \n\n- the decision splits curves that are successfully used are strictly concave curve\n\n- Gini curve\n > $\\sum\\limits_{c}\\hat{p}_{c}(1-\\hat{p}_{c})$",
"_____no_output_____"
],
[
"#### Regression Tree - Extension for decision tree\n- So far we used decision tree for classification\n- Decision trees can also be used for regression trees\n- Example: Amount of snowfall\n - Instead of predicting class, you predict mean of the \n \n- For Region $R_{m}$, the prediction will be\n> Predict $\\hat{y}_{m} = \\frac{\\sum\\limits_{i \\in R_{m}}Y_{i}}{|R_{m}|}$\n - sum all the values within the region and average them\n \n\n<img src=\"images/10_regressionTrees.png\" width=400 height=400> \n$\\tiny{\\text{YouTube-Stanford-CS229-Andrew Ng/Raphael Townshend}}$ \n\n\nThe loss will be\n> $L_{squared} = \\frac{\\sum\\limits_{i \\in R_{m}} (y_{i} - \\hat{y}_{m})^{2} }{|R_{m}|}$",
"_____no_output_____"
],
[
"#### Categorical Variables\n- can ask questions on any form of subset, is location in northern hemisphere?\n- $location \\in \\{N\\}$\n- if there are q categories, the possible number of splits would be $2^{q}$, which very quickly becomes intractable\n\n",
"_____no_output_____"
],
[
"#### Regularization of DTs\n- if you carry on the process of splits, you can split region for each datapoint and that will be case of overfitting\n- Decision trees are high variance models\n- So we need to regularize the decision tree models\n- Heuristics for regularization \n - If you have a minimum leaf size, stop\n - max depth\n - max number of nodes\n - min decrease in loss\n - Before split, the loss is: $L(R_{p})$\n - After split, the loss is: $L(R_{1}) + L(R_{2})$\n - if after split, the loss is not great enough, we might conclude that it didn't gain us anything\n - but there might be some correlation between variables \n - pruning\n - you grow up your full tree and check which nodes to prune out\n - you have a validation set that you use and you evaluate what your misclassification error is on the validation set, for each example for each leaf",
"_____no_output_____"
],
[
"#### Runtime\n- n train examples\n- f features\n- d depth of tree\n\n##### Test time O(d)\nd < log n\n\n##### Train time\n- Each point is part of O(d) nodes\n- Cost of point at each node is O(f)\n - for binary features, the cost will be f\n - for quantitative features, sort and scan linearly, the cost will be f, as well\n- Total cost is O(nfd)\n - where data matrix size is nf\n - and depth is log n\n - so cost is fairly fast training time",
"_____no_output_____"
],
[
"#### Downside of DT\n- it does not have additive structure\n- in the example below we get a very rough estimation of decision boundary\n- decision trees have problems where the features are interacting additively with one another\n\n<img src=\"images/10_noAdditiveStructure.png\" width=400 height=400> \n$\\tiny{\\text{YouTube-Stanford-CS229-Andrew Ng/Raphael Townshend}}$ \n",
"_____no_output_____"
],
[
"#### DT - Recap\n- Pos\n - Easy to explain\n - Interpretable\n - can deal with categorical variable\n - generally fast\n\n- Neg\n - high variance problems - generally leads to overfitting\n - Not additive\n - Low predictive accuracy\n\n- We can make it lot better with ensembling",
"_____no_output_____"
],
[
"### Ensembling\n- take $X_{i}'s$ which are random variables that are independent identically distributed (i.i.d.) \n> $Var(X_{i}) = \\sigma^{2}$\n> $Var(\\bar{X}) = Var\\left(\\frac{1}{n}\\sum\\limits_{i}X_{i}\\right) = \\frac{\\sigma^{2}}{n}$ \n- which means each independent rv is decreasing the variance of your model\n\n- If we drop the independence assumption, so now $X_{i}'s$ are only i.d. X's are correlated by $\\rho$\n- So the variance of mean will be:\n> $Var(\\bar{X}) = \\rho \\sigma^{2} + \\frac{1-\\rho}{n} \\sigma^{2}$ \n - if they are fully correlated ($\\rho = 1$), it becomes $Var(\\bar{X}) = \\sigma^{2}$\n - if there is no correlation($\\rho = 0$), it becomes $Var(\\bar{X}) = \\frac{\\sigma^{2}}{n} $\n - there would be interest in models with large n so the second term goes down. Also have models that are decorrelated so the first term goes down",
"_____no_output_____"
],
[
"#### Ways to ensemble\n- different algorithms, not really helpful\n- use different training sets, not really helpful\n- Bagging - Random Forest\n- Boosting - Adaboost, xgboost",
"_____no_output_____"
],
[
"### Bagging\n- Bootstrap aggregation\n - bootstrapping is a method used in statistics to measure uncertainty\n- Say that a true population is P\n- Training set $S \\sim P$\n- Assume population is the training sample P = S\n- Bootstrap samples Z \\sim S\n - Z is sampled from S. We take a training sample S with cardinality N. We sample N times from S with replacement, because we are assuming that S is a population and we are sampling from a population\n - Take model and then train on all these separate bootstrap samples\n \n<br> \n\n#### Bootstrap aggregation\n - we will train separate models separately and then average their outputs\n - Say we have bootstrap samples $Z_{1},...,Z_{M}$\n - We train model $G_{m}$ on $Z_{m}$ and define \n > Aggregate Predictor $G(m) = \\frac{\\sum\\limits_{m=1}{M}G_{m}(x)}{M}$\n - This process is called bagging\n ",
"_____no_output_____"
],
[
"#### Bias-Variance Analysis\n> $Var(\\bar{X}) = \\rho \\sigma^{2} + \\frac{1-\\rho}{n} \\sigma^{2}$ \n- Bootstrapping is driving down $\\rho$\n- But what about the second term\n - With the increase in bootstrap samples, the M term increases, driving down the second term\n- A nice property about bootstrapping is that increasing the number of bootstrap models does not cause overfit than before. \n- More M causes less variance\n- But the bias of the model increases\n - because of the random subsampling from S, it causes model to be less complex as we are drawing less data, and increases the bias",
"_____no_output_____"
],
[
"#### Decision Trees + Bagging\n- DT have high variance, low bias\n- this makes DT ideal fit for bagging ",
"_____no_output_____"
],
[
"### Random Forest \n- RF is a version of decision trees and bagging\n- the random forest introduces even more randomization into each individual decision tree\n- 1st - Earlier we learnt, bootstrapping drives down $\\rho$\n- 2nd - But if we can further decorrelate the random variables, we can drive down the variance even further\n- At each split for RF, we consider only a fraction of your total features\n- 1st - Decreasing $\\rho$ in $Var(\\bar{X})$ \n- 2nd - Say in a classification problem, we have found a very strong predictor that gives very good performance on its own (in ski example - the latitude split), and we use that predictor first at the first split. That causes all your models to be very highly correlated. So we should try to decorrelate the models",
"_____no_output_____"
],
[
"### Boosting\n- In bagging we tried to reduce variance\n- Boosting is opposite. In boosting we try to reduce bias\n- Is additive \n- In bagging, we took average of number of variables\n- In boosting, we train one model and then add it into the ensemble and then keep adding in as prediction\n- Decision stump - ask one question at a time\n - the reason behind this is: we are decreasing bias by restricting the tree depth to be only 1\n - this causes the bias to increase and decrease the variance \n- Say we make a split and make some misclassifications. \n- we identify those mistakes and increase the weights\n- in the next iteration, it works on the modified sets - because of more weights on misclassfied samples, split might pick this weighted decision boundary \n\n<img src=\"images/10_boosting.png\" width=400 height=400> \n$\\tiny{\\text{YouTube-Stanford-CS229-Andrew Ng/Raphael Townshend}}$ \n\n",
"_____no_output_____"
],
[
"#### Adaboost\n- Determine for classifier $G_{m}$ a weight $\\alpha_{m}$ proportional, which is log odds\n> $log\\left( \\frac{1-err_{m}}{err_{m}}\\right)$\n- Total classifier\n> $G(x) = \\sum\\limits_{m}\\alpha_{m}G_{m}$\n- each $G_{m}$ is trained on re-weighted training set\n\n- Similar mechanism is used to derive algorithm like XGBoost or gradient boosting machines that allow us to reweight the examples we are getting right or wrong in dynamic fashion and then adding them in additive fashion to your model",
"_____no_output_____"
]
]
] |
[
"markdown"
] |
[
[
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown"
]
] |
cbd8ff166384f36ea42f0e8bf84d5bfd0f59e969
| 18,240 |
ipynb
|
Jupyter Notebook
|
train.ipynb
|
okamo-to/FixMatch
|
358a45c6efcde979aad9c609aa1b74e32f145d1a
|
[
"MIT"
] | null | null | null |
train.ipynb
|
okamo-to/FixMatch
|
358a45c6efcde979aad9c609aa1b74e32f145d1a
|
[
"MIT"
] | null | null | null |
train.ipynb
|
okamo-to/FixMatch
|
358a45c6efcde979aad9c609aa1b74e32f145d1a
|
[
"MIT"
] | null | null | null | 47.5 | 905 | 0.516064 |
[
[
[
"import argparse\nimport logging\nimport math\nimport os\nimport random\nimport shutil\nimport time\nfrom collections import OrderedDict\n\nimport numpy as np\nimport torch\nimport torch.nn.functional as F\nimport torch.optim as optim\nfrom torch.optim.lr_scheduler import LambdaLR\nfrom torch.utils.data import DataLoader, RandomSampler, SequentialSampler\nfrom torch.utils.data.distributed import DistributedSampler\nfrom torch.utils.tensorboard import SummaryWriter\nfrom tqdm import tqdm\n\nfrom dataset.custom import DATASET_GETTERS\nfrom utils import AverageMeter, accuracy\n\nlogger = logging.getLogger(__name__)\nbest_acc = 0\n\n\ndef save_checkpoint(state, is_best, checkpoint, filename='checkpoint.pth.tar'):\n filepath = os.path.join(checkpoint, filename)\n torch.save(state, filepath)\n if is_best:\n shutil.copyfile(filepath, os.path.join(checkpoint,\n 'model_best.pth.tar'))\n\n\ndef set_seed(args):\n random.seed(args.seed)\n np.random.seed(args.seed)\n torch.manual_seed(args.seed)\n if args.n_gpu > 0:\n torch.cuda.manual_seed_all(args.seed)\n\n\ndef get_cosine_schedule_with_warmup(optimizer,\n num_warmup_steps,\n num_training_steps,\n num_cycles=7./16.,\n last_epoch=-1):\n def _lr_lambda(current_step):\n if current_step < num_warmup_steps:\n return float(current_step) / float(max(1, num_warmup_steps))\n no_progress = float(current_step - num_warmup_steps) / \\\n float(max(1, num_training_steps - num_warmup_steps))\n return max(0., math.cos(math.pi * num_cycles * no_progress))\n\n return LambdaLR(optimizer, _lr_lambda, last_epoch)\n\n\ndef interleave(x, size):\n s = list(x.shape)\n return x.reshape([-1, size] + s[1:]).transpose(0, 1).reshape([-1] + s[1:])\n\n\ndef de_interleave(x, size):\n s = list(x.shape)\n return x.reshape([size, -1] + s[1:]).transpose(0, 1).reshape([-1] + s[1:])\n\n\ndef main():\n parser = argparse.ArgumentParser(description='PyTorch FixMatch Training')\n parser.add_argument('--gpu-id', default='0', type=int,\n help='id(s) for CUDA_VISIBLE_DEVICES')\n parser.add_argument('--num-workers', type=int, default=4,\n help='number of workers')\n parser.add_argument('--dataset', default='cifar10', type=str,\n choices=['cifar10', 'cifar100'],\n help='dataset name')\n parser.add_argument('--num-labeled', type=int, default=4000,\n help='number of labeled data')\n parser.add_argument(\"--expand-labels\", action=\"store_true\",\n help=\"expand labels to fit eval steps\")\n parser.add_argument('--arch', default='wideresnet', type=str,\n choices=['wideresnet', 'resnext'],\n help='dataset name')\n parser.add_argument('--total-steps', default=2**20, type=int,\n help='number of total steps to run')\n parser.add_argument('--eval-step', default=1024, type=int,\n help='number of eval steps to run')\n parser.add_argument('--start-epoch', default=0, type=int,\n help='manual epoch number (useful on restarts)')\n parser.add_argument('--batch-size', default=64, type=int,\n help='train batchsize')\n parser.add_argument('--lr', '--learning-rate', default=0.03, type=float,\n help='initial learning rate')\n parser.add_argument('--warmup', default=0, type=float,\n help='warmup epochs (unlabeled data based)')\n parser.add_argument('--wdecay', default=5e-4, type=float,\n help='weight decay')\n parser.add_argument('--nesterov', action='store_true', default=True,\n help='use nesterov momentum')\n parser.add_argument('--use-ema', action='store_true', default=True,\n help='use EMA model')\n parser.add_argument('--ema-decay', default=0.999, type=float,\n help='EMA decay rate')\n parser.add_argument('--mu', default=7, type=int,\n help='coefficient of unlabeled batch size')\n parser.add_argument('--lambda-u', default=1, type=float,\n help='coefficient of unlabeled loss')\n parser.add_argument('--T', default=1, type=float,\n help='pseudo label temperature')\n parser.add_argument('--threshold', default=0.95, type=float,\n help='pseudo label threshold')\n parser.add_argument('--out', default='result',\n help='directory to output the result')\n parser.add_argument('--resume', default='', type=str,\n help='path to latest checkpoint (default: none)')\n parser.add_argument('--seed', default=None, type=int,\n help=\"random seed\")\n parser.add_argument(\"--amp\", action=\"store_true\",\n help=\"use 16-bit (mixed) precision through NVIDIA apex AMP\")\n parser.add_argument(\"--opt_level\", type=str, default=\"O1\",\n help=\"apex AMP optimization level selected in ['O0', 'O1', 'O2', and 'O3'].\"\n \"See details at https://nvidia.github.io/apex/amp.html\")\n parser.add_argument(\"--local_rank\", type=int, default=-1,\n help=\"For distributed training: local_rank\")\n parser.add_argument('--no-progress', action='store_true',\n help=\"don't use progress bar\")\n\n args = parser.parse_args()\n global best_acc\n\n def create_model(args):\n if args.arch == 'wideresnet':\n import models.wideresnet as models\n model = models.build_wideresnet(depth=args.model_depth,\n widen_factor=args.model_width,\n dropout=0,\n num_classes=args.num_classes)\n elif args.arch == 'resnext':\n import models.resnext as models\n model = models.build_resnext(cardinality=args.model_cardinality,\n depth=args.model_depth,\n width=args.model_width,\n num_classes=args.num_classes)\n logger.info(\"Total params: {:.2f}M\".format(\n sum(p.numel() for p in model.parameters())/1e6))\n return model\n\n if args.local_rank == -1:\n device = torch.device('cuda', args.gpu_id)\n args.world_size = 1\n args.n_gpu = torch.cuda.device_count()\n else:\n torch.cuda.set_device(args.local_rank)\n device = torch.device('cuda', args.local_rank)\n torch.distributed.init_process_group(backend='nccl')\n args.world_size = torch.distributed.get_world_size()\n args.n_gpu = 1\n\n args.device = device\n\n logging.basicConfig(\n format=\"%(asctime)s - %(levelname)s - %(name)s - %(message)s\",\n datefmt=\"%m/%d/%Y %H:%M:%S\",\n level=logging.INFO if args.local_rank in [-1, 0] else logging.WARN)\n\n logger.warning(\n f\"Process rank: {args.local_rank}, \"\n f\"device: {args.device}, \"\n f\"n_gpu: {args.n_gpu}, \"\n f\"distributed training: {bool(args.local_rank != -1)}, \"\n f\"16-bits training: {args.amp}\",)\n\n logger.info(dict(args._get_kwargs()))\n args.dataset == 'custom'\n args.num_classes = 8\n \n if args.seed is not None:\n set_seed(args)\n\n if args.local_rank in [-1, 0]:\n os.makedirs(args.out, exist_ok=True)\n args.writer = SummaryWriter(args.out)\n\n\n \n if args.arch == 'wideresnet':\n args.model_depth = 28\n args.model_width = 2\n elif args.arch == 'resnext':\n args.model_cardinality = 4\n args.model_depth = 28\n args.model_width = 4\n\n if args.local_rank not in [-1, 0]:\n torch.distributed.barrier()\n\n labeled_dataset, unlabeled_dataset, test_dataset = DATASET_GETTERS[args.dataset](\n args, './data')\n if args.local_rank == 0:\n torch.distributed.barrier()\n\n train_sampler = RandomSampler if args.local_rank == -1 else DistributedSampler\n\n labeled_trainloader = DataLoader(\n labeled_dataset,\n sampler=train_sampler(labeled_dataset),\n batch_size=args.batch_size,\n num_workers=args.num_workers,\n drop_last=True)\n\n unlabeled_trainloader = DataLoader(\n unlabeled_dataset,\n sampler=train_sampler(unlabeled_dataset),\n batch_size=args.batch_size*args.mu,\n num_workers=args.num_workers,\n drop_last=True)\n\n test_loader = DataLoader(\n test_dataset,\n sampler=SequentialSampler(test_dataset),\n batch_size=args.batch_size,\n num_workers=args.num_workers)\n\n if args.local_rank not in [-1, 0]:\n torch.distributed.barrier()\n\n model = create_model(args)\n\n if args.local_rank == 0:\n torch.distributed.barrier()\n\n model.to(args.device)\n\n no_decay = ['bias', 'bn']\n grouped_parameters = [\n {'params': [p for n, p in model.named_parameters() if not any(\n nd in n for nd in no_decay)], 'weight_decay': args.wdecay},\n {'params': [p for n, p in model.named_parameters() if any(\n nd in n for nd in no_decay)], 'weight_decay': 0.0}\n ]\n optimizer = optim.SGD(grouped_parameters, lr=args.lr,\n momentum=0.9, nesterov=args.nesterov)\n\n args.epochs = math.ceil(args.total_steps / args.eval_step)\n scheduler = get_cosine_schedule_with_warmup(\n optimizer, args.warmup, args.total_steps)\n\n if args.use_ema:\n from models.ema import ModelEMA\n ema_model = ModelEMA(args, model, args.ema_decay)\n\n args.start_epoch = 0\n\n if args.resume:\n logger.info(\"==> Resuming from checkpoint..\")\n assert os.path.isfile(\n args.resume), \"Error: no checkpoint directory found!\"\n args.out = os.path.dirname(args.resume)\n checkpoint = torch.load(args.resume)\n best_acc = checkpoint['best_acc']\n args.start_epoch = checkpoint['epoch']\n model.load_state_dict(checkpoint['state_dict'])\n if args.use_ema:\n ema_model.ema.load_state_dict(checkpoint['ema_state_dict'])\n optimizer.load_state_dict(checkpoint['optimizer'])\n scheduler.load_state_dict(checkpoint['scheduler'])\n\n if args.amp:\n from apex import amp\n model, optimizer = amp.initialize(\n model, optimizer, opt_level=args.opt_level)\n\n if args.local_rank != -1:\n model = torch.nn.parallel.DistributedDataParallel(\n model, device_ids=[args.local_rank],\n output_device=args.local_rank, find_unused_parameters=True)\n\n logger.info(\"***** Running training *****\")\n logger.info(f\" Task = {args.dataset}@{args.num_labeled}\")\n logger.info(f\" Num Epochs = {args.epochs}\")\n logger.info(f\" Batch size per GPU = {args.batch_size}\")\n logger.info(\n f\" Total train batch size = {args.batch_size*args.world_size}\")\n logger.info(f\" Total optimization steps = {args.total_steps}\")\n\n model.zero_grad()\n train(args, labeled_trainloader, unlabeled_trainloader, test_loader,\n model, optimizer, ema_model, scheduler)\n \ndef test(args, test_loader, model, epoch):\n batch_time = AverageMeter()\n data_time = AverageMeter()\n losses = AverageMeter()\n top1 = AverageMeter()\n top5 = AverageMeter()\n end = time.time()\n\n if not args.no_progress:\n test_loader = tqdm(test_loader,\n disable=args.local_rank not in [-1, 0])\n\n with torch.no_grad():\n for batch_idx, (inputs, targets) in enumerate(test_loader):\n data_time.update(time.time() - end)\n model.eval()\n\n inputs = inputs.to(args.device)\n targets = targets.to(args.device)\n outputs = model(inputs)\n loss = F.cross_entropy(outputs, targets)\n\n prec1, prec5 = accuracy(outputs, targets, topk=(1, 5))\n losses.update(loss.item(), inputs.shape[0])\n top1.update(prec1.item(), inputs.shape[0])\n top5.update(prec5.item(), inputs.shape[0])\n batch_time.update(time.time() - end)\n end = time.time()\n if not args.no_progress:\n test_loader.set_description(\"Test Iter: {batch:4}/{iter:4}. Data: {data:.3f}s. Batch: {bt:.3f}s. Loss: {loss:.4f}. top1: {top1:.2f}. top5: {top5:.2f}. \".format(\n batch=batch_idx + 1,\n iter=len(test_loader),\n data=data_time.avg,\n bt=batch_time.avg,\n loss=losses.avg,\n top1=top1.avg,\n top5=top5.avg,\n ))\n if not args.no_progress:\n test_loader.close()\n\n logger.info(\"top-1 acc: {:.2f}\".format(top1.avg))\n logger.info(\"top-5 acc: {:.2f}\".format(top5.avg))\n return losses.avg, top1.avg\n\n\nif __name__ == '__main__':\n main()\n",
"_____no_output_____"
]
]
] |
[
"code"
] |
[
[
"code"
]
] |
cbd90780920208be79b7e40f5b9f3eda05afa35d
| 9,673 |
ipynb
|
Jupyter Notebook
|
notebooks/pingutests/owslib.wsp.ipynb
|
Zeitsperre/birdhouse-docs
|
1789c524ea19de38c0ebd920fe9817323e86e565
|
[
"Apache-2.0"
] | null | null | null |
notebooks/pingutests/owslib.wsp.ipynb
|
Zeitsperre/birdhouse-docs
|
1789c524ea19de38c0ebd920fe9817323e86e565
|
[
"Apache-2.0"
] | null | null | null |
notebooks/pingutests/owslib.wsp.ipynb
|
Zeitsperre/birdhouse-docs
|
1789c524ea19de38c0ebd920fe9817323e86e565
|
[
"Apache-2.0"
] | null | null | null | 31.819079 | 318 | 0.563734 |
[
[
[
"from owslib.wps import WebProcessingService, monitorExecution, printInputOutput\n",
"_____no_output_____"
],
[
"wps = WebProcessingService(url=\"http://localhost:8094/wps\", verbose=False)",
"_____no_output_____"
],
[
"print wps.identification.title",
"Emu\n"
],
[
"for process in wps.processes:\n print '%s : \\t %s' % (process.identifier, process.abstract, )",
"helloworld : \t Welcome user and say hello ...\nultimatequestionprocess : \t Numerical solution that is the answer to Life, Universe and Everything. The process is an improvement to Deep Tought computer (therefore version 2.0) since it no longer takes 7.5 milion years, but only a few seconds to give a response, with an update of status every 10 seconds.\ndummyprocess : \t The Dummy process is used for testing the WPS structure. The process will accept 2 input numbers and will return the XML result with an add one and subtract one operation\nwordcount : \t Counts words in a given text ...\ninout : \t Just testing data types like date, datetime etc ...\nmultiplesources : \t Process with multiple different sources ...\nchomsky : \t Generates a random chomsky text ...\nzonal_mean : \t zonal mean in NetCDF File.\n"
],
[
"# get infos about process inout\np = wps.describeprocess(identifier='inout')",
"_____no_output_____"
],
[
"for input in p.dataInputs:\n printInputOutput(input)",
" identifier=int, title=Integer, abstract=This is an Integer, data type=//www.w3.org/TR/xmlschema-2/#integer\n Any value allowed\n Default Value: 10 \n minOccurs=0, maxOccurs=1\n identifier=string, title=String, abstract=This is a String, data type=//www.w3.org/TR/xmlschema-2/#string\n Any value allowed\n Default Value: nothing important \n minOccurs=0, maxOccurs=1\n identifier=float, title=Float, abstract=This is a Float, data type=//www.w3.org/TR/xmlschema-2/#float\n Any value allowed\n Default Value: 3.14 \n minOccurs=0, maxOccurs=1\n identifier=boolean, title=Boolean, abstract=This is a Boolean, data type=//www.w3.org/TR/xmlschema-2/#boolean\n Any value allowed\n Default Value: False \n minOccurs=0, maxOccurs=1\n identifier=date, title=Date, abstract=This is a Date: 2013-07-10, data type=//www.w3.org/TR/xmlschema-2/#string\n Any value allowed\n Default Value: 2013-07-11 \n minOccurs=0, maxOccurs=1\n identifier=stringChoice, title=String Choice, abstract=Choose a string, data type=//www.w3.org/TR/xmlschema-2/#string\n Allowed Value: one\n Allowed Value: two\n Allowed Value: three\n Default Value: one \n minOccurs=0, maxOccurs=3\n identifier=intRequired, title=Integer Required, abstract=This is an required Integer, data type=//www.w3.org/TR/xmlschema-2/#integer\n Any value allowed\n Default Value: None \n minOccurs=1, maxOccurs=1\n identifier=stringMoreThenOne, title=More then One, abstract=This is a more then one String (0-2), data type=//www.w3.org/TR/xmlschema-2/#string\n Any value allowed\n Default Value: None \n minOccurs=0, maxOccurs=2\n identifier=xml_url, title=XML File, abstract=URL of XML File, data type=ComplexData\n Supported Value: mimeType=text/xml, encoding=None, schema=None\n Default Value: mimeType=text/xml, encoding=None, schema=None \n minOccurs=0, maxOccurs=2\n identifier=dummybbox, title=Dummy BBox, abstract=This is a BBox: (minx,miny,maxx,maxy), data type=//www.w3.org/TR/xmlschema-2/#string\n Any value allowed\n Default Value: 0,-90,180,90 \n minOccurs=0, maxOccurs=1\n"
],
[
"for input in p.dataInputs:\n print '**************'\n print 'identifier:', input.identifier\n print 'title:', input.title\n print 'abstract:', input.abstract\n print 'dataType:', input.dataType\n print 'defaultValue:', input.defaultValue\n print 'allowedValues:', input.allowedValues\n print 'supportedValues:', input.supportedValues\n print 'minOccurs:', input.minOccurs\n print 'maxOccurs:', input.maxOccurs",
"**************\nidentifier: int\ntitle: Integer\nabstract: This is an Integer\ndataType: //www.w3.org/TR/xmlschema-2/#integer\ndefaultValue: 10\nallowedValues: []\nsupportedValues: []\nminOccurs: 0\nmaxOccurs: 1\n**************\nidentifier: string\ntitle: String\nabstract: This is a String\ndataType: //www.w3.org/TR/xmlschema-2/#string\ndefaultValue: nothing important\nallowedValues: []\nsupportedValues: []\nminOccurs: 0\nmaxOccurs: 1\n**************\nidentifier: float\ntitle: Float\nabstract: This is a Float\ndataType: //www.w3.org/TR/xmlschema-2/#float\ndefaultValue: 3.14\nallowedValues: []\nsupportedValues: []\nminOccurs: 0\nmaxOccurs: 1\n**************\nidentifier: boolean\ntitle: Boolean\nabstract: This is a Boolean\ndataType: //www.w3.org/TR/xmlschema-2/#boolean\ndefaultValue: False\nallowedValues: []\nsupportedValues: []\nminOccurs: 0\nmaxOccurs: 1\n**************\nidentifier: date\ntitle: Date\nabstract: This is a Date: 2013-07-10\ndataType: //www.w3.org/TR/xmlschema-2/#string\ndefaultValue: 2013-07-11\nallowedValues: []\nsupportedValues: []\nminOccurs: 0\nmaxOccurs: 1\n**************\nidentifier: stringChoice\ntitle: String Choice\nabstract: Choose a string\ndataType: //www.w3.org/TR/xmlschema-2/#string\ndefaultValue: one\nallowedValues: ['one', 'two', 'three']\nsupportedValues: []\nminOccurs: 0\nmaxOccurs: 3\n**************\nidentifier: intRequired\ntitle: Integer Required\nabstract: This is an required Integer\ndataType: //www.w3.org/TR/xmlschema-2/#integer\ndefaultValue: None\nallowedValues: []\nsupportedValues: []\nminOccurs: 1\nmaxOccurs: 1\n**************\nidentifier: stringMoreThenOne\ntitle: More then One\nabstract: This is a more then one String (0-2)\ndataType: //www.w3.org/TR/xmlschema-2/#string\ndefaultValue: None\nallowedValues: []\nsupportedValues: []\nminOccurs: 0\nmaxOccurs: 2\n**************\nidentifier: xml_url\ntitle: XML File\nabstract: URL of XML File\ndataType: ComplexData\ndefaultValue: <owslib.wps.ComplexData object at 0x7f3600b8cbd0>\nallowedValues: []\nsupportedValues: [<owslib.wps.ComplexData object at 0x7f3600b8cc10>]\nminOccurs: 0\nmaxOccurs: 2\n**************\nidentifier: dummybbox\ntitle: Dummy BBox\nabstract: This is a BBox: (minx,miny,maxx,maxy)\ndataType: //www.w3.org/TR/xmlschema-2/#string\ndefaultValue: 0,-90,180,90\nallowedValues: []\nsupportedValues: []\nminOccurs: 0\nmaxOccurs: 1\n"
]
]
] |
[
"code"
] |
[
[
"code",
"code",
"code",
"code",
"code",
"code",
"code"
]
] |
cbd9082326b59f0f4416d533bee3b75de3815675
| 124,655 |
ipynb
|
Jupyter Notebook
|
docs/beta/notebooks/Guide_for_Authors.ipynb
|
BharathMonash/fuzzingbook
|
c734bccc4515b2561f1bfa2313f73d02eede2057
|
[
"MIT"
] | null | null | null |
docs/beta/notebooks/Guide_for_Authors.ipynb
|
BharathMonash/fuzzingbook
|
c734bccc4515b2561f1bfa2313f73d02eede2057
|
[
"MIT"
] | null | null | null |
docs/beta/notebooks/Guide_for_Authors.ipynb
|
BharathMonash/fuzzingbook
|
c734bccc4515b2561f1bfa2313f73d02eede2057
|
[
"MIT"
] | 1 |
2021-01-26T02:30:59.000Z
|
2021-01-26T02:30:59.000Z
| 50.838091 | 42,388 | 0.738582 |
[
[
[
"# Guide for Authors",
"_____no_output_____"
]
],
[
[
"print('Welcome to \"The Fuzzing Book\"!')",
"Welcome to \"The Fuzzing Book\"!\n"
]
],
[
[
"This notebook compiles the most important conventions for all chapters (notebooks) of \"The Fuzzing Book\".",
"_____no_output_____"
],
[
"## Organization of this Book",
"_____no_output_____"
],
[
"### Chapters as Notebooks\n\nEach chapter comes in its own _Jupyter notebook_. A single notebook (= a chapter) should cover the material (text and code, possibly slides) for a 90-minute lecture.\n\nA chapter notebook should be named `Topic.ipynb`, where `Topic` is the topic. `Topic` must be usable as a Python module and should characterize the main contribution. If the main contribution of your chapter is a class `FooFuzzer`, for instance, then your topic (and notebook name) should be `FooFuzzer`, such that users can state\n\n```python\nfrom FooFuzzer import FooFuzzer\n```\n\nSince class and module names should start with uppercase letters, all non-notebook files and folders start with lowercase letters. this may make it easier to differentiate them. The special notebook `index.ipynb` gets converted into the home pages `index.html` (on fuzzingbook.org) and `README.md` (on GitHub).\n\nNotebooks are stored in the `notebooks` folder.",
"_____no_output_____"
],
[
"### Output Formats\n\nThe notebooks by themselves can be used by instructors and students to toy around with. They can edit code (and text) as they like and even run them as a slide show.\n\nThe notebook can be _exported_ to multiple (non-interactive) formats:\n\n* HTML – for placing this material online.\n* PDF – for printing\n* Python – for coding\n* Slides – for presenting\n\nThe included Makefile can generate all of these automatically (and a few more).\n\nAt this point, we mostly focus on HTML and Python, as we want to get these out quickly; but you should also occasionally ensure that your notebooks can (still) be exported into PDF. Other formats (Word, Markdown) are experimental.",
"_____no_output_____"
],
[
"## Sites\n\nAll sources for the book end up on the [Github project page](https://github.com/uds-se/fuzzingbook). This holds the sources (notebooks), utilities (Makefiles), as well as an issue tracker.\n\nThe derived material for the book ends up in the `docs/` folder, from where it is eventually pushed to the [fuzzingbook website](http://www.fuzzingbook.org/). This site allows to read the chapters online, can launch Jupyter notebooks using the binder service, and provides access to code and slide formats. Use `make publish` to create and update the site.",
"_____no_output_____"
],
[
"### The Book PDF\n\nThe book PDF is compiled automatically from the individual notebooks. Each notebook becomes a chapter; references are compiled in the final chapter. Use `make book` to create the book.",
"_____no_output_____"
],
[
"## Creating and Building",
"_____no_output_____"
],
[
"### Tools you will need\n\nTo work on the notebook files, you need the following:\n\n1. Jupyter notebook. The easiest way to install this is via the [Anaconda distribution](https://www.anaconda.com/download/).\n\n2. Once you have the Jupyter notebook installed, you can start editing and coding right away by starting `jupyter notebook` (or `jupyter lab`) in the topmost project folder.\n\n3. If (like me) you don't like the Jupyter Notebook interface, I recommend [Jupyter Lab](https://jupyterlab.readthedocs.io/en/stable/), the designated successor to Jupyter Notebook. Invoke it as `jupyter lab`. It comes with a much more modern interface, but misses autocompletion and a couple of extensions. I am running it [as a Desktop application](http://christopherroach.com/articles/jupyterlab-desktop-app/) which gets rid of all the browser toolbars.\nOn the Mac, there is also the [Pineapple app](https://nwhitehead.github.io/pineapple/), which integrates a nice editor with a local server. This is easy to use, but misses a few features; also, it hasn't seen updates since 2015.\n\n4. To create the entire book (with citations, references, and all), you also need the [ipybublish](https://github.com/chrisjsewell/ipypublish) package. This allows you to create the HTML files, merge multiple chapters into a single PDF or HTML file, create slides, and more. The Makefile provides the essential tools for creation.\n",
"_____no_output_____"
],
[
"### Version Control\n\nWe use git in a single strand of revisions. Feel free branch for features, but eventually merge back into the main \"master\" branch. Sync early; sync often. Only push if everything (\"make all\") builds and passes.\n\nThe Github repo thus will typically reflect work in progress. If you reach a stable milestone, you can push things on the fuzzingbook.org web site, using `make publish`.",
"_____no_output_____"
],
[
"#### nbdime\n\nThe [nbdime](https://github.com/jupyter/nbdime) package gives you tools such as `nbdiff` (and even better, `nbdiff-web`) to compare notebooks against each other; this ensures that cell _contents_ are compared rather than the binary format.\n\n`nbdime config-git --enable` integrates nbdime with git such that `git diff` runs the above tools; merging should also be notebook-specific.",
"_____no_output_____"
],
[
"#### nbstripout\n\nNotebooks in version control _should not contain output cells,_ as these tend to change a lot. (Hey, we're talking random output generation here!) To have output cells automatically stripped during commit, install the [nbstripout](https://github.com/kynan/nbstripout) package and use\n\n```\nnbstripout --install\n```\n\nto set it up as a git filter. The `notebooks/` folder comes with a `.gitattributes` file already set up for `nbstripout`, so you should be all set.\n\nNote that _published_ notebooks (in short, anything under the `docs/` tree _should_ have their output cells included, such that users can download and edit notebooks with pre-rendered output. This folder contains a `.gitattributes` file that should explicitly disable `nbstripout`, but it can't hurt to check.\n\nAs an example, the following cell \n\n1. _should_ have its output included in the [HTML version of this guide](https://www.fuzzingbook.org/beta/html/Guide_for_Authors.html);\n2. _should not_ have its output included in [the git repo](https://github.com/uds-se/fuzzingbook/blob/master/notebooks/Guide_for_Authors.ipynb) (`notebooks/`);\n3. _should_ have its output included in [downloadable and editable notebooks](https://github.com/uds-se/fuzzingbook/blob/master/docs/beta/notebooks/Guide_for_Authors.ipynb) (`docs/notebooks/` and `docs/beta/notebooks/`).",
"_____no_output_____"
]
],
[
[
"import random",
"_____no_output_____"
],
[
"random.random()",
"_____no_output_____"
]
],
[
[
"### Inkscape and GraphViz\n\nCreating derived files uses [Inkscape](https://inkscape.org/en/) and [Graphviz](https://www.graphviz.org/) – through its [Python wrapper](https://pypi.org/project/graphviz/) – to process SVG images. These tools are not automatically installed, but are available on pip, _brew_ and _apt-get_ for all major distributions.",
"_____no_output_____"
],
[
"### LaTeX Fonts\n\nBy default, creating PDF uses XeLaTeX with a couple of special fonts, which you can find in the `fonts/` folder; install these fonts system-wide to make them accessible to XeLaTeX.\n\nYou can also run `make LATEX=pdflatex` to use `pdflatex` and standard LaTeX fonts instead.",
"_____no_output_____"
],
[
"### Creating Derived Formats (HTML, PDF, code, ...)\n\nThe [Makefile](../Makefile) provides rules for all targets. Type `make help` for instructions.\n\nThe Makefile should work with GNU make and a standard Jupyter Notebook installation. To create the multi-chapter book and BibTeX citation support, you need to install the [iPyPublish](https://github.com/chrisjsewell/ipypublish) package (which includes the `nbpublish` command).",
"_____no_output_____"
],
[
"### Creating a New Chapter\n\nTo create a new chapter for the book,\n\n1. Set up a new `.ipynb` notebook file as copy of [Template.ipynb](Template.ipynb).\n2. Include it in the `CHAPTERS` list in the `Makefile`.\n3. Add it to the git repository.",
"_____no_output_____"
],
[
"## Teaching a Topic\n\nEach chapter should be devoted to a central concept and a small set of lessons to be learned. I recommend the following structure:\n\n* Introduce the problem (\"We want to parse inputs\")\n* Illustrate it with some code examples (\"Here's some input I'd like to parse\")\n* Develop a first (possibly quick and dirty) solution (\"A PEG parser is short and often does the job\"_\n* Show that it works and how it works (\"Here's a neat derivation tree. Look how we can use this to mutate and combine expressions!\")\n* Develop a second, more elaborated solution, which should then become the main contribution. (\"Here's a general LR(1) parser that does not require a special grammar format. (You can skip it if you're not interested)\")\n* Offload non-essential extensions to later sections or to exercises. (\"Implement a universal parser, using the Dragon Book\")\n\nThe key idea is that readers should be able to grasp the essentials of the problem and the solution in the beginning of the chapter, and get further into details as they progress through it. Make it easy for readers to be drawn in, providing insights of value quickly. If they are interested to understand how things work, they will get deeper into the topic. If they just want to use the technique (because they may be more interested in later chapters), having them read only the first few examples should be fine for them, too.\n\nWhatever you introduce should be motivated first, and illustrated after. Motivate the code you'll be writing, and use plenty of examples to show what the code just introduced is doing. Remember that readers should have fun interacting with your code and your examples. Show and tell again and again and again.",
"_____no_output_____"
],
[
"### Special Sections",
"_____no_output_____"
],
[
"#### Quizzes",
"_____no_output_____"
],
[
"You can have _quizzes_ as part of the notebook. These are created using the `quiz()` function. Its arguments are\n\n* The question\n* A list of options\n* The correct answer(s) - either\n * the single number of the one single correct answer (starting with 1)\n * a list of numbers of correct answers (multiple choices)\n \nTo make the answer less obvious, you can specify it as a string containing an arithmetic expression evaluating to the desired number(s). The expression will remain in the code (and possibly be shown as hint in the quiz).",
"_____no_output_____"
]
],
[
[
"from bookutils import quiz",
"_____no_output_____"
],
[
"# A single-choice quiz\nquiz(\"The color of the sky is\", ['blue', 'red', 'black'], '5 - 4')",
"_____no_output_____"
],
[
"# A multiple-choice quiz\nquiz(\"What is this book?\", ['Novel', 'Friendly', 'Useful'], ['5 - 4', '1 + 1', '27 / 9'])",
"_____no_output_____"
]
],
[
[
"Cells that contain only the `quiz()` call will not be rendered (but the quiz will).",
"_____no_output_____"
],
[
"#### Synopsis",
"_____no_output_____"
],
[
"Each chapter should have a section named \"Synopsis\" at the very end:\n\n```\n## Synopsis\n\nThis is the text of the synopsis.\n```",
"_____no_output_____"
],
[
"This section is evaluated at the very end of the notebook. It should summarize the most important functionality (classes, methods, etc.) together with examples. In the derived HTML and PDF files, it is rendered at the beginning, such that it can serve as a quick reference",
"_____no_output_____"
],
[
"#### Excursions",
"_____no_output_____"
],
[
"There may be longer stretches of text (and code!) that are too special, too boring, or too repetitve to read. You can mark such stretches as \"Excursions\" by enclosing them in MarkDown cells that state:\n\n```\n#### Excursion: TITLE\n```\n\nand\n\n```\n#### End of Excursion\n```",
"_____no_output_____"
],
[
"Stretches between these two markers get special treatment when rendering:\n\n* In the resulting HTML output, these blocks are set up such that they are shown on demand only.\n* In printed (PDF) versions, they will be replaced by a pointer to the online version.\n* In the resulting slides, they will be omitted right away.",
"_____no_output_____"
],
[
"Here is an example of an excursion:",
"_____no_output_____"
],
[
"#### Excursion: Fine points on Excursion Cells",
"_____no_output_____"
],
[
"Note that the `Excursion` and `End of Excursion` cells must be separate cells; they cannot be merged with others.",
"_____no_output_____"
],
[
"#### End of Excursion",
"_____no_output_____"
],
[
"### Ignored Code\n\nIf a code cell starts with\n```python\n# ignore\n```\nthen the code will not show up in rendered input. Its _output_ will, however. \nThis is useful for cells that create drawings, for instance - the focus should be on the result, not the code.\n\nThis also applies to cells that start with a call to `display()` or `quiz()`.",
"_____no_output_____"
],
[
"### Ignored Cells\n\nYou can have _any_ cell not show up at all (including its output) in any rendered input by adding the following meta-data to the cell:\n```json\n{\n \"ipub\": {\n \"ignore\": true\n}\n```\n",
"_____no_output_____"
],
[
"*This* text, for instance, does not show up in the rendered version.",
"_____no_output_____"
],
[
"## Coding",
"_____no_output_____"
],
[
"### Set up\n\nThe first code block in each notebook should be",
"_____no_output_____"
]
],
[
[
"import bookutils",
"_____no_output_____"
]
],
[
[
"This sets up stuff such that notebooks can import each other's code (see below). This import statement is removed in the exported Python code, as the .py files would import each other directly.",
"_____no_output_____"
],
[
"Importing `bookutils` also sets a fixed _seed_ for random number generation. This way, whenever you execute a notebook from scratch (restarting the kernel), you get the exact same results; these results will also end up in the derived HTML and PDF files. (If you run a notebook or a cell for the second time, you will get more random results.)",
"_____no_output_____"
],
[
"### Coding Style and Consistency\n\nHere's a few rules regarding coding style.",
"_____no_output_____"
],
[
"#### Use Python 3\n\nWe use Python 3 (specifically, Python 3.6) for all code. If you can, try to write code that can be easily backported to Python 2.",
"_____no_output_____"
],
[
"#### Follow Python Coding Conventions\n\nWe use _standard Python coding conventions_ according to [PEP 8](https://www.python.org/dev/peps/pep-0008/).\n\nYour code must pass the `pycodestyle` style checks which you get by invoking `make style`. A very easy way to meet this goal is to invoke `make reformat`, which reformats all code accordingly. The `code prettify` notebook extension also allows you to automatically make your code (mostly) adhere to PEP 8.",
"_____no_output_____"
],
[
"#### One Cell per Definition\n\nUse one cell for each definition or example. During importing, this makes it easier to decide which cells to import (see below).",
"_____no_output_____"
],
[
"#### Identifiers\n\nIn the book, this is how we denote `variables`, `functions()` and `methods()`, `Classes`, `Notebooks`, `variables_and_constants`, `EXPORTED_CONSTANTS`, `files`, `folders/`, and `<grammar-elements>`.",
"_____no_output_____"
],
[
"#### Quotes\n\nIf you have the choice between quoting styles, prefer \n* double quotes (`\"strings\"`) around strings that are used for interpolation or that are natural language messages, and \n* single quotes (`'characters'`) for single characters and formal language symbols that a end user would not see.",
"_____no_output_____"
],
[
"#### Read More\n\nBeyond simple syntactical things, here's a [very nice guide](https://docs.python-guide.org/writing/style/) to get you started writing \"pythonic\" code.",
"_____no_output_____"
],
[
"### Importing Code from Notebooks\n\nTo import the code of individual notebooks, you can import directly from .ipynb notebook files.",
"_____no_output_____"
]
],
[
[
"from Fuzzer import fuzzer",
"_____no_output_____"
],
[
"fuzzer(100, ord('0'), 10)",
"_____no_output_____"
]
],
[
[
"**Important**: When importing a notebook, the module loader will **only** load cells that start with\n\n* a function definition (`def`)\n* a class definition (`class`)\n* a variable definition if all uppercase (`ABC = 123`)\n* `import` and `from` statements\n\nAll other cells are _ignored_ to avoid recomputation of notebooks and clutter of `print()` output.",
"_____no_output_____"
],
[
"Exported Python code will import from the respective .py file instead. The exported Python code is set up such that only the above items will be imported.",
"_____no_output_____"
],
[
"If importing a module prints out something (or has other side effects), that is an error. Use `make check-imports` to check whether your modules import without output.",
"_____no_output_____"
],
[
"Import modules only as you need them, such that you can motivate them well in the text.",
"_____no_output_____"
],
[
"### Imports and Dependencies\n\nTry to depend on as few other notebooks as possible. This will not only ease construction and reconstruction of the code, but also reduce requirements for readers, giving then more flexibility in navigating through the book.\n\nWhen you import a notebook, this will show up as a dependency in the [Sitemap](00_Table_of_Contents.ipynb). If the imported module is not critical for understanding, and thus should not appear as a dependency in the sitemap, mark the import as \"minor dependency\" as follows:",
"_____no_output_____"
]
],
[
[
"from Reducer import DeltaDebuggingReducer # minor dependency",
"_____no_output_____"
]
],
[
[
"### Design and Architecture\n\nStick to simple functions and data types. We want our readers to focus on functionality, not Python. You are encouraged to write in a \"pythonic\" style, making use of elegant Python features such as list comprehensions, sets, and more; however, if you do so, be sure to explain the code such that readers familiar with, say, C or Java can still understand things.",
"_____no_output_____"
],
[
"### Incomplete Examples\n\nWhen introducing examples for students to complete, use the ellipsis `...` to indicate where students should add code, as in here:",
"_____no_output_____"
]
],
[
[
"def student_example():\n x = some_value()\n # Now, do something with x\n ...",
"_____no_output_____"
]
],
[
[
"The ellipsis is legal code in Python 3. (Actually, it is an `Ellipsis` object.)",
"_____no_output_____"
],
[
"### Introducing Classes\n\nDefining _classes_ can be a bit tricky, since all of a class must fit into a single cell. This defeats the incremental style preferred for notebooks. By defining a class _as a subclass of itself_, though, you can avoid this problem.",
"_____no_output_____"
],
[
"Here's an example. We introduce a class `Foo`:",
"_____no_output_____"
]
],
[
[
"class Foo:\n def __init__(self):\n pass\n\n def bar(self):\n pass",
"_____no_output_____"
]
],
[
[
"Now we could discuss what `__init__()` and `bar()` do, or give an example of how to use them:",
"_____no_output_____"
]
],
[
[
"f = Foo()\nf.bar()",
"_____no_output_____"
]
],
[
[
"We now can introduce a new `Foo` method by subclassing from `Foo` into a class which is _also_ called `Foo`:",
"_____no_output_____"
]
],
[
[
"class Foo(Foo):\n def baz(self):\n pass",
"_____no_output_____"
]
],
[
[
"This is the same as if we had subclassed `Foo` into `Foo_1` with `Foo` then becoming an alias for `Foo_1`. The original `Foo` class is overshadowed by the new one:",
"_____no_output_____"
]
],
[
[
"new_f = Foo()\nnew_f.baz()",
"_____no_output_____"
]
],
[
[
"Note, though, that _existing_ objects keep their original class:",
"_____no_output_____"
]
],
[
[
"from ExpectError import ExpectError",
"_____no_output_____"
],
[
"with ExpectError():\n f.baz()",
"Traceback (most recent call last):\n File \"<ipython-input-17-19c0eadbd4d4>\", line 2, in <module>\n f.baz()\nAttributeError: 'Foo' object has no attribute 'baz' (expected)\n"
]
],
[
[
"## Helpers\n\nThere's a couple of notebooks with helpful functions, including [Timer](Timer.ipynb), [ExpectError and ExpectTimeout](ExpectError.ipynb). Also check out the [Coverage](Coverage.ipynb) class.",
"_____no_output_____"
],
[
"### Quality Assurance\n\nIn your code, make use of plenty of assertions that allow to catch errors quickly. These assertions also help your readers understand the code.",
"_____no_output_____"
],
[
"### Issue Tracker\n\nThe [Github project page](https://github.com/uds-se/fuzzingbook) allows to enter and track issues.",
"_____no_output_____"
],
[
"## Writing Text\n\nText blocks use Markdown syntax. [Here is a handy guide](https://github.com/adam-p/markdown-here/wiki/Markdown-Cheatsheet).\n",
"_____no_output_____"
],
[
"### Sections\n\nAny chapter notebook must begin with `# TITLE`, and sections and subsections should then follow by `## SECTION` and `### SUBSECTION`.\n\nSections should start with their own block, to facilitate cross-referencing.\n",
"_____no_output_____"
],
[
"### Highlighting\n\nUse\n\n* _emphasis_ (`_emphasis_`) for highlighting,\n* *emphasis* (`*emphasis*`) for highlighting terms that will go into the index,\n* `backticks` for code and other verbatim elements.",
"_____no_output_____"
],
[
"### Hyphens and Dashes\n\nUse \"–\" for em-dashes, \"-\" for hyphens, and \"$-$\" for minus.",
"_____no_output_____"
],
[
"### Quotes\n\nUse standard typewriter quotes (`\"quoted string\"`) for quoted text. The PDF version will automatically convert these to \"smart\" (e.g. left and right) quotes.",
"_____no_output_____"
],
[
"### Lists and Enumerations\n\nYou can use bulleted lists:\n\n* Item A\n* Item B\n\nand enumerations:\n\n1. item 1\n1. item 2\n\nFor description lists, use a combination of bulleted lists and highlights:\n\n* **PDF** is great for reading offline\n* **HTML** is great for reading online\n",
"_____no_output_____"
],
[
"\n### Math\n\nLaTeX math formatting works, too.\n\n`$x = \\sum_{n = 1}^{\\infty}\\frac{1}{n}$` gets you\n$x = \\sum_{n = 1}^{\\infty}\\frac{1}{n}$.\n",
"_____no_output_____"
],
[
"### Inline Code\n\nPython code normally goes into its own cells, but you can also have it in the text:\n\n```python\ns = \"Python syntax highlighting\"\nprint(s)\n```",
"_____no_output_____"
],
[
"### Images\n\nTo insert images, use Markdown syntax `{width=100%}` inserts a picture from the `PICS` folder.",
"_____no_output_____"
],
[
"{width=100%}",
"_____no_output_____"
],
[
"All pictures go to `PICS/`, both in source as well as derived formats; both are stored in git, too. (Not all of us have all tools to recreate diagrams, etc.)",
"_____no_output_____"
],
[
"### Footnotes\n\nMarkdown supports footnotes, as in [^footnote]. These are rendered as footnotes in HTML and PDF, _but not within Jupyter_; hence, readers may find them confusing. So far, the book makes no use of footnotes, and uses parenthesized text instead.\n\n[^footnote]: Test, [Link](https://www.fuzzingbook.org).",
"_____no_output_____"
],
[
"### Floating Elements and References\n\n\\todo[inline]{I haven't gotten this to work yet -- AZ}\n\nTo produce floating elements in LaTeX and PDF, edit the metadata of the cell which contains it. (In the Jupyter Notebook Toolbar go to View -> Cell Toolbar -> Edit Metadata and a button will appear above each cell.) This allows you to control placement and create labels.",
"_____no_output_____"
],
[
"#### Floating Figures\n\nEdit metadata as follows:\n\n```json\n{\n\"ipub\": {\n \"figure\": {\n \"caption\": \"Figure caption.\",\n \"label\": \"fig:flabel\",\n \"placement\": \"H\",\n\t\"height\":0.4,\n \"widefigure\": false,\n }\n }\n}\n```\n\n- all tags are optional\n- height/width correspond to the fraction of the page height/width, only one should be used (aspect ratio will be maintained automatically)\n- `placement` is optional and constitutes using a placement arguments for the figure (e.g. \\begin{figure}[H]). See [Positioning_images_and_tables](https://www.sharelatex.com/learn/Positioning_images_and_tables).\n- `widefigure` is optional and constitutes expanding the figure to the page width (i.e. \\begin{figure*}) (placement arguments will then be ignored)",
"_____no_output_____"
],
[
"#### Floating Tables\n\nFor **tables** (e.g. those output by `pandas`), enter in cell metadata:\n\n```json\n{\n\"ipub\": {\n \"table\": {\n\t \"caption\": \"Table caption.\",\n\t \"label\": \"tbl:tlabel\",\n\t \"placement\": \"H\",\n \"alternate\": \"gray!20\"\n\t }\n }\n}\n```\n\n- `caption` and `label` are optional\n- `placement` is optional and constitutes using a placement arguments for the table (e.g. \\begin{table}[H]). See [Positioning_images_and_tables](https://www.sharelatex.com/learn/Positioning_images_and_tables).\n- `alternate` is optional and constitutes using alternating colors for the table rows (e.g. \\rowcolors{2}{gray!25}{white}). See (https://tex.stackexchange.com/a/5365/107738)[https://tex.stackexchange.com/a/5365/107738].\n- if tables exceed the text width, in latex, they will be shrunk to fit \n\n\n#### Floating Equations\n\nFor **equations** (e.g. those output by `sympy`), enter in cell metadata:\n\n```json\n{\n \"ipub\": {\n\t \"equation\": {\n \"environment\": \"equation\",\n\t \"label\": \"eqn:elabel\"\n\t }\n }\n}\n```\n\n- environment is optional and can be 'none' or any of those available in [amsmath](https://www.sharelatex.com/learn/Aligning_equations_with_amsmath); 'equation', 'align','multline','gather', or their \\* variants. Additionally, 'breqn' or 'breqn\\*' will select the experimental [breqn](https://ctan.org/pkg/breqn) environment to *smart* wrap long equations. \n- label is optional and will only be used if the equation is in an environment\n\n\n#### References\n\nTo reference a floating object, use `\\cref`, e.g. \\cref{eq:texdemo}",
"_____no_output_____"
],
[
"### Cross-Referencing\n\n#### Section References\n\n* To refer to sections in the same notebook, use the header name as anchor, e.g. \n`[Code](#Code)` gives you [Code](#Code). For multi-word titles, replace spaces by hyphens (`-`), as in [Using Notebooks as Modules](#Using-Notebooks-as-Modules).\n\n* To refer to cells (e.g. equations or figures), you can define a label as cell metadata. See [Floating Elements and References](#Floating-Elements-and-References) for details.\n\n* To refer to other notebooks, use a Markdown cross-reference to the notebook file, e.g. [the \"Fuzzing\" chapter](Fuzzer.ipynb). A special script will be run to take care of these links. Reference chapters by name, not by number.",
"_____no_output_____"
],
[
"### Citations\n\nTo cite papers, cite in LaTeX style. The text",
"_____no_output_____"
]
],
[
[
"print(r\"\\cite{Purdom1972}\")",
"\\cite{Purdom1972}\n"
]
],
[
[
"is expanded to \\cite{Purdom1972}, which in HTML and PDF should be a nice reference.\nThe keys refer to BibTeX entries in [fuzzingbook.bib](fuzzingbook.bib). \n* LaTeX/PDF output will have a \"References\" section appended.\n* HTML output will link to the URL field from the BibTeX entry. Be sure it points to the DOI.",
"_____no_output_____"
],
[
"### Todo's\n\n* To mark todo's, use `\\todo{Thing to be done}.` \\todo{Expand this}",
"_____no_output_____"
],
[
"### Tables\n\nTables with fixed contents can be produced using Markdown syntax:\n\n| Tables | Are | Cool |\n| ------ | ---:| ----:|\n| Zebra | 2 | 30 |\n| Gnu | 20 | 400 |\n",
"_____no_output_____"
],
[
"If you want to produce tables from Python data, the `PrettyTable` package (included in the book) allows to [produce tables with LaTeX-style formatting.](http://blog.juliusschulz.de/blog/ultimate-ipython-notebook)",
"_____no_output_____"
]
],
[
[
"from bookutils import PrettyTable as pt",
"_____no_output_____"
],
[
"import numpy as np",
"_____no_output_____"
],
[
"data = np.array([[1, 2, 30], [2, 3, 400]])\npt.PrettyTable(data, [r\"$\\frac{a}{b}$\", r\"$b$\",\n r\"$c$\"], print_latex_longtable=False)",
"_____no_output_____"
]
],
[
[
"### Plots and Data\n\nIt is possible to include plots in notebooks. Here is an example of plotting a function:",
"_____no_output_____"
]
],
[
[
"%matplotlib inline\n\nimport matplotlib.pyplot as plt\n\nx = np.linspace(0, 3 * np.pi, 500)\nplt.plot(x, np.sin(x ** 2))\nplt.title('A simple chirp');",
"_____no_output_____"
]
],
[
[
"And here's an example of plotting data:",
"_____no_output_____"
]
],
[
[
"%matplotlib inline\n\nimport matplotlib.pyplot as plt\ndata = [25, 36, 57]\nplt.plot(data)\nplt.title('Increase in data');",
"_____no_output_____"
]
],
[
[
"Plots are available in all derived versions (HTML, PDF, etc.) Plots with `plotly` are even nicer (and interactive, even in HTML), However, at this point, we cannot export them to PDF, so `matplotlib` it is.",
"_____no_output_____"
],
[
"## Slides\n\nYou can set up the notebooks such that they also can be presented as slides. In the browser, select View -> Cell Toolbar -> Slideshow. You can then select a slide type for each cell:\n\n* `New slide` starts a new slide with the cell (typically, every `## SECTION` in the chapter)\n* `Sub-slide` starts a new sub-slide which you navigate \"down\" to (anything in the section)\n* `Fragment` is a cell that gets revealed after a click (on the same slide)\n* `Skip` is skipped during the slide show (e.g. `import` statements; navigation guides)\n* `Notes` goes into presenter notes\n\nTo create slides, do `make slides`; to view them, change into the `slides/` folder and open the created HTML files. (The `reveal.js` package has to be in the same folder as the slide to be presented.)\n\nThe ability to use slide shows is a compelling argument for teachers and instructors in our audience.",
"_____no_output_____"
],
[
"(Hint: In a slide presentation, type `s` to see presenter notes.)",
"_____no_output_____"
],
[
"## Writing Tools\n\nWhen you're editing in the browser, you may find these extensions helpful:\n\n### Jupyter Notebook\n\n[Jupyter Notebook Extensions](https://github.com/ipython-contrib/jupyter_contrib_nbextensions) is a collection of productivity-enhancing tools (including spellcheckers).\n\nI found these extensions to be particularly useful:\n\n * Spell Checker (while you're editing)\n \n * Table of contents (for quick navigation)\n\n * Code prettify (to produce \"nice\" syntax)\n \n * Codefolding\n \n * Live Markdown Preview (while you're editing)\n\n### Jupyter Lab\n\nExtensions for _Jupyter Lab_ are much less varied and less supported, but things get better. I am running\n\n * [Spell Checker](https://github.com/ijmbarr/jupyterlab_spellchecker)\n \n * [Table of Contents](https://github.com/jupyterlab/jupyterlab-toc)\n \n * [JupyterLab-LSP](https://towardsdatascience.com/jupyterlab-2-0-edd4155ab897) providing code completion, signatures, style checkers, and more.\n ",
"_____no_output_____"
],
[
"## Interaction\n\nIt is possible to include interactive elements in a notebook, as in the following example:",
"_____no_output_____"
],
[
"```python\ntry:\n from ipywidgets import interact, interactive, fixed, interact_manual\n\n x = interact(fuzzer, char_start=(32, 128), char_range=(0, 96))\nexcept ImportError:\n pass\n```",
"_____no_output_____"
],
[
"Note that such elements will be present in the notebook versions only, but not in the HTML and PDF versions, so use them sparingly (if at all). To avoid errors during production of derived files, protect against `ImportError` exceptions as in the above example.",
"_____no_output_____"
],
[
"## Read More\n\nHere is some documentation on the tools we use:\n\n1. [Markdown Cheatsheet](https://github.com/adam-p/markdown-here/wiki/Markdown-Cheatsheet) - general introduction to Markdown\n\n1. [iPyPublish](https://github.com/chrisjsewell/ipypublish) - rich set of tools to create documents with citations and references\n",
"_____no_output_____"
],
[
"\n\n## Alternative Tool Sets\n\nWe don't currently use these, but they are worth learning:\n\n1. [Making Publication-Ready Python Notebooks](http://blog.juliusschulz.de/blog/ultimate-ipython-notebook) - Another tool set on how to produce book chapters from notebooks\n\n1. [Writing academic papers in plain text with Markdown and Jupyter notebook](https://sylvaindeville.net/2015/07/17/writing-academic-papers-in-plain-text-with-markdown-and-jupyter-notebook/) - Alternate ways on how to generate citations\n\n1. [A Jupyter LaTeX template](https://gist.github.com/goerz/d5019bedacf5956bcf03ca8683dc5217#file-revtex-tplx) - How to define a LaTeX template\n\n1. [Boost Your Jupyter Notebook Productivity](https://towardsdatascience.com/jupyter-notebook-hints-1f26b08429ad) - a collection of hints for debugging and profiling Jupyter notebooks\n\n",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown"
] |
[
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown"
]
] |
cbd9141a3312b80e672a57b4ce4d1406545b2028
| 167,362 |
ipynb
|
Jupyter Notebook
|
Appendix.ipynb
|
andrew-dang/whisky_predictions
|
8c7717596d236dce2dcb5fddadc8ec83dfbc6ad1
|
[
"MIT"
] | null | null | null |
Appendix.ipynb
|
andrew-dang/whisky_predictions
|
8c7717596d236dce2dcb5fddadc8ec83dfbc6ad1
|
[
"MIT"
] | null | null | null |
Appendix.ipynb
|
andrew-dang/whisky_predictions
|
8c7717596d236dce2dcb5fddadc8ec83dfbc6ad1
|
[
"MIT"
] | null | null | null | 64.494027 | 490 | 0.494533 |
[
[
[
"# Appendix \nA. Creating custom list of stop words. ",
"_____no_output_____"
]
],
[
[
"# Basic data science packages\nimport pandas as pd \nimport numpy as np\nimport matplotlib.pyplot as plt\n\nimport joblib\nfrom sklearn.feature_extraction import text\n",
"_____no_output_____"
],
[
"# load data \ndata = pd.read_pickle('data/cleaned_data.pkl')\ndata.head()",
"_____no_output_____"
]
],
[
[
"***\n## Adding adjectives and proper nouns to list of stop words\nSome adjectives are positive or negative by definition. These words are probably highly predictive of a high or low rating, but do not tell us anything else about the whisky specifically. Therefore, to learn more about whisky specific language, these words should be ignored when tokenizing. Likewise, the name of the distiller in a review does not tell us anything about the whisky. These too will be added to the list of stopwords. We will start with creating a list of adjectives. ",
"_____no_output_____"
]
],
[
[
"adj_to_remove = [\n 'best', 'good', 'love', 'incredible', 'remarkable', 'excellent', 'stunning', 'great', 'fantastic', 'wonderful',\n 'outstanding', 'superb', 'magnificent', 'exceptional', 'marvelous', 'superior', 'awesome', \n 'bad', 'terrible', 'worst', 'poor', 'unpleasant', 'inferior', 'unsatisfactory', 'inadequate', 'lousy', 'atrocious', \n 'deficient', 'awful'\n]",
"_____no_output_____"
]
],
[
[
"We can inspect the names of the whiskys to see the names of each distiller, and then manually add them to a list",
"_____no_output_____"
]
],
[
[
"with pd.option_context('display.max_rows', None, 'display.max_columns', None):\n print(data['name'].sort_values())",
"935 Big Peat Christmas Edition 2017, 54.1%\n1418 Dewar’s White Label, 40%\n740 Exclusive Malts 2002 (distilled at Miltonduff...\n1058 GlenDronach Cask Strength (batch 5), 55.3%\n1409 J. Mossman Platinum Crown 15 year old, 40%\n69 Lagavulin 12 year old (Diageo Special Release...\n718 Port Askaig 110° Proof, 55%\n722 Teaninich 17 year old (Diageo Special Release...\n538 That Boutique-y Whisky Company Blended Whisky...\n421 A. D. Rattray (distilled at Bowmore), 18 year ...\n669 A.D. Rattray (distilled at Highland Park) 19 y...\n1297 A.D. Rattray (distilled at Littlemill) 22 year...\n43 Aberfeldy 14 year old Single Cask, 58.1%\n1614 Aberfeldy 16 year old, 40%\n1197 Aberfeldy 18 year old, 40%\n181 Aberfeldy Single Cask (Cask No. 5) 16 year old...\n1924 Aberfeldy, 12 year old, 40%\n780 Aberfeldy, 21 year old, 40%\n1612 Aberlour 12 year old, 40%\n982 Aberlour 12 year old, 40%\n1611 Aberlour 12 year old, 43%\n779 Aberlour 16 year old, 40%\n396 Aberlour 16 year old, 43%\n286 Aberlour 18 year old, 43%\n180 Aberlour 18 year old, 43%\n580 Aberlour A’Bunadh (Batch 61), 60.8%\n395 Aberlour a'bunadh, 59.9%\n397 Aberlour a'bunadh, 60%\n1613 Aberlour a’Bunadh (batch 50), 59.6%\n2067 Abhainn Dearg Single Malt Special Edition, 46%\n299 Adelphi (distilled at BenRiach) 1977 34 year o...\n422 Adelphi (distilled at Bowmore) 12 year old 200...\n1454 Adelphi (distilled at Bunnahabhain) 1989 25 ye...\n440 Adelphi (distilled at Clynelish), 1989, 10 yea...\n1364 Adelphi (distilled at Dailuaine) 1983 27 year ...\n1047 Adelphi (distilled at Glen Grant) 1985 25 year...\n1140 Adelphi (distilled at Glen Moray) 27 year old,...\n854 Adelphi (distilled at Glenrothes) 1990 20 year...\n2010 Adelphi (distilled at Glenrothes) 7 year old, ...\n230 Adelphi (distilled at Linkwood) 1984 26 year o...\n1697 Adelphi (distilled at Longmorn) 1985, 49.2%\n1972 Adelphi (distilled at Longmorn) 20 year old 19...\n495 Adelphi (distilled at Macallan) 14 year old 19...\n685 Adelphi (distilled at Macallan) 22 year old, 56%\n500 Adelphi (distilled at Mortlach) 25 year old, 5...\n553 Adelphi (distilled at Mortlach) 26 year old, 5...\n2113 Adelphi (distilled at The Glenlivet) 1978 34 y...\n1149 Adelphi Breath of Islay 13 year old, 56.6%\n1287 Adelphi Fascadale 12 year old Batch 5 (distill...\n377 Ailsa Bay, 48.9%\n381 Alexander Murray & Co. The Monumental 18 year ...\n568 Alexander Murray & Co. The Monumental 30 year ...\n1133 Anchor Bay, 40%\n584 Ar4 Elements of Islay (distilled at Ardbeg), 5...\n182 Ardbeg Airidh Nam Beist, 1990 Vintage, 46%\n183 Ardbeg Alligator, 51.2%\n91 Ardbeg An Oa\n287 Ardbeg Ardbog, 52.1%\n1615 Ardbeg Auriverdes, 49.9%\n5 Ardbeg Corryvreckan, 57.1%\n290 Ardbeg Dark Cove Committee Edition, 55%\n985 Ardbeg Dark Cove, 46.5%\n583 Ardbeg Day, 56.7%\n2085 Ardbeg Galileo, 49%\n292 Ardbeg Grooves, 46%\n782 Ardbeg Kelpie, 46%\n585 Ardbeg Kildalton, 46%\n400 Ardbeg Perpetuum, 47.4%\n582 Ardbeg Rollercoaster, 57.3%\n399 Ardbeg Supernova (2010 Release), 60.1%\n289 Ardbeg Supernova 2015 Release, 54.3%\n288 Ardbeg Supernova Committee bottling 2014, 55%\n581 Ardbeg Supernova, 58.9%\n586 Ardbeg Twenty One, 46%\n291 Ardbeg Twenty Something, 46.3%\n18 Ardbeg Uigeadail, 54.2%\n90 Ardbeg, 10 year old, 46%\n17 Ardbeg, 1974 Vintage, Cask #3145, 49.9%\n44 Ardbeg, 1974 Vintage, Cask #3524, 49.9%\n184 Ardmore 100th Anniversary 12 year old, 40%\n986 Ardmore Traditional Cask 1998\n2086 Ardmore Traditional Cask, 46%\n1787 Ardmore, 25 year old, 51.4%\n401 Ardmore, 30 year old, 53.7%\n1616 Arran 12 year old Cask Strength, 53.6%\n1200 Arran 14 Year Old, 46%\n592 Arran 18 year old, 46%\n593 Arran 18 year old, 46%\n1789 Arran Machrie Moor 5th Edition, 46%\n1617 Arran Machrie Moor Cask Strength (Batch 1), 58.4%\n785 Arran Malt 16 year old, 46%\n590 Arran Malt 17 year old, 46%\n594 Arran Malt Amarone Wine Finish, 50%\n1423 Arran Malt Orkney Bere, 56.2%\n589 Arran Malt The Millennium Casks, 53.5%\n989 Arran Premium Bourbon Single Cask #2096, 52%\n784 Arran Premium Sherry Single Cask #1979, 51.1%\n1201 Arran Sherry Premium Single Cask 1997 #217, 53.2%\n2088 Arran Single Island Malt, Non-chill-filtered, 46%\n1202 Arran Smugglers’ Series Vol. 1 The Illicit Sti...\n991 Arran The Bothy Quarter Cask, 55.7%\n588 Arran The Devil’s Punch Bowl Chapter II: Angel...\n591 Arran The Devil’s Punchbowl III The Fiendish F...\n988 Arran The Eagle 1999, 46%\n987 Arran The Peacock, 1996 Vintage, 46%\n783 Arran ‘The Devil’s Punch Bowl,’ 52.3%\n2020 Auchentoshan 10 year old, 43%\n2168 Auchentoshan 17 year old, Bordeaux Wine Finish...\n2216 Auchentoshan 1962, 41 year old, 40.3%\n786 Auchentoshan 1975, 45.6%\n992 Auchentoshan 1977, 49.0%\n595 Auchentoshan 1979 Oloroso Sherry Matured, 50.5%\n1425 Auchentoshan 1998, 54.6%\n1620 Auchentoshan 1999 Bordeaux Cask Matured, 58%\n293 Auchentoshan 21 year old, 43%\n1926 Auchentoshan 32 year old, 1973 vintage, 55.5%\n1791 Auchentoshan American Oak, 40%\n1431 Auchentoshan Blood Oak, 46%\n993 Auchentoshan Bourbon Matured 1975, 46.9%\n2090 Auchentoshan Classic, 40%\n1430 Auchentoshan Cooper’s Reserve 14 year old, 46%\n1205 Auchentoshan Heartwood, 43%\n994 Auchentoshan Silveroak 1990, 50.9%\n1427 Auchentoshan Springwood, 40%\n2089 Auchentoshan Three Wood, 43%\n1428 Auchentoshan Valinch 2012, 57.2%\n1426 Auchentoshan Valinch, 57.5%\n1622 Auchentoshan Virgin Oak, 46%\n1619 Auchentoshan, 12 year old, 40%\n1790 Auchentoshan, 16 year old, 53.7%\n1204 Auchentoshan, 18 year old, 43%\n403 Auchentoshan, 18 year old, 55.8%\n1424 Auchentoshan, 1966 Vintage, 37 year old, 48.1%\n404 Auchentoshan, 1966, 36 year old, 49.4%\n1203 Auchentoshan, 40 years old, 41.6%\n724 Auchroisk 20 year old, 58.1%\n1573 Auchroisk 25 year old, 51.2%\n922 Auchroisk 30 year old Special Release 2012, 54.7%\n1741 Auld Reekie, 46%\n1592 Aultmore 12 year old, 55%\n1397 Aultmore 21 year old, 46%\n954 Aultmore 25 year old, 46%\n1623 Balblair 16 year old, 40%\n597 Balblair 1969 Vintage, 41.4%\n995 Balblair 1975 Vintage, 46%\n1624 Balblair 1989 Vintage, 43%\n788 Balblair 1989, 43%\n598 Balblair 1990 (2nd release), 46%\n294 Balblair 1995, 46%\n1626 Balblair 1996 Vintage, 46%\n790 Balblair 1997, 46%\n1434 Balblair 1999 (2nd release), 46%\n596 Balblair 2000, 43%\n789 Balblair 2001 Vintage, 46%\n1625 Balblair 2002 Vintage, 46%\n1792 Balblair 2004 Bourbon Matured, 46%\n1433 Balblair 2004 Sherry Matured, 46%\n996 Balblair Vintage 2003, 46%\n997 Balblair Vintage 2005, 46%\n1432 Balblair, 1991 Vintage, 43%\n1210 Balblair, 1991, Duncan Taylor Rare Auld range,...\n787 Balblair, 1997 vintage, 43%\n927 Ballantine's Christmas Reserve 2011, 40%\n1368 Ballantine’s 12 year old, 40%\n928 Ballantine’s 17 year old Signature Distillery ...\n1367 Ballantine’s 17 year old Signature Distillery ...\n2068 Ballantine’s 17 year old Signature Distillery ...\n365 Ballantine’s 17 year old Signature Distillery ...\n729 Ballantine’s 17 year old, 40%\n261 Ballantine’s 21 year old Signature Oak Edition...\n929 Ballantine’s 21 year old, 40%\n728 Ballantine’s 30 year old, 40%\n2069 Ballantine’s Finest, 40%\n529 Ballantine’s Limited, 43%\n95 Balvenie 1961 55 year old, 41.7%\n19 Balvenie 1973 43 year old, 46.6%\n93 Balvenie 1973 Vintage, 30 year old, Cask #9219\n410 Balvenie 1981 35 year old, 43.8%\n999 Balvenie 1993 23 year old, 51.9%\n1214 Balvenie 2004 13 year old, 58.2%\n94 Balvenie 25 year old, 46.9%\n405 Balvenie 30 year old, 47.3%\n792 Balvenie 40 year old (Batch #2), 48.5%\n599 Balvenie Caribbean Rum Cask, 14 year old, 43%\n409 Balvenie DCS Compendium 1st Chapter 1968 46 ye...\n998 Balvenie DCS Compendium 1st Chapter 1978 37 ye...\n1436 Balvenie DCS Compendium 1st Chapter 1985 30 ye...\n1628 Balvenie DCS Compendium 1st Chapter 1997 17 ye...\n2092 Balvenie DCS Compendium 1st Chapter 2005 9 yea...\n1213 Balvenie DoubleWood 17 year old, 43%\n1927 Balvenie Madeira Cask 21 year old, 40%\n794 Balvenie Peat Week 14 year old, 48.3%\n1435 Balvenie Peated Cask, 17 year old, 43%\n297 Balvenie PortWood 21 year old, 40%\n1627 Balvenie Single Barrel Sherry Cask 15 year old...\n793 Balvenie Single Barrel Traditional Oak 25 year...\n185 Balvenie Tun 1401 (Batch #2), 50.6%\n407 Balvenie Tun 1401 Batch 5, 50.1%\n408 Balvenie Tun 1401 Batch 9, 49.3%\n46 Balvenie Tun 1401 Batch No. 3, 50.3%\n298 Balvenie Tun 1509 (Batch 3), 52.2%\n296 Balvenie Tun 1509, 47.1%\n92 Balvenie, Cask #17893, 1974 vintage, 52.8%\n406 Balvenie, Madeira Cask, 17 year old, 43%\n795 Ben Nevis 15 year old 1998 (Cask #590), 57.3%\n1928 Ben Nevis 1998, Duncan Taylor NC2 range, 46%\n1215 Ben Nevis, 10 year old, 46%\n1004 BenRiach 'Curiositas,' 10 year old, 46%\n1007 BenRiach 10 year old Curiositas, 46%\n797 BenRiach 10 year old, 43%\n603 BenRiach 22 year old Moscatel Wood Finish, 46%\n415 BenRiach 25 year old, 46.8%\n1799 BenRiach 35 year old, 42.5%\n2093 BenRiach 40 year old, 50%\n600 BenRiach Authenticus 21 year old, 46%\n796 BenRiach Authenticus 25 year old, 46%\n414 BenRiach Authenticus Peated 30 year old, 46%\n602 BenRiach Cask Strength (Batch 1), 57.2%\n605 BenRiach Cask Strength (Batch 2), 60.6%\n604 BenRiach Four Cask Matured 21 year old, 46%\n2095 BenRiach Horizons 12 year old, 50%\n1632 BenRiach Peated Quarter Casks, 46%\n798 BenRiach Pedro Ximenez Sherry Wood Finish 15 y...\n186 BenRiach Pedro Ximinez Finish, (Cask #7165), 1...\n413 BenRiach Septendecim 17 year old, 46%\n601 BenRiach Solstice 15 year old, 50%\n2096 BenRiach Solstice 17 year old 2nd batch, 50%\n1006 BenRiach Tawny Port Wood Finish 21 year old, 46%\n1003 BenRiach, 12 year old, 1994 vintage, 56%\n411 BenRiach, 25 year old, 45%\n1630 BenRiach, 30 year old, 45%\n412 BenRiach, Cask #4469, 1976 vintage, 30 year ol...\n1438 BenRiach, Maderensis Fumosis (Madeira Wood Fin...\n1798 BenRiach, Sauternes Wood Finish, 16 year old, 46%\n1631 Benriach, 12 year old, 1994 vintage, 59.5%\n1634 Benrinnes 21 year old (Diageo Special Release ...\n1441 Benromach 15 year old, 43%\n1008 Benromach 15 year old, 43%\n187 Benromach 1974 (cask 1583), 49.1%\n300 Benromach 1975, 49.9%\n301 Benromach 1977, 56%\n416 Benromach 30 year old, 43%\n188 Benromach Heritage 1973, 48.9%\n801 Benromach Heritage Collection 35 year old, 43%\n1637 Benromach Hermitage 2007, 45%\n1010 Benromach Organic 2017 (distilled 2010), 43%\n2233 Benromach Organic, 43%\n1636 Benromach Origins 12 year old, Batch 2, 50%\n2237 Benromach Origins, Batch #1, 1999 vintage, 50%\n1930 Benromach Peat Smoke, 46%\n1440 Benromach Port Wood Finish 22 year old, 45%\n1009 Benromach Sassicaia Wood Finish 2009, 45%\n2022 Benromach Traditional, 40%\n417 Benromach Triple Distilled 2009, 50%\n800 Benromach, 10 year old, 43%\n2023 Benromach, 18year old, 40%\n799 Benromach, 24 year old, 58.6%\n1635 Benromach, 25 year old, 43%\n2217 Benromach, Tokaji Finish, 45%\n1931 Benromach, cask #613, 1999 vintage, 58.6%\n1811 Berry Bros. & Rudd (distilled at Bunnahabhain)...\n2061 Berry Bros. & Rudd (distilled at Glen Moray) 2...\n1841 Berry Bros. & Rudd (distilled at Imperial) 17 ...\n1401 Berry Bros. & Rudd Exceptional Casks (distille...\n524 Berry Bros. & Rudd, distilled at Balmenach, 19...\n2033 Berry Brothers & Rudd (distilled at Cragganmor...\n448 Berry Brothers & Rudd (distilled at Glen Grant...\n687 Berry Brothers & Rudd (distilled at Mortlach) ...\n960 Berrys’ Islay Reserve, 46%\n865 Berry’s Own Selection (distilled at Jura) 1976...\n154 Big Peat 25 year old, 52.1%\n534 Big Peat Christmas Edition 2014, 55.7%\n266 Big Peat Christmas Edition 2015, 53.8%\n265 Big Peat Small Batch, 53.6%\n87 Black Bottle, 10 year old, 43%\n1786 Black Bottle, 40%\n1 Black Bowmore, 1964 vintage, 42 year old, 40.5%\n1730 Black Bull 12 year old, 50%\n913 Black Bull 21 year old, 50%\n75 Black Bull 40 year old 7th release, 47.6%\n1350 Black Bull 8 year old, 50%\n1132 Black Bull Kyloe, 50%\n719 Black Bull, 12 year old, 50%\n74 Black Bull, 30 year old, 50%\n256 Black Bull, 40 year old, 40.2%\n918 Black Grouse Alpha Edition, 40%\n1805 Blackadder 'Raw Cask' (distilled at Bruichladd...\n610 Blackadder (distilled at Bowmore), 27 years ol...\n871 Blackadder (distilled at Laphroaig), 12 year o...\n2035 Blackadder Raw Cask #6437 (distilled at Glen G...\n1041 Blackadder Raw Cask (distilled at Edradour), 1...\n1299 Blackadder Single Cask #1101 (distilled at Lon...\n873 Blackadder Single Cask #30013 (distilled at Li...\n1220 Bladnoch 1993, Gordon & MacPhail Connoisseurs’...\n1442 Bladnoch 9 year old, 46%\n1012 Bladnoch Adela 15 year old, 46.7%\n1222 Bladnoch Samsara, 46.7%\n804 Bladnoch Talia 25 year old, 49.2%\n418 Blair Athol 23 year old (Diageo Special Releas...\n1639 Blair Athol, 55.8%\n1369 Blue Hanger 10th Blended Malt, 45.6%\n264 Blue Hanger 11th Limited Release, 45.6%\n262 Blue Hanger 4th Release Berry Bros. & Rudd, 45.6%\n76 Blue Hanger 6th Release Berry Bros. & Rudd, 45.6%\n263 Blue Hanger 7th Release Limited Edition, 45.6%\n1370 Blue Hanger 8th Release, 45.6%\n530 Blue Hanger 9th Release, 45.6%\n1812 Bn6 Elements of Islay (distilled at Bunnahabha...\n2163 Borders, 51.7%\n1801 Bowmore 'Wine Cask Matured', 1992 vintage, 16 ...\n49 Bowmore 1966 50 year old, 41.5%\n1449 Bowmore 1999 PX Hand-bottled, 56.1%\n1225 Bowmore 23 year old Port Wood, 50.8%\n2 Bowmore 46 year old (distilled 1964), 42.9%\n20 Bowmore 50 year old (distilled 1961), 40.7%\n1450 Bowmore 9 year old, 40%\n1445 Bowmore Bourbon Cask, 1964 Vintage, 43%\n609 Bowmore Claret (56%)\n420 Bowmore Darkest, 15 year old, 43%\n1015 Bowmore Devil’s Cask Batch 2, 56.3%\n423 Bowmore Devil’s Casks 10 year old, 56.9%\n607 Bowmore Fino Cask, 1964 Vintage, 49.4%\n1643 Bowmore Fèis Ìle 2016, 54.9%\n424 Bowmore Laimrig 15 year old, 53.7%\n1016 Bowmore Mizunara Cask Finish, 53.9%\n189 Bowmore Oloroso Sherry Cask, 1964 Vintage\n805 Bowmore Tempest (Second Release), 10 year old,...\n2098 Bowmore Tempest Batch 3 (10 year old), 55.6%\n1224 Bowmore Tempest IV/Dorus Mor, 55.1%\n1642 Bowmore The Devil’s Casks III, 56.7%\n305 Bowmore Vault Edition Atlantic Sea Salt (First...\n1226 Bowmore Vintage Fèis Ìle 2016 Edition 25 year ...\n608 Bowmore Voyage, 56%\n96 Bowmore, 16 year old, 1989 vintage, 51.8%\n2198 Bowmore, 16 year old, 1990 vintage, 53.8%\n303 Bowmore, 18 year old, 43%\n352 Bowmore, 18 year old, 43%\n1013 Bowmore, 1991 Vintage, 16 year old, 53.1%\n302 Bowmore, 25 year old, 43%\n419 Bowmore, 34 year old, 1971 vintage, 51%\n1444 Bowmore, 37 year old, 1968 vintage, 43.4%\n7 Bowmore, 40 year old, 44.8%\n1230 Br5 Elements of Islay (distilled at Bruichladd...\n52 Brora 1978 35 year old (Diageo Special Release...\n24 Brora 2010, 54.3%\n100 Brora 34 year old (Diageo Special Releases 201...\n50 Brora 35 year old Special Release 2012, 48.1%\n25 Brora 35 year old, 49.9%\n98 Brora 37 year old (Diageo Special Releases 201...\n99 Brora 38 year old, 48.6%\n51 Brora 40 year old Single Cask 1972 Vintage, 59.1%\n306 Brora, 25 year old, 56.3%\n23 Brora, 30 year old (2009 Release), 53.2%\n21 Brora, 30 year old, 55.7%\n190 Bruichladdich 'PC-5', 63.5%\n1935 Bruichladdich 'Peat', 46%\n1648 Bruichladdich 'Rocks', 46%\n2199 Bruichladdich 'Waves', 46%\n755 Bruichladdich (Port Charlotte) PC7, 61%\n430 Bruichladdich 10 year old, 46%\n426 Bruichladdich 15 year old, 2nd Edition, 46%\n428 Bruichladdich 1973 vintage, 40.2% ABV\n102 Bruichladdich 1984, 43.7%\n191 Bruichladdich 1985, 48.7%\n308 Bruichladdich 1986, 44.6%\n1804 Bruichladdich 20 year old 'Second Edition', 46%\n1803 Bruichladdich 20 year old, 3rd Edition, 46%\n1231 Bruichladdich 2001 15 year old (2016-1881-135-...\n427 Bruichladdich 3D, The Big Peat, 50%\n615 Bruichladdich Bere Barley 2008, 50%\n1936 Bruichladdich Bere Barley 2009, 50%\n307 Bruichladdich Black Art 05.1 1992, 48.4%\n614 Bruichladdich Black Art 4 23 year old, 49.2%\n1646 Bruichladdich Full Strength, 1989 vintage, 13 ...\n1017 Bruichladdich Infinity, 55.5%\n431 Bruichladdich Islay Barley 2007 6 year old, 50%\n433 Bruichladdich Islay Barley 2010, 50%\n1649 Bruichladdich Islay Barley 5 year old, 50%\n2027 Bruichladdich Laddie Eight 8 year old, 50%\n53 Bruichladdich Legacy 6, 34 year old, 41%\n101 Bruichladdich Legacy III, 35 year old, 1968 vi...\n1934 Bruichladdich Legacy Two, 1965 vintage, 41.8% ABV\n1645 Bruichladdich Links, 14 year old, 46% ABV\n1808 Bruichladdich Links, Torrey Pine, 15 year old,...\n1810 Bruichladdich Micro Provenance 2007 (Cask 14),...\n2017 Bruichladdich Octomore (Edition 01.1), 5 years...\n1921 Bruichladdich Octomore 07.4 Virgin Oak, 61.2%\n760 Bruichladdich Octomore 08.1, 59.3%\n566 Bruichladdich Octomore 08.3, 61.2%\n565 Bruichladdich Octomore 10 Second Limited Editi...\n1777 Bruichladdich Octomore 6.1 5 year old, 57%\n1406 Bruichladdich Octomore 7.1, 59.1%\n1604 Bruichladdich Octomore 7.2, 58.5%\n964 Bruichladdich Octomore 7.3, 63%\n2164 Bruichladdich Octomore ‘Comus’ 4.2 2007 5 year...\n756 Bruichladdich Port Charlotte PC12, 58.7%\n1184 Bruichladdich Port Charlotte Scottish Barley, 50%\n1229 Bruichladdich Scottish Barley, 50%\n1020 Bruichladdich Sherry Edition, Manzanilla, 1998...\n1227 Bruichladdich Sherry Edition, Oloroso, 1998 vi...\n616 Bruichladdich The Laddie Ten Second Limited Ed...\n1644 Bruichladdich WMD II, 46%\n1806 Bruichladdich, 12 year old, 2nd Edition, 46%\n1807 Bruichladdich, 12 year old, 46%\n2170 Bruichladdich, 16 year old, 1989 vintage, 50%\n2026 Bruichladdich, 18 year old, 1986 vintage, 57.5%\n1019 Bruichladdich, PC6, 61.6%\n563 Bruichladdich, Port Charlotte “PC-7,” Cask #12...\n1605 Buchanan’s Deluxe 12 year old, 40%\n1606 Buchanan’s Master, 40%\n380 Buchanan’s Red Seal, 40%\n762 Buchanan’s Special Reserve 18 year old, 40%\n1021 Bunnahabhain 12 year old, 43%\n192 Bunnahabhain 18 year old, 43%\n435 Bunnahabhain 25 year old, 43%\n1023 Bunnahabhain 40 year old, 41.7%\n437 Bunnahabhain 46 year old Eich Bhana Lìr, 41.8%\n1651 Bunnahabhain Amontillado 16 year old, 54%\n1455 Bunnahabhain Ceobanach, 46.3%\n1233 Bunnahabhain Cruach-Mhóna, 50%\n2101 Bunnahabhain Moine Dram an Stiùireadair 10 yea...\n2102 Bunnahabhain Moine PX 12 year old, 52.7%\n811 Bunnahabhain Pedro Ximénez Cask Finish 15 year...\n810 Bunnahabhain Toiteach, 46%\n1232 Bunnahabhain Westering Home 17 year old Feis I...\n2100 Bunnahabhain ‘Over 12 year old’ Feis Ile 2012 ...\n618 Bunnahabhain, 12 year old, 46.3%,\n54 Bunnahabhain, 35 year old, 1971 vintage, 44.9%\n1448 Bw3 Elements of Islay (distilled at Bowmore), ...\n611 Cadenhead\n1768 Cadenhead Authentic Collection 28 year old (di...\n1638 Cadenhead's, (distilled at Blair Athol), 1989,...\n628 Cadenhead's, (distilled at Dalmore), 1987, 59.9%\n1272 Cadenhead's, (distilled at Glenlivet), 12 year...\n875 Cadenhead's, (distilled at Littlemill), 1989, ...\n1310 Cadenhead's, (distilled at Rosebank), 11 year ...\n1572 Cadenhead’s (distilled at\\t Auchroisk) 12 year...\n2014 Cadenhead’s (distilled at BenRiach) 1996, 46.5%\n1933 Cadenhead’s (distilled at Bowmore) 11 year old...\n1748 Cadenhead’s (distilled at Braes of Glenlivet) ...\n1597 Cadenhead’s (distilled at Bunnahabhain) 2005, ...\n1769 Cadenhead’s (distilled at Caol Ila) 1984, 50.7%\n1026 Cadenhead’s (distilled at Caol Ila) 22 year ol...\n2105 Cadenhead’s (distilled at Caol Ila) 29 year ol...\n1403 Cadenhead’s (distilled at Caol lla) 2000, 54.4%\n2161 Cadenhead’s (distilled at Dailuaine) 1996, 57.5%\n1821 Cadenhead’s (distilled at Glen Grant) 15 year ...\n1254 Cadenhead’s (distilled at Glendullan) 17 year ...\n1770 Cadenhead’s (distilled at Glentauchers) 1989, ...\n374 Cadenhead’s (distilled at Linkwood) 1989, 51.1%\n1695 Cadenhead’s (distilled at Littlemill) 36 year ...\n562 Caledonian ‘The Cally’ 1974 40 year old, 53.3%\n730 Cambus 40 year old, 52.7%\n816 Caol Ila 'Unpeated Style,' 10 year old (2009 R...\n438 Caol Ila 12 year old Feis Ile bottling 2012, 6...\n623 Caol Ila 12 year old, 43%\n818 Caol Ila 12 year old, 57.6%\n1234 Caol Ila 14 year old Special Release 2012, 59.3%\n1235 Caol Ila 15 year old, 61.5%\n819 Caol Ila 18 year old (Diageo Special Releases ...\n1027 Caol Ila 2002 Feis Ile 2014 bottling, 55.5%\n814 Caol Ila 25 year old, 59.4%\n439 Caol Ila 30 year old (Diageo Special Release 2...\n2104 Caol Ila Distillers Edition 1998, 43%\n310 Caol Ila Feis Ile Bottling 2013, 56.5%\n2103 Caol Ila Moch, 43%\n624 Caol Ila Stitchell Reserve, 59.6%\n2030 Caol Ila Unpeated 12 year old Special Release ...\n1028 Caol Ila Unpeated 15 year old (Diageo Special ...\n625 Caol Ila Unpeated 1997 17 year old (Diageo Spe...\n812 Caol Ila, 12 year old, 43%\n2028 Caol Ila, 18 year old, 43%\n817 Caol Ila, 25 Year Old, 43%\n1652 Caol Ila, 58.4%\n1410 Cardhu 12 year old, 45%\n1608 Cardhu 18 year old, 40%\n971 Cardhu 21 year old, 54.2%\n1922 Cardhu Amber Rock, 40%\n1781 Cardhu, 12 year old, 40%\n767 Cardhu, 1997 vintage, 57.3%\n969 Carlyle, 40%\n1002 Carn Mor (distilled at Ben Nevis) 1997, 46%\n983 Cask & Thistle (distilled at Aberfeldy), 29 ye...\n1663 Cask & Thistle (distilled at Glen Grant), 15 y...\n1832 Cask & Thistle (distilled at Glenlivet), 30 ye...\n1964 Cask & Thistle (distilled at Imperial), 28 yea...\n537 Chapter 7 (undisclosed Highland distillery), 5...\n967 Chapter 7 2008 (distilled at Allt-a-Bhainne) 9...\n968 Chapter 7 2008 (distilled at Aultmore) 9 year ...\n1228 Chieftain’s (distilled at Bruichladdich) 22 ye...\n1887 Chieftain’s (distilled at Glen Moray), 1989 vi...\n1079 Chieftain’s (distilled at Glenturret) 21 year ...\n1309 Chieftain’s (distilled at Port Ellen), 1982 vi...\n242 Chieftain’s (distilled at Springbank), 40 year...\n891 Chieftain’s Choice (distilled at Rosebank) 20 ...\n1195 Chivas Regal 12 year old, 40%\n177 Chivas Regal 18 year old Ultimate Cask Collect...\n42 Chivas Regal 18 year old Ultimate Cask Collect...\n176 Chivas Regal 18 year old, 40%\n777 Chivas Regal 25 year old, 40%\n1415 Chivas Regal Extra, 40%, \n1194 Chivas Regal The Chivas Brothers’ Blend 12 yea...\n4 Chivas Regal Ultis, 40%\n16 Chivas, 18 year old, 40%\n980 Chivas, 25 year old, 40%\n1653 Ci6 Elements of Islay (distilled at Caol Ila),...\n2193 Clan Denny (distilled at Girvan) 1992 21 year ...\n2070 Clan Denny (distilled at Girvan) 20 year old, ...\n2148 Clan Denny (distilled at Girvan) 21 year old 1...\n1148 Clan Denny (distilled at Port Dundas) 1992 21 ...\n933 Clan Denny (distilled at Strathclyde) 27 year ...\n531 Clan Denny Islay, 46%\n731 Clan Denny Speyside, 46%\n2225 Clansman, 40%\n778 Classic Cask 23 year old Caribbean Rum Barrels...\n981 Classic Cask 23 year old Oloroso Sherry Butt, 43%\n285 Classic Cask 23 year old Original Cask, 43%\n578 Classic Cask 23 year old Port Pipe, 43%\n1462 Clynelish 14 year old, 43%\n1030 Clynelish 1993, Gordon & MacPhail Connoisseurs...\n1461 Clynelish Distillers Edition 1992, 46%\n1236 Clynelish Distillers Edition 1993, 43%\n822 Clynelish Select Reserve (Diageo Special Relea...\n823 Clynelish Select Reserve (Diageo Special Relea...\n1029 Clynelish, 14 year old, 46%\n312 Clynelish, 57.3%\n79 Collectivum XXVIII (Diageo Special Releases 20...\n277 Compass Box 'Lady Luck', 46%\n1609 Compass Box 'The Spice Tree', 46%\n278 Compass Box 3 year old Deluxe, 49.2%\n572 Compass Box Asyla, 40%\n769 Compass Box Asyla, 43%\n386 Compass Box Delilah's, 40%\n768 Compass Box Eleuthera, 46%\n972 Compass Box Enlightenment, 46%\n13 Compass Box Flaming Heart (10th Anniversary bo...\n167 Compass Box Flaming Heart 2012 Edition, 48.9%\n81 Compass Box Flaming Heart 2015 Edition 15th An...\n280 Compass Box Great King Street Artist's Blend, 43%\n974 Compass Box Great King Street Experimental OO-...\n1411 Compass Box Great King Street Experimental TR-...\n770 Compass Box Great King Street Glasgow Blend, 43%\n169 Compass Box Great King Street New York Blend, 45%\n388 Compass Box Hedonism Maximus\n173 Compass Box Hedonism Quindecimus, 46%\n975 Compass Box Hedonism, 43%\n976 Compass Box Last Vatted Grain, 46%\n383 Compass Box Last Vatted Malt, 53.7%\n165 Compass Box Monster, 54.9%\n172 Compass Box No Name, 48.9%\n1782 Compass Box Oak Cross, 43%\n385 Compass Box Oak Cross, 43%\n80 Compass Box Oak Cross, 46%\n168 Compass Box Peat Monster 10th Anniversary Limi...\n39 Compass Box Phenomenology, 46%\n82 Compass Box Spice Tree Extravaganza, 46%\n279 Compass Box Spice Tree, 46%\n171 Compass Box The Circus, 49%\n83 Compass Box The Double Single 2017 Release, 46%\n973 Compass Box The Entertainer, 46%\n3 Compass Box The General, 53.4%\n384 Compass Box The Lost Blend, 46%\n14 Compass Box The Peat Monster 10th Anniversary ...\n571 Compass Box The Peat Monster Cask Strength Mag...\n38 Compass Box The Peat Monster, 46%\n387 Compass Box The Peat Monster, 46%\n170 Compass Box This Is Not A Luxury Whisky, 53.1%\n276 Compass Box, Canto Cask 46, 53.2%\n164 Compass Box, Flaming Heart, Batch #2, 48.9%\n166 Compass Box, The Peat Monster, Reserve Edition...\n1974 Connoisseurs Choice (distilled at Mannochmore)...\n151 Convalmore 1977 28 year old, 57.9%\n923 Convalmore 32 year old (Diageo Special Release...\n528 Convalmore 36 year old, 58%\n1407 Copper Dog, 40%\n441 Cragganmore 10 year old, 60.1%\n1034 Cragganmore 12 year old, 43%\n2219 Cragganmore 21 year old, 56%\n313 Cragganmore 25 year old (Diageo Special Releas...\n626 Cragganmore Distillers Edition 1997, 43%\n1033 Cragganmore Distillers Edition, 1992 vintage, 40%\n105 Cragganmore Triple Matured Edition, 48%\n104 Cragganmore, 29 year old, 52.9%\n627 Cragganmore, 55.7%\n548 Craigellachie 13 year old, 46%\n1168 Craigellachie 17 year old, 46%\n745 Craigellachie 19 year old, 46%\n371 Craigellachie 23 year old, 46%\n746 Craigellachie 31 year old, 52.2%\n1520 Creative Whisky Co. Exclusive Malts (distilled...\n1713 Cuatro Series Fino, 46%\n1331 Cuatro Series Manzanilla, 46%\n1542 Cuatro Series Oloroso, 46%\n1332 Cuatro Series Pedro Ximenez, 46%\n713 Cutty Sark 18 year old, 43%\n520 Cutty Sark Art Deco Limited Edition 33 year ol...\n2136 Cutty Sark Blended Malt, 40%\n714 Cutty Sark Cask Strength, 51.4%\n519 Cutty Sark Prohibition, 50%\n712 Cutty Sark Storm, 40%\n360 Cutty Sark Tam o' Shanter 25 year old, 46.5%\n148 Cutty Sark, 25 year old, 45.7%\n1544 Cù Bòcan 2005 Vintage, 50%\n1543 Cù Bòcan Limited Edition Virgin Oak, 46%\n1330 Cù Bòcan Sherry Cask, 46%\n241 D&M (Distilled at Scapa), 19 year old, 1989 Vi...\n926 Dailuaine 16 year old Flora & Fauna, 43%\n152 Dailuaine 1980 34 year old (Diageo Special Rel...\n106 Dalmore 1973 Vintage Gonzalez Byass Sherry Cas...\n1939 Dalmore 1995 The Distillery Manager’s Exclusiv...\n828 Dalmore 30 year old, 45%\n195 Dalmore 35 year old, 40%\n197 Dalmore 40 year old, 42%\n109 Dalmore 50 year old, 40%\n444 Dalmore Ceti 30 year old, 45%\n827 Dalmore Distillery Exclusive 2015, 48%\n318 Dalmore Port Collection Vintage 1996, 45%\n196 Dalmore Quintessence, 45%\n1941 Dalmore Valour, 40%\n1242 Dalwhinnie 15 year old, 43%\n1244 Dalwhinnie 25 year old (Diageo Special Release...\n1243 Dalwhinnie 25 year old Special Release 2012, 5...\n1039 Dalwhinnie Distillers Edition 1992, 43%\n1040 Dalwhinnie Distillers Edition 1997, 43%\n1658 Dalwhinnie Distillers Edition, 1990 vintage, 43%\n1659 Dalwhinnie Winter’s Gold, 43%\n1038 Dalwhinnie, 29 year old, 57.8%\n2080 Darkness! (distilled at Invergordon) 23 year o...\n1595 Darkness! (distilled at North British) 18 year...\n1817 Deanston 18 year old, 46.3%\n1944 Deanston 20 year old Oloroso, 55.3%\n446 Deanston 40 year old, 45.6%\n2034 Deanston Organic 15 year old, 46.3%\n1943 Deanston Virgin Oak, 46.3%\n1942 Deanston, 12 year old, 46.3%\n1816 Deanston, 30 year old, 46.7%\n1886 Deerstalker, 18 year old, 46%\n2162 Deveron 10 year old, 40%\n2082 Deveron 18 year old, 40%\n283 Dewar's Founder's Reserve, 18 year old, 40%\n178 Dewar's Signature, 43%\n576 Dewar's Special Reserve, 12 year old, 43%\n577 Dewar’s 12 year old The Ancestor, 40%\n393 Dewar’s 15 year old The Monarch, 40%\n1417 Dewar’s 18 year old The Vintage, 40%\n89 Dewar’s 25 year old, 40%\n284 Dewar’s 30 year old Ne Plus Ultra, 40%\n1416 Dewar’s White Label Scratched Cask, 40%\n85 Diamond Jubilee by John Walker & Sons, 42.5%\n2241 Distillery Select 'Craiglodge' (distilled at L...\n1298 Distillery Select 'Craiglodge' (distilled at L...\n2238 Distillery Select 'Croftengea' (distilled at L...\n1970 Distillery Select 'Croftengea' (distilled at L...\n2244 Distillery Select 'Inchmoan' (distilled at Loc...\n587 Double Barrel Ardbeg & Craigellachie, 46%\n2149 Double Barrel Caol Ila/Tamdhu, 46%\n1896 Double Barrel Laphroaig/Mortlach, 46%\n934 Double Barrel Ledaig/Bowmore, 46%\n1740 Double Barrel Talisker/Craigellachie, 46%\n1206 Douglas Laing (distilled at Auchentoshan) 11 y...\n1371 Douglas Laing (distilled at Glen Mhor) 30 year...\n1499 Douglas Laing (distilled at Glenturret) 18 yea...\n1382 Douglas Laing 21 year old (distilled at Highla...\n153 Douglas Laing Big Peat Christmas 2013, 54.9%\n533 Douglas Laing Big Peat, 46%\n943 Douglas Laing Directors’ Cut (distilled at Ros...\n1570 Douglas Laing Director’s Cut (distilled at Tea...\n1117 Douglas Laing Director’s Cut (distilled at Tom...\n370 Douglas Laing Extra Old Particular (distilled ...\n1694 Douglas Laing Old & Rare (distilled at Littlem...\n1502 Douglas Laing Old Malt Cask (distilled at High...\n1535 Douglas Laing Old Malt Cask (distilled at Rose...\n1312 Douglas Laing Old Malt Cask (distilled at Roya...\n1848 Douglas Laing Old Malt Cask series (distilled ...\n1223 Douglas Laing Old Particular (distilled at Bla...\n2025 Douglas Laing Old Particular (distilled at Bow...\n1900 Douglas Laing Old Particular (distilled at Bow...\n2151 Douglas Laing Old Particular (distilled at Bun...\n1381 Douglas Laing Old Particular (distilled at Bun...\n268 Douglas Laing Old Particular (distilled at Cam...\n1405 Douglas Laing Old Particular (distilled at Cam...\n158 Douglas Laing Old Particular (distilled at Car...\n1750 Douglas Laing Old Particular (distilled at Cra...\n1584 Douglas Laing Old Particular (distilled at Dal...\n1901 Douglas Laing Old Particular (distilled at Dal...\n1751 Douglas Laing Old Particular (distilled at Dea...\n1772 Douglas Laing Old Particular (distilled at Gir...\n1385 Douglas Laing Old Particular (distilled at Gir...\n1156 Douglas Laing Old Particular (distilled at Gle...\n2001 Douglas Laing Old Particular (distilled at Gle...\n1155 Douglas Laing Old Particular (distilled at Gle...\n2074 Douglas Laing Old Particular (distilled at Gle...\n541 Douglas Laing Old Particular (distilled at Gle...\n1157 Douglas Laing Old Particular (distilled at Gle...\n1379 Douglas Laing Old Particular (distilled at Hig...\n1904 Douglas Laing Old Particular (distilled at Inv...\n872 Douglas Laing Old Particular (distilled at Lap...\n2003 Douglas Laing Old Particular (distilled at Mac...\n944 Douglas Laing Old Particular (distilled at Mil...\n542 Douglas Laing Old Particular (distilled at Mil...\n2083 Douglas Laing Old Particular (distilled at Nor...\n1773 Douglas Laing Old Particular (distilled at Nor...\n2226 Douglas Laing Old Particular (distilled at Por...\n1386 Douglas Laing Old Particular (distilled at Por...\n2208 Douglas Laing Old Particular (distilled at Str...\n1774 Douglas Laing Old Particular (distilled at Str...\n754 Douglas Laing Old Particular (distilled at Str...\n2152 Douglas Laing Old Particular (distilled at Tam...\n1903 Douglas Laing Old Particular 10 year old (Leda...\n1383 Douglas Laing Old Particular 1988 (distilled a...\n1384 Douglas Laing Old Particular 1995 (distilled a...\n1688 Douglas Laing Old Particular 1996 (distilled a...\n1052 Douglas Laing Old Particular 21 year old (dist...\n1325 Douglas Laing Old Particular Ledaig 1993 (dist...\n1360 Douglas Laing Old Particular Range (distilled ...\n2154 Douglas Laing Premier Barrel (distilled at Gle...\n2155 Douglas Laing Premier Barrel (distilled at Tal...\n1902 Douglas Laing Provenanace (distilled at Longmo...\n1422 Douglas Laing Provenance (distilled at Arran) ...\n1443 Douglas Laing Provenance (distilled at Blair A...\n1752 Douglas Laing Provenance (distilled at Caol Il...\n1585 Douglas Laing Provenance (distilled at Dailuai...\n2002 Douglas Laing Provenance (distilled at Fetterc...\n1483 Douglas Laing Provenance (distilled at Glen Ga...\n540 Douglas Laing Provenance (distilled at Glenrot...\n1380 Douglas Laing Provenance (distilled at Laphroa...\n2153 Douglas Laing Provenance (distilled at Mortlac...\n2132 Douglas Laing Provenance (distilled at Taliske...\n2000 Douglas Laing Provenance 8 year old (distilled...\n2131 Douglas Laing Provenance Young & Feisty (disti...\n736 Douglas Laing Queen of the Hebrides (distilled...\n945 Douglas Laing Rock Oyster Cask Strength, 57.4%\n1159 Douglas Laing Scallywag Cask Strength (batch 2...\n2075 Douglas Laing Scallywag, 46%\n2194 Douglas Laing Single Minded (distilled at Jura...\n2004 Douglas Laing Single Minded Speyside 15 year o...\n1753 Douglas Laing Single Minded Speyside 24 year o...\n1754 Douglas Laing Xtra Old Particular (distilled a...\n1183 Douglas Laing Xtra Old Particular (distilled a...\n543 Douglas Laing Xtra Old Particular (distilled a...\n622 Douglas of Drumlanrig (distilled at Caol Ila),...\n1456 Dun Bheagan (distilled at Caol Ila) 1993 Vinta...\n1540 Dun Bheagan (distilled at Springbank), Cask #5...\n33 Dun Bheagan (distilled at Springbank), Cask No...\n1795 Duncan Taylor 'Rarest of the Rare' (distilled ...\n1080 Duncan Taylor 'Rarest of the Rare' (distilled ...\n1685 Duncan Taylor 'Rarest of the Rare' (distilled ...\n1846 Duncan Taylor 'Rarest of the Rare' (distilled ...\n2127 Duncan Taylor 'Rarest of the Rare' (distilled ...\n634 Duncan Taylor 'Rarest of the Rare' Glencraig (...\n1196 Duncan Taylor (distilled at Aberlour) Dimensio...\n1621 Duncan Taylor (distilled at Auchentoshan) 13 y...\n1593 Duncan Taylor (distilled at Aultmore) 6 year o...\n1217 Duncan Taylor (distilled at Benriach), 34 year...\n1640 Duncan Taylor (distilled at Blair Athol) 22 ye...\n1446 Duncan Taylor (distilled at Bowmore), 21 year ...\n304 Duncan Taylor (distilled at Bowmore), Cask #85...\n429 Duncan Taylor (distilled at Bruichladdich), 34...\n1916 Duncan Taylor (distilled at Bunnahabhain) 25 y...\n1022 Duncan Taylor (distilled at Bunnahabhain) Rare...\n1453 Duncan Taylor (distilled at Bunnahabhain), 199...\n434 Duncan Taylor (distilled at Bunnahabhain), 36 ...\n617 Duncan Taylor (distilled at Bunnahabhain), cas...\n2240 Duncan Taylor (distilled at Cameronbridge), Ca...\n1237 Duncan Taylor (distilled at Clynelish) 1990, 5...\n1819 Duncan Taylor (distilled at Glen Grant) 1972 V...\n199 Duncan Taylor (distilled at Glen Grant), Cask ...\n1756 Duncan Taylor (distilled at Glenallachie) 6 ye...\n468 Duncan Taylor (distilled at Glenlivet) 1968 Vi...\n208 Duncan Taylor (distilled at Glenlivet), 1987 v...\n653 Duncan Taylor (distilled at Glenlivet), cask #...\n1280 Duncan Taylor (distilled at Glentauchers) 8 ye...\n1181 Duncan Taylor (distilled at Imperial) 1995, 48%\n37 Duncan Taylor (distilled at Invergordon), 40 y...\n2018 Duncan Taylor (distilled at Invergordon), Cask...\n1853 Duncan Taylor (distilled at Longmorn) 18 year ...\n1524 Duncan Taylor (distilled at Longmorn), 29 year...\n710 Duncan Taylor (distilled at Mosstowie), cask #...\n364 Duncan Taylor (distilled at Royal Lochnagar) 2...\n506 Duncan Taylor (distilled at Springbank) 13 yea...\n143 Duncan Taylor (distilled at Strathisla), 35 ye...\n244 Duncan Taylor (distilled at Strathisla), 36 ye...\n514 Duncan Taylor (distilled at Tamdhu), Cask # 73...\n1113 Duncan Taylor (distilled at Tamnavulin), 37 ye...\n961 Duncan Taylor 12 year old, 40%\n1599 Duncan Taylor 18 year old, 40%\n682 Duncan Taylor Dimensions Range (distilled at L...\n2126 Duncan Taylor Dimensions Range (distilled at M...\n907 Duncan Taylor Dimensions Range (distilled at T...\n1600 Duncan Taylor Five Star, 40%\n1354 Duncan Taylor NC2 (distilled at Balmenach) 200...\n1050 Duncan Taylor NC2 (distilled at Glen Scotia), ...\n1503 Duncan Taylor NC2 (distilled at Imperial) 1997...\n1598 Duncan Taylor Octave (distilled at Glen Grant)...\n478 Duncan Taylor Peerless (distilled at Glenrothe...\n2166 Duncan Taylor Rare Auld (distilled at Aberlour...\n273 Duncan Taylor Rare Auld (distilled at Girvan) ...\n2205 Duncan Taylor Rare Auld (distilled at Glen Mor...\n532 Duncan Taylor The Big Smoke 60, 60%\n1182 Duncan Taylor The Big Smoke, 46%\n179 Duncan Taylor “NC2” (distilled at Aberlour), 1...\n2081 Duncan Taylor, Octave (distilled at Aultmore) ...\n1307 Duncansby Head, 46%\n1873 Eades Highland 'Double Malt', 46%\n1987 Eades Islay 'Double Malt', 46%\n1986 Eades Speyside 'Double Malt', 46%\n1553 Eades “Double Malt” Highland (Second Edition),...\n1123 Eades “Double Malt” Islay (Second Edition), 46%\n1660 Edradour 'Ballechin (#4),' 46%\n1043 Edradour 10 year old, 40%\n1466 Edradour 1999 Natural Cask Strength cask #295,...\n1469 Edradour 9 year old 2006 Barolo Cask Matured (...\n1246 Edradour Ballechin #5, 46%,\n1467 Edradour Ballechin 10 year old, 46%\n1662 Edradour Ballechin Sauternes Cask Matured, 46%\n2242 Edradour Barolo Finish, 11 year old, 57.1%\n2231 Edradour Bordeaux Finish, 11 year old, 55.8%\n1042 Edradour Caledonia, 12 year old, 46%, \n1245 Edradour Madeira Finish, 11 year old, 59.6%\n1661 Edradour Port Cask, 6th Release, cask #383, 46%\n198 Edradour Port Wood Finish, 1983 Vintage, Cask ...\n1468 Edradour The Fairy Flag 15 year old, 46%\n1945 Edradour, cask #06/538/4, 1996 vintage, 11 yea...\n1650 Elements of Islay Br4 (distilled at Bruichladd...\n2099 Elements of Islay Bw1 (distilled at Bowmore), ...\n808 Elements of Islay Pl1 (distilled at Bruichladd...\n269 Exclusive Blend 1991 21 year old, 46%\n160 Exclusive Blend 1993 23 year old, 50%\n741 Exclusive Malts (Distilled at Benrinnes) 2006 ...\n738 Exclusive Malts (distilled at Ardmore) 14 year...\n1164 Exclusive Malts (distilled at Benrinnes) 20 ye...\n806 Exclusive Malts (distilled at Bowmore) 2001, 5...\n809 Exclusive Malts (distilled at Bunnahabhain) 26...\n270 Exclusive Malts (distilled at Cameronbridge) 1...\n948 Exclusive Malts (distilled at Clynelish) 21 ye...\n1389 Exclusive Malts (distilled at Dailuaine) 2007 ...\n1940 Exclusive Malts (distilled at Dalmore) 13 year...\n1387 Exclusive Malts (distilled at Dalmore) 18 year...\n546 Exclusive Malts (distilled at Glen Elgin) 21 y...\n947 Exclusive Malts (distilled at Glen Moray) 9 ye...\n1279 Exclusive Malts (distilled at Glenrothes) 1996...\n1166 Exclusive Malts (distilled at Highland Park) 2...\n1165 Exclusive Malts (distilled at Highland Park) 2...\n161 Exclusive Malts (distilled at Invergordon) 43 ...\n679 Exclusive Malts (distilled at Linkwood) 14 yea...\n232 Exclusive Malts (distilled at Longmorn) 28 yea...\n950 Exclusive Malts (distilled at Macduff) 2006 11...\n742 Exclusive Malts (distilled at Royal Brackla) 2...\n1823 Exclusive Malts 1991 (distilled at Glencadam),...\n1932 Exclusive Malts 1992 (distilled at Bladnoch), ...\n1674 Exclusive Malts 1993 (distilled at Glen Garioc...\n949 Exclusive Malts 2005 Ledaig (distilled at Tobe...\n1906 Exclusive Malts 2005 Ledaig (distilled at Tobe...\n547 Exclusive Malts 2006 (distilled at Linkwood) 1...\n1160 Exclusive Malts North Highland 17 year old 199...\n1905 Exclusive Malts Speyside 10 year old 2003 Cask...\n1161 Exclusive Malts Speyside 10 year old 2004, 57.7%\n159 Exclusive Malts Speyside 25 year old 1989 Cask...\n763 Exclusive Regions Islay (undisclosed Islay sin...\n1778 Exclusive Regions Single Grain (distilled at N...\n1413 Famous Grouse Mellow Gold, 40%\n2062 Fettercairn Fasque, 42%\n527 Fettercairn Fior, 42%\n2210 Girvan No. 4 Apps, 42%\n952 Girvan Patent Still 25 year old, 42%\n750 Girvan Patent Still 30 year old, 42%\n1175 Girvan Patent Still Proof Strength, 57.1%\n1395 Glen Deveron 16 year old, 40%\n1590 Glen Deveron 20 year old, 40%\n551 Glen Deveron 30 year old, 40%\n1045 Glen Elgin 12 year old, 43%\n631 Glen Elgin 18 year old (Diageo Special Release...\n1044 Glen Elgin, 1998 vintage, 61.1%\n1067 Glen Garioch 15 year old The Renaissance Chapt...\n1486 Glen Garioch 1797 Founder’s Reserve, 48%\n458 Glen Garioch 1986, 54.6%\n643 Glen Garioch 1991, 54.7%\n1481 Glen Garioch 1994 Vintage, 53.9%\n459 Glen Garioch Cask #992 14 year old 1998, 54.6%\n644 Glen Garioch Cask No. 986 13 year old, 55%\n1485 Glen Garioch Single Cask #990 1998, 56.1%\n204 Glen Garioch The Renaissance 3rd Chapter 17 ye...\n1065 Glen Garioch Vintage 1995, 55.3%\n1482 Glen Garioch Vintage 1997, 56.7%\n645 Glen Garioch Vintage 1999, 56.3%\n1484 Glen Garioch Virgin Oak, 48%\n1829 Glen Garioch Wine Cask Matured 1998, 48%\n1264 Glen Garioch Wine Cask Matured, 48%\n1263 Glen Garioch, 1991, Duncan Taylor Rare Auld ra...\n1666 Glen Grant 12 year old, 43%\n1947 Glen Grant 170th Anniversary, 48%\n319 Glen Grant 25 year old, 43%\n1048 Glen Grant 50 year old, 54.4%\n449 Glen Grant Five Decades, 46%\n2036 Glen Grant, 10 year old, 43%\n1471 Glen Grant, 16 year old, 43%\n2060 Glen Moray 12 year old, 40%\n917 Glen Moray 12 year old, 40%\n526 Glen Moray 15 year old, 40%\n525 Glen Moray 18 year old, 47.2%\n150 Glen Moray 25 year old Port Cask Finish, 43%\n1355 Glen Moray Elgin Classic Chardonnay Cask Finis...\n1356 Glen Moray Elgin Classic Peated, 40%\n1571 Glen Moray Elgin Classic Port Cask Finish, 40%\n1357 Glen Moray Elgin Classic Sherry Cask Finish, 40%\n1141 Glen Moray Elgin Classic, 40%\n2144 Glen Moray Elgin Classic, 40%\n259 Glen Moray Mastery, 52.3%\n2143 Glen Moray Peated Spirit Batch #1 Cask 141, 60.6%\n1667 Glen Ord, Milroy’s of Soho Single Cask, 1999, 46%\n1473 Glen Scotia 10 year old, 46%\n1474 Glen Scotia 12 year old, 46%\n1248 Glen Scotia 15 year old, 46%\n1051 Glen Scotia 16 year old, 46%\n1247 Glen Scotia 18 year old, 46%\n321 Glen Scotia 18 year old, 46%\n633 Glen Scotia 21 year old, 46%\n320 Glen Scotia 25 year old, 48.8%\n1476 Glen Scotia Distillery Edition, 56.9%\n1668 Glen Scotia Double Cask, 46%\n1475 Glen Scotia Heavily Peated 10 year old, 50%\n1053 Glen Scotia Victoriana, 51.5%\n1948 Glen Scotia, 12 year old, 43%\n2173 Glen Spey 21 year old, 50.4%\n1766 Glen Turner Heritage Double Wood, 40%\n832 GlenDronach 14 year old Virgin Oak, 46%\n111 GlenDronach 21 year old Parliament, 48%\n323 GlenDronach 25 year old Grandeur (batch 7), 50.6%\n451 GlenDronach Cask Strength (Batch 6), 56.1%\n112 GlenDronach Grandeur 24 year old (Batch 9), 48.7%\n57 GlenDronach Grandeur, 31 year old, 45.8%\n452 GlenDronach Kingsman Edition 1991, 48.2%\n635 GlenDronach Peated Port Wood, 46%\n1253 GlenDronach Peated, 46%\n1057 GlenDronach Sauternes Finish, 46%, 14 year old\n324 GlenDronach Single Cask #5476, 49.9%\n1251 GlenDronach, 12 year old, 46%\n110 GlenDronach, 15 year old, 46%\n1670 GlenDronach, 18 year old, 46%\n1825 GlenDronach, 1971 vintage, 39 year old, oloros...\n1950 GlenDronach, 1978 vintage, 31 year old, oloros...\n831 GlenDronach, 1989 vintage, 19 year old, 58.7%\n1826 GlenDronach, 1989 vintage, 20 year old, pedro ...\n450 GlenDronach, 1990 vintage, 20 year old, oloros...\n1252 GlenDronach, 1991 vintage, 18 year old, sherry...\n2111 GlenDronach, 1993 vintage, 17 year old, oloros...\n2109 Glenburgie 15 year old, 54.5%\n322 Glencadam 25 year old The Remarkable, 46%\n1055 Glencadam 30 year old, 46%\n1250 Glencadam Origin 1825, 40%\n1249 Glencadam The Exceptional 21 year old, 46%\n2110 Glencadam The Rather Delicate 10 year old, 46%\n1822 Glencadam The Rather Enriched Oloroso Finish 1...\n1949 Glencadam The Rather Refined Port Wood Finish ...\n2037 Glencadam, 15 year old, 40%\n1056 Glendronach, 12 year old, 40% ABV\n56 Glendronach, 33 year old, 40%\n1479 Glenfarclas 105, 60%\n456 Glenfarclas 17 year old, 43%\n326 Glenfarclas 175th Anniversary, 43%\n114 Glenfarclas 1953, 47.2%\n26 Glenfarclas 1968 Vintage, 43% ABV\n636 Glenfarclas 50 year old, 1955 vintage, 44.4%\n457 Glenfarclas 50 year old, 41.1%\n2174 Glenfarclas Family Cask 1961 (Cask 1325), 47%\n2038 Glenfarclas Family Cask 1962 (Cask 2639), 48.9%\n1255 Glenfarclas Family Cask 1964 (Cask 4719), 48.5%\n327 Glenfarclas Family Cask 1970 (Cask 140), 57.1%\n453 Glenfarclas Family Cask 1971 (Cask 3546), 51.1%\n113 Glenfarclas Family Cask 1972 (Cask 3551), 44.7%\n1059 Glenfarclas Family Cask 1981 (Cask 57), 50.8%\n637 Glenfarclas Family Cask 1982 (Cask 4567), 56%\n1060 Glenfarclas Family Cask 1984 (Cask 6030), 51%\n454 Glenfarclas Family Cask 1996 (Cask 1306), 55.6%\n9 Glenfarclas Family Casks 1954 Cask #1260, 47.2%\n1061 Glenfarclas Family Casks 1969 Cask #2545, 57.5%\n638 Glenfarclas Family Casks 1974 Cask #8579, 57.2%\n1256 Glenfarclas Family Casks 1978 Cask #4004, 41.3%\n1671 Glenfarclas Family Casks 1983 Cask #31, 46.1%\n1257 Glenfarclas Family Casks 1984 Cask #6032, 47%\n833 Glenfarclas Family Casks 1985 Cask #2591, 45.4%\n58 Glenfarclas Family Casks 1987 Cask #3829, 48%\n201 Glenfarclas Family Casks 1988 Cask #434, 53.4%\n455 Glenfarclas Family Casks 1990 Cask #1362, 51.4%\n639 Glenfarclas £511. 19s. 0d.\n1478 Glenfarclas, (Cask #128), 1981 Vintage, 27 yea...\n325 Glenfarclas, 1974 vintage, 57.4%\n27 Glenfarclas, 40 year old, 46%\n1258 Glenfiddich 12 year old, 40%\n2039 Glenfiddich 125th Anniversary, 43%\n1262 Glenfiddich 15 year old, 40%\n1261 Glenfiddich 26 year old, 43.1%\n641 Glenfiddich Cask of Dreams, 48.8%\n1673 Glenfiddich IPA Experiment, 43%\n1062 Glenfiddich Over 38 year old, 40%\n1064 Glenfiddich Project XX, 47%\n203 Glenfiddich Rare Oak 25 year old, 43%\n1259 Glenfiddich Reserve Cask, 40%\n1260 Glenfiddich Select Cask, 40%\n1063 Glenfiddich Vintage Cask, 40%\n330 Glenfiddich Winter Storm 21 year old, 43%\n115 Glenfiddich, 1973 Vintage, Cask No. 9874, 46.5%\n640 Glenfiddich, 21 year old, Gran Reserva Caribbe...\n202 Glenfiddich, 50 year old, 46.1%\n1672 Glenfiddich, Age of Discovery, 19 year old, 40%\n329 Glenfiddich, Cask #4414, 1977 Vintage, 54.1%\n1828 Glenfiddich, Snow Phoenix, 47.6%\n1480 Glengarioch, 1979 Vintage, Cask #6489, 52.1%\n642 Glengarioch, 21 year old, 43%\n331 Glenglassaugh 30 year old, 44.8%\n461 Glenglassaugh 35 Year Old The Chosen Few, 49.6%\n205 Glenglassaugh 37 year old, 56%\n462 Glenglassaugh 40 year old, 42.5%\n1267 Glenglassaugh Aged 37 Years, 56%\n1068 Glenglassaugh Aged Over 30 Years, 43%\n1675 Glenglassaugh Evolution, 57.2%\n834 Glenglassaugh Massandra Connection 35 year old...\n646 Glenglassaugh Massandra Connection 41 year old...\n1676 Glenglassaugh Octaves Classic, 44%\n1952 Glenglassaugh Octaves Peated, 44%\n835 Glenglassaugh Peated Port Wood Finish, 46%\n1951 Glenglassaugh Revival, 46%\n1487 Glenglassaugh Torfa, 40%\n460 Glenglassaugh, 26 year old, 46%\n1266 Glenglassaugh, 30 year old, 43%\n28 Glenglassaugh, 40 year old, 44.6%\n1830 Glenglassaugh, 44 year old, 1960 vintage, 41.7%\n1489 Glengoyne 10 year old, 40%\n1270 Glengoyne 15 year old, 43%\n836 Glengoyne 17 year old, 43%\n837 Glengoyne 18 year old, 43%\n1677 Glengoyne 1996 Vintage, 43%\n464 Glengoyne 25 year old, 48%\n647 Glengoyne 26 year old single cask #384 distill...\n207 Glengoyne 30 year old, 46.8%\n206 Glengoyne 35 year old, 46.8%\n463 Glengoyne Christmas Cask, 60.6%\n1269 Glengoyne Distiller’s Gold 15 year old, 40%\n1488 Glengoyne Scottish Oak Wood Finish, 43%\n1490 Glengoyne Teapot Dram (Batch 3), 59.4%\n1268 Glengoyne, 12 year old, 57.2%\n1069 Glengoyne, 21 year old, 43%\n59 Glengoyne, 40 year old, 45.9%\n1914 Glenkeir Treasures (Ledaig, distilled at Tober...\n2160 Glenkeir Treasures (distilled at Fettercairn) ...\n2040 Glenkinchie 12 year old, 43%\n648 Glenkinchie 20 year old, 55.1%\n1491 Glenkinchie 2000 Amontillado Finish Distillers...\n1271 Glenkinchie 24 year old, 57.5%\n1679 Glenkinchie Distillers Edition 1996, 43%\n1678 Glenkinchie Distillers Edition, 1991 vintage, ...\n1831 Glenkinchie, 12 years old, 43%\n2112 Glenkinchie, 59.3%\n842 Glenlivet 18 year old, 43%\n211 Glenlivet 21 year old Archive, 43%\n651 Glenlivet Archive, 21 year old, 43%\n1835 Glenlivet Nadurra Islay Cask, 61.5%\n846 Glenmorangie 15 Year Old Nectar d’Or, 46%\n472 Glenmorangie 1981 Sauternes, Wood Finish, 46%\n1074 Glenmorangie A Midwinter Night's Dram, 43%\n847 Glenmorangie Artein, 46%\n60 Glenmorangie Astar 2017, 52.5%\n120 Glenmorangie Astar, 57.1%\n336 Glenmorangie Bacalta, 46%\n1957 Glenmorangie Burgundy Wood Finish, 43%\n850 Glenmorangie Cadboll, 43%\n475 Glenmorangie Companta, 46%\n1680 Glenmorangie Dornoch Limited Edition, 43%\n1275 Glenmorangie Duthac, 43%\n474 Glenmorangie Ealanta Private Edition, 46%\n845 Glenmorangie Finealta, 46%\n213 Glenmorangie Grand Vintage Malt 1990, 43%\n2176 Glenmorangie Lasanta, 46%\n471 Glenmorangie Margaux Cask Finish, 18 year old,...\n657 Glenmorangie Milsean, 46%\n656 Glenmorangie Nectar D'or, 46%\n119 Glenmorangie Original, 10 year old, 43%\n121 Glenmorangie Pride 1981, 56.7%\n1073 Glenmorangie Quinta Ruban, 46%\n844 Glenmorangie Signet, 46%\n473 Glenmorangie Sonnalta PX, 46%\n658 Glenmorangie Spìos, 46%\n1681 Glenmorangie Taghta, 46%\n654 Glenmorangie Tain L'Hermitage Wood Finish, 197...\n849 Glenmorangie Tarlogan, 43%\n848 Glenmorangie Tùsail, 46%\n843 Glenmorangie, 15 year old, 43%\n655 Glenmorangie, 18 year old, 43%\n1072 Glenmorangie, 1977 Vintage, 43%\n335 Glenmorangie, 25 year old, 43%\n212 Glenmorangie, Cote de Nuits Finish, 1975, 43%\n1077 Glenrothes 1978, 43%\n1278 Glenrothes 1979 Editor’s Cask #3828, 52.1%\n855 Glenrothes 1988, 43%\n662 Glenrothes 1995 American Oak, 45%\n1277 Glenrothes 1995, 43%\n2220 Glenrothes 1996 Editor’s Cask #9973, 57%\n1836 Glenrothes 2001 (bottled 2012), 43%\n214 Glenrothes 32 year old, 1972 Vintage, 43%\n856 Glenrothes Elders’ Reserve, 43%\n857 Glenrothes Manse Reserve, 43%\n479 Glenrothes Minister’s Reserve, 43%\n2177 Glenrothes Peated Cask Reserve, 40%\n1682 Glenrothes Select Reserve, 40%\n1497 Glenrothes Sherry Cask Reserve, 40%\n216 Glenrothes Vintage 1988 2nd Edition, 44.1%\n661 Glenrothes Vintage Reserve, 40%\n1078 Glenrothes Vintage Reserve, 40%\n853 Glenrothes, 1985 vintage, 43%\n337 Glenrothes, 1987 vintage, 43%\n851 Glenrothes, 1991 Vintage, 43%\n1076 Glenrothes, 1994 Vintage, $43%\n1495 Glenrothes, 20 year old, 1984 Vintage, 43%\n2114 Glenrothes, Cask # 13458, 1979 vintage, 43%\n1498 Glenturret 10 year old, 40%\n1500 Glenturret 26 year old 1986 (Hunter Laing/The ...\n859 Glenturret Fly’s 16 Masters Edition, 44%\n1960 Glenturret Peated Edition, 46%\n1837 Glenturret Sherry Edition, 46%\n663 Glenturret The Brock Malloy Edition, 46.8%\n1961 Glenturret Triple Wood, 46%\n1282 Glenugie, Chivas Brothers, Deoch an Doras bott...\n664 Glenury Royal 40 year old, 59.4%\n6 Gold Bowmore, 1964 vintage, 42.4% \n606 Gordon & MacPhail 'Connoisseurs Choice' (disti...\n1459 Gordon & MacPhail 'Connoisseurs Choice' (disti...\n1274 Gordon & MacPhail ('Smith's Glenlivet'), 21 ye...\n1618 Gordon & MacPhail (distilled at Arran) 2006, 46%\n2169 Gordon & MacPhail (distilled at Banff), 28 yea...\n1451 Gordon & MacPhail (distilled at Brora), 24 yea...\n813 Gordon & MacPhail (distilled at Caol Ila) 15 y...\n1937 Gordon & MacPhail (distilled at Caol Ila), 12 ...\n621 Gordon & MacPhail (distilled at Caol Ila), 14 ...\n1024 Gordon & MacPhail (distilled at Caol Ila), 8 y...\n1813 Gordon & MacPhail (distilled at Caol Ila), cas...\n2172 Gordon & MacPhail (distilled at Caol Ila), cas...\n200 Gordon & MacPhail (distilled at Glen Grant) 60...\n447 Gordon & MacPhail (distilled at Glen Grant), 1...\n1669 Gordon & MacPhail (distilled at Glenburgie) 19...\n1477 Gordon & MacPhail (distilled at Glencadam), 31...\n1955 Gordon & MacPhail (distilled at Glenlossie), 2...\n1281 Gordon & MacPhail (distilled at Glentauchers) ...\n1684 Gordon & MacPhail (distilled at Glentauchers),...\n2045 Gordon & MacPhail (distilled at Highland Park)...\n1693 Gordon & MacPhail (distilled at Ledaig), cask ...\n1847 Gordon & MacPhail (distilled at Linkwood) 15 y...\n1696 Gordon & MacPhail (distilled at Longmorn) 30 y...\n1851 Gordon & MacPhail (distilled at Longmorn), 15 ...\n494 Gordon & MacPhail (distilled at Macallan) 1970...\n1863 Gordon & MacPhail (distilled at Miltonduff) Ca...\n1976 Gordon & MacPhail (distilled at Mortlach) 15 y...\n134 Gordon & MacPhail (distilled at Mortlach), 11 ...\n239 Gordon & MacPhail (distilled at Mortlach), 36 ...\n1302 Gordon & MacPhail (distilled at Mortlach), cas...\n347 Gordon & MacPhail (distilled at Old Pulteney),...\n2236 Gordon & MacPhail (distilled at Port Ellen), 1...\n1100 Gordon & MacPhail (distilled at Pulteney) 21 y...\n1701 Gordon & MacPhail (distilled at Rosebank) 1989...\n2185 Gordon & MacPhail (distilled at Rosebank), cas...\n372 Gordon & MacPhail (distilled at Strathisla) 19...\n701 Gordon & MacPhail (distilled at Strathisla) 40...\n245 Gordon & MacPhail (distilled at Strathisla), 1...\n1706 Gordon & MacPhail (distilled at Strathisla), C...\n1324 Gordon & MacPhail (distilled at Tamnavulin), 4...\n251 Gordon & MacPhail (distilled at Tomintoul), ca...\n1171 Gordon & MacPhail 10 year old (distilled at Gl...\n549 Gordon & MacPhail 1954 (distilled at Glen Gran...\n550 Gordon & MacPhail 1977 (distilled at Strathisl...\n1588 Gordon & MacPhail 8 year old (distilled at Pul...\n1170 Gordon & MacPhail Cask Strength 1995 (distille...\n748 Gordon & MacPhail Cask Strength 2005 (distille...\n1393 Gordon & MacPhail Connoisseurs Choice (distill...\n1169 Gordon & MacPhail Connoisseurs Choice (distill...\n1589 Gordon & MacPhail Connoisseurs Choice 1997 (di...\n1758 Gordon & MacPhail Connoisseurs Choice 1999 (Le...\n1908 Gordon & MacPhail Connoisseurs’ Choice 1998 Le...\n1394 Gordon & MacPhail Distillery Labels (distilled...\n1172 Gordon & MacPhail Distillery Labels 10 year ol...\n469 Gordon & MacPhail Generations: The Glenlivet 7...\n688 Gordon & MacPhail Private Collection (distille...\n1211 Gordon & MacPhail Private Collection Balblair ...\n1516 Gordon & MacPhail Private Collection Ledaig 19...\n210 Gordon & MacPhail Private Collection: The Glen...\n470 Gordon & MacPhail Private Collection: The Glen...\n2175 Gordon & MacPhail Private Collection: The Glen...\n1492 Gordon & MacPhail Private Collection: The Glen...\n1833 Gordon & MacPhail Private Collection: The Glen...\n1283 Gordon & MacPhail Rare Old (distilled at Glenu...\n659 Gordon & MacPhail Reserve (distilled at Glenro...\n1420 Gordon & MacPhail Single Cask #12274 (distille...\n2128 Gordon & MacPhail Single Cask #1535 (distilled...\n1647 Gordon & MacPhail Single Cask #1961 (distilled...\n1946 Gordon & MacPhail Single Cask #5054 (distilled...\n1025 Gordon & MacPhail “Private Collection” Madeira...\n1460 Gordon & MacPhail “Private Collection” Port Wo...\n1273 Gordon & MacPhail, 'Smith's Glenlivet', 21 yea...\n1396 Grand Macnish 12 year old, 40%\n2222 Grand Macnish 150th Anniversary Edition, 40%\n1174 Grand Macnish Black Edition, 40%\n2008 Grand Macnish Sherry Cask 15 year old, 40%\n1911 Grand Macnish Six Cask Edition, 40%\n1189 Grangestone 12 year old Highland single malt, 40%\n1408 Grangestone Sherry Finish, 40%\n1897 Grant's Voyager, 40%\n1742 Grant’s Signature, 40%\n2223 H5 Iced Single Grain, 40%\n1179 Haig Club, 40%\n1154 Hankey Bannister Heritage Blend, 46%\n1641 Hart Brothers (distilled at Bowmore) 11 year o...\n375 Hart Brothers (distilled at Glen Grant) 22 yea...\n1820 Hart Brothers (distilled at Glen Grant), 29 ye...\n2042 Hart Brothers (distilled at Glenturret), 10 ye...\n1290 Hart Brothers (distilled at Inchgower), 26 yea...\n342 Hart Brothers (distilled at Laphroaig), 1990 v...\n1363 Hazelburn 10 year old, 46%\n1146 Hazelburn 12 year old, 46%\n1737 Hazelburn 8 year old, 46%\n1147 Hazelburn 9 year old Barolo Cask, 57.9%\n1994 Hazelburn CV, 46%\n924 Hazelburn Rundlets & Kilderkins 10 year old 3r...\n1704 Hazelburn, 47.4%\n1771 Hepburn’s Choice (distilled at Blair Athol) 12...\n2196 High Commissioner, 40%\n1178 Highland Journey, 46.2%\n219 Highland Park 'Hjarta', 12 years old, 58.1%\n481 Highland Park 12 Year Old, 40%\n480 Highland Park 15 year old, 43%\n67 Highland Park 18 Year Old, 43%\n221 Highland Park 1971 Vintage, 46.7%\n668 Highland Park 1976 Vintage, 49.1%\n64 Highland Park 30 Year Old, 48.1%\n1840 Highland Park Ambassador's Choice 10 year old,...\n1689 Highland Park Dark Origins, 46.8%\n1687 Highland Park Einar, 40%\n483 Highland Park Fire, 45.2%\n1289 Highland Park Freya 15 year old, 51.2%\n1087 Highland Park Full Volume, 47.2%\n1286 Highland Park Harald, 40%\n1084 Highland Park ICE Edition, 53.9%\n1083 Highland Park Loki 15 year old, 48.7%\n1086 Highland Park Magnus, 40%\n863 Highland Park Odin 16 year old, 55.8%\n1839 Highland Park Svein, 40%\n222 Highland Park The Dark 17 year old, 52.9%\n482 Highland Park Thor 16 year old, 52.1%\n1085 Highland Park Valkyrie, 45.9%\n2180 Highland Park, 19 year old, 1986 vintage, 53.8%\n65 Highland Park, 1964 vintage, 42.2%\n127 Highland Park, 1968 vintage, 45.6%\n66 Highland Park, 1970 vintage, 48%\n125 Highland Park, 21 year old, 47.5%\n2179 Highland Park, 24 year old, 1980 vintage, 58%\n1285 Highland Park, 24 year old, 1981 Vintage, Cask...\n1838 Highland Park, 31 year old, 1974 vintage, cask...\n63 Highland Park, 32 year old, 1973 Vintage, Cask...\n61 Highland Park, 34 year old, 1971 vintage (Cask...\n126 Highland Park, 40 year old, 48.3%\n2178 Highland Park, Cask #10132, 1990 vintage, 15 y...\n2116 Highland Park, Cask #10140, 1990 vintage, 15 y...\n1081 Highland Park, Cask #10146, 1990 vintage, 15 y...\n30 Highland Park, Cask #13308, 1973 vintage, 33 y...\n860 Highland Park, Cask #1673, 1992 vintage, 13 ye...\n2044 Highland Park, Cask #2310, 1991 vintage, 14 ye...\n1686 Highland Park, Cask #2498, 1986 vintage, 19 ye...\n1284 Highland Park, Cask #686, 1983 vintage, 22 yea...\n124 Highland Park, Cask #691, 1983 vintage, 23 yea...\n2243 Highland Park, Cask #7380, 1981 vintage, 25 ye...\n62 Highland Park, Cask #7957, 1977 vintage, 29 ye...\n218 Highland Park, Cask #8998, 1974 vintage, 31 ye...\n220 Highland Park, “Saint Magnus,” 55%\n1762 Highland Queen 12 year old, 40%\n1177 Highland Queen 1561 30 year old, 40%\n2079 Highland Queen 8 year old, 40%\n752 Highland Queen Majesty 15 year old, 46%\n2212 Highland Queen, 40%\n937 House of Hazelwood 18 year old, 40%\n1898 House of Hazelwood 21 year old, 40%\n536 House of Hazelwood 25 year old, 40%\n956 Hunter Laing Old Malt Cask 1996 (distilled at ...\n1221 Hunter Laing Old Malt Cask 20 year old (distil...\n1965 Inchgower 14 year old, 43%\n764 Inchmoan Vintage 1992, 48.6%\n2048 Inchmurrin 12 year old, 46%\n1966 Inchmurrin 12 year old, 46%\n2200 Inchmurrin 15 year old, 46%\n2119 Inchmurrin 18 year old, 46%\n2181 Inchmurrin 18 year old, 46%\n1842 Inchmurrin 18 year old, 46%\n2046 Inchmurrin 21 year old, 46%\n2047 Inchmurrin Madeira Finish, 46%\n2120 Inchmurrin Madeira Wood Finish, 46%\n2118 Inchmurrin, 12 year old, 43%\n1574 Inverleven, Chivas Brothers, Deoch an Doras bo...\n1351 Islay Mist, 12 year old, 40%\n1733 Islay Mist, 17 year old, 40%\n2191 Islay Mist, 8 year old, 40%\n1293 Isle of Jura Superstition, 45%\n1292 Isle of Jura, 16 year old, 43%\n2182 Isle of Jura, 21 year old, 43%\n1731 Isle of Skye, 8 year old, 43%\n1607 J&B Rare, 40%\n970 J. Mossman Gold Crown 12 year old, 40%\n765 J. Mossman Pink Gold Crown 18 year old, 40%\n1582 James Brookes Blended Whisky, 40%\n1779 Jamie Stewart, 40%\n815 Jewels of Scotland (distilled at Caol Ila), 19...\n925 Jewels of Scotland (distilled at Dailuaine), 1...\n34 Jewels of Scotland (distilled at Springbank), ...\n1780 John Barr Reserve, 40%\n1011 John MacDougall's Selection (distilled at Blad...\n1800 John McDougall's Selection (distilled at Bladn...\n281 John Walker & Sons Private Collection 2014 Edi...\n1193 John Walker & Sons Private Collection 2015 Edi...\n86 John Walker & Sons Private Collection 2016 Edi...\n174 John Walker & Sons Private Collection 2017 Edi...\n775 Johnnie Walker Black Label 12 year old, 40%\n776 Johnnie Walker Black Label The Director’s Cut,...\n1923 Johnnie Walker Blenders’ Batch No. 1 Red Rye F...\n175 Johnnie Walker Blenders’ Batch No. 2 Bourbon C...\n391 Johnnie Walker Blenders’ Batch No. 3 Triple Gr...\n575 Johnnie Walker Blenders’ Batch Sherry Cask Fin...\n979 Johnnie Walker Blenders’ Batch Wine Cask, 40%,\n15 Johnnie Walker Blue Anniversary, 60%\n282 Johnnie Walker Blue Label Ghost and Rare, 46%\n0 Johnnie Walker Blue Label, 40%\n390 Johnnie Walker Double Black, 40%\n1414 Johnnie Walker Explorers' Club Royal Route, 40%\n1784 Johnnie Walker Explorers’ Club Collection The ...\n978 Johnnie Walker Gold, 18 year old, 40%\n573 Johnnie Walker Green Label, 43%\n389 Johnnie Walker Island Green, 43%\n84 Johnnie Walker Odyssey, 40%\n1610 Johnnie Walker Red Label, 40%\n574 Johnnie Walker Select Cask Rye Cask Finish, 46%\n40 Johnnie Walker The Casks Edition, 55.8%\n774 Johnnie Walker The Spice Road, 40%\n2121 Jura 10 Year Old, 40%\n962 Jura 10, 40%\n1187 Jura 16 year old Diurachs’ Own, 43%\n484 Jura 1977 Vintage, 46%\n1506 Jura 1984, 44%\n671 Jura Camas an Staca 30 year old, 44%\n2049 Jura Elixir, 40%\n1504 Jura Feith A’ Chaorainn 1976 Vintage, 46%\n378 Jura One and All, 51%\n864 Jura Prophecy, 46%\n759 Jura Tastival, 51%\n1505 Jura Turas-Mara, 42%\n258 Kilchoman 'Inaugural Release,' 46%\n1568 Kilchoman (Summer 2010 Release), 46%\n1136 Kilchoman 100% Islay (distilled 2008, bottled ...\n1137 Kilchoman 100% Islay 3rd Edition, 50%\n1882 Kilchoman 100% Islay 4th Edition, 59%\n721 Kilchoman 100% Islay 7th Edition, 50%\n2058 Kilchoman 2006 vintage 5 year old, 46%\n362 Kilchoman 2007, 46%\n363 Kilchoman 2009 Vintage, 46%\n1138 Kilchoman Coull Point, 46%\n915 Kilchoman Fèis Ìle 2016 bottling (distilled 20...\n1569 Kilchoman Loch Gorm 2016 (distilled 2010), 46%\n523 Kilchoman Loch Gorm 2017, 46%\n1881 Kilchoman Machir Bay 2013 Release, 46%\n916 Kilchoman Red Wine Cask Matured, 50%\n1735 Kilchoman Sanaig, 46%\n1352 Kilchoman Sauternes Cask Matured, 50%\n1734 Kilchoman Single Cask Release Cask #370 (disti...\n2204 Kilchoman Single Cask Sherry Cask (Cask #85), ...\n720 Kilchoman Spring 2011 release, 46%\n522 Kilchoman, Autumn 2009 release, 46%\n1135 Kilchoman, Winter 2010 Release, 46%\n921 Kilkerran 12 year old, 46%\n1359 Kilkerran Work In Progress 5 Bourbon, 46%\n1142 Kilkerran Work in Progress 4 (2012 Release), 46%\n1143 Kilkerran Work in Progress 5 Sherry, 46%\n919 Kilkerran Work in Progress 6 Bourbon Matured, 46%\n1144 Kilkerran Work in Progress 6 Sherry Matured, 46%\n1145 Kilkerran Work in Progress VII – Bourbon, 54.1%\n920 Kilkerran Work in Progress VII – Sherry, 46%\n1317 Kilkerran, 60.3%\n394 King's Crest, 25 year old, 40%\n1419 King's Crest, 25 year, 40%\n1603 Kininvie 23 year old (Batch 3), 42.6%\n1834 Kirkland Signature (distilled at Glenlivet) 40...\n1912 Kirkland Signature Highland 18 year old Sherry...\n2228 Knockando 12 year old, 43%\n2232 Knockando 25 year old Special Release 2011, 43%\n2013 Label 5 Classic Black, 40%\n1398 Label 5 Extra Premium 12 year old, 40%\n959 Label 5 Extra Rare 18 year old, 40%\n1764 Label 5 Gold Heritage, 40%\n1219 Lady of the Glen (distilled at Benrinnes) 14 y...\n1185 Ladyburn 41 year old, 40%\n1843 Lagavulin 12 year old (Diageo Special Release ...\n1512 Lagavulin 12 year old (Diageo Special Release ...\n1507 Lagavulin 12 year old Special Release 2011, 57.5%\n1508 Lagavulin 12 year old Special Release 2012, 56.1%\n1294 Lagavulin 12 year old, 55.1%\n68 Lagavulin 12 year old, 56.5%\n1513 Lagavulin 12 year old, 57.7%\n223 Lagavulin 16 year old, 43%\n486 Lagavulin 18 year old Fèis Ìle 2016 bottling, ...\n128 Lagavulin 1993 Islay Jazz Festival bottling (b...\n129 Lagavulin 1995 Feis Ile 2014 bottling, 54.7%\n224 Lagavulin 21 year old Special Release 2012, 52%\n130 Lagavulin 25 year old, 50.9%\n1509 Lagavulin 37 year old, 51%\n869 Lagavulin 8 year old, 48%\n2050 Lagavulin Cask #1716 (distilled 1998) Feis Ile...\n868 Lagavulin Distillers Edition 1995, 43%\n225 Lagavulin Feis Ile 2013 bottling (distilled 19...\n1511 Lagavulin Triple Matured Edition, 48%\n485 Lagavulin, 12 year old (2009 Release), 57.9% \n1690 Lagavulin, 12 year, 57.8%\n672 Lagavulin, 21 year, 56.5%\n866 Lagavulin, 30 year old, 52.6%\n339 Lagavulin, 52.5%\n338 Lagavulin, Distillers' Edition, 1991 Vintage, 43%\n1209 Langside Distillers (distilled at Auchentoshan...\n1692 Laphroaig 15 year old, 40%\n1515 Laphroaig 15 year old, 43%\n489 Laphroaig 32 year old, 46.5%\n2123 Laphroaig Cairdeas Madeira Cask, 51.6%\n1691 Laphroaig Cairdeas, 50.5%\n1296 Laphroaig Lore, 48%\n870 Laphroaig Original Cask Strength, 10 year old,...\n226 Laphroaig Original Cask, 10 year old, 57.3%\n2122 Laphroaig PX cask, 48%\n341 Laphroaig Quarter Cask, 48%\n1968 Laphroaig Select, 40%\n70 Laphroaig Triple Wood, 48%\n488 Laphroaig Triple Wood, 48%\n487 Laphroaig, 18 year old, 48%\n31 Laphroaig, 25 year old (2008 Edition), 51.2%\n340 Laphroaig, 30 year old, 43%\n1088 Laphroaig, 40 year old, 42.4%\n753 Last Drop Distillers (distilled at Glen Garioc...\n1517 Ledaig 18 year old, 46.3%\n678 Ledaig 19 year old, 51%\n677 Ledaig 42 year old, 46.3%\n2183 Ledaig Sherry Finish, 42%\n2115 Ledaig, 10 year old, 43%\n1845 Ledaig, 10 year old, 46.3%\n1510 Lg4 Elements of Islay (distilled at Lagavulin)...\n1518 Linkwood 37 year old, 50.3%\n1089 Linkwood, 1996 vintage, 58.2%\n1519 Littlemill 21 year old Burns Malt, 55.1%\n1521 Littlemill 21 year old Second Release, 47%\n1522 Littlemill 25 year old Private Cellar Edition,...\n2184 Littlemill, 12 year old, 43%\n876 Loch Lomond 12 year old, 46%\n490 Loch Lomond 18 year old, 46%\n2227 Loch Lomond Original, 40%\n877 Loch Lomond, 46%\n22 Lombard 'Jewels of Scotland' (distilled at Bro...\n1969 Lombard 'Jewels of Scotland' (distilled at Loc...\n1732 Lombard Gold Label, 40%\n350 Lombard Jewels of Scotland (distilled at Sprin...\n1991 Lombard Speyside, 8 year old, 43%\n1552 Lonach (distilled at Carsebridge), 1063 vintag...\n840 Lonach (distilled at Glenlivet), 34 year old, ...\n246 Lonach (distilled at Strathisla), 42 year old,...\n681 Longmorn 16 year old, 48%\n680 Longmorn 16 year old, 48%\n491 Longmorn 16 year old, 48%\n343 Longmorn 23 year old, 48%\n2124 Longmorn The Distiller’s Choice, 40%\n231 Longrow 18 year old, 46%\n132 Longrow 18 year old, 46%\n131 Longrow 1974, 25 year old, 46%\n1850 Longrow C.V., 46%\n1091 Longrow Red 11 year old Port Cask, 51.8%\n878 Longrow Red, 52.1%\n2203 Longrow Tokaji Wood, 10 year old, 55.6%\n1971 Longrow, 10 year old, 50%\n1523 Longrow, 14 year old, 46%\n2052 Longrow, Gaja Borolo Finish, 7 year old, 55.8%\n1913 Lord Elcho 15 year old, 40%\n1176 Lord Elcho, 40%\n675 Lp4 Elements of Islay (distilled at Laphroaig)...\n676 Lp5 Elements of Islay (distilled at Laphroaig)...\n2229 MacNaMara Blended, 40%\n2214 MacNaMara Rum Cask Finish, 40%\n1757 MacPhail’s Collection (distilled at Glenturret...\n2165 MacQueen’s 3 year old, 40%\n1973 Macallan 12 year old, 40%\n238 Macallan 18 year old Sherry Oak 2017 Edition, 43%\n11 Macallan 1976 Vintage, 29 year old, cask #1135...\n1858 Macallan Amber, 40%\n883 Macallan Classic Cut 2017 Edition, 58.4%\n1860 Macallan Coronation Cask American Oak #190950,...\n882 Macallan Coronation Cask Spanish Oak #190952, ...\n1094 Macallan Diamond Jubilee, 52%\n493 Macallan Easter Elchies, 52.3%\n1095 Macallan Edition No. 1, 48%\n499 Macallan Edition No.3, 48.3%\n1859 Macallan Gold, 40%\n345 Macallan Masters of Photography 3rd release 19...\n1301 Macallan Masters of Photography 3rd release 19...\n1857 Macallan Masters of Photography 3rd release 19...\n1529 Macallan Masters of Photography 3rd release 19...\n237 Macallan Royal Wedding, 46.8%\n496 Macallan Ruby, 43%\n344 Macallan Sherry Oak 18 year old, 43%\n881 Macallan Sienna, 43%\n1909 Macduff 16 year old, 46%\n619 Mackillop's Choice (distilled at Caol Ila), 15...\n2029 Mackillop's Choice (distilled at Caol Ila), 26...\n820 Mackillop's Choice (distilled at Caperdonich),...\n829 Mackillop's Choice (distilled at Glen Ord), 21...\n465 Mackillop's Choice (distilled at Glenlivet), 2...\n1963 Mackillop's Choice (distilled at Imperial), 23...\n1525 Mackillop's Choice (distilled at Longmorn) 196...\n1526 Mackillop's Choice (distilled at Macallan) 197...\n1698 Mackillop's Choice (distilled at Macallan), 14...\n686 Mackillop's Choice (distilled at Mortlach), 22...\n353 Mackillop's Choice (distilled at Tomintoul), 3...\n841 Mackillop’s Choice (distilled at Glenlivet), 1...\n163 Mackillop’s Choice (distilled at Mortlach) 199...\n1531 Mackillop’s Choice (distilled at Mortlach), 19...\n1765 Mackillop’s Choice (distilled at Tormore) 1988...\n260 Mackinlay’s Rare Old Highland whisky, 47.3%\n1096 Mannochmore 25 year old, 53.4%\n1997 Master of Malt 10 year old, 47.5%\n1738 Master of Malt Boxes Blend, 40.9%\n2167 Master of Malt Darkness! Ardbeg 21 year old PX...\n1633 Master of Malt Darkness! Benrinnes 15 year old...\n1929 Master of Malt Darkness! Benrinnes 15 year old...\n821 Master of Malt Darkness! Clynelish 16 year old...\n1861 Master of Malt Darkness! Macallan 15 year old ...\n2221 Master of Malt Reference Series I, 47.5%\n1365 Master of Malt Reference Series II, 47.5%\n1996 Master of Malt Reference Series II, 47.5%\n1576 Master of Malt Reference Series III, 47.5%\n1739 Master of Malt Secret Bottlings Series 8 year ...\n1366 Master of Malt Single Cask Series (distilled a...\n1894 Master of Malt Single Cask Series (distilled a...\n1995 Master of Malt Speyside Single Malt 30 year ol...\n2147 Master of Malt Speyside Single Malt 40 year ol...\n2084 McDougall's Selection, 13 year old, 54.7%\n1862 Miltonduff 19 year old, 51.3%\n1999 Monarch of The Glen 8 year old, 40%\n2073 Monarch of the Glen, 40%\n1527 Montgomerie's (distilled at Macallan) 1989 Vin...\n1439 Montgomerie's Single Cask #910 (distilled at B...\n963 Moon Harbour Pier #2 Peated Edition, 47.1%\n346 Mortlach 16 year old, 43%\n135 Mortlach 18 year old, 43.4%\n1303 Mortlach 25 year old, 43.4%\n501 Mortlach Rare Old, 43.4%\n502 Mortlach Special Strength, 49%\n1530 Mortlach, 1997 vintage, 57.1%\n667 Murray McDavid \n673 Murray McDavid 'Leapfrog' (distilled at Laphro...\n103 Murray McDavid 'Mission' (distilled at Clyneli...\n1463 Murray McDavid 'Mission' (distilled at Dallas ...\n1265 Murray McDavid 'Mission' (distilled at Glengla...\n332 Murray McDavid 'Mission' (distilled at Glenliv...\n867 Murray McDavid 'Mission' (distilled at Lagavul...\n1311 Murray McDavid 'Mission' (distilled at Royal B...\n398 Murray McDavid (Distilled at Ardbeg) 1990, 8 y...\n1018 Murray McDavid (distilled at Bruichladdich) 19...\n1814 Murray McDavid (distilled at Caperdonich) 1968...\n1470 Murray McDavid (distilled at Glen Grant) 1969,...\n1956 Murray McDavid (distilled at Glenlossie) 1975,...\n1082 Murray McDavid (distilled at Highland Park) 19...\n1501 Murray McDavid (distilled at Highland Park), 1...\n1967 Murray McDavid (distilled at Isle of Jura, 198...\n229 Murray McDavid (distilled at Linkwood) 1973, 46%\n2051 Murray McDavid (distilled at Littlemill), 1991...\n1849 Murray McDavid (distilled at Lochside) 1981, 46%\n1300 Murray McDavid (distilled at Macallan), 1990 v...\n1699 Murray McDavid (distilled at Macallan), 1997 v...\n1702 Murray McDavid (distilled at Rosebank) 1992, 1...\n142 Murray McDavid (distilled at Springbank) 1965,...\n1705 Murray McDavid (distilled at Strathisla) 1976,...\n2032 Murray McDavid Mission IV (distilled at Clynel...\n1054 Murray McDavid Mission IV (distilled at Glen S...\n1824 Murray McDavid Mission IV (distilled at Glendr...\n666 Murray McDavid Mission IV (distilled at Highla...\n874 Murray McDavid Mission IV (distilled at Linlit...\n1326 Murray McDavid Mission IV (distilled at Tomint...\n1532 Noss Head, 46%\n1985 Oban 14 year old, 43%\n1551 Oban 1999 Montilla Fino Finish Distillers Edit...\n711 Oban 21 year old, 58.5%\n1723 Oban Distillers Edition 1996, 43%\n255 Oban Little Bay, 43%\n357 Oban, 18 year old, 43%\n358 Oban, 2000 vintage, 58.7%\n1550 Oban, Distiller's Edition, 1992 vintage, 43%\n1452 Octomore Islay Barley 2009, 64%\n781 Old Malt Cask (distilled at Ardbeg) 1992 Vinta...\n504 Old Malt Cask (distilled at Port Ellen), 25 ye...\n1134 Old Masters Freemason Whisky, 40%\n2016 Old Particular (distilled at Cambus) 27 year o...\n1919 Old Particular (distilled at Girvan) 25 year o...\n1304 Old Pulteney (Cask #4931), 1990 Vintage, 57.8%\n1099 Old Pulteney 12 year old, 40%\n888 Old Pulteney 1990 Vintage, 46%\n890 Old Pulteney 25 year old, 46%\n1306 Old Pulteney 30 year old (2013 Edition), 40.1%\n1533 Old Pulteney 35 year old, 42.5%\n348 Old Pulteney 40 Year Old, 51.3%\n889 Old Pulteney 40 year old, 51.3%\n1308 Old Pulteney Dunnet Head, 46%\n887 Old Pulteney Navigator, 46%\n1305 Old Pulteney Row to the Pole, 40%\n689 Old Pulteney WK209 ‘Good Hope,’ 46%\n886 Old Pulteney WK217, 46%\n885 Old Pulteney WK499, 52%\n1098 Old Pulteney, 17 year old, 46%\n1977 Old Pulteney, 21 year old, 46%\n744 Peat Elements of Islay, 59.3%\n2024 Peerless (distilled at Bowmore) 1966, 36 year ...\n632 Peerless (distilled at Glen Grant) 1970, 46.1%\n839 Peerless (distilled at Glenlivet) 1968, 34 yea...\n1276 Peerless (distilled at Glenrothes) 1967, 35 ye...\n665 Peerless (distilled at Highland Park) 1967, 40.3%\n508 Peerless (distilled at Strathisla) 1967, 35 ye...\n1101 Pentland Skerries, 46%,\n1601 Pittyvaich 1989 25 year old (Diageo Special Re...\n735 Poit Dhubh 21 year old Blended Malt, 43%\n2057 Port Askaig 12 year old, 45.8%\n521 Port Askaig 15 year old Sherry Cask, 45.8%\n2190 Port Askaig 19 year old, 58.6%\n2140 Port Askaig 30 year old Cask Strength, 51.1%\n361 Port Askaig, 17 year old, 45.8% \n2141 Port Askaig, 19 year old Cask Strength, 50.4%\n1567 Port Askaig, 25 year old, 45.8%\n564 Port Charlotte 10 Second Limited Edition, 50%\n1775 Port Charlotte 2007 CC:01, 57.8%\n432 Port Charlotte Islay Barley 2008, 50%\n2011 Port Dundas 12 year old, 40%\n1760 Port Dundas 18 year old, 43%\n555 Port Dundas 52 year old (Diageo Special Releas...\n138 Port Ellen 1978 35 year old (Diageo Special Re...\n693 Port Ellen 1983 32 year old (Diageo Special Re...\n137 Port Ellen 31 year old, 54.6%\n240 Port Ellen 32 year old (11th release) Special ...\n1700 Port Ellen 32 year old, 52.5%\n1102 Port Ellen 34 year old, 55%\n140 Port Ellen 37 year old (Diageo Special Release...\n139 Port Ellen 37 year old, 55.2%\n503 Port Ellen, 1979 vintage, 28 year old, 53.8%\n691 Port Ellen, 29 year old, 55.3%\n136 Port Ellen, 30 year old (2009 Release), 57.7%, \n1291 Provenance (distilled at Inchgower), 14 year o...\n690 Provenance (distilled at Port Ellen), 21 year ...\n275 Pure Scot Virgin Oak 43, 43%\n2213 Raasay While We Wait, 46%\n862 Ragnvald, 44.6%\n535 Rare Cask Reserves Blended Reserve 26 year old...\n2230 Robert Burns Single Malt, 40%\n1190 Rock Oyster 18 year old, 46.8%\n382 Rock Oyster Sherry Edition, 46.8%\n1158 Rock Oyster, 46.8%\n2209 Ron Burgundy “Great Odin’s Raven” Special Rese...\n892 Rosebank 1992 21 year old (Diageo Special Rele...\n694 Rosebank 21 year old 1990 (Diageo Special Rele...\n1313 Royal Brackla 12 year old, 40%\n893 Royal Brackla 16 year old, 40%\n1104 Royal Brackla 21 year old, 40%\n1139 Royal Lochnagar 'Selected Reserve,' 43%\n2059 Royal Lochnagar 12 year old, 40%\n2142 Royal Lochnagar Distillers Edition 1998, 40%\n1883 Royal Lochnagar Distillers Edition 2000 (batch...\n1353 Royal Lochnagar Triple Matured Edition, 48%\n1186 Royal Mile Whiskies 40 year old, 47.1%\n392 Royal Salute 21 year old, 40%\n157 Royal Salute 32 year old Union of the Crowns, 43%\n77 Royal Salute 62 Gun Salute, 43%\n369 Royal Salute Special Batch 21 year old Polo Ed...\n88 Royal Salute Stone of Destiny, 38 year old, 40%\n1749 Royal Salute The Diamond Tribute, 40%\n1378 Royal Salute The Eternal Reserve, 40%\n558 SIA, 43%\n737 Scallywag 13 year old, 46%\n544 Scallywag Cask Strength, 53.6%\n695 Scapa 25 year old, 54%\n1105 Scapa Glansa, 40%\n2129 Scapa Skiren, 40%\n1536 Scapa, 14 year old, 40%\n1537 Scapa, 16 year old, 40%\n709 Scotch Malt Whisky Society (28:23) 21 year old...\n803 Scotch Malt Whisky Society (50:42) 18 year old...\n556 Scotch Malt Whisky Society Berber Whiskey With...\n78 Scotch Malt Whisky Society Hunting Hound on Ho...\n309 Scott\n802 Scott Selection (distilled at Bladnoch) vintag...\n1046 Scott Selection (distilled at Glen Grant) 1973...\n683 Scott Selection (distilled at Macallan) 1973, ...\n518 Scott's Selection (Distilled at 'North of Scot...\n1120 Scott's Selection (distilled at 'North of Scot...\n2021 Scott's Selection (distilled at Auchentoshan),...\n613 Scott's Selection (distilled at Bruichladdich)...\n1457 Scott's Selection (distilled at Caol Ila), 198...\n1818 Scott's Selection (distilled at Glen Elgin), 1...\n1664 Scott's Selection (distilled at Glen Grant), 2...\n650 Scott's Selection (distilled at Glenlivet), 27...\n861 Scott's Selection (distilled at Highland Park)...\n2053 Scott's Selection (distilled at Longmorn), 33 ...\n879 Scott's Selection (distilled at Macallan), 30 ...\n1097 Scott's Selection (distilled at Mortlach) 1961...\n2054 Scott's Selection (distilled at Old Pulteney),...\n569 Shackleton, 40%\n257 Sheep Dip 'Old Hebridean,' 1990 vintage, 40%\n149 Sheep Dip Islay Blended Malt, 40%\n1581 Shieldaig 12 year old, 40%\n1377 Shieldaig 14 year old, 40%\n1998 Shieldaig Classic, 40%\n2055 Signatory 'Craigduff' (distilled at Strathisla...\n2108 Signatory 'Glencraig' (distilled at Glenburgie...\n1472 Signatory 'Glenisla' (distilled at Glen Keith)...\n1975 Signatory 'Mosstowie' (distilled at Miltonduff...\n1815 Signatory 'Un-chillfiltered' (distilled at Cly...\n442 Signatory 'Unchillfiltered' (distilled at Crag...\n2107 Signatory 'Unchillfiltered' (distilled at Edra...\n2201 Signatory 'Unchillfiltered' (distilled at Link...\n579 Signatory (distilled at Aberlour), 20 year old...\n1925 Signatory (distilled at Auchentoshan), 12 year...\n723 Signatory (distilled at Balmenach) 1988, 55.6%\n1000 Signatory (distilled at Ben Nevis), Cask #7438...\n1218 Signatory (distilled at BenRiach), 12 year old...\n1802 Signatory (distilled at Bowmore), 11 year old,...\n1014 Signatory (distilled at Bowmore), 16 year old,...\n47 Signatory (distilled at Bowmore), 35 year old,...\n807 Signatory (distilled at Brora) 18 year old, 19...\n425 Signatory (distilled at Brora), 24 year old, 1...\n436 Signatory (distilled at Bunnahabhain), Cask #2...\n2218 Signatory (distilled at Bunnahabhain), cask #5...\n1587 Signatory (distilled at Cameronbridge) 1995, 43%\n311 Signatory (distilled at Caol Ila) 1984, 54.7%\n620 Signatory (distilled at Caol Ila) 1989, 43%\n1458 Signatory (distilled at Caol Ila), 14 year old...\n2171 Signatory (distilled at Caol Ila), 15 year old...\n2106 Signatory (distilled at Dallas Dhu), 26 year o...\n1049 Signatory (distilled at Glen Ord), 1998 Vintag...\n838 Signatory (distilled at Glenlivet), 25 year ol...\n215 Signatory (distilled at Glenrothes), 30 year o...\n1958 Signatory (distilled at Glenrothes), Cask #108...\n2235 Signatory (distilled at Kinclaith), Cask #3014...\n228 Signatory (distilled at Laphroaig) 1998, 60.8%\n1514 Signatory (distilled at Laphroaig), 13 year ol...\n1844 Signatory (distilled at Laphroaig), 7 year old...\n1854 Signatory (distilled at Macallan), 14 year old...\n505 Signatory (distilled at Port Ellen), 1982 vint...\n1534 Signatory (distilled at Rosebank), 12 year old...\n1103 Signatory (distilled at Rosebank), 14 year old...\n1978 Signatory (distilled at Rosebank), 15 year old...\n1392 Signatory Cask Strength Collection (distilled ...\n1391 Signatory Un-Chillfiltered Collection 11 year ...\n1167 Signatory Vintage 10 year old (distilled at Ed...\n696 Signatory, (distilled at Springbank) 1969 Vint...\n1288 Sigurd, 43%\n766 Single Cask Nation (distilled at Tormore) 21 y...\n2078 Single Malts of Scotland (distilled at Allt-a-...\n2077 Single Malts of Scotland (distilled at BenRiac...\n2007 Single Malts of Scotland (distilled at Bunnaha...\n1655 Single Malts of Scotland (distilled at Craggan...\n2195 Single Malts of Scotland (distilled at Craigel...\n1665 Single Malts of Scotland (distilled at Glen Gr...\n2157 Single Malts of Scotland (distilled at Glenall...\n1959 Single Malts of Scotland (distilled at Glenrot...\n1852 Single Malts of Scotland (distilled at Longmor...\n1092 Single Malts of Scotland (distilled at Longmor...\n1719 Single Malts of Scotland (distilled at Tormore...\n1890 Singleton of Dufftown 15 year old, 43%\n2063 Singleton of Dufftown Reserve Collection Trini...\n1993 Singleton of Dufftown Reserve Collection Unité...\n2065 Singleton of Glen Ord 12 year old, 40%\n2064 Singleton of Glen Ord 15 year old, 40%\n1892 Singleton of Glen Ord 18 year old, 40%\n2145 Singleton of Glen Ord Signature, 40%\n1891 Singleton of Glendullan 12 year old, 40%\n727 Singleton of Glendullan 18 year old, 40%\n1362 Singleton of Glendullan 38 year old (Diageo Sp...\n1747 Sir Edward’s 12 year old Blended Whisky, 40%\n2072 Sir Edward’s Blended Whisky, 40%\n1583 Sir Edward’s Smoky, 40%\n1992 Smokehead, 43%\n1884 Smoking Ember, 40%\n1180 Sovereign (distilled at Cameronbridge) 23 year...\n2211 Spey Royal Choice, 46%\n1575 Speyburn 10 year old, 40%\n726 Speyburn 25 year old, 46%\n1889 Speyburn Arranta, 46%\n1361 Speyburn Bradan Orach, 40%\n697 Springbank 11 year old, 58%\n899 Springbank 12 year old Cask Strength 2014 rele...\n1107 Springbank 14 year old Amontillado Cask (#305)...\n897 Springbank 14 year old Fino Cask (#265), 55.3%\n1318 Springbank 14 year old Manzanilla Cask (#259),...\n1319 Springbank 14 year old Oloroso Cask (#268), 56.2%\n243 Springbank 15 year old, 46%\n507 Springbank 21 year old 2014 release, 46%\n1865 Springbank C.V., 46%\n1108 Springbank Calvados Wood 12 year old, 52.7%\n1320 Springbank Green 12 year old, 46%\n1109 Springbank Green 13 year old, 46%\n351 Springbank Local Barley 16 year old, 54.3%\n699 Springbank Marrying Strength 10 year old, 49.9%\n698 Springbank Refill Sherry 17 year old, 57%\n898 Springbank Rundlets & Kilderkins, 49.4%\n700 Springbank Sherry Wood 17 year old, 52.3%\n1703 Springbank, 10 year old, 46%\n894 Springbank, 10 year old, 50%\n1539 Springbank, 10 year old, 58.7%\n1314 Springbank, 12 year old, 50% ABV\n1315 Springbank, 15 year old, 46%\n896 Springbank, 18 year old, 46%\n895 Springbank, 18 year old, 46%\n141 Springbank, 35 year old, 46%\n1106 Springbank, 40 year old, 46%\n1316 Springbank, Oloroso Sherry Butt, 12 year old, ...\n1538 Springbank, Rum Wood Finish, 16 year old, 1991...\n2130 Springbank, cask #97/613, 1997 vintage, 10 yea...\n955 Storm, 43%\n2009 Strathclyde 12 year old, 62.1%\n509 Strathisla 16 year old, 55.3%\n1761 Strathmill 25 year old (Diageo Special Release...\n1736 Stronachie, 12 year old, 43%\n1763 Syndicate 58/6 Premium, 43%\n513 Talisker 10 year old, 45.8%\n510 Talisker 10 year old, 45.8%\n1707 Talisker 175th Anniversary bottling, 45.8%\n145 Talisker 18 year old, 45.8%\n249 Talisker 1985, 56.1%\n900 Talisker 2003 Amoroso Finish Distillers Editio...\n1322 Talisker 25 year old, 45.8%\n702 Talisker 25 year old, 57.8%\n1110 Talisker 30 year old, 45.8%\n512 Talisker 30 year old, 57.3%\n248 Talisker 35 year old Special Release 2012, 54.6%\n705 Talisker Dark Storm, 45.8%\n1321 Talisker Distillers Edition 2000, 45.8%\n511 Talisker Distillers Edition, 1992 vintage, 45.6%\n1112 Talisker Neist Point, 45.8%\n1323 Talisker Port Ruighe, 45.8%\n1708 Talisker Skye, 45.8%\n704 Talisker Storm, 45.8%\n1111 Talisker Triple Matured Edition, 48%\n1980 Talisker, 20 year old, 59.7%\n247 Talisker, 25 year old (2009 Release), 54.8%\n703 Talisker, 30 year old, 50.7%\n144 Talisker, 30 year old, 51.9%\n706 Tamdhu 10 year old, 40%\n250 Tamdhu Batch Strength No. 002, 58.5%\n515 Tamdhu Batch Strength, 58.8%\n942 Te Bheag Blended Whisky, 40%\n1885 Teaninich, 1996 vintage, 55.3%\n1577 That Boutique-y Springbank Batch 2, 53.10%\n984 That Boutique-y Whisky Company (distilled at A...\n732 That Boutique-y Whisky Company (distilled at A...\n1429 That Boutique-y Whisky Company (distilled at A...\n1375 That Boutique-y Whisky Company (distilled at A...\n367 That Boutique-y Whisky Company (distilled at B...\n1745 That Boutique-y Whisky Company (distilled at C...\n1580 That Boutique-y Whisky Company (distilled at C...\n938 That Boutique-y Whisky Company (distilled at C...\n1743 That Boutique-y Whisky Company (distilled at G...\n155 That Boutique-y Whisky Company (distilled at G...\n1746 That Boutique-y Whisky Company (distilled at I...\n1153 That Boutique-y Whisky Company (distilled at I...\n941 That Boutique-y Whisky Company (distilled at I...\n2150 That Boutique-y Whisky Company (distilled at I...\n1150 That Boutique-y Whisky Company (distilled at K...\n1899 That Boutique-y Whisky Company (distilled at L...\n156 That Boutique-y Whisky Company (distilled at M...\n1578 That Boutique-y Whisky Company (distilled at M...\n539 That Boutique-y Whisky Company (distilled at N...\n1744 That Boutique-y Whisky Company (distilled at N...\n2207 That Boutique-y Whisky Company (distilled at N...\n1373 That Boutique-y Whisky Company (distilled at S...\n1374 That Boutique-y Whisky Company (distilled at T...\n2206 That Boutique-y Whisky Company Blended Malt #2...\n733 That Boutique-y Whisky Company Blended Malt #2...\n1376 That Boutique-y Whisky Company Blended Malt #2...\n1579 That Boutique-y Whisky Company Blended Malt, 54%\n1152 That Boutique-y Whisky Company Blended Whisky ...\n368 That Boutique-y Whisky Company Blended Whisky ...\n939 That Boutique-y Whisky Company Blended Whisky ...\n940 That Boutique-y Whisky Company Blended Whisky ...\n734 That Boutique-y Whisky Company Blended Whisky ...\n1151 That Boutique-y Whisky Company Islay Blended M...\n2071 The Annasach Reserve 25 year old, 46%\n1724 The Antiquary 12 year old, 40%\n1338 The Antiquary 21 year old, 43%\n1122 The Antiquary 35 year old, 46%\n1872 The Antiquary, 12 year old, 40%\n1121 The Antiquary, 1977 vintage, 46%\n73 The Antiquary, 21 year old, 43%\n2135 The Antiquary, 40%\n990 The Arran Malt Orkney Bere, 46%\n2019 The Arran Malt, 10 year old, 46%\n2215 The Arran Malt, 12 year old, 46%\n2197 The Arran Malt, 43%\n2087 The Arran Malt, Limited Edition, 58.4%\n402 The Arran Malt, Madeira Wine Cask Finish (2008...\n1199 The Arran Malt, Pomerol Wine Cask Finish (Limi...\n1198 The Arran Malt, Single Bourbon Cask, (Cask#180...\n1421 The Arran Malt, Single Sherry Cask, (Cask#69),...\n1788 The Arran Malt, St. Emilion Wine Cask Finish (...\n1212 The Balvenie Golden Cask, 14 year old, 47.5%\n1793 The Balvenie New Wood, 17 year old, 43%\n2091 The Balvenie Rum Cask, 17 year old, 43%\n1794 The Balvenie Signature, Batch #1, 12 year old,...\n295 The Balvenie Single Barrel, 15 year old (Cask ...\n45 The Balvenie Vintage Cask, 1976 vintage, 53%\n791 The Balvenie, 17 year old, 43%\n2094 The BenRiach 12 year old, 46%\n1797 The BenRiach 16 year old, 46%\n1005 The BenRiach 20 year old, 46%\n1796 The BenRiach Arumaticus, 12 year old, 46%\n1629 The BenRiach Herodotus, 12 year old, 46%\n1216 The BenRiach Importanticus, 12 year old, 46%\n1888 The Black Grouse Alpha, 40%\n1358 The Black Grouse, 40%\n932 The Clan Denny (distilled at Strathclyde) 9 ye...\n930 The Clan Denny Single Grain (distilled at Camb...\n1962 The Classic Cask (distilled at Highland Park),...\n1404 The Cooper’s Choice (distilled at Ben Nevis) 1...\n328 The Coronation (distilled at Glenfarclas) 1953...\n825 The Dalmore 1263 King Alexander III, 40%\n316 The Dalmore 1969 aged 42 years, 45.5%\n629 The Dalmore 1973 aged 38 years, 48.1%\n108 The Dalmore 1978, 46%\n443 The Dalmore 1979 aged 33 years, 48.5%\n1465 The Dalmore 1980 aged 31 years, 52.1%\n630 The Dalmore 1991 aged 20 years, 57.9%\n1241 The Dalmore 1992 aged 19 years, 53.8%\n317 The Dalmore 21 year old, 42%\n445 The Dalmore 25 year old, 42%\n194 The Dalmore Cigar Malt Reserve, 44%\n826 The Dalmore Cromartie, 45%\n1239 The Dalmore Dee Dram, 40%\n315 The Dalmore Distillery Exclusive 1991, 59%\n1657 The Dalmore Gran Reserva, 40%\n314 The Dalmore Mackenzie, 1992 Vintage, 46%\n1464 The Dalmore Spey Dram, 40%\n55 The Dalmore Stillman's Dram 28 Year Old, 45%\n1036 The Dalmore Tay Dram, 40%\n1240 The Dalmore Tweed Dram, 40%\n1037 The Dalmore Valour, 40%\n1656 The Dalmore, 12 year old, 40%\n824 The Dalmore, 15 year old, 40%\n1035 The Dalmore, 18 year old, 43%\n1938 The Dalmore, 1974 Vintage, 42%\n193 The Dalmore, 1981 Vintage, Amoroso Finish, 42%\n107 The Dalmore, 40 year old, 40%\n8 The Dalmore, 50 year old, 52.8%\n1917 The Deveron 12 year old, 40%\n2156 The Epicurean, 46.2%\n751 The Exceptional Grain, 43%\n951 The Exclusive Blend 35 year old 1980, 46%\n2005 The Exclusive Malts (distilled at Caledonian)...\n739 The Exclusive Malts (distilled at Invergordon...\n1586 The Exclusive Malts (distilled at Auchroisk) 1...\n946 The Exclusive Malts (distilled at Ben Nevis) 1...\n1162 The Exclusive Malts (distilled at BenRiach) 17...\n2076 The Exclusive Malts (distilled at Bowmore) 15 ...\n1031 The Exclusive Malts (distilled at Clynelish) 1...\n1388 The Exclusive Malts (distilled at Girvan) 27 y...\n545 The Exclusive Malts (distilled at Glen Garioch...\n570 The Exclusive Malts (distilled at Invergordon)...\n1295 The Exclusive Malts (distilled at Laphroaig) 2...\n674 The Exclusive Malts (distilled at Laphroaig) 8...\n227 The Exclusive Malts (distilled at Laphroaig) C...\n1390 The Exclusive Malts (distilled at Port Dundas)...\n1755 The Exclusive Malts (distilled at Strathclyde)...\n1907 The Exclusive Malts (distilled at Tomatin) 10 ...\n2006 The Exclusive Malts Blend of Malts 20 year old...\n1163 The Exclusive Malts Islay 7 year old 2007 (cas...\n773 The Famous Grouse 40 year old, 47.3%\n772 The Famous Grouse Celebration, 40%\n977 The Famous Grouse Double Matured 16 year old V...\n1412 The Famous Grouse Malt Whisky, 10 year old, 40%\n1192 The Famous Grouse Malt Whisky, 12 year old, 40%\n771 The Famous Grouse Malt Whisky, 18 year old, 43%\n1783 The Famous Jubilee, 40%\n2012 The Fat Trout, 40%\n271 The Feathery, 40%\n274 The Five Distinguished and Rare 39 year old, 51%\n1493 The Glenlivet 12 year old, 40%\n117 The Glenlivet Archive, 21 year old, 43%\n334 The Glenlivet Cellar Collection 1980 (bottled ...\n116 The Glenlivet Cellar Collection, 1964 vintage,...\n1070 The Glenlivet Cellar Collection, 1967, 46%\n10 The Glenlivet Cellar Collection, 1969 vintage,...\n652 The Glenlivet Cellar Collection, 1972 vintage,...\n29 The Glenlivet Cellar Collection, 1973 vintage,...\n649 The Glenlivet Cellar Collection, 1983 Vintage,...\n1954 The Glenlivet Founder’s Reserve, 40%\n209 The Glenlivet Founder’s Reserve, 55.6%\n1071 The Glenlivet French Oak Reserve 15 year old, 40%\n467 The Glenlivet French Oak, 12 year old, 40%\n2041 The Glenlivet Guardians Single Cask 18 year ol...\n1953 The Glenlivet Nadurra Oloroso, 48%\n333 The Glenlivet Nadurra Triumph 1991, 48%\n466 The Glenlivet Nàdurra, 16 year old, 57.2%\n118 The Glenlivet XXV, 25 year old, 43%\n858 The Glenrothes (Vintage Cask #3) 1998 Vintage,...\n1683 The Glenrothes Alba Reserve, 40%\n123 The Glenrothes “John Ramsay,” 46.7%\n852 The Glenrothes, 1974 Vintage, 29 year old, 43%\n122 The Glenrothes, 1975 vintage, 43%\n477 The Glenrothes, 1979 Vintage, 43%\n660 The Glenrothes, 1985 Vintage, 43%\n476 The Glenrothes, 1987, 43%\n1075 The Glenrothes, 1989, 43%\n1494 The Glenrothes, 1992 Vintage, 43% ABV\n1496 The Glenrothes, 1998 vintage, 43%\n217 The Glenturret Jamieson's Jigger Edition, 43.1%\n757 The Golden Age Blend, 44.3%\n758 The Half Century Blend, 45.5%\n41 The John Walker, 40%\n12 The Last Drop (distilled at Lochside) 1972 (ca...\n359 The Last Drop 1965 Blend 48 year old, 48.6%\n147 The Last Drop 1971, 47.2%\n146 The Last Drop 50 year old Double Matured, 51.8%\n36 The Last Drop 50 year old, 50.9%\n35 The Last Drop, 1960 vintage, 48 year old, 52%\n554 The Lost Distilleries Blend (Batch Four), 50.9%\n1591 The Lost Distilleries Blend (batch 6), 49.3%\n2224 The Lost Distillery Company Auchnagie (batch 2...\n1767 The Lost Distillery Company Classic Selection ...\n1402 The Lost Distillery Company Classic Selection ...\n560 The Lost Distillery Company Classic Selection ...\n1596 The Lost Distillery Company Gerston (batch 2/1...\n1915 The Lost Distillery Company Stratheden (batch ...\n2043 The MacPhail's Collection (distilled at Glentu...\n965 The MacPhail’s Collection 8 year old (distille...\n72 The Macallan 1824 Limited Release, 48%\n492 The Macallan 1861 Replica, 42.7%\n235 The Macallan Cask Strength, 58.6%\n497 The Macallan Double Cask 12 year old, 40%\n498 The Macallan Edition No. 2, 48.2%\n32 The Macallan Estate Reserve, 45.7%\n1855 The Macallan Fine Oak, 10 year old, 40%\n236 The Macallan Fine Oak, 15 year old, 43%\n71 The Macallan Fine Oak, 17 year old, 43%\n684 The Macallan Fine Oak, 21 year old, 43%\n880 The Macallan Fine Oak, 30 year old, 43%\n2234 The Macallan Lalique Decanter, 55 year old, 40.1%\n1528 The Macallan Select Oak, 40%\n1856 The Macallan Whisky Maker's Edition, 42.8%\n233 The Macallan, 15 year old, 43%\n2125 The Macallan, 1841 Replica Bottling, 41.7%\n1093 The Macallan, 1951 Vintage, 48.8%\n234 The Macallan, 1961 Vintage, 54.1%\n133 The Macallan, 50 year old, 43%\n747 The Maltman (distilled at Ben Nevis) 17 year o...\n2202 The Maltman (distilled at Linkwood) 18 year ol...\n349 The McGibbons Provenance (distilled at Port El...\n725 The Naked Grouse, 40%\n557 The Pearls of Scotland (distilled at Invergord...\n957 The Pearls of Scotland (distilled at North of ...\n162 The Scotch Malt Whisky Society Pigs in Plaster...\n2192 The Singleton of Dufftown 28 year old, 52.3%\n2066 The Singleton of Glendullan Classic, 40%\n2146 The Singleton of Glendullan Double Matured, 40%\n1893 The Singleton of Glendullan Master’s Art, 40%\n1827 The Singleton of Glendullan, 12 year old, 40%\n1895 The Sovereign (distilled at Cambus) 26 year ol...\n272 The Sovereign (distilled at Cambus) 30 year ol...\n931 The Sovereign (distilled at Cambus) 40 year ol...\n366 The Sovereign (distilled at Cambus) 50 year ol...\n958 The Sovereign (distilled at Dumbarton) 1964 50...\n561 The Sovereign (distilled at Dumbarton) 50 year...\n2159 The Sovereign (distilled at Invergordon) 50 ye...\n1594 The Sovereign (distilled at North British) 25 ...\n2158 The Sovereign (distilled at Port Dundas) 25 ye...\n953 The Sovereign (distilled at Port Dundas) 26 ye...\n1979 The Speyside, 12 year old, 43% ABV\n749 The Tweeddale 12 year old Batch 3, 46%\n1910 The Tweeddale 14 year old Batch 4, 46%\n552 The Tweeddale Last Centennial, 52.2%\n1173 The Tweeddale Single Lowland Malt 14 year old,...\n743 The Whisky Exchange (distilled at Bunnahabhain...\n1090 The Whisky Exchange (distilled at Linkwood) 16...\n1809 The Whisky Exchange Elements of Islay Br2 (dis...\n914 The Whisky Exchange Elements of Islay Kh1 (dis...\n692 The Whisky Exchange Elements of Islay Pe5 (dis...\n97 The Whisky Exchange Masterpieces Range (distil...\n670 Thorfinn, 45.1%\n1400 Timorous Beastie 10 year old, 46.8%\n559 Timorous Beastie 18 year old, 46.8%\n373 Timorous Beastie 40 year old, 54.7%\n1399 Timorous Beastie, 46.8%\n1709 Tobermory 15 year old, 46.3%\n901 Tobermory 21 year old, 53.8%\n1114 Tobermory, 10 year old, 40%\n1714 Tomatin 12 year old French Oak Finish, 46%\n2133 Tomatin 12 year old, 43%\n1329 Tomatin 14 year old Port Wood Finish, 46%\n1328 Tomatin 15 year old, 52%\n254 Tomatin 1971 Warehouse 6 Collection, 45.8%\n1118 Tomatin 1988 Vintage, 46%\n1711 Tomatin 30 year old, 46%\n906 Tomatin 36 year old, 46%\n1717 Tomatin Cabernet Sauvignon 2002, 46%\n1718 Tomatin Caribbean Rum 2007, 46%\n1867 Tomatin Cask Strength Edition, 57.5%\n1716 Tomatin Contrasting Casks Bourbon Matured, 46%\n1333 Tomatin Contrasting Casks Sherry Matured, 46%\n708 Tomatin Cù Bòcan 1988 Limited Edition, 51.5%\n1119 Tomatin Cù Bòcan 1989 Limited Edition, 53.2%\n1334 Tomatin Cù Bòcan The Bourbon Edition, 46%\n1541 Tomatin Cù Bòcan, 46%\n1981 Tomatin Decades, 46%\n1712 Tomatin Legacy, 43%\n1335 Tomatin Oloroso Sherry 1995, 46%\n1866 Tomatin, 15 year old, 43%\n1115 Tomatin, 18 year old, 43%\n904 Tomatin, 18 year old, 46%\n253 Tomatin, 1973 Vintage (Cask #25602), 44%\n905 Tomatin, 1982 Vintage (Cask #92), 57%,\n2187 Tomatin, 25 year old, 43%\n1116 Tomatin, 30 year old, 1976 vintage, 49.3%\n355 Tomatin, 40 year old, 1967 vintage, 42.9%\n1715 Tomatin, 57.2%\n2186 Tomatin,12 year old, 43%\n1710 Tomintoul 14 year old, 46%\n517 Tomintoul 25 year old, 43%\n354 Tomintoul Five Decades, 50%\n252 Tomintoul Reserve 37 year old, 43%\n516 Tomintoul Stillman's Dram, 33 year old, 45%\n2239 Tomintoul, 'With a peaty tang', 40%\n1327 Tomintoul, 16 year old, 40% ABV\n902 Tomintoul, 27 year old, 40% ABV\n707 Tomintoul, 33 year old, 43%\n903 Tomintoul, Oloroso Sherry Cask Finish, 12 year...\n2134 Tormore 14 year old, 43%\n2188 Tormore 16 year old, 48%\n1982 Tormore, 12 year old, 40% ABV\n2117 Trader Joe's (distilled at Imperial), 12 year ...\n1437 Traditional Ben Nevis, 46%\n1191 Treacle Chest, 46%\n356 Tullibardine 1952, 40.2%\n1720 Tullibardine 1988, 46%\n1337 Tullibardine 20 year old, 43%\n1984 Tullibardine 225 Sauternes, 43%\n1870 Tullibardine 228 Burgundy, 43%\n1548 Tullibardine 25 year old, 43%\n1721 Tullibardine 500 Sherry, 43%\n1547 Tullibardine Aged Oak Edition, 46%\n1722 Tullibardine Custodians’ Release 1970, 40.5%\n1869 Tullibardine John Black 6 Single Cask 1993, 55.1%\n1545 Tullibardine Sherry Wood Finish, 1993 vintage,...\n1871 Tullibardine Sovereign, 43%\n1549 Tullibardine The Murray, 56.1%\n1336 Tullibardine Vintage 1993, 46%\n1546 Tullibardine, 1988 vintage, 46%\n1983 Tullibardine, 1992 Vintage, 46%\n1868 Tullibardine, 1993 Vintage, 40%\n1602 Usquaebach 15 year old, 43%\n376 Usquaebach An Ard Ri Cask Strength, 57.1%\n1918 Usquaebach Old Rare Stone Flagon, 43%\n2015 Usquaebach Reserve, 43%\n1340 Wemyss Lord Elcho 15 year old, 40%\n912 Wemyss Malt Caribbean Crème (distilled at Inve...\n1339 Wemyss Malts (distilled at Auchentoshan) ‘Lemo...\n2056 Wemyss Malts (distilled at Aultmore) 20 year o...\n1130 Wemyss Malts (distilled at Blair Athol) Autumn...\n2137 Wemyss Malts (distilled at Bowmore) 1982 Lochi...\n1875 Wemyss Malts (distilled at Bunnahabhain) 1991 ...\n1876 Wemyss Malts (distilled at Bunnahabhain) 1991 ...\n2031 Wemyss Malts (distilled at Caol Ila) 16 year o...\n2138 Wemyss Malts (distilled at Caol Ila) 1982 Smok...\n1556 Wemyss Malts (distilled at Clynelish) “A Day a...\n1558 Wemyss Malts (distilled at Clynelish) “A Day a...\n1557 Wemyss Malts (distilled at Glen Scotia) 'Straw...\n910 Wemyss Malts (distilled at Glenrothes) 1988 Ar...\n1131 Wemyss Malts (distilled at Invergordon) Mocha ...\n1127 Wemyss Malts (distilled at Tormore) 1988 Flora...\n1238 Wemyss Malts Apple Basket 1997 (distilled at C...\n1342 Wemyss Malts At Anchor in a Cove 1991 (distill...\n1727 Wemyss Malts Banquet of Fruits 1994 (distilled...\n1877 Wemyss Malts Bench with a Sea View (distilled ...\n1066 Wemyss Malts Brandy Casket (distilled at Glen ...\n716 Wemyss Malts Cacao Geyser (distilled at Bowmor...\n1560 Wemyss Malts Candied Fruit (distilled at Auche...\n1726 Wemyss Malts Cayenne Cocoa Bean 1997 (distille...\n1341 Wemyss Malts Chocolate Honeycomb (distilled at...\n1344 Wemyss Malts Foraged Fruit Fool (distilled at ...\n1559 Wemyss Malts Fruit Bonbons (distilled at Glen ...\n1125 Wemyss Malts Ginger Spice (distilled at Glenro...\n1126 Wemyss Malts Heathery Smoke (distilled at Caol...\n717 Wemyss Malts Kiln Embers, 46%\n1878 Wemyss Malts Kirsch Gateau (distilled at Bunna...\n1759 Wemyss Malts Lead on Macduff! 2002 (distilled ...\n1208 Wemyss Malts Lemon Zest 1998 (distilled at Auc...\n1561 Wemyss Malts Melon Vine 1994 (distilled at Abe...\n830 Wemyss Malts Merchant’s Mahogany Chest (distil...\n1347 Wemyss Malts Nuts about Pears 1991 (distilled ...\n1562 Wemyss Malts Peaches and Cream 1989 (distilled...\n1349 Wemyss Malts Peat Chimney, 46%\n2097 Wemyss Malts Rhubarb Royale (distilled at Benr...\n1729 Wemyss Malts Rosy Apple Brûlée (distilled at ...\n909 Wemyss Malts Salted Caramels (distilled at Gle...\n1345 Wemyss Malts Seville Bazaar 1991 (distilled at...\n1343 Wemyss Malts Snuffed Candle (distilled at Aber...\n1879 Wemyss Malts Spice King, 46%\n1654 Wemyss Malts Spiced Chocolate Cup (distilled a...\n1128 Wemyss Malts Stem Ginger Preserve (distilled a...\n1563 Wemyss Malts Summer Fruit Cup 1998 (distilled ...\n1207 Wemyss Malts Tarte au Citron (distilled at Auc...\n1346 Wemyss Malts The Highland Mariner (distilled a...\n1566 Wemyss Malts The Hive, 46%\n612 Wemyss Malts The Rockpool (distilled at Bowmor...\n1032 Wemyss Malts Toffee Glaze 1997 (distilled at C...\n1564 Wemyss Malts Toffee Tuile (distilled at Aberfe...\n1348 Wemyss Malts Vanilla Summer Single Cask, 46%\n1728 Wemyss Malts Velvet Fig, 46%\n1880 Wemyss Malts Vintage Strawberry Punnet (distil...\n1129 Wemyss Malts Waffles and Ice Cream (distilled ...\n1725 Wemyss Malts ‘Caribbean Fruits,’ 46%\n908 Wemyss Malts ‘Honey Spice’ (distilled at Bunna...\n1874 Wemyss Malts ‘Lemon Grove’ (distilled at Cragg...\n1555 Wemyss Malts ‘Rum ‘n’ Raisin’ Single Cask, 46%\n715 Wemyss Malts ‘Winter Spice’ Single Cask, 44.4%\n1565 Wemyss The Hive 12 year old, 40%\n1990 Wemyss The Peat Chimney 12 year old, 40%\n2139 Wemyss The Spice King 12 Year Old, 40%\n911 Wemyss Vintage Malt Ginger Compote, 1996, 46%\n1988 Wemyss Vintage Malts 'The Peat Chimney,' 8 yea...\n2189 Wemyss Vintage Malts 'The Smooth Gentleman,' 8...\n1989 Wemyss Vintage Malts 'The Spice King,' 8 year ...\n1124 Wemyss “A Matter of Smoke” 15 year old, 46%\n1447 Whisky Galore (distilled at Bowmore), 16 year ...\n1864 Whisky Galore (distilled at Old Pulteney), 16 ...\n48 White Bowmore, 1964 vintage, 43 year old, 42.8%\n1001 Wild Scotsman (distilled at Ben Nevis), 13 yea...\n884 Wild Scotsman (distilled at Mortlach), 1994 vi...\n1785 Wild Scotsman, 15 year old, 46%\n267 William Grant Rare Cask Reserves Ghosted Reser...\n1372 William Grant’s 18 year old Rare, 40%\n936 William Grant’s 25 year old, 40%\n1920 Wolfburn Single Malt, 46%\n1776 Wolfburn The Kylver Series, 46%\n761 anCnoc 2002, 46%\n379 anCnoc 35 year old Second Release, 41%\n567 anCnoc Cutter, 46%\n966 anCnoc Flaughter, 46%\n1188 anCnoc Rutter, 46%\n1554 “Double Malt” Speyside (Second Edition), 46%\nName: name, dtype: object\n"
],
[
"# actual proper nouns that I need to remove - manually inspect proper nouns list and names of the whiskeus\nproper_nouns_list = [\n 'aberfeldy', 'aberlour', 'alltabhainne', 'ardbeg', 'auchentoshan', 'auchentoshans', 'asyla', 'alexander', 'ardmore',\n 'arran', 'auchroisk', 'aultmore', 'askaig', 'antiquary', 'adelphi', \n 'ballechin', 'balvenie', 'benachie', 'benriachs', 'bladnoch', 'bladnocha', 'brora', 'broras', 'bowmore', 'bull',\n 'bruichladdich', 'bruichladdichit', 'bruichladdichs', 'bunnahabhain', 'balblair', 'ballantine', 'nevis', 'benriach', \n 'benrinnes', 'benromach', 'balmenach', 'blackadder', 'blair', 'boutique', 'box', 'binnys', \"binny's\",\n 'cardhu', 'chivas', 'clynelishs', 'clynelish', 'craigellachie', 'cragganmore', 'cadenhead', 'caol', 'ila', \n 'chieftain', 'compass', 'cuatro', 'cutty', 'collection',\n 'deanston', 'dailuaine', 'dalmore', 'dalwhinnie', 'dewars', 'dewar', 'deveron', 'douglas', 'duncan', 'dufftown', \n 'edradour', 'edradours', 'ellen', 'editors', \n 'farclas', \n 'garioch', 'gariochs', 'glenallechie','glenburgie', 'glencadam', 'glencraig', 'glendronach', 'glendronachs', \n 'glenfiddich', 'glenfiddichs', 'glengoyne', 'glenisla', 'glenkeir', 'glenrothes', 'glenlivet', 'glenturret', \n 'glenfarclas', 'glenglassaugh', 'glenkinchie', 'glenmorangie', 'glenugie', 'gordon',\n 'hart', 'hazelburn', 'highland', 'hazelwood', 'hunter', \n 'inchmurrin', \n 'johnnie', 'jura', 'juras',\n 'keith', 'kensington', 'kilchomanfree', 'kilchomans', 'kildalton', 'kinchie', 'kininvie', 'kirkland', \n 'lochnager', 'lochranza', 'lagavulin', 'littlemill', 'linkwood', 'longmorn', 'linlithgow', 'laphroig', 'ledaig',\n 'lomand', 'lombard', 'lonach', 'longrow', \n 'macduff', 'macmillans', 'magdelene', 'macallan', 'mortlach', 'monnochmore', 'macdougall', 'mossman', 'mackillops',\n 'mackinlays', 'master', 'murray', 'mcdavid', \n 'oban', \n 'park', 'pulteney', 'peerless', \n 'scapa', 'shackleton', 'shieldaig', 'skye', 'springbank', 'springbanks', 'strathclyde', 'strathisla', 'scotia',\n 'signatory', 'scotts', 'singleton', 'speyburn', 'strathmill', 'sovereign', \n 'talisker', 'tomintoul', 'turasmara', 'teaninich', 'taylor', 'tobermory', 'tomatin', 'tormore', 'tullibardine', \n 'uigeadail', 'usquaebach', \n 'valinch', \n 'walker', 'wemyss'\n ]",
"_____no_output_____"
],
[
"# Adding adjectives and names of distillers to the list of stop words\nmy_stop_words = text.ENGLISH_STOP_WORDS.union(proper_nouns_list, adj_to_remove)\n\n# Saving custom stop words to list\njoblib.dump(my_stop_words, 'data/my_stop_words.pkl')",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
]
] |
cbd9258ff21099cc0dbe31e9d91083d600f68e42
| 68,171 |
ipynb
|
Jupyter Notebook
|
ipythons/executions.ipynb
|
KarizCache/serverless
|
c5735afee29e104f3909f3b0140e993d461a5420
|
[
"MIT"
] | null | null | null |
ipythons/executions.ipynb
|
KarizCache/serverless
|
c5735afee29e104f3909f3b0140e993d461a5420
|
[
"MIT"
] | null | null | null |
ipythons/executions.ipynb
|
KarizCache/serverless
|
c5735afee29e104f3909f3b0140e993d461a5420
|
[
"MIT"
] | null | null | null | 127.42243 | 45,716 | 0.806017 |
[
[
[
"import os\nimport json\nimport re\nimport ast\nimport json\nfrom graphviz import Digraph\nimport pandas as pd\n\n# color the graph\nimport graph_tool.all as gt\nimport copy\nimport matplotlib.colors as mcolors\nimport sys\nimport matplotlib.colors as mcolors\n\n\n\nimport matplotlib.pyplot as plt\nimport numpy as np\nimport seaborn as sns\nsns.set_style(\"whitegrid\")\nimport pandas as pd\nimport os\nimport json\nimport itertools\nimport matplotlib\nimport matplotlib.ticker as ticker\n\nworker_color = {'10.255.23.108': '#e41a1c',\n '10.255.23.109': '#984ea3',\n '10.255.23.110': '#ff7f00',\n '10.255.23.115': '#4daf4a'}\n\nresults_dir = '/local0/serverless/serverless/ipythons/plots'\nstats_dir='/opt/dask-distributed/benchmark/stats'",
"/usr/lib/python3/dist-packages/graph_tool/draw/cairo_draw.py:1495: RuntimeWarning: Error importing Gtk module: cannot import name '_gi' from partially initialized module 'gi' (most likely due to a circular import) (/usr/lib/python3/dist-packages/gi/__init__.py); GTK+ drawing will not work.\n warnings.warn(msg, RuntimeWarning)\n"
],
[
"# get all benchmarks\ndef get_benchmarks():\n benchmarks = {}\n for _file in os.listdir(stats_dir):\n try:\n app = _file.split('.', 1)[0]\n assert os.path.isfile(os.path.join(stats_dir, f'{app}.g')) \\\n and os.path.isfile(os.path.join(stats_dir, f'{app}.json')) \\\n and os.path.isfile(os.path.join(stats_dir, f'{app}.colors'))\n\n bnch, scheduler, _ = app.split('_', 2)\n #scheduler = 'vanilla'\n \n benchmarks[app] = {'app': app, 'scheduler': scheduler, 'benchmark': bnch}\n except AssertionError:\n pass\n return benchmarks\n\n",
"_____no_output_____"
],
[
"\nworker_color = {'10.255.23.108': '#e41a1c',\n '10.255.23.109': '#984ea3',\n '10.255.23.110': '#ff7f00',\n '10.255.23.115': '#4daf4a'}\n\n\n \ndef color_assignment(benchmark, task_style):\n cfile = f'/opt/dask-distributed/benchmark/stats/{benchmark}.colors'\n with open(cfile, 'r') as cfd:\n raw = cfd.read().split('\\n')\n for ln in raw:\n if not ln: \n continue\n task_name, actual = ln.split(',')\n if task_name not in task_style:\n task_style[task_name] = {}\n task_style[task_name]['actual'] = actual\n \n \ndef build_graph(benchmark, task_style):\n css_colors = list(mcolors.CSS4_COLORS.keys())\n gfile = f'/opt/dask-distributed/benchmark/stats/{benchmark}.g'\n\n with open(gfile, 'r') as fd:\n raw = fd.read().split('\\n')\n g = gt.Graph(directed=True)\n vid_to_vx = {}\n name_to_vid = {}\n\n g.vertex_properties['name'] = g.new_vertex_property(\"string\")\n g.vertex_properties['color'] = g.new_vertex_property(\"string\")\n g.vertex_properties['worker'] = g.new_vertex_property(\"string\")\n g.vertex_properties['icolor'] = g.new_vertex_property(\"int\")\n g.vertex_properties['simcolor'] = g.new_vertex_property(\"string\")\n g.vertex_properties['isimcolor'] = g.new_vertex_property(\"string\")\n for ln in raw:\n if ln.startswith('v'):\n _, vid, name = ln.split(',', 2)\n v = g.add_vertex()\n vid_to_vx[vid] = v\n name_to_vid[name] = vid\n\n g.vp.name[v] = name\n try:\n g.vp.icolor[v] = int(task_style[name]['actual'])\n #if g.vp.icolor[v] >= len(css_colors):\n #g.vp.color[v] = mcolors.CSS4_COLORS[css_colors[0]]\n #else:\n g.vp.color[v] = mcolors.CSS4_COLORS[css_colors[int(task_style[name]['actual'])]]\n except KeyError:\n print(f'Keyerror for {name}')\n raise NameError('Error')\n #g.vp.color[v] = 'yellow'\n #g.vp.icolor[v] = 2\n\n for ln in raw: \n if ln.startswith('e'):\n _, vsrc, vdst, _ = ln.split(',', 3)\n g.add_edge(vid_to_vx[vsrc], vid_to_vx[vdst])\n return g\n \nkeys = list(mcolors.CSS4_COLORS.keys())\ndef update_runtime_state(benchmark, g, task_style):\n print('process', benchmark)\n tasks = []\n jfile = f'/opt/dask-distributed/benchmark/stats/{benchmark}.json'\n with open(jfile, 'r') as fd:\n stats = ast.literal_eval(fd.read())\n\n #print(json.dumps(stats, indent=4))\n print('stat size is', len(stats))\n \n min_ts = sys.maxsize\n for s in stats:\n task_style[s]['output_size'] = stats[s]['msg']['nbytes']\n task_style[s]['input_size'] = 0\n task_style[s]['remote_read'] = 0\n task_style[s]['local_read'] = 0\n task_style[s]['worker'] = stats[s]['worker'].split(':')[1].replace('/', '')\n \n startsstops = stats[s]['msg']['startstops']\n for ss in startsstops:\n if ss['action'] == 'inputsize': continue\n if ss['action'] == 'compute':\n task_style[s]['compute_end'] = ss['stop']\n task_style[s]['compute_start'] = ss['start']\n task_style[s]['runtime'] = ss['stop'] - ss['start']\n if ss['start'] < min_ts: min_ts = ss['start']\n if ss['stop'] < min_ts: min_ts = ss['stop']\n \n print(min_ts)\n for s in stats:\n startsstops = stats[s]['msg']['startstops']\n min_start = sys.maxsize\n max_end = 0\n\n transfer_stop = 0\n transfer_start = sys.maxsize\n for ss in startsstops:\n if ss['action'] == 'inputsize': continue\n if ss['start'] < min_start: min_start = ss['start']\n if ss['stop'] > max_end: max_end = ss['stop']\n \n if ss['action'] == 'compute':\n compute_stop = ss['stop']\n compute_start = ss['start']\n run_time = ss['stop'] - ss['start']\n \n if ss['action'] == 'transfer':\n #print(ss['start'] - min_ts, ss['stop'] - min_ts)\n if ss['start'] < transfer_start:\n transfer_start = ss['start']\n if ss['stop'] > transfer_stop:\n transfer_stop = ss['stop']\n #print('transfer start', ss['start'] - min_ts, \n # 'transfer stop', ss['stop'] - min_ts)\n \n \n if transfer_stop == 0: \n transfer_stop = compute_start\n transfer_start = compute_start\n #print('***transfer start', transfer_start - min_ts, \n # '****transfer stop', transfer_stop - min_ts)\n tasks.append({'name': s, 'start_ts': min_start - min_ts, 'end_ts': max_end - min_ts, \n 'compute_start': compute_start - min_ts, 'compute_stop': compute_stop - min_ts,\n 'transfer_start': transfer_start - min_ts, 'transfer_stop': transfer_stop - min_ts,\n 'worker': stats[s]['worker'].split(':')[1].replace('/', '')})\n #print('\\n')\n \n \n \n #total amount of data accessed, data accessed remotely, data accessed locally\n for v in g.vertices():\n for vi in v.in_neighbors():\n task_style[g.vp.name[v]]['input_size'] += task_style[g.vp.name[vi]]['output_size']\n if task_style[g.vp.name[v]]['worker'] == task_style[g.vp.name[vi]]['worker']:\n task_style[g.vp.name[v]]['local_read'] += task_style[g.vp.name[vi]]['output_size']\n else:\n task_style[g.vp.name[v]]['remote_read'] += task_style[g.vp.name[vi]]['output_size']\n \n for v in g.vertices():\n g.vp.worker[v] = task_style[g.vp.name[v]]['worker']\n #g.vp.color[v] = colors[task_style[g.vp.name[v]]['worker']]\n \n #Check the slack for the prefetching\n bw = 10*(1<<27) # 10 Gbps (1<<30)/(1<<3)\n not_from_remote = 0\n for v in g.vertices():\n parents_end = []\n for vi in v.in_neighbors():\n parents_end.append(task_style[g.vp.name[vi]]['compute_end'])\n \n if len(parents_end):\n max_end = max(parents_end)\n \n for vi in v.in_neighbors():\n if max_end == task_style[g.vp.name[vi]]['compute_end'] and task_style[g.vp.name[vi]]['worker'] != task_style[g.vp.name[v]]['worker']:\n #print(f'Slack come from local chain')\n not_from_remote += 1\n \n\n #print(f'slack for {g.vp.name[v]}: {round(1000*(max(parents_end) - min(parents_end)), 2)}msec',\n # '\\t runtime:', round(1000*task_style[g.vp.name[vi]]['runtime'], 4), 'msec',\n # '\\t remote read', task_style[g.vp.name[v]]['remote_read']/bw)\n #print(not_from_remote)\n return tasks\n\n\n\ndef plot_graph(g):\n policy = b.split('_')[1]\n print('policy is', policy)\n dg = Digraph('G', filename=f'{b}.gv', format='png')\n for v in g.vertices():\n #print(g.vp.name[v])\n \n vname = g.vp.name[v].split('-', 1)\n vname = vname[1] if len(vname) > 1 else vname[0]\n \n \n #dg.attr('node', shape='ellipse', style='filled', color=g.vp.color[v])\n dg.attr('node', shape='ellipse', style=\"filled,solid\",\n penwidth=\"3\",\n fillcolor=g.vp.color[v] if \"chaincolor\" in policy else \"#f0f0f0\",\n color=worker_color[g.vp.worker[v]])\n \n color = '-'\n if 'chaincolor' in policy:\n color = g.vp.icolor[v]\n dg.node(f'{vname}, color({color})')\n #print(g.vp.name[v], g.vp.icolor[v], g.vp.worker[v])\n for e in g.edges():\n vname = g.vp.name[e.source()].split('-', 1)\n sname = vname[1] if len(vname) > 1 else vname[0]\n \n vname = g.vp.name[e.target()].split('-', 1)\n tname = vname[1] if len(vname) > 1 else vname[0]\n \n dg.edge(f'{sname}, color({g.vp.icolor[e.source()] if \"chaincolor\" in policy else \"-\"})', \n f'{tname}, color({g.vp.icolor[e.target()] if \"chaincolor\" in policy else \"-\"})')\n dg.view(f'/local0/serverless/serverless/ipythons/plots/{b}', quiet=False)\n\n \ndef plot_graph2(g):\n policy = b.split('_')[1]\n dg = Digraph('G', filename=f'{b.split(\"_\")[0]}.gv', format='png')\n for v in g.vertices():\n vname = g.vp.name[v].split('-', 1)\n vname = vname[1] if len(vname) > 1 else vname[0]\n \n dg.attr('node', shape='ellipse', style='filled', color='#252525')\n dg.attr('node', shape='ellipse', style=\"filled,solid\",\n penwidth=\"3\",\n fillcolor= \"#f0f0f0\",\n color='#252525')\n \n dg.node(f'{vname}')\n for e in g.edges():\n vname = g.vp.name[e.source()].split('-', 1)\n sname = vname[1] if len(vname) > 1 else vname[0]\n \n vname = g.vp.name[e.target()].split('-', 1)\n tname = vname[1] if len(vname) > 1 else vname[0]\n \n dg.edge(f'{sname}', \n f'{tname}')\n\n dg.view(f'/local0/serverless/serverless/ipythons/plots/{b.split(\"_\")[0]}', quiet=False)\n\n \n\ndef format_xticks(x, pos=None):\n return x\n #return str(int(x*1000))\n \ndef plot_gannt_chart(tasks, benchmark):\n #_max = df['runtime'].max()\n #print(benchmark, df['runtime'].max())\n sns.set_style(\"ticks\")\n sns.set_context(\"paper\", font_scale=1)\n sns.set_context(rc = {'patch.linewidth': 1.5, 'patch.color': 'black'})\n\n plt.rc('font', family='serif')\n\n fig, ax = plt.subplots(figsize=(10,8))\n ax.set_xlabel('Time (second)', fontsize=16)\n\n sns.despine()\n ax.yaxis.grid(color='#99999910', linestyle=(0, (5, 10)), linewidth=0.4)\n\n ax.set_axisbelow(True)\n ax.tick_params(axis='both', which='major', labelsize=14)\n ax.yaxis.set_ticks_position('both')\n ax.xaxis.set_major_formatter(ticker.FuncFormatter(format_xticks))\n ax.set_yticklabels([])\n #ax.set_xlim([0, _max])\n\n # Setting graph attribute\n ax.grid(True)\n \n \n base = 0\n size = 8\n margin = 3\n workers_load={}\n for ts in tasks:\n #print(json.dumps(ts, indent=4))\n #print(ts['name'], ts['start_ts'], ts['end_ts'], ts['worker'])\n if ts['worker'] not in workers_load:\n workers_load[ts['worker']] = []\n #workers_load[ts['worker']].append((ts['start_ts'], ts['end_ts'] - ts['start_ts']))\n ax.broken_barh([(ts['transfer_start'], ts['transfer_stop'] - ts['transfer_start'])], (base, size), \n edgecolors =worker_color[ts['worker']], facecolors = worker_color[ts['worker']])\n ax.broken_barh([(ts['compute_start'], ts['compute_stop'] - ts['compute_start'])], (base, size), \n edgecolors = worker_color[ts['worker']], facecolors='#f0f0f0')\n \n vname = ts['name'].split('-', 1)\n sname = vname[1] if len(vname) > 1 else vname[0]\n \n ax.text(x=ts['compute_start'] + (ts['compute_stop'] - ts['compute_start'])/2, \n y=base + size/2,\n s=sname, \n ha='center', \n va='center',\n color='black')\n \n base += (size + margin)\n jobname, policy, _ = benchmark.split('_')\n #ax.axes.yaxis.set_visible(False)\n ax.set_ylabel(f'{benchmark}', fontsize=14)\n ax.set_title(f'{jobname}, {policy}', fontsize=16)\n leg1 = ax.legend(['10.255.23.108', '10.255.23.109', '10.255.23.109', '10.255.23.115'], \n title='Communication cost', ncol = 1, loc='lower center')\n leg1.legendHandles[0].set_color(worker_color['10.255.23.108'])\n leg1.legendHandles[1].set_color(worker_color['10.255.23.109'])\n leg1.legendHandles[2].set_color(worker_color['10.255.23.110'])\n leg1.legendHandles[3].set_color(worker_color['10.255.23.115'])\n\n leg2 = ax.legend(['10.255.23.108', '10.255.23.109', '10.255.23.109', '10.255.23.115'], \n title='Computation cost',loc='lower right')\n leg2.legendHandles[0].set_edgecolor(worker_color['10.255.23.108'])\n leg2.legendHandles[1].set_edgecolor(worker_color['10.255.23.109'])\n leg2.legendHandles[2].set_edgecolor(worker_color['10.255.23.110'])\n leg2.legendHandles[3].set_edgecolor(worker_color['10.255.23.115'])\n leg2.legendHandles[0].set_facecolor('#f0f0f0')\n leg2.legendHandles[1].set_facecolor('#f0f0f0')\n leg2.legendHandles[2].set_facecolor('#f0f0f0')\n leg2.legendHandles[3].set_facecolor('#f0f0f0')\n \n #end_2_end = df[df['exp'] == benchmark]['runtime']\n #ax.vlines(end_2_end, ymin=0, ymax=base, label = end_2_end, linestyles='--', color='#bdbdbd')\n \n ax.add_artist(leg1)\n fig.savefig(f'{os.path.join(\"/local0/serverless/serverless/ipythons/plots\",benchmark)}.gannt.png', \n format='png', dpi=200)\n plt.show()\n\ndef plot_runtime(name, df):\n sns.set_style(\"ticks\")\n \n sns.set_context(\"paper\", font_scale=1)\n sns.set_context(rc = {'patch.linewidth': 1.5, 'patch.color': 'black'})\n\n plt.rc('font', family='serif')\n\n fig, ax = plt.subplots(figsize=(8,5))\n sns.barplot(x='scheduler', y='runtime', #, hue = 'scheduler', \n order = ['optimal', 'vanilla', 'optplacement', 'chaincolorrr', 'chaincolorch', 'random'],\n palette = ['#fbb4ae','#fdb462', '#8dd3c7', '#80b1d3', '#bebada' , '#fb8072'],\n data=df, ax=ax)\n sns.despine()\n\n ax.yaxis.grid(color='#99999910', linestyle=(0, (5, 10)), linewidth=0.4)\n\n ax.set_axisbelow(True)\n ax.tick_params(axis='x', which='major', labelsize=14, rotation=15)\n ax.tick_params(axis='y', which='major', labelsize=14)\n ax.yaxis.set_ticks_position('both')\n \n ax.set_ylabel('Runtime (sec)', fontsize=16)\n ax.set_xlabel(f'Benchmark: {df[\"benchmark\"].unique()[0]}', fontsize=16)\n #ax.legend(fontsize=14, ncol=3)\n\n plt.tight_layout()\n fig.savefig(f'/local0/serverless/serverless-sim/results/runtime_{name}.png', \n format='png', dpi=200)\n plt.show()\n\n\n#dt = pd.read_csv('/local0/serverless/task-bench/dask/stats.csv')\n#dt.head(5)\n#gdf = dt.groupby('name')\n\nbenchmarks = ['tree.w10.s8.1G1B_vanilla_3d70b39c'] #get_benchmarks()\n\n#for name, df in gdf:\n #plot_runtime(name, df)\n #continue\n #if 'spread' not in name: continue\nfor b in benchmarks:\n #if name not in b: continue \n task_style = {}\n color_assignment(b, task_style)\n g = build_graph(b, task_style)\n tasks = update_runtime_state(b, g, task_style)\n plot_graph(g)\n plot_graph2(g)\n plot_gannt_chart(tasks, b)\n #break\n #break",
"process tree.w10.s8.1G1B_vanilla_3d70b39c\nstat size is 25\n1625147405.2534475\npolicy is vanilla\n"
]
]
] |
[
"code"
] |
[
[
"code",
"code",
"code"
]
] |
cbd92818c6194b7a69fe7bb8938b0c53df2dd201
| 10,261 |
ipynb
|
Jupyter Notebook
|
Python_Basics.ipynb
|
Antonyjovin24/Business-analytics-with-Python
|
b056eef6bc9c829f2a745dba1cfda729b6158411
|
[
"Apache-2.0"
] | null | null | null |
Python_Basics.ipynb
|
Antonyjovin24/Business-analytics-with-Python
|
b056eef6bc9c829f2a745dba1cfda729b6158411
|
[
"Apache-2.0"
] | null | null | null |
Python_Basics.ipynb
|
Antonyjovin24/Business-analytics-with-Python
|
b056eef6bc9c829f2a745dba1cfda729b6158411
|
[
"Apache-2.0"
] | null | null | null | 21.244306 | 67 | 0.381444 |
[
[
[
"4*5",
"_____no_output_____"
],
[
"7-9",
"_____no_output_____"
],
[
"5/6",
"_____no_output_____"
],
[
"x=10",
"_____no_output_____"
],
[
"print(x)\n\n",
"10\n"
],
[
"x=10\ny=12\nx*y",
"_____no_output_____"
],
[
"x/y",
"_____no_output_____"
],
[
"x=\"Jovin\"\nprint(x)",
"Jovin\n"
],
[
"x=10\ny=20.5\nz=\"Jovin\"",
"_____no_output_____"
],
[
"type(x)\n",
"_____no_output_____"
],
[
"type(y)",
"_____no_output_____"
],
[
"type(z)",
"_____no_output_____"
],
[
"x=\"Jovin\"\ny=\"Rusail\"\nx+y",
"_____no_output_____"
],
[
"x=input(\"Enter a number\")",
"Enter a number2442\n"
],
[
"print(x)",
"2442\n"
],
[
"x=input(\"Enter the first number\")\ny=input(\"Enter the second number\")",
"Enter the first number2442\nEnter the second number5002\n"
],
[
"print(x)\nprint(y)\nprint(x,y)\nx+y ",
"2442\n5002\n2442 5002\n"
],
[
"int(x)+int(y)",
"_____no_output_____"
],
[
"",
"_____no_output_____"
]
]
] |
[
"code"
] |
[
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
]
] |
cbd928ce30cb326592936bd858385eae798d7229
| 361,521 |
ipynb
|
Jupyter Notebook
|
lessons/CRISP_DM/How To Break Into the Field - Solution .ipynb
|
43piRcubed/DSND_Term2
|
5ecca7168b1b3e8c8b2b64a6360513b28d7c31b3
|
[
"MIT"
] | null | null | null |
lessons/CRISP_DM/How To Break Into the Field - Solution .ipynb
|
43piRcubed/DSND_Term2
|
5ecca7168b1b3e8c8b2b64a6360513b28d7c31b3
|
[
"MIT"
] | null | null | null |
lessons/CRISP_DM/How To Break Into the Field - Solution .ipynb
|
43piRcubed/DSND_Term2
|
5ecca7168b1b3e8c8b2b64a6360513b28d7c31b3
|
[
"MIT"
] | null | null | null | 394.243184 | 288,024 | 0.911073 |
[
[
[
"### How To Break Into the Field\n\nNow you have had a closer look at the data, and you saw how I approached looking at how the survey respondents think you should break into the field. Let's recreate those results, as well as take a look at another question.",
"_____no_output_____"
]
],
[
[
"import numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\nimport seaborn as sns\nimport HowToBreakIntoTheField as t\n%matplotlib inline\n\ndf = pd.read_csv('./survey_results_public.csv')\nschema = pd.read_csv('./survey_results_schema.csv')\ndf.head()",
"_____no_output_____"
],
[
"schema.head()",
"_____no_output_____"
]
],
[
[
"#### Question 1\n\n**1.** In order to understand how to break into the field, we will look at the **CousinEducation** field. Use the **schema** dataset to answer this question. Write a function called **get_description** that takes the **schema dataframe** and the **column** as a string, and returns a string of the description for that column.",
"_____no_output_____"
]
],
[
[
"def get_description(column_name, schema=schema):\n '''\n INPUT - schema - pandas dataframe with the schema of the developers survey\n column_name - string - the name of the column you would like to know about\n OUTPUT - \n desc - string - the description of the column\n '''\n desc = list(schema[schema['Column'] == column_name]['Question'])[0]\n return desc\n\n#test your code\n#Check your function against solution - you shouldn't need to change any of the below code\nget_description(df.columns[0]) # This should return a string of the first column description",
"_____no_output_____"
],
[
"#Check your function against solution - you shouldn't need to change any of the below code\ndescrips = set(get_description(col) for col in df.columns)\nt.check_description(descrips)",
"Nice job it looks like your function works correctly!\n"
]
],
[
[
"The question we have been focused on has been around how to break into the field. Use your **get_description** function below to take a closer look at the **CousinEducation** column.",
"_____no_output_____"
]
],
[
[
"get_description('CousinEducation')",
"_____no_output_____"
]
],
[
[
"#### Question 2\n\n**2.** Provide a pandas series of the different **CousinEducation** status values in the dataset. Store this pandas series in **cous_ed_vals**. If you are correct, you should see a bar chart of the proportion of individuals in each status. If it looks terrible, and you get no information from it, then you followed directions. However, we should clean this up!",
"_____no_output_____"
]
],
[
[
"cous_ed_vals = df.CousinEducation.value_counts()#Provide a pandas series of the counts for each CousinEducation status\n\ncous_ed_vals # assure this looks right",
"_____no_output_____"
],
[
"# The below should be a bar chart of the proportion of individuals in your ed_vals\n# if it is set up correctly.\n\n(cous_ed_vals/df.shape[0]).plot(kind=\"bar\");\nplt.title(\"Formal Education\");",
"_____no_output_____"
]
],
[
[
"We definitely need to clean this. Above is an example of what happens when you do not clean your data. Below I am using the same code you saw in the earlier video to take a look at the data after it has been cleaned.",
"_____no_output_____"
]
],
[
[
"possible_vals = [\"Take online courses\", \"Buy books and work through the exercises\", \n \"None of these\", \"Part-time/evening courses\", \"Return to college\",\n \"Contribute to open source\", \"Conferences/meet-ups\", \"Bootcamp\",\n \"Get a job as a QA tester\", \"Participate in online coding competitions\",\n \"Master's degree\", \"Participate in hackathons\", \"Other\"]\n\ndef clean_and_plot(df, title='Method of Educating Suggested', plot=True):\n '''\n INPUT \n df - a dataframe holding the CousinEducation column\n title - string the title of your plot\n axis - axis object\n plot - bool providing whether or not you want a plot back\n \n OUTPUT\n study_df - a dataframe with the count of how many individuals\n Displays a plot of pretty things related to the CousinEducation column.\n '''\n study = df['CousinEducation'].value_counts().reset_index()\n study.rename(columns={'index': 'method', 'CousinEducation': 'count'}, inplace=True)\n study_df = t.total_count(study, 'method', 'count', possible_vals)\n\n study_df.set_index('method', inplace=True)\n if plot:\n (study_df/study_df.sum()).plot(kind='bar', legend=None);\n plt.title(title);\n plt.show()\n props_study_df = study_df/study_df.sum()\n return props_study_df\n \nprops_df = clean_and_plot(df)",
"_____no_output_____"
]
],
[
[
"#### Question 4\n\n**4.** I wonder if some of the individuals might have bias towards their own degrees. Complete the function below that will apply to the elements of the **FormalEducation** column in **df**. ",
"_____no_output_____"
]
],
[
[
"def higher_ed(formal_ed_str):\n '''\n INPUT\n formal_ed_str - a string of one of the values from the Formal Education column\n \n OUTPUT\n return 1 if the string is in (\"Master's degree\", \"Doctoral\", \"Professional degree\")\n return 0 otherwise\n \n '''\n if formal_ed_str in (\"Master's degree\", \"Doctoral\", \"Professional degree\"):\n return 1\n else:\n return 0\n \n\ndf[\"FormalEducation\"].apply(higher_ed)[:5] #Test your function to assure it provides 1 and 0 values for the df",
"_____no_output_____"
],
[
"# Check your code here\ndf['HigherEd'] = df[\"FormalEducation\"].apply(higher_ed)\nhigher_ed_perc = df['HigherEd'].mean()\nt.higher_ed_test(higher_ed_perc)",
"Nice job! That's right. The percentage of individuals in these three groups is 0.2302376714480159.\n"
]
],
[
[
"#### Question 5\n\n**5.** Now we would like to find out if the proportion of individuals who completed one of these three programs feel differently than those that did not. Store a dataframe of only the individual's who had **HigherEd** equal to 1 in **ed_1**. Similarly, store a dataframe of only the **HigherEd** equal to 0 values in **ed_0**.\n\nNotice, you have already created the **HigherEd** column using the check code portion above, so here you only need to subset the dataframe using this newly created column.",
"_____no_output_____"
]
],
[
[
"ed_1 = df[df['HigherEd'] == 1] # Subset df to only those with HigherEd of 1\ned_0 = df[df['HigherEd'] == 0] # Subset df to only those with HigherEd of 0\n\n\nprint(ed_1['HigherEd'][:5]) #Assure it looks like what you would expect\nprint(ed_0['HigherEd'][:5]) #Assure it looks like what you would expect",
"4 1\n6 1\n7 1\n9 1\n14 1\nName: HigherEd, dtype: int64\n0 0\n1 0\n2 0\n3 0\n5 0\nName: HigherEd, dtype: int64\n"
],
[
"#Check your subset is correct - you should get a plot that was created using pandas styling\n#which you can learn more about here: https://pandas.pydata.org/pandas-docs/stable/style.html\n\ned_1_perc = clean_and_plot(ed_1, 'Higher Formal Education', plot=False)\ned_0_perc = clean_and_plot(ed_0, 'Max of Bachelors Higher Ed', plot=False)\n\ncomp_df = pd.merge(ed_1_perc, ed_0_perc, left_index=True, right_index=True)\ncomp_df.columns = ['ed_1_perc', 'ed_0_perc']\ncomp_df['Diff_HigherEd_Vals'] = comp_df['ed_1_perc'] - comp_df['ed_0_perc']\ncomp_df.style.bar(subset=['Diff_HigherEd_Vals'], align='mid', color=['#d65f5f', '#5fba7d'])",
"_____no_output_____"
]
],
[
[
"#### Question 6\n\n**6.** What can you conclude from the above plot? Change the dictionary to mark **True** for the keys of any statements you can conclude, and **False** for any of the statements you cannot conclude.",
"_____no_output_____"
]
],
[
[
"sol = {'Everyone should get a higher level of formal education': False, \n 'Regardless of formal education, online courses are the top suggested form of education': True,\n 'There is less than a 1% difference between suggestions of the two groups for all forms of education': False,\n 'Those with higher formal education suggest it more than those who do not have it': True}\n\nt.conclusions(sol)",
"Nice job that looks right!\n"
]
],
[
[
"This concludes another look at the way we could compare education methods by those currently writing code in industry.",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown"
] |
[
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
]
] |
cbd92ef3e48acfb0bfb8ef6aca8cf4fc2cc67c3a
| 8,484 |
ipynb
|
Jupyter Notebook
|
examples/tests/pylab-switch/Pylab Switching.ipynb
|
breisfeld/ipython
|
70e2c414014f3323d8a52fbcc94ee9e3a92d5d5f
|
[
"BSD-3-Clause-Clear"
] | 2 |
2015-02-10T18:00:31.000Z
|
2015-05-01T02:53:46.000Z
|
examples/tests/pylab-switch/Pylab Switching.ipynb
|
breisfeld/ipython
|
70e2c414014f3323d8a52fbcc94ee9e3a92d5d5f
|
[
"BSD-3-Clause-Clear"
] | null | null | null |
examples/tests/pylab-switch/Pylab Switching.ipynb
|
breisfeld/ipython
|
70e2c414014f3323d8a52fbcc94ee9e3a92d5d5f
|
[
"BSD-3-Clause-Clear"
] | 1 |
2020-05-03T10:25:12.000Z
|
2020-05-03T10:25:12.000Z
| 78.555556 | 6,154 | 0.822372 |
[
[
[
"empty"
]
]
] |
[
"empty"
] |
[
[
"empty"
]
] |
cbd933ff0449e4c83971eadb5b117241b6500f58
| 28,117 |
ipynb
|
Jupyter Notebook
|
notebooks/modelExperiments111920.ipynb
|
thejonathanvancetrance/Alfalfa
|
ac4f74610c601c2e74513d094b9138f6d6c696fa
|
[
"MIT"
] | null | null | null |
notebooks/modelExperiments111920.ipynb
|
thejonathanvancetrance/Alfalfa
|
ac4f74610c601c2e74513d094b9138f6d6c696fa
|
[
"MIT"
] | null | null | null |
notebooks/modelExperiments111920.ipynb
|
thejonathanvancetrance/Alfalfa
|
ac4f74610c601c2e74513d094b9138f6d6c696fa
|
[
"MIT"
] | null | null | null | 36.515584 | 338 | 0.440018 |
[
[
[
"The notebook is meant to help the user experiment with different models and features. This notebook assumes that there is a saved csv called 'filteredAggregateData.csv' somewhere on your local harddrive. The location must be specified below.",
"_____no_output_____"
],
[
"The cell imports all of the relevant packages.",
"_____no_output_____"
]
],
[
[
"############## imports\n# general\nimport statistics\nimport datetime\nfrom sklearn.externals import joblib # save and load models\nimport random\n# data manipulation and exploration\nimport pandas as pd\nimport numpy as np\nimport matplotlib.pyplot as plt\nimport matplotlib\n\n## machine learning stuff\n# preprocessing\nfrom sklearn import preprocessing\n# feature selection\nfrom sklearn.feature_selection import SelectKBest, SelectPercentile\nfrom sklearn.feature_selection import f_regression\n# pipeline\nfrom sklearn.pipeline import Pipeline\n# train/testing\nfrom sklearn.model_selection import train_test_split, KFold, GridSearchCV, cross_val_score \n# error calculations\nfrom sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score\n# models\nfrom sklearn.linear_model import LinearRegression # linear regression\nfrom sklearn.linear_model import BayesianRidge #bayesisan ridge regression\nfrom sklearn.svm import SVR # support vector machines regression\nfrom sklearn.gaussian_process import GaussianProcessRegressor # import GaussianProcessRegressor\nfrom sklearn.neighbors import KNeighborsRegressor # k-nearest neightbors for regression\nfrom sklearn.neural_network import MLPRegressor # neural network for regression\nfrom sklearn.tree import DecisionTreeRegressor # decision tree regressor\nfrom sklearn.ensemble import RandomForestRegressor # random forest regression\nfrom sklearn.ensemble import AdaBoostRegressor # adaboost for regression\n# saving models\n# from sklearn.externals import joblib\nimport joblib\n",
"_____no_output_____"
]
],
[
[
"Imports the API. 'APILoc' is the location of 'API.py' on your local harddrive.",
"_____no_output_____"
]
],
[
[
"# import the API\nAPILoc = r\"C:\\Users\\thejo\\Documents\\school\\AI in AG research\\API\"\n\nimport sys\nsys.path.insert(0, APILoc)\n\nfrom API import *",
"_____no_output_____"
]
],
[
[
"Load the dataset. Note that the location of the dataset must be specified.",
"_____no_output_____"
]
],
[
[
"# get aggregate data\naggDataLoc = r'C:\\Users\\thejo\\Documents\\school\\AI in AG research\\experiment\\aggregateData_MS_KY_GA.csv'\n#aggDataLoc = r'C:\\Users\\thejo\\Documents\\school\\AI in AG research\\experiment\\aggregateDataWithVariety.csv'\n#targetDataLoc = r'C:\\Users\\thejo\\Documents\\school\\AI in AG research\\experiment\\aggregateData_GAonly_Annual_final.csv'\n\naggDf = pd.read_csv(aggDataLoc)\n#aggDf = aggDf.drop(\"Unnamed: 0\",axis=1)\n#targetDf = pd.read_csv(targetDataLoc)\n#targetDf = targetDf.drop(\"Unnamed: 0\",axis=1)\n",
"_____no_output_____"
]
],
[
[
"Test to see if the dataset was loaded properly. A table of the first 5 datapoints should appear.",
"_____no_output_____"
]
],
[
[
"aggDf.head()\n#targetDf.head()",
"_____no_output_____"
]
],
[
[
"Filter out features that will not be made available for feature selection. All of the features in the list 'XColumnsToKeep' will be made available for feature selection. The features to include are: <br>\n\"Julian Day\" <br>\n\"Time Since Sown (Days)\" <br>\n\"Time Since Last Harvest (Days)\" <br>\n\"Total Radiation (MJ/m^2)\" <br>\n\"Total Rainfall (mm)\" <br>\n\"Avg Air Temp (C)\" <br>\n\"Avg Min Temp (C)\" <br>\n\"Avg Max Temp (C)\"<br>\n\"Avg Soil Moisture (%)\"<br>\n\"Day Length (hrs)\"<br>\n\"Percent Cover (%)\"<br>",
"_____no_output_____"
]
],
[
[
"# filter out the features that will not be used by the machine learning models\n\n# the features to keep:\n# xColumnsToKeep = [\"Julian Day\", \"Time Since Sown (Days)\", \"Time Since Last Harvest (Days)\", \"Total Radiation (MJ/m^2)\",\n# \"Total Rainfall (mm)\", \"Avg Air Temp (C)\", \"Avg Min Temp (C)\", \"Avg Max Temp (C)\",\n# \"Avg Soil Moisture (%)\", \"Day Length (hrs)\"], \"Percent Cover (%)\"]\n\nxColumnsToKeep = [\"Julian Day\", \"Time Since Sown (Days)\", \"Total Radiation (MJ/m^2)\",\n \"Total Rainfall (mm)\", \"Avg Air Temp (C)\", \"Avg Min Temp (C)\", \"Avg Max Temp (C)\",\n \"Avg Soil Moisture (%)\"]\n\n\n#xColumnsToKeep = [\"Julian Day\", \"Time Since Sown (Days)\", \"Total Radiation (MJ/m^2)\", \"Total Rainfall (mm)\"]\n\n# the target to keep\nyColumnsToKeep = [\"Yield (tons/acre)\"]\n\n# get a dataframe containing the features and the targets\nxDf = aggDf[xColumnsToKeep]\n#yDf = targetDf[yColumnsToKeep]\nyDf = aggDf[yColumnsToKeep]\n\n# reset the index\nxDf = xDf.reset_index(drop=True)\nyDf = yDf.reset_index(drop=True)\n\npd.set_option('display.max_rows', 2500)\npd.set_option('display.max_columns', 500)\n\nxCols = list(xDf)",
"_____no_output_____"
]
],
[
[
"Test to see if the features dataframe and the target dataframe were successfully made.",
"_____no_output_____"
]
],
[
[
"xDf.head()",
"_____no_output_____"
],
[
"yDf.head()",
"_____no_output_____"
]
],
[
[
"Lets now define the parameters that will be used to run the machine learning experiments. Note that parameter grids could be made that will allow sci-kit learn to use a 5-fold gridsearch to find the model's best hyperparameters. The parameter grids that are defined here will specify the possible values for the grid search. <br>\n<br>\nOnce the parameter grids are defined, a list of tuples must also be defined. The tuples must take the form of: <br>\n(sci-kit learn model, appropriate parameter grid, name of the file to be saved). <br>\n<br>\nThen the number of iterations should be made. This is represented by the variable 'N'. Each model will be evaluated N times (via N-fold cross validation), and the average results of the models over those N iterations will be returned. <br>\n<br>\n'workingDir' is the directory in which all of the results will be saved. <br>\n<br>\n'numFeatures' is the number of features that will be selected (via feature selection).",
"_____no_output_____"
]
],
[
[
"\n# hide the warnings because training the neural network caues lots of warnings.\nimport warnings\nwarnings.filterwarnings('ignore')\n\n# make the parameter grids for sklearn's gridsearchcv\nrfParamGrid = {\n 'model__n_estimators': [5, 10, 25, 50, 100], # Number of estimators\n 'model__max_depth': [5, 10, 15, 20], # Maximum depth of the tree\n 'model__criterion': [\"mae\"]\n }\nknnParamGrid ={\n 'model__n_neighbors':[2,5,10],\n 'model__weights': ['uniform', 'distance'],\n 'model__leaf_size': [5, 10, 30, 50] \n }\nsvrParamGrid = {\n 'model__kernel': ['linear', 'poly', 'rbf', 'sigmoid'],\n 'model__C': [0.1, 1.0, 5.0, 10.0],\n 'model__gamma': [\"scale\", \"auto\"],\n 'model__degree': [2,3,4,5]\n }\nnnParamGrid = {\n 'model__hidden_layer_sizes':[(3), (5), (10), (3,3), (5,5), (7,7)],\n 'model__solver': ['sgd', 'adam'],\n 'model__learning_rate' : ['constant', 'invscaling', 'adaptive'],\n 'model__learning_rate_init': [0.1, 0.01, 0.001] \n }\n\nlinRegParamGrid = {}\n\nbayesParamGrid={\n 'model__n_iter':[100,300,500]\n }\n\ndtParamGrid = {\n 'model__criterion': ['mae'],\n 'model__max_depth': [5,10,25,50,100]\n }\n\naModelList = [(RandomForestRegressor(), rfParamGrid, \"rfTup.pkl\"),\n (KNeighborsRegressor(), knnParamGrid, \"knnTup.pkl\"),\n (SVR(), svrParamGrid, \"svrTup.pkl\"),\n #(MLPRegressor(), nnParamGrid, \"nnTup.pkl\")]#,\n (LinearRegression(), linRegParamGrid, \"linRegTup.pkl\"),\n (BayesianRidge(), bayesParamGrid, \"bayesTup.pkl\"),\n (DecisionTreeRegressor(), dtParamGrid, \"dtTup.pkl\")]\n\nN = 10\nworkingDir = r\"C:\\Users\\thejo\\Documents\\school\\AI in AG research\\experiment\"\nnumFeatures = 8 # 11\n",
"_____no_output_____"
]
],
[
[
"This cell will run the tests and save the results.",
"_____no_output_____"
]
],
[
[
"saveMLResults(N, xDf, yDf, aModelList, workingDir, numFeatures, printResults=True)",
"model: rfTup\nAvg MAE: 546289.95\nAvg R squared: 0.082\nBest MAE: 1668.613\nBest R squared: 0.644\nParameters of the best model: {'model__criterion': 'mae', 'model__max_depth': 20, 'model__n_estimators': 10}\nFeatures selected by best model: ['Julian Day', 'Time Since Sown (Days)', 'Total Radiation (MJ/m^2)', 'Total Rainfall (mm)', 'Avg Air Temp (C)', 'Avg Min Temp (C)', 'Avg Max Temp (C)', 'Avg Soil Moisture (%)']\n \nmodel: knnTup\nAvg MAE: 546595.773\nAvg R squared: -1.1\nBest MAE: 2371.145\nBest R squared: 0.606\nParameters of the best model: {'model__leaf_size': 5, 'model__n_neighbors': 2, 'model__weights': 'uniform'}\nFeatures selected by best model: ['Julian Day', 'Time Since Sown (Days)', 'Total Radiation (MJ/m^2)', 'Total Rainfall (mm)', 'Avg Air Temp (C)', 'Avg Min Temp (C)', 'Avg Max Temp (C)', 'Avg Soil Moisture (%)']\n \nmodel: svrTup\nAvg MAE: 545885.93\nAvg R squared: -0.197\nBest MAE: 1557.557\nBest R squared: 0.863\nParameters of the best model: {'model__C': 10.0, 'model__degree': 2, 'model__gamma': 'scale', 'model__kernel': 'rbf'}\nFeatures selected by best model: ['Julian Day', 'Time Since Sown (Days)', 'Total Radiation (MJ/m^2)', 'Total Rainfall (mm)', 'Avg Air Temp (C)', 'Avg Min Temp (C)', 'Avg Max Temp (C)', 'Avg Soil Moisture (%)']\n \nmodel: linRegTup\nAvg MAE: 5261549.3\nAvg R squared: -79188014.247\nBest MAE: 5447810.775\nBest R squared: -0.033\nParameters of the best model: {}\nFeatures selected by best model: ['Julian Day', 'Time Since Sown (Days)', 'Total Radiation (MJ/m^2)', 'Total Rainfall (mm)', 'Avg Air Temp (C)', 'Avg Min Temp (C)', 'Avg Max Temp (C)', 'Avg Soil Moisture (%)']\n \nmodel: bayesTup\nAvg MAE: 2480509.91\nAvg R squared: -9082338.845\nBest MAE: 5446424.692\nBest R squared: -0.033\nParameters of the best model: {'model__n_iter': 100}\nFeatures selected by best model: ['Julian Day', 'Time Since Sown (Days)', 'Total Radiation (MJ/m^2)', 'Total Rainfall (mm)', 'Avg Air Temp (C)', 'Avg Min Temp (C)', 'Avg Max Temp (C)', 'Avg Soil Moisture (%)']\n \nmodel: dtTup\nAvg MAE: 546570.144\nAvg R squared: -0.112\nBest MAE: 532.903\nBest R squared: 0.925\nParameters of the best model: {'model__criterion': 'mae', 'model__max_depth': 50}\nFeatures selected by best model: ['Julian Day', 'Time Since Sown (Days)', 'Total Radiation (MJ/m^2)', 'Total Rainfall (mm)', 'Avg Air Temp (C)', 'Avg Min Temp (C)', 'Avg Max Temp (C)', 'Avg Soil Moisture (%)']\n \n"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
]
] |
cbd93a56910a7f4afd1066accd76acd2ba6cb435
| 144,890 |
ipynb
|
Jupyter Notebook
|
4idl_dihedral/06_grest_production/.ipynb_checkpoints/parameter_4idlDIHED-checkpoint.ipynb
|
matsunagalab/paper_higashida2021
|
7a7786edb970db292340566f3a0420df3265d1f6
|
[
"BSD-3-Clause"
] | null | null | null |
4idl_dihedral/06_grest_production/.ipynb_checkpoints/parameter_4idlDIHED-checkpoint.ipynb
|
matsunagalab/paper_higashida2021
|
7a7786edb970db292340566f3a0420df3265d1f6
|
[
"BSD-3-Clause"
] | null | null | null |
4idl_dihedral/06_grest_production/.ipynb_checkpoints/parameter_4idlDIHED-checkpoint.ipynb
|
matsunagalab/paper_higashida2021
|
7a7786edb970db292340566f3a0420df3265d1f6
|
[
"BSD-3-Clause"
] | null | null | null | 287.480159 | 131,462 | 0.913217 |
[
[
[
"empty"
]
]
] |
[
"empty"
] |
[
[
"empty"
]
] |
cbd961d65ebb5802345e788fbb9e2996e5f20ca8
| 6,806 |
ipynb
|
Jupyter Notebook
|
fyp/Test ML.ipynb
|
Ninad998/deepstylometry-python
|
1cc9776d6624311a31f081852da9cbf08f23c350
|
[
"MIT"
] | 1 |
2021-05-20T13:05:36.000Z
|
2021-05-20T13:05:36.000Z
|
fyp/Test ML.ipynb
|
ntungare/deepstylometry-python
|
1cc9776d6624311a31f081852da9cbf08f23c350
|
[
"MIT"
] | null | null | null |
fyp/Test ML.ipynb
|
ntungare/deepstylometry-python
|
1cc9776d6624311a31f081852da9cbf08f23c350
|
[
"MIT"
] | 1 |
2018-12-11T19:11:51.000Z
|
2018-12-11T19:11:51.000Z
| 28.596639 | 107 | 0.459741 |
[
[
[
"import pandas as pd\ndf = pd.read_csv('queryset_CNN.csv')\nprint(df.shape)\nprint(df.dtypes)\npreds = []",
"(32, 3)\ndoc_id int64\nauthor_id int64\nauthorList object\ndtype: object\n"
],
[
"pred = []\nfor index, row in df.iterrows():\n doc_id = row.doc_id\n\n author_id = row.author_id\n\n import ast\n authorList = ast.literal_eval(row.authorList)\n \n candidate = len(authorList)\n \n algo = \"tfidf_svc\"\n\n test = algo # change before run\n\n level = \"word\"\n\n iterations = 30\n\n dropout = 0.5\n\n samples = 3200\n\n dimensions = 200\n\n loc = authorList.index(author_id)\n\n printstate = ((\"doc_id = %s, candidate = %s, \") % (str(doc_id), str(candidate)))\n printstate += ((\"samples = %s, \") % (str(samples)))\n printstate += ((\"test = %s\") % (str(test)))\n\n print(\"Current test: %s\" % (str(printstate)))\n \n from sshtunnel import SSHTunnelForwarder\n with SSHTunnelForwarder(('144.214.121.15', 22),\n ssh_username='ninadt',\n ssh_password='Ninad123',\n remote_bind_address=('localhost', 3306),\n local_bind_address=('localhost', 3300)):\n import UpdateDB as db\n case = db.checkOldML(doc_id = doc_id, candidate = candidate, samples = samples,\n test = test, port = 3300)\n \n if case == False:\n \n print(\"Running: %12s\" % (str(printstate)))\n\n import StyloML as Stylo\n (labels_index, train_acc, val_acc, samples) = Stylo.getResults(\n algo, \n doc_id = doc_id, authorList = authorList[:], \n samples = samples)\n\n (labels_index, testY, predY, samples) = Stylo.getTestResults(\n algo, labels_index = labels_index,\n doc_id = doc_id, authorList = authorList[:], \n samples = samples)\n \n loc = testY\n\n test_acc = predY[loc]\n\n test_bin = 0\n\n if(predY.tolist().index(max(predY)) == testY):\n test_bin = 1\n \n from sshtunnel import SSHTunnelForwarder\n with SSHTunnelForwarder(('144.214.121.15', 22),\n ssh_username='ninadt',\n ssh_password='Ninad123',\n remote_bind_address=('localhost', 3306),\n local_bind_address=('localhost', 3300)):\n import UpdateDB as db\n case = db.updateresultOldML(doc_id = doc_id, candidate = candidate, samples = samples,\n train_acc = train_acc, val_acc = val_acc,\n test_acc = test_acc, test_bin = test_bin,\n test = test, port = 3300)\n \n del Stylo\n \n import time\n time.sleep(10)\n \n from IPython.display import clear_output\n\n clear_output()\n\n else:\n print(\"Skipped: %12s\" % (str(printstate)))",
"Current test: doc_id = 85, candidate = 3, samples = 3200, test = tfidf_svc\nExecution completed\nRunning: doc_id = 85, candidate = 3, samples = 3200, test = tfidf_svc\nAlgo: tfidf_svc\nExecution completed\nRead completed\nNumber of rows: 124\nauthor_id int64\ndoc_content object\ndtype: object\nData Frame created: Shape: (11578, 2)\nAuthor: 44 Size: 4746\nAuthor: 64 Size: 5106\nAuthor: 82 Size: 1726\nMin: 1726\nMax: 5106\nAuthors [44, 64, 82].\nFound 5178 texts.\nFound 5178 labels.\n"
],
[
"# import matplotlib.pyplot as plt\n# # summarize history for accuracy\n# plt.plot(history.history['acc'])\n# plt.plot(history.history['val_acc'])\n# plt.title('model accuracy')\n# plt.ylabel('accuracy')\n# plt.xlabel('epoch')\n# plt.legend(['train', 'test'], loc='upper left')\n# plt.show()",
"_____no_output_____"
],
[
"# # summarize history for loss\n# plt.plot(history.history['loss'])\n# plt.plot(history.history['val_loss'])\n# plt.title('model loss')\n# plt.ylabel('loss')\n# plt.xlabel('epoch')\n# plt.legend(['train', 'test'], loc='upper left')\n# plt.show()",
"_____no_output_____"
],
[
"%tb",
"_____no_output_____"
],
[
"",
"_____no_output_____"
]
]
] |
[
"code"
] |
[
[
"code",
"code",
"code",
"code",
"code",
"code"
]
] |
cbd965ecf82313834403226fe6cf4a713b11f356
| 82,553 |
ipynb
|
Jupyter Notebook
|
numpy_tutorial.ipynb
|
ViacheslavKolupaev/cs224u
|
1c47fdd53036225a0cf6b186e3a4905737e00eb2
|
[
"Apache-2.0"
] | null | null | null |
numpy_tutorial.ipynb
|
ViacheslavKolupaev/cs224u
|
1c47fdd53036225a0cf6b186e3a4905737e00eb2
|
[
"Apache-2.0"
] | null | null | null |
numpy_tutorial.ipynb
|
ViacheslavKolupaev/cs224u
|
1c47fdd53036225a0cf6b186e3a4905737e00eb2
|
[
"Apache-2.0"
] | null | null | null | 29.845625 | 15,132 | 0.487844 |
[
[
[
"# Introduction to NumPy\n\nThis notebook is the first half of a special session on NumPy and PyTorch for CS 224U. \n\nWhy should we care about NumPy? \n- It allows you to perform tons of operations on vectors and matrices. \n- It makes things run faster than naive for-loop implementations (a.k.a. vectorization). \n- We use it in our class (see files prefixed with `np_` in your cs224u directory). \n- It's used a ton in machine learning / AI. \n- Its arrays are often inputs into other important Python packages' functions. \n\nIn Jupyter notebooks, NumPy documentation is two clicks away: Help -> NumPy reference. ",
"_____no_output_____"
]
],
[
[
"__author__ = 'Will Monroe, Chris Potts, and Lucy Li'",
"_____no_output_____"
],
[
"import numpy as np",
"_____no_output_____"
]
],
[
[
"# Vectors\n## Vector Initialization",
"_____no_output_____"
]
],
[
[
"np.zeros(5)",
"_____no_output_____"
],
[
"np.ones(5)",
"_____no_output_____"
],
[
"# convert list to numpy array\nnp.array([1,2,3,4,5])",
"_____no_output_____"
],
[
"# convert numpy array to list\nnp.ones(5).tolist()",
"_____no_output_____"
],
[
"# one float => all floats\nnp.array([1.0,2,3,4,5])",
"_____no_output_____"
],
[
"# same as above\nnp.array([1,2,3,4,5], dtype='float')",
"_____no_output_____"
],
[
"# spaced values in interval\nnp.array([x for x in range(20) if x % 2 == 0])",
"_____no_output_____"
],
[
"# same as above\nnp.arange(0,20,2)",
"_____no_output_____"
],
[
"# random floats in [0, 1)\nnp.random.random(10)",
"_____no_output_____"
],
[
"# random integers\nnp.random.randint(5, 15, size=10)",
"_____no_output_____"
]
],
[
[
"## Vector indexing",
"_____no_output_____"
]
],
[
[
"x = np.array([10,20,30,40,50])",
"_____no_output_____"
],
[
"x[0]",
"_____no_output_____"
],
[
"# slice\nx[0:2]",
"_____no_output_____"
],
[
"x[0:1000]",
"_____no_output_____"
],
[
"# last value\nx[-1]",
"_____no_output_____"
],
[
"# last value as array\nx[[-1]]",
"_____no_output_____"
],
[
"# last 3 values\nx[-3:]",
"_____no_output_____"
],
[
"# pick indices\nx[[0,2,4]]",
"_____no_output_____"
]
],
[
[
"## Vector assignment\n\nBe careful when assigning arrays to new variables! ",
"_____no_output_____"
]
],
[
[
"#x2 = x # try this line instead\nx2 = x.copy()",
"_____no_output_____"
],
[
"x2[0] = 10\n\nx2",
"_____no_output_____"
],
[
"x2[[1,2]] = 10\n\nx2",
"_____no_output_____"
],
[
"x2[[3,4]] = [0, 1]\n\nx2",
"_____no_output_____"
],
[
"# check if the original vector changed\nx",
"_____no_output_____"
]
],
[
[
"## Vectorized operations",
"_____no_output_____"
]
],
[
[
"x.sum()",
"_____no_output_____"
],
[
"x.mean()",
"_____no_output_____"
],
[
"x.max()",
"_____no_output_____"
],
[
"x.argmax()",
"_____no_output_____"
],
[
"np.log(x)",
"_____no_output_____"
],
[
"np.exp(x)",
"_____no_output_____"
],
[
"x + x # Try also with *, -, /, etc.",
"_____no_output_____"
],
[
"x + 1",
"_____no_output_____"
]
],
[
[
"## Comparison with Python lists\n\nVectorizing your mathematical expressions can lead to __huge__ performance gains. The following example is meant to give you a sense for this. It compares applying `np.log` to each element of a list with 10 million values with the same operation done on a vector.",
"_____no_output_____"
]
],
[
[
"# log every value as list, one by one\ndef listlog(vals):\n return [np.log(y) for y in vals]",
"_____no_output_____"
],
[
"# get random vector\nsamp = np.random.random_sample(int(1e7))+1\nsamp",
"_____no_output_____"
],
[
"%time _ = np.log(samp)",
"CPU times: user 270 ms, sys: 54.6 ms, total: 324 ms\nWall time: 326 ms\n"
],
[
"%time _ = listlog(samp)",
"CPU times: user 13.4 s, sys: 1.4 s, total: 14.8 s\nWall time: 14.8 s\n"
]
],
[
[
"# Matrices\n\nThe matrix is the core object of machine learning implementations. ",
"_____no_output_____"
],
[
"## Matrix initialization",
"_____no_output_____"
]
],
[
[
"np.array([[1,2,3], [4,5,6]])",
"_____no_output_____"
],
[
"np.array([[1,2,3], [4,5,6]], dtype='float')",
"_____no_output_____"
],
[
"np.zeros((3,5))",
"_____no_output_____"
],
[
"np.ones((3,5))",
"_____no_output_____"
],
[
"np.identity(3)",
"_____no_output_____"
],
[
"np.diag([1,2,3])",
"_____no_output_____"
]
],
[
[
"## Matrix indexing",
"_____no_output_____"
]
],
[
[
"X = np.array([[1,2,3], [4,5,6]])\nX",
"_____no_output_____"
],
[
"X[0]",
"_____no_output_____"
],
[
"X[0,0]",
"_____no_output_____"
],
[
"# get row\nX[0, : ]",
"_____no_output_____"
],
[
"# get column\nX[ : , 0]",
"_____no_output_____"
],
[
"# get multiple columns\nX[ : , [0,2]]",
"_____no_output_____"
]
],
[
[
"## Matrix assignment",
"_____no_output_____"
]
],
[
[
"# X2 = X # try this line instead\nX2 = X.copy()\n\nX2",
"_____no_output_____"
],
[
"X2[0,0] = 20\n\nX2",
"_____no_output_____"
],
[
"X2[0] = 3\n\nX2",
"_____no_output_____"
],
[
"X2[: , -1] = [5, 6]\n\nX2",
"_____no_output_____"
],
[
"# check if original matrix changed\nX",
"_____no_output_____"
]
],
[
[
"## Matrix reshaping",
"_____no_output_____"
]
],
[
[
"z = np.arange(1, 7)\n\nz",
"_____no_output_____"
],
[
"z.shape",
"_____no_output_____"
],
[
"Z = z.reshape(2,3)\n\nZ",
"_____no_output_____"
],
[
"Z.shape",
"_____no_output_____"
],
[
"Z.reshape(6)",
"_____no_output_____"
],
[
"# same as above\nZ.flatten()",
"_____no_output_____"
],
[
"# transpose\nZ.T",
"_____no_output_____"
]
],
[
[
"## Numeric operations",
"_____no_output_____"
]
],
[
[
"A = np.array(range(1,7), dtype='float').reshape(2,3)\n\nA",
"_____no_output_____"
],
[
"B = np.array([1, 2, 3])\nB",
"_____no_output_____"
],
[
"# not the same as A.dot(B)\nA * B",
"_____no_output_____"
],
[
"A + B",
"_____no_output_____"
],
[
"A / B",
"_____no_output_____"
],
[
"# matrix multiplication\nA.dot(B)",
"_____no_output_____"
],
[
"B.dot(A.T)",
"_____no_output_____"
],
[
"A.dot(A.T)",
"_____no_output_____"
],
[
"# outer product\n# multiplying each element of first vector by each element of the second\nnp.outer(B, B)",
"_____no_output_____"
]
],
[
[
"The following is a practical example of numerical operations on NumPy matrices. \n\nIn our class, we have a shallow neural network implemented in `np_shallow_neural_network.py`. See how the forward and backward passes use no for loops, and instead takes advantage of NumPy's ability to vectorize manipulations of data. \n\n```python\ndef forward_propagation(self, x):\n h = self.hidden_activation(x.dot(self.W_xh) + self.b_xh)\n y = softmax(h.dot(self.W_hy) + self.b_hy)\n return h, y\n\ndef backward_propagation(self, h, predictions, x, labels):\n y_err = predictions.copy()\n y_err[np.argmax(labels)] -= 1 # backprop for cross-entropy error: -log(prediction-for-correct-label)\n d_b_hy = y_err\n h_err = y_err.dot(self.W_hy.T) * self.d_hidden_activation(h)\n d_W_hy = np.outer(h, y_err)\n d_W_xh = np.outer(x, h_err)\n d_b_xh = h_err\n return d_W_hy, d_b_hy, d_W_xh, d_b_xh\n```\n\nThe forward pass essentially computes the following: \n $$h = f(xW_{xh} + b_{xh})$$\n $$y = \\text{softmax}(hW_{hy} + b_{hy}),$$\nwhere $f$ is `self.hidden_activation`. \n\nThe backward pass propagates error by computing local gradients and chaining them. Feel free to learn more about backprop [here](http://cs231n.github.io/optimization-2/), though it is not necessary for our class. Also look at this [neural networks case study](http://cs231n.github.io/neural-networks-case-study/) to see another example of how NumPy can be used to implement forward and backward passes of a simple neural network. ",
"_____no_output_____"
],
[
"## Going beyond NumPy alone\n\nThese are examples of how NumPy can be used with other Python packages. ",
"_____no_output_____"
],
[
"### Pandas\nWe can convert numpy matrices to Pandas dataframes. In the following example, this is useful because it allows us to label each row. You may have noticed this being done in our first unit on distributed representations. ",
"_____no_output_____"
]
],
[
[
"import pandas as pd",
"_____no_output_____"
],
[
"count_df = pd.DataFrame(\n np.array([\n [1,0,1,0,0,0],\n [0,1,0,1,0,0],\n [1,1,1,1,0,0],\n [0,0,0,0,1,1],\n [0,0,0,0,0,1]], dtype='float64'),\n index=['gnarly', 'wicked', 'awesome', 'lame', 'terrible'])\ncount_df",
"_____no_output_____"
]
],
[
[
"### Scikit-learn\n\nIn `sklearn`, NumPy matrices are the most common input and output and thus a key to how the library's numerous methods can work together. Many of the cs224u's model built by Chris operate just like `sklearn` ones, such as the classifiers we used for our sentiment analysis unit. ",
"_____no_output_____"
]
],
[
[
"from sklearn.linear_model import LogisticRegression\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.metrics import classification_report\nfrom sklearn import datasets",
"_____no_output_____"
],
[
"iris = datasets.load_iris()\nX = iris.data\ny = iris.target\nprint(type(X))\nprint(\"Dimensions of X:\", X.shape)\nprint(type(y))\nprint(\"Dimensions of y:\", y.shape)",
"<class 'numpy.ndarray'>\nDimensions of X: (150, 4)\n<class 'numpy.ndarray'>\nDimensions of y: (150,)\n"
],
[
"# split data into train/test\nX_iris_train, X_iris_test, y_iris_train, y_iris_test = train_test_split(\n X, y, train_size=0.7, test_size=0.3)\nprint(\"X_iris_train:\", type(X_iris_train))\nprint(\"y_iris_train:\", type(y_iris_train))\nprint()\n\n# start up model\nmaxent = LogisticRegression(fit_intercept=True, \n solver='liblinear', \n multi_class='auto')\n\n# train on train set\nmaxent.fit(X_iris_train, y_iris_train)\n\n# predict on test set\niris_predictions = maxent.predict(X_iris_test)\nfnames_iris = iris['feature_names']\ntnames_iris = iris['target_names']\n\n# how well did our model do?\nprint(classification_report(y_iris_test, iris_predictions, target_names=tnames_iris))",
"X_iris_train: <class 'numpy.ndarray'>\ny_iris_train: <class 'numpy.ndarray'>\n\n precision recall f1-score support\n\n setosa 1.00 1.00 1.00 13\n versicolor 1.00 0.94 0.97 16\n virginica 0.94 1.00 0.97 16\n\n accuracy 0.98 45\n macro avg 0.98 0.98 0.98 45\nweighted avg 0.98 0.98 0.98 45\n\n"
]
],
[
[
"### SciPy\n\nSciPy contains what may seem like an endless treasure trove of operations for linear algebra, optimization, and more. It is built so that everything can work with NumPy arrays. ",
"_____no_output_____"
]
],
[
[
"from scipy.spatial.distance import cosine\nfrom scipy.stats import pearsonr\nfrom scipy import linalg",
"_____no_output_____"
],
[
"# cosine distance\na = np.random.random(10)\nb = np.random.random(10)\ncosine(a, b)",
"_____no_output_____"
],
[
"# pearson correlation (coeff, p-value)\npearsonr(a, b)",
"_____no_output_____"
],
[
"# inverse of matrix\nA = np.array([[1,3,5],[2,5,1],[2,3,8]])\nlinalg.inv(A)",
"_____no_output_____"
]
],
[
[
"To learn more about how NumPy can be combined with SciPy and Scikit-learn for machine learning, check out this [notebook tutorial](https://github.com/cgpotts/csli-summer/blob/master/advanced_python/intro_to_python_ml.ipynb) by Chris Potts and Will Monroe. (You may notice that over half of this current notebook is modified from theirs.) Their tutorial also has some interesting exercises in it! ",
"_____no_output_____"
],
[
"### Matplotlib",
"_____no_output_____"
]
],
[
[
"import matplotlib.pyplot as plt",
"_____no_output_____"
],
[
"a = np.sort(np.random.random(30))\nb = a**2\nc = np.log(a)\nplt.plot(a, b, label='y = x^2')\nplt.plot(a, c, label='y = log(x)')\nplt.legend()\nplt.title(\"Some functions\")\nplt.show()",
"_____no_output_____"
],
[
"",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown",
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code",
"code"
]
] |
cbd9791ff822066249b77b39bf27c4b001acce34
| 881 |
ipynb
|
Jupyter Notebook
|
CHEM1035_Chemistry_For_Engineers/.ipynb_checkpoints/test-checkpoint.ipynb
|
robraddi/tu_chem
|
18b8247d6c00e33f15f040a57a32b5fc2372137a
|
[
"MIT"
] | 1 |
2021-04-29T04:26:42.000Z
|
2021-04-29T04:26:42.000Z
|
CHEM1035_Chemistry_For_Engineers/.ipynb_checkpoints/test-checkpoint.ipynb
|
robraddi/tu_chem
|
18b8247d6c00e33f15f040a57a32b5fc2372137a
|
[
"MIT"
] | null | null | null |
CHEM1035_Chemistry_For_Engineers/.ipynb_checkpoints/test-checkpoint.ipynb
|
robraddi/tu_chem
|
18b8247d6c00e33f15f040a57a32b5fc2372137a
|
[
"MIT"
] | 1 |
2020-12-03T17:47:05.000Z
|
2020-12-03T17:47:05.000Z
| 20.97619 | 230 | 0.552781 |
[
[
[
"<iframe width=\"1000\" height=\"1000\" align=\"center\" src=\"https://www.falstad.com/qmatom/\" frameborder=\"0\" allow=\"accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture\" allowfullscreen></iframe>\n\n\n",
"_____no_output_____"
]
]
] |
[
"markdown"
] |
[
[
"markdown"
]
] |
cbd98324fe1f52fb8d24ee2081231b105d6b7ca1
| 378,627 |
ipynb
|
Jupyter Notebook
|
ml-clustering-and-retrieval/week-5/5_lda_blank.ipynb
|
zomansud/coursera
|
8b63eda4194241edc0c493fb74ca6834c9d0792d
|
[
"MIT"
] | null | null | null |
ml-clustering-and-retrieval/week-5/5_lda_blank.ipynb
|
zomansud/coursera
|
8b63eda4194241edc0c493fb74ca6834c9d0792d
|
[
"MIT"
] | null | null | null |
ml-clustering-and-retrieval/week-5/5_lda_blank.ipynb
|
zomansud/coursera
|
8b63eda4194241edc0c493fb74ca6834c9d0792d
|
[
"MIT"
] | 1 |
2021-08-10T20:05:24.000Z
|
2021-08-10T20:05:24.000Z
| 143.691461 | 178,716 | 0.648971 |
[
[
[
"# Latent Dirichlet Allocation for Text Data\n\nIn this assignment you will\n\n* apply standard preprocessing techniques on Wikipedia text data\n* use GraphLab Create to fit a Latent Dirichlet allocation (LDA) model\n* explore and interpret the results, including topic keywords and topic assignments for documents\n\nRecall that a major feature distinguishing the LDA model from our previously explored methods is the notion of *mixed membership*. Throughout the course so far, our models have assumed that each data point belongs to a single cluster. k-means determines membership simply by shortest distance to the cluster center, and Gaussian mixture models suppose that each data point is drawn from one of their component mixture distributions. In many cases, though, it is more realistic to think of data as genuinely belonging to more than one cluster or category - for example, if we have a model for text data that includes both \"Politics\" and \"World News\" categories, then an article about a recent meeting of the United Nations should have membership in both categories rather than being forced into just one.\n\nWith this in mind, we will use GraphLab Create tools to fit an LDA model to a corpus of Wikipedia articles and examine the results to analyze the impact of a mixed membership approach. In particular, we want to identify the topics discovered by the model in terms of their most important words, and we want to use the model to predict the topic membership distribution for a given document. ",
"_____no_output_____"
],
[
"**Note to Amazon EC2 users**: To conserve memory, make sure to stop all the other notebooks before running this notebook.",
"_____no_output_____"
],
[
"## Text Data Preprocessing\nWe'll start by importing our familiar Wikipedia dataset.\n\nThe following code block will check if you have the correct version of GraphLab Create. Any version later than 1.8.5 will do. To upgrade, read [this page](https://turi.com/download/upgrade-graphlab-create.html).",
"_____no_output_____"
]
],
[
[
"import graphlab as gl\nimport numpy as np\nimport matplotlib.pyplot as plt \n\n%matplotlib inline\n\n'''Check GraphLab Create version'''\nfrom distutils.version import StrictVersion\nassert (StrictVersion(gl.version) >= StrictVersion('1.8.5')), 'GraphLab Create must be version 1.8.5 or later.'",
"This non-commercial license of GraphLab Create for academic use is assigned to [email protected] and will expire on September 18, 2017.\n"
],
[
"# import wiki data\nwiki = gl.SFrame('people_wiki.gl/')\nwiki",
"_____no_output_____"
]
],
[
[
"In the original data, each Wikipedia article is represented by a URI, a name, and a string containing the entire text of the article. Recall from the video lectures that LDA requires documents to be represented as a _bag of words_, which ignores word ordering in the document but retains information on how many times each word appears. As we have seen in our previous encounters with text data, words such as 'the', 'a', or 'and' are by far the most frequent, but they appear so commonly in the English language that they tell us almost nothing about how similar or dissimilar two documents might be. \n\nTherefore, before we train our LDA model, we will preprocess the Wikipedia data in two steps: first, we will create a bag of words representation for each article, and then we will remove the common words that don't help us to distinguish between documents. For both of these tasks we can use pre-implemented tools from GraphLab Create:",
"_____no_output_____"
]
],
[
[
"wiki_docs = gl.text_analytics.count_words(wiki['text'])\nwiki_docs = wiki_docs.dict_trim_by_keys(gl.text_analytics.stopwords(), exclude=True)",
"_____no_output_____"
],
[
"wiki_docs",
"_____no_output_____"
]
],
[
[
"## Model fitting and interpretation\nIn the video lectures we saw that Gibbs sampling can be used to perform inference in the LDA model. In this assignment we will use a GraphLab Create method to learn the topic model for our Wikipedia data, and our main emphasis will be on interpreting the results. We'll begin by creating the topic model using create() from GraphLab Create's topic_model module.\n\nNote: This may take several minutes to run.",
"_____no_output_____"
]
],
[
[
"topic_model = gl.topic_model.create(wiki_docs, num_topics=10, num_iterations=200)",
"_____no_output_____"
]
],
[
[
"GraphLab provides a useful summary of the model we have fitted, including the hyperparameter settings for alpha, gamma (note that GraphLab Create calls this parameter beta), and K (the number of topics); the structure of the output data; and some useful methods for understanding the results.",
"_____no_output_____"
]
],
[
[
"topic_model",
"_____no_output_____"
]
],
[
[
"It is certainly useful to have pre-implemented methods available for LDA, but as with our previous methods for clustering and retrieval, implementing and fitting the model gets us only halfway towards our objective. We now need to analyze the fitted model to understand what it has done with our data and whether it will be useful as a document classification system. This can be a challenging task in itself, particularly when the model that we use is complex. We will begin by outlining a sequence of objectives that will help us understand our model in detail. In particular, we will\n\n* get the top words in each topic and use these to identify topic themes\n* predict topic distributions for some example documents\n* compare the quality of LDA \"nearest neighbors\" to the NN output from the first assignment\n* understand the role of model hyperparameters alpha and gamma",
"_____no_output_____"
],
[
"## Load a fitted topic model\nThe method used to fit the LDA model is a _randomized algorithm_, which means that it involves steps that are random; in this case, the randomness comes from Gibbs sampling, as discussed in the LDA video lectures. Because of these random steps, the algorithm will be expected to yield slighty different output for different runs on the same data - note that this is different from previously seen algorithms such as k-means or EM, which will always produce the same results given the same input and initialization.\n\nIt is important to understand that variation in the results is a fundamental feature of randomized methods. However, in the context of this assignment this variation makes it difficult to evaluate the correctness of your analysis, so we will load and analyze a pre-trained model. \n\nWe recommend that you spend some time exploring your own fitted topic model and compare our analysis of the pre-trained model to the same analysis applied to the model you trained above.",
"_____no_output_____"
]
],
[
[
"topic_model = gl.load_model('lda_assignment_topic_model')",
"_____no_output_____"
]
],
[
[
"# Identifying topic themes by top words\n\nWe'll start by trying to identify the topics learned by our model with some major themes. As a preliminary check on the results of applying this method, it is reasonable to hope that the model has been able to learn topics that correspond to recognizable categories. In order to do this, we must first recall what exactly a 'topic' is in the context of LDA. \n\nIn the video lectures on LDA we learned that a topic is a probability distribution over words in the vocabulary; that is, each topic assigns a particular probability to every one of the unique words that appears in our data. Different topics will assign different probabilities to the same word: for instance, a topic that ends up describing science and technology articles might place more probability on the word 'university' than a topic that describes sports or politics. Looking at the highest probability words in each topic will thus give us a sense of its major themes. Ideally we would find that each topic is identifiable with some clear theme _and_ that all the topics are relatively distinct.\n\nWe can use the GraphLab Create function get_topics() to view the top words (along with their associated probabilities) from each topic.\n\n__Quiz Question:__ Identify the top 3 most probable words for the first topic. ",
"_____no_output_____"
]
],
[
[
"topic_model",
"_____no_output_____"
],
[
"topic_model.get_topics()",
"_____no_output_____"
]
],
[
[
"__ Quiz Question:__ What is the sum of the probabilities assigned to the top 50 words in the 3rd topic?",
"_____no_output_____"
]
],
[
[
"topic_model.get_topics(topic_ids=[2], num_words=50)['score'].sum()",
"_____no_output_____"
]
],
[
[
"Let's look at the top 10 words for each topic to see if we can identify any themes:",
"_____no_output_____"
]
],
[
[
"[x['words'] for x in topic_model.get_topics(output_type='topic_words', num_words=10)]",
"_____no_output_____"
]
],
[
[
"We propose the following themes for each topic:\n\n- topic 0: Science and research\n- topic 1: Team sports\n- topic 2: Music, TV, and film\n- topic 3: American college and politics\n- topic 4: General politics\n- topic 5: Art and publishing\n- topic 6: Business\n- topic 7: International athletics\n- topic 8: Great Britain and Australia\n- topic 9: International music\n\nWe'll save these themes for later:",
"_____no_output_____"
]
],
[
[
"themes = ['science and research','team sports','music, TV, and film','American college and politics','general politics', \\\n 'art and publishing','Business','international athletics','Great Britain and Australia','international music']",
"_____no_output_____"
]
],
[
[
"### Measuring the importance of top words\n\nWe can learn more about topics by exploring how they place probability mass (which we can think of as a weight) on each of their top words.\n\nWe'll do this with two visualizations of the weights for the top words in each topic:\n - the weights of the top 100 words, sorted by the size\n - the total weight of the top 10 words\n",
"_____no_output_____"
],
[
"Here's a plot for the top 100 words by weight in each topic:",
"_____no_output_____"
]
],
[
[
"for i in range(10):\n plt.plot(range(100), topic_model.get_topics(topic_ids=[i], num_words=100)['score'])\nplt.xlabel('Word rank')\nplt.ylabel('Probability')\nplt.title('Probabilities of Top 100 Words in each Topic')",
"_____no_output_____"
]
],
[
[
"In the above plot, each line corresponds to one of our ten topics. Notice how for each topic, the weights drop off sharply as we move down the ranked list of most important words. This shows that the top 10-20 words in each topic are assigned a much greater weight than the remaining words - and remember from the summary of our topic model that our vocabulary has 547462 words in total!\n\n\nNext we plot the total weight assigned by each topic to its top 10 words: ",
"_____no_output_____"
]
],
[
[
"top_probs = [sum(topic_model.get_topics(topic_ids=[i], num_words=10)['score']) for i in range(10)]\n\nind = np.arange(10)\nwidth = 0.5\n\nfig, ax = plt.subplots()\n\nax.bar(ind-(width/2),top_probs,width)\nax.set_xticks(ind)\n\nplt.xlabel('Topic')\nplt.ylabel('Probability')\nplt.title('Total Probability of Top 10 Words in each Topic')\nplt.xlim(-0.5,9.5)\nplt.ylim(0,0.15)\nplt.show()",
"_____no_output_____"
]
],
[
[
"Here we see that, for our topic model, the top 10 words only account for a small fraction (in this case, between 5% and 13%) of their topic's total probability mass. So while we can use the top words to identify broad themes for each topic, we should keep in mind that in reality these topics are more complex than a simple 10-word summary.\n\nFinally, we observe that some 'junk' words appear highly rated in some topics despite our efforts to remove unhelpful words before fitting the model; for example, the word 'born' appears as a top 10 word in three different topics, but it doesn't help us describe these topics at all.",
"_____no_output_____"
],
[
"# Topic distributions for some example documents\n\nAs we noted in the introduction to this assignment, LDA allows for mixed membership, which means that each document can partially belong to several different topics. For each document, topic membership is expressed as a vector of weights that sum to one; the magnitude of each weight indicates the degree to which the document represents that particular topic.\n\nWe'll explore this in our fitted model by looking at the topic distributions for a few example Wikipedia articles from our data set. We should find that these articles have the highest weights on the topics whose themes are most relevant to the subject of the article - for example, we'd expect an article on a politician to place relatively high weight on topics related to government, while an article about an athlete should place higher weight on topics related to sports or competition.",
"_____no_output_____"
],
[
"Topic distributions for documents can be obtained using GraphLab Create's predict() function. GraphLab Create uses a collapsed Gibbs sampler similar to the one described in the video lectures, where only the word assignments variables are sampled. To get a document-specific topic proportion vector post-facto, predict() draws this vector from the conditional distribution given the sampled word assignments in the document. Notice that, since these are draws from a _distribution_ over topics that the model has learned, we will get slightly different predictions each time we call this function on a document - we can see this below, where we predict the topic distribution for the article on Barack Obama:",
"_____no_output_____"
]
],
[
[
"obama = gl.SArray([wiki_docs[int(np.where(wiki['name']=='Barack Obama')[0])]])\npred1 = topic_model.predict(obama, output_type='probability')\npred2 = topic_model.predict(obama, output_type='probability')\nprint(gl.SFrame({'topics':themes, 'predictions (first draw)':pred1[0], 'predictions (second draw)':pred2[0]}))",
"+--------------------------+---------------------------+\n| predictions (first draw) | predictions (second draw) |\n+--------------------------+---------------------------+\n| 0.0403225806452 | 0.0215053763441 |\n| 0.0591397849462 | 0.0483870967742 |\n| 0.0215053763441 | 0.0241935483871 |\n| 0.10752688172 | 0.155913978495 |\n| 0.604838709677 | 0.545698924731 |\n| 0.0134408602151 | 0.0215053763441 |\n| 0.0483870967742 | 0.0645161290323 |\n| 0.0510752688172 | 0.0618279569892 |\n| 0.0188172043011 | 0.0322580645161 |\n| 0.0349462365591 | 0.0241935483871 |\n+--------------------------+---------------------------+\n+-------------------------------+\n| topics |\n+-------------------------------+\n| science and research |\n| team sports |\n| music, TV, and film |\n| American college and politics |\n| general politics |\n| art and publishing |\n| Business |\n| international athletics |\n| Great Britain and Australia |\n| international music |\n+-------------------------------+\n[10 rows x 3 columns]\n\n"
]
],
[
[
"To get a more robust estimate of the topics for each document, we can average a large number of predictions for the same document:",
"_____no_output_____"
]
],
[
[
"def average_predictions(model, test_document, num_trials=100):\n avg_preds = np.zeros((model.num_topics))\n for i in range(num_trials):\n avg_preds += model.predict(test_document, output_type='probability')[0]\n avg_preds = avg_preds/num_trials\n result = gl.SFrame({'topics':themes, 'average predictions':avg_preds})\n result = result.sort('average predictions', ascending=False)\n return result",
"_____no_output_____"
],
[
"print average_predictions(topic_model, obama, 100)",
"+---------------------+-------------------------------+\n| average predictions | topics |\n+---------------------+-------------------------------+\n| 0.59561827957 | general politics |\n| 0.138709677419 | American college and politics |\n| 0.0526075268817 | Business |\n| 0.0476612903226 | team sports |\n| 0.0408333333333 | science and research |\n| 0.0337634408602 | international athletics |\n| 0.0252150537634 | Great Britain and Australia |\n| 0.022876344086 | art and publishing |\n| 0.0222849462366 | international music |\n| 0.0204301075269 | music, TV, and film |\n+---------------------+-------------------------------+\n[10 rows x 2 columns]\n\n"
]
],
[
[
"__Quiz Question:__ What is the topic most closely associated with the article about former US President George W. Bush? Use the average results from 100 topic predictions.",
"_____no_output_____"
]
],
[
[
"bush = gl.SArray([wiki_docs[int(np.where(wiki['name']=='George W. Bush')[0])]])\nprint average_predictions(topic_model, bush, 100)",
"+---------------------+-------------------------------+\n| average predictions | topics |\n+---------------------+-------------------------------+\n| 0.428567251462 | general politics |\n| 0.190614035088 | American college and politics |\n| 0.101081871345 | Business |\n| 0.0594444444444 | science and research |\n| 0.0485964912281 | art and publishing |\n| 0.045730994152 | team sports |\n| 0.0383040935673 | Great Britain and Australia |\n| 0.0367251461988 | international athletics |\n| 0.0295906432749 | music, TV, and film |\n| 0.0213450292398 | international music |\n+---------------------+-------------------------------+\n[10 rows x 2 columns]\n\n"
]
],
[
[
"__Quiz Question:__ What are the top 3 topics corresponding to the article about English football (soccer) player Steven Gerrard? Use the average results from 100 topic predictions.",
"_____no_output_____"
]
],
[
[
"gerrard = gl.SArray([wiki_docs[int(np.where(wiki['name']=='Steven Gerrard')[0])]])\nprint average_predictions(topic_model, gerrard, 100)",
"+---------------------+-------------------------------+\n| average predictions | topics |\n+---------------------+-------------------------------+\n| 0.48896 | team sports |\n| 0.17488 | Great Britain and Australia |\n| 0.12724 | international athletics |\n| 0.03732 | international music |\n| 0.03196 | general politics |\n| 0.03176 | music, TV, and film |\n| 0.02948 | Business |\n| 0.02772 | art and publishing |\n| 0.0264 | American college and politics |\n| 0.02428 | science and research |\n+---------------------+-------------------------------+\n[10 rows x 2 columns]\n\n"
]
],
[
[
"# Comparing LDA to nearest neighbors for document retrieval\n\nSo far we have found that our topic model has learned some coherent topics, we have explored these topics as probability distributions over a vocabulary, and we have seen how individual documents in our Wikipedia data set are assigned to these topics in a way that corresponds with our expectations. \n\nIn this section, we will use the predicted topic distribution as a representation of each document, similar to how we have previously represented documents by word count or TF-IDF. This gives us a way of computing distances between documents, so that we can run a nearest neighbors search for a given document based on its membership in the topics that we learned from LDA. We can contrast the results with those obtained by running nearest neighbors under the usual TF-IDF representation, an approach that we explored in a previous assignment. \n\nWe'll start by creating the LDA topic distribution representation for each document:",
"_____no_output_____"
]
],
[
[
"wiki['lda'] = topic_model.predict(wiki_docs, output_type='probability')",
"_____no_output_____"
]
],
[
[
"Next we add the TF-IDF document representations:",
"_____no_output_____"
]
],
[
[
"wiki['word_count'] = gl.text_analytics.count_words(wiki['text'])\nwiki['tf_idf'] = gl.text_analytics.tf_idf(wiki['word_count'])",
"_____no_output_____"
]
],
[
[
"For each of our two different document representations, we can use GraphLab Create to compute a brute-force nearest neighbors model:",
"_____no_output_____"
]
],
[
[
"model_tf_idf = gl.nearest_neighbors.create(wiki, label='name', features=['tf_idf'],\n method='brute_force', distance='cosine')\nmodel_lda_rep = gl.nearest_neighbors.create(wiki, label='name', features=['lda'],\n method='brute_force', distance='cosine')",
"_____no_output_____"
]
],
[
[
"Let's compare these nearest neighbor models by finding the nearest neighbors under each representation on an example document. For this example we'll use Paul Krugman, an American economist:",
"_____no_output_____"
]
],
[
[
"model_tf_idf.query(wiki[wiki['name'] == 'Paul Krugman'], label='name', k=10)",
"_____no_output_____"
],
[
"model_lda_rep.query(wiki[wiki['name'] == 'Paul Krugman'], label='name', k=10)",
"_____no_output_____"
]
],
[
[
"Notice that that there is no overlap between the two sets of top 10 nearest neighbors. This doesn't necessarily mean that one representation is better or worse than the other, but rather that they are picking out different features of the documents. \n\nWith TF-IDF, documents are distinguished by the frequency of uncommon words. Since similarity is defined based on the specific words used in the document, documents that are \"close\" under TF-IDF tend to be similar in terms of specific details. This is what we see in the example: the top 10 nearest neighbors are all economists from the US, UK, or Canada. \n\nOur LDA representation, on the other hand, defines similarity between documents in terms of their topic distributions. This means that documents can be \"close\" if they share similar themes, even though they may not share many of the same keywords. For the article on Paul Krugman, we expect the most important topics to be 'American college and politics' and 'science and research'. As a result, we see that the top 10 nearest neighbors are academics from a wide variety of fields, including literature, anthropology, and religious studies.\n\n\n__Quiz Question:__ Using the TF-IDF representation, compute the 5000 nearest neighbors for American baseball player Alex Rodriguez. For what value of k is Mariano Rivera the k-th nearest neighbor to Alex Rodriguez? (Hint: Once you have a list of the nearest neighbors, you can use `mylist.index(value)` to find the index of the first instance of `value` in `mylist`.)\n\n__Quiz Question:__ Using the LDA representation, compute the 5000 nearest neighbors for American baseball player Alex Rodriguez. For what value of k is Mariano Rivera the k-th nearest neighbor to Alex Rodriguez? (Hint: Once you have a list of the nearest neighbors, you can use `mylist.index(value)` to find the index of the first instance of `value` in `mylist`.)",
"_____no_output_____"
]
],
[
[
"nn_tfidf = model_tf_idf.query(wiki[wiki['name'] == 'Alex Rodriguez'], label='name', k=5000)\nnn_lda = model_lda_rep.query(wiki[wiki['name'] == 'Alex Rodriguez'], label='name', k=5000)",
"_____no_output_____"
],
[
"nn_tfidf[nn_tfidf['reference_label'] == 'Mariano Rivera']",
"_____no_output_____"
],
[
"nn_lda[nn_lda['reference_label'] == 'Mariano Rivera']",
"_____no_output_____"
]
],
[
[
"# Understanding the role of LDA model hyperparameters\n\nFinally, we'll take a look at the effect of the LDA model hyperparameters alpha and gamma on the characteristics of our fitted model. Recall that alpha is a parameter of the prior distribution over topic weights in each document, while gamma is a parameter of the prior distribution over word weights in each topic. \n\nIn the video lectures, we saw that alpha and gamma can be thought of as smoothing parameters when we compute how much each document \"likes\" a topic (in the case of alpha) or how much each topic \"likes\" a word (in the case of gamma). In both cases, these parameters serve to reduce the differences across topics or words in terms of these calculated preferences; alpha makes the document preferences \"smoother\" over topics, and gamma makes the topic preferences \"smoother\" over words.\n\nOur goal in this section will be to understand how changing these parameter values affects the characteristics of the resulting topic model.\n\n__Quiz Question:__ What was the value of alpha used to fit our original topic model? ",
"_____no_output_____"
]
],
[
[
"topic_model",
"_____no_output_____"
]
],
[
[
"__Quiz Question:__ What was the value of gamma used to fit our original topic model? Remember that GraphLab Create uses \"beta\" instead of \"gamma\" to refer to the hyperparameter that influences topic distributions over words.",
"_____no_output_____"
],
[
"We'll start by loading some topic models that have been trained using different settings of alpha and gamma. Specifically, we will start by comparing the following two models to our original topic model:\n - tpm_low_alpha, a model trained with alpha = 1 and default gamma\n - tpm_high_alpha, a model trained with alpha = 50 and default gamma",
"_____no_output_____"
]
],
[
[
"tpm_low_alpha = gl.load_model('lda_low_alpha')\ntpm_high_alpha = gl.load_model('lda_high_alpha')",
"_____no_output_____"
]
],
[
[
"### Changing the hyperparameter alpha\n\nSince alpha is responsible for smoothing document preferences over topics, the impact of changing its value should be visible when we plot the distribution of topic weights for the same document under models fit with different alpha values. In the code below, we plot the (sorted) topic weights for the Wikipedia article on Barack Obama under models fit with high, original, and low settings of alpha.",
"_____no_output_____"
]
],
[
[
"a = np.sort(tpm_low_alpha.predict(obama,output_type='probability')[0])[::-1]\nb = np.sort(topic_model.predict(obama,output_type='probability')[0])[::-1]\nc = np.sort(tpm_high_alpha.predict(obama,output_type='probability')[0])[::-1]\nind = np.arange(len(a))\nwidth = 0.3\n\ndef param_bar_plot(a,b,c,ind,width,ylim,param,xlab,ylab):\n fig = plt.figure()\n ax = fig.add_subplot(111)\n\n b1 = ax.bar(ind, a, width, color='lightskyblue')\n b2 = ax.bar(ind+width, b, width, color='lightcoral')\n b3 = ax.bar(ind+(2*width), c, width, color='gold')\n\n ax.set_xticks(ind+width)\n ax.set_xticklabels(range(10))\n ax.set_ylabel(ylab)\n ax.set_xlabel(xlab)\n ax.set_ylim(0,ylim)\n ax.legend(handles = [b1,b2,b3],labels=['low '+param,'original model','high '+param])\n\n plt.tight_layout()\n \nparam_bar_plot(a,b,c,ind,width,ylim=1.0,param='alpha',\n xlab='Topics (sorted by weight of top 100 words)',ylab='Topic Probability for Obama Article')",
"_____no_output_____"
]
],
[
[
"Here we can clearly see the smoothing enforced by the alpha parameter - notice that when alpha is low most of the weight in the topic distribution for this article goes to a single topic, but when alpha is high the weight is much more evenly distributed across the topics.\n\n__Quiz Question:__ How many topics are assigned a weight greater than 0.3 or less than 0.05 for the article on Paul Krugman in the **low alpha** model? Use the average results from 100 topic predictions.",
"_____no_output_____"
]
],
[
[
"paul = gl.SArray([wiki_docs[int(np.where(wiki['name']=='Paul Krugman')[0])]])\npaul_nn_low = np.sort(tpm_low_alpha.predict(paul,output_type='probability')[0])\npaul_nn_low",
"_____no_output_____"
]
],
[
[
"__Quiz Question:__ How many topics are assigned a weight greater than 0.3 or less than 0.05 for the article on Paul Krugman in the **high alpha** model? Use the average results from 100 topic predictions.",
"_____no_output_____"
]
],
[
[
"paul_nn_high = np.sort(tpm_high_alpha.predict(paul,output_type='probability')[0])\npaul_nn_high",
"_____no_output_____"
]
],
[
[
"### Changing the hyperparameter gamma\n\nJust as we were able to see the effect of alpha by plotting topic weights for a document, we expect to be able to visualize the impact of changing gamma by plotting word weights for each topic. In this case, however, there are far too many words in our vocabulary to do this effectively. Instead, we'll plot the total weight of the top 100 words and bottom 1000 words for each topic. Below, we plot the (sorted) total weights of the top 100 words and bottom 1000 from each topic in the high, original, and low gamma models.",
"_____no_output_____"
],
[
"Now we will consider the following two models:\n - tpm_low_gamma, a model trained with gamma = 0.02 and default alpha\n - tpm_high_gamma, a model trained with gamma = 0.5 and default alpha",
"_____no_output_____"
]
],
[
[
"del tpm_low_alpha\ndel tpm_high_alpha\ntpm_low_gamma = gl.load_model('lda_low_gamma')\ntpm_high_gamma = gl.load_model('lda_high_gamma')",
"_____no_output_____"
],
[
"a_top = np.sort([sum(tpm_low_gamma.get_topics(topic_ids=[i], num_words=100)['score']) for i in range(10)])[::-1]\nb_top = np.sort([sum(topic_model.get_topics(topic_ids=[i], num_words=100)['score']) for i in range(10)])[::-1]\nc_top = np.sort([sum(tpm_high_gamma.get_topics(topic_ids=[i], num_words=100)['score']) for i in range(10)])[::-1]\n\na_bot = np.sort([sum(tpm_low_gamma.get_topics(topic_ids=[i], num_words=547462)[-1000:]['score']) for i in range(10)])[::-1]\nb_bot = np.sort([sum(topic_model.get_topics(topic_ids=[i], num_words=547462)[-1000:]['score']) for i in range(10)])[::-1]\nc_bot = np.sort([sum(tpm_high_gamma.get_topics(topic_ids=[i], num_words=547462)[-1000:]['score']) for i in range(10)])[::-1]\n\nind = np.arange(len(a))\nwidth = 0.3\n \nparam_bar_plot(a_top, b_top, c_top, ind, width, ylim=0.6, param='gamma',\n xlab='Topics (sorted by weight of top 100 words)', \n ylab='Total Probability of Top 100 Words')\n\nparam_bar_plot(a_bot, b_bot, c_bot, ind, width, ylim=0.0002, param='gamma',\n xlab='Topics (sorted by weight of bottom 1000 words)',\n ylab='Total Probability of Bottom 1000 Words')",
"_____no_output_____"
]
],
[
[
"From these two plots we can see that the low gamma model results in higher weight placed on the top words and lower weight placed on the bottom words for each topic, while the high gamma model places relatively less weight on the top words and more weight on the bottom words. Thus increasing gamma results in topics that have a smoother distribution of weight across all the words in the vocabulary.",
"_____no_output_____"
],
[
"__Quiz Question:__ For each topic of the **low gamma model**, compute the number of words required to make a list with total probability 0.5. What is the average number of words required across all topics? (HINT: use the get\\_topics() function from GraphLab Create with the cdf\\_cutoff argument).",
"_____no_output_____"
]
],
[
[
"tpm_low_gamma",
"_____no_output_____"
],
[
"count = 0\nfor i, topic in enumerate(themes):\n c = len(tpm_low_gamma.get_topics(topic_ids=[i], cdf_cutoff=0.5, num_words=547462))\n count += c\n print \"# of words required for topic - \" + topic + \" = \" + str(c)\n \ncount / float(len(themes))",
"# of words required for topic - science and research = 184\n# of words required for topic - team sports = 163\n# of words required for topic - music, TV, and film = 329\n# of words required for topic - American college and politics = 139\n# of words required for topic - general politics = 189\n# of words required for topic - art and publishing = 153\n# of words required for topic - Business = 295\n# of words required for topic - international athletics = 482\n# of words required for topic - Great Britain and Australia = 164\n# of words required for topic - international music = 426\n"
]
],
[
[
"__Quiz Question:__ For each topic of the **high gamma model**, compute the number of words required to make a list with total probability 0.5. What is the average number of words required across all topics? (HINT: use the get\\_topics() function from GraphLab Create with the cdf\\_cutoff argument).",
"_____no_output_____"
]
],
[
[
"count = 0\nfor i, topic in enumerate(themes):\n c = len(tpm_high_gamma.get_topics(topic_ids=[i], cdf_cutoff=0.5, num_words=547462))\n count += c\n print \"# of words required for topic - \" + topic + \" = \" + str(c)\n \ncount / float(len(themes))",
"# of words required for topic - science and research = 318\n# of words required for topic - team sports = 519\n# of words required for topic - music, TV, and film = 701\n# of words required for topic - American college and politics = 545\n# of words required for topic - general politics = 379\n# of words required for topic - art and publishing = 384\n# of words required for topic - Business = 572\n# of words required for topic - international athletics = 467\n# of words required for topic - Great Britain and Australia = 1303\n# of words required for topic - international music = 574\n"
]
],
[
[
"We have now seen how the hyperparameters alpha and gamma influence the characteristics of our LDA topic model, but we haven't said anything about what settings of alpha or gamma are best. We know that these parameters are responsible for controlling the smoothness of the topic distributions for documents and word distributions for topics, but there's no simple conversion between smoothness of these distributions and quality of the topic model. In reality, there is no universally \"best\" choice for these parameters. Instead, finding a good topic model requires that we be able to both explore the output (as we did by looking at the topics and checking some topic predictions for documents) and understand the impact of hyperparameter settings (as we have in this section).",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown"
] |
[
[
"markdown",
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
]
] |
cbd9851b54e64b0ba83fe6c02c49dd0e223d7e6b
| 13,416 |
ipynb
|
Jupyter Notebook
|
cvicenia/tyzden-07/IAU_071_nlp.ipynb
|
FIIT-IAU/2021-2022
|
1ee71c0db8f4c27401a2b2428886c0e7b4df3b97
|
[
"Apache-2.0"
] | 8 |
2021-09-20T15:33:02.000Z
|
2022-02-03T00:48:50.000Z
|
cvicenia/tyzden-07/IAU_071_nlp.ipynb
|
FIIT-IAU/2021-2022
|
1ee71c0db8f4c27401a2b2428886c0e7b4df3b97
|
[
"Apache-2.0"
] | null | null | null |
cvicenia/tyzden-07/IAU_071_nlp.ipynb
|
FIIT-IAU/2021-2022
|
1ee71c0db8f4c27401a2b2428886c0e7b4df3b97
|
[
"Apache-2.0"
] | 4 |
2021-10-05T10:58:59.000Z
|
2021-12-17T20:24:30.000Z
| 30.421769 | 150 | 0.576998 |
[
[
[
"# -*- coding: utf-8 -*-\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or\n# implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n#",
"_____no_output_____"
]
],
[
[
"# Natural Language Processing (NLP) with machine learning (ML)\n\n**Preprocessing of textual data**",
"_____no_output_____"
],
[
"### Download NLTK data - we need to do this only one time\nThe download process can last longer (with GUI) and all data packages are bigger size of 3.3 GB\n\nUncomment the *nltk.download()* line to download all! It open a new download window, which requires to click !",
"_____no_output_____"
]
],
[
[
"import nltk\n# nltk.download()\n# nltk.download('punkt')\n# nltk.download('stopwords')\n# nltk.download('averaged_perceptron_tagger') # Part-of-Speech Tagging (POS)\n# nltk.download('tagsets')\n# nltk.download('maxent_ne_chunker') # Name Entity Recognition (NER)\n# nltk.download('words')",
"_____no_output_____"
]
],
[
[
"### Tokenization",
"_____no_output_____"
]
],
[
[
"import string\nimport re\nfrom nltk.tokenize import word_tokenize\nfrom nltk.corpus import stopwords\n\n# load data\nfilename = 'data/metamorphosis_clean.txt'\nfile = open(filename, 'rt')\ntext = file.read()\nfile.close()\n\n# split into words\ntokens = word_tokenize(text)\n\n# convert to lower case\ntokens = [w.lower() for w in tokens]\n\n# prepare regex for char filtering\nre_punc = re.compile('[%s]' % re.escape(string.punctuation)) \n\n# remove punctuation from each word\nstripped = [re_punc.sub('', w) for w in tokens]\n\n# remove remaining tokens that are not alphabetic\nwords = [word for word in stripped if word.isalpha()]\n\n# filter out stop words\nstop_words = set(stopwords.words('english'))\nwords = [w for w in words if not w in stop_words] \nprint(words[:100])",
"_____no_output_____"
]
],
[
[
"### TF-IDF with TfidfVectorizer",
"_____no_output_____"
]
],
[
[
"import pandas as pd\nfrom sklearn.feature_extraction.text import TfidfVectorizer\n\ndataset = [\n \"I enjoy reading about Machine Learning and Machine Learning is my PhD subject\",\n \"I would enjoy a walk in the park\",\n \"I was reading in the library\"\n]\n\nvectorizer = TfidfVectorizer(use_idf=True)\ntfIdf = vectorizer.fit_transform(dataset)\n\ndf = pd.DataFrame(tfIdf[0].T.todense(), index=vectorizer.get_feature_names(), columns=[\"TF-IDF\"])\ndf = df.sort_values('TF-IDF', ascending=False)\n\nprint (df.head(25))",
"_____no_output_____"
]
],
[
[
"### TF-IDF with TfidfTransformer",
"_____no_output_____"
]
],
[
[
"from sklearn.feature_extraction.text import TfidfTransformer\nfrom sklearn.feature_extraction.text import CountVectorizer\n\ntransformer = TfidfTransformer(use_idf=True)\ncountVectorizer = CountVectorizer()\n\nwordCount = countVectorizer.fit_transform(dataset)\nnewTfIdf = transformer.fit_transform(wordCount)\n\ndf = pd.DataFrame(newTfIdf[0].T.todense(), index=countVectorizer.get_feature_names(), columns=[\"TF-IDF\"])\ndf = df.sort_values('TF-IDF', ascending=False)\nprint (df.head(25))",
"_____no_output_____"
]
],
[
[
"### Cosine similarity\nURL: https://stackoverflow.com/questions/12118720/python-tf-idf-cosine-to-find-document-similarity",
"_____no_output_____"
]
],
[
[
"from sklearn.datasets import fetch_20newsgroups\nfrom sklearn.feature_extraction.text import TfidfVectorizer\nfrom sklearn.metrics.pairwise import linear_kernel\n\n# twenty dataset\ntwenty = fetch_20newsgroups()\ntfidf = TfidfVectorizer().fit_transform(twenty.data)\n\n# cosine similarity\ncosine_similarities = linear_kernel(tfidf[0:1], tfidf).flatten()\n\n# top-5 related documents\nrelated_docs_indices = cosine_similarities.argsort()[:-5:-1]\nprint(related_docs_indices)\nprint(cosine_similarities[related_docs_indices])\n\n# print the first result to check\nprint(twenty.data[0])\nprint(twenty.data[958])",
"_____no_output_____"
]
],
[
[
"### Text classification\n\nURL https://towardsdatascience.com/machine-learning-nlp-text-classification-using-scikit-learn-python-and-nltk-c52b92a7c73a",
"_____no_output_____"
]
],
[
[
"import numpy as np\nfrom sklearn.datasets import fetch_20newsgroups\nfrom sklearn.feature_extraction.text import CountVectorizer\nfrom sklearn.feature_extraction.text import TfidfTransformer\nfrom sklearn.pipeline import Pipeline\n\n# twenty dataset\ntwenty_train = fetch_20newsgroups(subset='train', shuffle=True)\ntwenty_test = fetch_20newsgroups(subset='test', shuffle=True)\n\nprint(twenty_train.target_names)\n# print(\"\\n\".join(twenty_train.data[0].split(\"\\n\")[:3]))",
"_____no_output_____"
]
],
[
[
"### Multinomial Naive Bayes",
"_____no_output_____"
]
],
[
[
"from sklearn.naive_bayes import MultinomialNB\n\n# Bag-of-words\ncount_vect = CountVectorizer()\nX_train_counts = count_vect.fit_transform(twenty_train.data)\nX_test_counts = count_vect.transform(twenty_test.data)\n\n# TF-IDF\ntransformer = TfidfTransformer()\nX_train_tfidf = transformer.fit_transform(X_train_counts)\nX_test_tfidf = transformer.transform(X_test_counts)\n\n# Naive Bayes (NB) for text classification\nclf = MultinomialNB().fit(X_train_tfidf, twenty_train.target)\n\n# Performance of the model\npredicted = clf.predict(X_test_tfidf)\nnp.mean(predicted == twenty_test.target)",
"_____no_output_____"
]
],
[
[
"### Pipeline\nThe above code with Multinomial Naive Bayes can be written more ellegant with scikit-learn pipeline. The code will be shorter and more reliable.",
"_____no_output_____"
]
],
[
[
"text_clf = Pipeline([('vect', CountVectorizer(stop_words='english')),\n ('tfidf', TfidfTransformer()),\n ('clf', MultinomialNB()),\n ])\ntext_clf = text_clf.fit(twenty_train.data, twenty_train.target)\n\n# Performance of the model\npredicted = text_clf.predict(twenty_test.data)\nnp.mean(predicted == twenty_test.target)",
"_____no_output_____"
]
],
[
[
"### GridSearchCV with Naive Bayes\nWe want and need to optimize the pipeline by hyper-parameter tunning. We may get some better classification results.",
"_____no_output_____"
]
],
[
[
"from sklearn.model_selection import GridSearchCV\n\nparameters = {'vect__ngram_range': [(1, 1), (1, 2)],\n 'tfidf__use_idf': (True, False),\n 'clf__alpha': (1e-2, 1e-3),\n }\ngs_clf = GridSearchCV(text_clf, parameters, n_jobs=-1)\ngs_clf = gs_clf.fit(twenty_train.data, twenty_train.target)\n\nprint(gs_clf.best_score_)\nprint(gs_clf.best_params_)",
"_____no_output_____"
]
],
[
[
"### SGDClassifier\nWe are trying another classifier called SGDClassifier instead of the previous Multinomial Naive Bayes. \n\nLet see if this new classifier acts better incomparison with and without optimization.",
"_____no_output_____"
]
],
[
[
"from sklearn.linear_model import SGDClassifier\n\ntext_clf_svm = Pipeline([('vect', CountVectorizer()),\n ('tfidf', TfidfTransformer()),\n ('clf-svm', SGDClassifier(loss='hinge', \n penalty='l2',\n alpha=1e-3, \n random_state=42)),\n ])\ntext_clf_svm = text_clf_svm.fit(twenty_train.data, twenty_train.target)\n\n# Performance of the model\npredicted_svm = text_clf_svm.predict(twenty_test.data)\nnp.mean(predicted_svm == twenty_test.target)",
"_____no_output_____"
]
],
[
[
"### GridSearchCV with SVM\n\nHere a more classifiers, e.g., SVM.\n\nWe are going to try SVM with Grid Search optimization.",
"_____no_output_____"
]
],
[
[
"from sklearn.model_selection import GridSearchCV\n\nparameters_svm = {'vect__ngram_range': [(1, 1), (1, 2)],\n 'tfidf__use_idf': (True, False),\n 'clf-svm__alpha': (1e-2, 1e-3),\n }\ngs_clf_svm = GridSearchCV(text_clf_svm, parameters_svm, n_jobs=-1)\ngs_clf_svm = gs_clf_svm.fit(twenty_train.data, twenty_train.target)\n\nprint(gs_clf_svm.best_score_)\nprint(gs_clf_svm.best_params_)",
"_____no_output_____"
]
],
[
[
"### Stemming\n\nStemming can improve classifier results too. Let see if it works in our case example with Multinomial Naive Bayes.",
"_____no_output_____"
]
],
[
[
"from nltk.stem.snowball import SnowballStemmer\nstemmer = SnowballStemmer(\"english\", ignore_stopwords=True)\n\nclass StemmedCountVectorizer(CountVectorizer):\n def build_analyzer(self):\n analyzer = super(StemmedCountVectorizer, self).build_analyzer()\n return lambda doc: ([stemmer.stem(w) for w in analyzer(doc)])\n \nstemmed_count_vect = StemmedCountVectorizer(stop_words='english')\n\ntext_mnb_stemmed = Pipeline([('vect', stemmed_count_vect), \n ('tfidf', TfidfTransformer()), \n ('mnb', MultinomialNB(fit_prior=False))])\n\ntext_mnb_stemmed = text_mnb_stemmed.fit(twenty_train.data, twenty_train.target)\n\npredicted_mnb_stemmed = text_mnb_stemmed.predict(twenty_test.data)\n\nnp.mean(predicted_mnb_stemmed == twenty_test.target)",
"_____no_output_____"
]
]
] |
[
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
]
] |
cbd98773cc7f7f11c39734133012338cd7f0962c
| 239,821 |
ipynb
|
Jupyter Notebook
|
base_algos/R/SVM.ipynb
|
vivekec/machineLearning
|
588a29218fbce0d0cc05dd044f6351e0c6e0aee4
|
[
"MIT"
] | null | null | null |
base_algos/R/SVM.ipynb
|
vivekec/machineLearning
|
588a29218fbce0d0cc05dd044f6351e0c6e0aee4
|
[
"MIT"
] | null | null | null |
base_algos/R/SVM.ipynb
|
vivekec/machineLearning
|
588a29218fbce0d0cc05dd044f6351e0c6e0aee4
|
[
"MIT"
] | 1 |
2021-06-20T19:22:16.000Z
|
2021-06-20T19:22:16.000Z
| 384.328526 | 49,910 | 0.918915 |
[
[
[
"library(e1071)\nlibrary(ggplot2)",
"_____no_output_____"
],
[
"# reading data\nsvm_data <- read.csv(\"../../datasets/knn.csv\")\nhead(svm_data)",
"_____no_output_____"
],
[
"# building SVM model\nsvm_model <- svm(level~., data = svm_data, kernel = \"linear\")\nsvm_model",
"_____no_output_____"
],
[
"# visualizing the model\noptions(repr.plot.width=6, repr.plot.height=5)\nplot(svm_model, svm_data)",
"_____no_output_____"
],
[
"# accuracy of the model\naddmargins(table(predict(svm_model,svm_data),svm_data$level))\n# 100% accurate",
"_____no_output_____"
],
[
"# Slack Variable\nsvm_model$epsilon",
"_____no_output_____"
],
[
"# Regularization parameter C controls margin length. It is inversely proportional to margin length.\nsvm_model$cost # smaller value is acceptable",
"_____no_output_____"
],
[
"# Comparing SVM model on different values of C\nsvm_c_10000 <- svm(level~., data = svm_data, kernel = \"linear\", cost = 10000)\nplot(svm_c_10000, svm_data)",
"_____no_output_____"
],
[
"svm_c_0.1 <- svm(level~., data = svm_data, kernel = \"linear\", cost = 0.1)\nplot(svm_c_0.1, svm_data)",
"_____no_output_____"
],
[
"# Best value of C - cross-validation\ntune_c <- tune(svm, level~., \n data = svm_data, \n kernel = \"linear\",\n ranges = list(cost = c(0.001, 0.01, 0.1, 1, 10, 100, 1000, 10000)))\ntune_c$best.model # Best value of C = 0.1",
"_____no_output_____"
],
[
"# Gamma is a hyper-parameter used to estimate the classification error.\n# Best value of Gamma can also be found out using CV.\ntune_gamma <- tune(svm, level~., \n data = svm_data, \n kernel = \"linear\",\n ranges = list(gamma = c(0.001, 0.01, 0.1, 1, 3, 5, 7, 10)))\ntune_gamma$best.model # Best value of gamma = 0.001",
"_____no_output_____"
],
[
"plot(tune_gamma) # No misclassified value on all the provided values to gamma.",
"_____no_output_____"
]
],
[
[
"# Non-linear kernel",
"_____no_output_____"
]
],
[
[
"svm_data <- read.csv(\"../../datasets/svm.csv\")\nhead(svm_data)\nlevels(svm_data$y)",
"_____no_output_____"
],
[
"# visualizing the data\nCategory <- svm_data$y\nggplot() +\n geom_point(aes(svm_data$X2, svm_data$X1, col = Category), cex = 2) +\n xlab(\"X2\") + ylab(\"X1\") + ggtitle(\"Category prediction using X1 and X2\") +\n theme_bw()",
"_____no_output_____"
],
[
"# builing SVM non-linear model - using tune for degree of polynomial\ntune_svm <- tune(svm, y~., data = svm_data, kernel = \"polynomial\",\n ranges = list(degree = c(1, 2, 3, 4, 5)))\ntune_svm$best.model\n# So, a polynomial of order 2 is best suited to classify given data",
"_____no_output_____"
],
[
"svm_model <- svm(y~., data = svm_data, kernel = \"polynomial\", degree = 2)\n\nplot(svm_model, svm_data)",
"_____no_output_____"
]
],
[
[
"# Multiple class classification using SVM",
"_____no_output_____"
]
],
[
[
"iris_df <- subset.data.frame(iris, select = c(\"Petal.Length\", \"Petal.Width\", \"Species\"))\n\nsvm_model <- svm(Species~., data = iris_df, kernel = \"radial\")\nplot(svm_model, iris_df)",
"_____no_output_____"
]
]
] |
[
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
]
] |
cbd9a9f3b4a502cb7f6fc3f422e396019514086d
| 380,675 |
ipynb
|
Jupyter Notebook
|
Evo/Assignment1/SimulatedAnnealing.ipynb
|
TheFebrin/Machine-Learning
|
3e58b89315960e7d4896e44075a8105fcb78f0c0
|
[
"MIT"
] | null | null | null |
Evo/Assignment1/SimulatedAnnealing.ipynb
|
TheFebrin/Machine-Learning
|
3e58b89315960e7d4896e44075a8105fcb78f0c0
|
[
"MIT"
] | null | null | null |
Evo/Assignment1/SimulatedAnnealing.ipynb
|
TheFebrin/Machine-Learning
|
3e58b89315960e7d4896e44075a8105fcb78f0c0
|
[
"MIT"
] | null | null | null | 198.785901 | 135,188 | 0.8975 |
[
[
[
"import numpy as np\nimport matplotlib.pyplot as plt\nimport time\nimport seaborn as sns\nimport urllib.request\nfrom tqdm import tqdm\nimport pandas as pd\n\n%matplotlib inline",
"_____no_output_____"
]
],
[
[
"# Quadratic Assignment Problem",
"_____no_output_____"
],
[
"## 1. Read data",
"_____no_output_____"
],
[
"Popular QAP with loss function minimums: <br>\n- Nug12 12 578 (OPT) (12,7,9,3,4,8,11,1,5,6,10,2) \n- Nug14 14 1014 (OPT) (9,8,13,2,1,11,7,14,3,4,12,5,6,10) \n- Nug15 15 1150 (OPT) (1,2,13,8,9,4,3,14,7,11,10,15,6,5,12) \n- Nug16a 16 1610 (OPT) (9,14,2,15,16,3,10,12,8,11,6,5,7,1,4,13) \n- Nug16b 16 1240 (OPT) (16,12,13,8,4,2,9,11,15,10,7,3,14,6,1,5) \n- Nug17 17 1732 (OPT) (16,15,2,14,9,11,8,12,10,3,4,1,7,6,13,17,5) \n- Nug18 18 1930 (OPT) (10,3,14,2,18,6,7,12,15,4,5,1,11,8,17,13,9,16) \n- Nug20 20 2570 (OPT) (18,14,10,3,9,4,2,12,11,16,19,15,20,8,13,17,5,7,1,6) \n- Nug21 21 2438 (OPT) (4,21,3,9,13,2,5,14,18,11,16,10,6,15,20,19,8,7,1,12,17) \n- Nug22 22 3596 (OPT) (2,21,9,10,7,3,1,19,8,20,17,5,13,6,12,16,11,22,18,14,15) \n- Nug24 24 3488 (OPT) (17,8,11,23,4,20,15,19,22,18,3,14,1,10,7,9,16,21,24,12,6,13,5,2) \n- Nug25 25 3744 (OPT) (5,11,20,15,22,2,25,8,9,1,18,16,3,6,19,24,21,14,7,10,17,12,4,23,13) * \n- Nug27 27 5234 (OPT) (23,18,3,1,27,17,5,12,7,15,4,26,8,19,20,2,24,21,14,10,9,13,22,25,6,16,11) * \n- Nug28 28 5166 (OPT) (18,21,9,1,28,20,11,3,13,12,10,19,14,22,15,2,25,16,4,23,7,17,24,26,5,27,8,6) * \n- Nug30 30 6124 (OPT) (5 12 6 13 2 21 26 24 10 9 29 28 17 1 8 7 19 25 23 22 11 16 30 4 15 18 27 3 14 20)",
"_____no_output_____"
]
],
[
[
"def get_nug(no):\n QAP_INSTANCE_URL = f'http://anjos.mgi.polymtl.ca/qaplib/data.d/nug{no}.dat'\n qap_instance_file = urllib.request.urlopen(QAP_INSTANCE_URL)\n\n line = qap_instance_file.readline()\n n = int(line.decode()[:-1].split()[0])\n \n\n A = np.empty((n, n))\n qap_instance_file.readline()\n for i in range(n):\n line = qap_instance_file.readline()\n A[i, :] = list(map(int, line.decode()[:-1].split()))\n \n\n B = np.empty((n, n))\n qap_instance_file.readline()\n for i in range(n):\n line = qap_instance_file.readline()\n B[i, :] = list(map(int, line.decode()[:-1].split()))\n \n return n, A, B",
"_____no_output_____"
],
[
"n, A, B = get_nug(12)\n\nprint('Problem size: %d' % n)\nprint('Flow matrix:\\n', A)\nprint('Distance matrix:\\n', B)",
"Problem size: 12\nFlow matrix:\n [[0. 1. 2. 3. 1. 2. 3. 4. 2. 3. 4. 5.]\n [1. 0. 1. 2. 2. 1. 2. 3. 3. 2. 3. 4.]\n [2. 1. 0. 1. 3. 2. 1. 2. 4. 3. 2. 3.]\n [3. 2. 1. 0. 4. 3. 2. 1. 5. 4. 3. 2.]\n [1. 2. 3. 4. 0. 1. 2. 3. 1. 2. 3. 4.]\n [2. 1. 2. 3. 1. 0. 1. 2. 2. 1. 2. 3.]\n [3. 2. 1. 2. 2. 1. 0. 1. 3. 2. 1. 2.]\n [4. 3. 2. 1. 3. 2. 1. 0. 4. 3. 2. 1.]\n [2. 3. 4. 5. 1. 2. 3. 4. 0. 1. 2. 3.]\n [3. 2. 3. 4. 2. 1. 2. 3. 1. 0. 1. 2.]\n [4. 3. 2. 3. 3. 2. 1. 2. 2. 1. 0. 1.]\n [5. 4. 3. 2. 4. 3. 2. 1. 3. 2. 1. 0.]]\nDistance matrix:\n [[ 0. 5. 2. 4. 1. 0. 0. 6. 2. 1. 1. 1.]\n [ 5. 0. 3. 0. 2. 2. 2. 0. 4. 5. 0. 0.]\n [ 2. 3. 0. 0. 0. 0. 0. 5. 5. 2. 2. 2.]\n [ 4. 0. 0. 0. 5. 2. 2. 10. 0. 0. 5. 5.]\n [ 1. 2. 0. 5. 0. 10. 0. 0. 0. 5. 1. 1.]\n [ 0. 2. 0. 2. 10. 0. 5. 1. 1. 5. 4. 0.]\n [ 0. 2. 0. 2. 0. 5. 0. 10. 5. 2. 3. 3.]\n [ 6. 0. 5. 10. 0. 1. 10. 0. 0. 0. 5. 0.]\n [ 2. 4. 5. 0. 0. 1. 5. 0. 0. 0. 10. 10.]\n [ 1. 5. 2. 0. 5. 5. 2. 0. 0. 0. 5. 0.]\n [ 1. 0. 2. 5. 1. 4. 3. 5. 10. 5. 0. 2.]\n [ 1. 0. 2. 5. 1. 0. 3. 0. 10. 0. 2. 0.]]\n"
]
],
[
[
"## 2. Objective function",
"_____no_output_____"
]
],
[
[
"def qap_objective_function(p):\n s = 0.0\n for i in range(n):\n s += (A[i, :] * B[p, p[i]]).sum()\n return s",
"_____no_output_____"
],
[
"p = [11, 6, 8, 2, 3, 7, 10, 0, 4, 5, 9, 1]\nprint(qap_objective_function(p), p)",
"578.0 [11, 6, 8, 2, 3, 7, 10, 0, 4, 5, 9, 1]\n"
]
],
[
[
"# 3. Random Sampling",
"_____no_output_____"
]
],
[
[
"%%time\n\nT = 1000000\n\npermutations = np.empty((T, n), dtype=np.int64)\ncosts = np.zeros(T)\nfor i in tqdm(range(T)):\n permutations[i, :] = np.random.permutation(n)\n costs[i] = qap_objective_function(permutations[i, :])\n\np = permutations[costs.argmin(), :]\nprint(qap_objective_function(p), p)",
"100%|█████████████████████████████████████████████████████████████████████| 1000000/1000000 [01:30<00:00, 11082.18it/s]\n"
],
[
"plt.figure()\nplt.hist(costs, bins=100, edgecolor='black')\nplt.show()\n\nprint(costs.mean(), costs.std())",
"_____no_output_____"
]
],
[
[
"## 4. Simulated Annealing",
"_____no_output_____"
]
],
[
[
"def qap_objective_function(p):\n s = 0.0\n for i in range(n):\n s += (A[i, :] * B[p, p[i]]).sum()\n return s",
"_____no_output_____"
],
[
"q = p.copy()\n\ni, j = 4, 8\nq[i], q[j] = q[j], q[i]",
"_____no_output_____"
],
[
"qap_objective_function(q)",
"_____no_output_____"
],
[
"qap_objective_function(p)",
"_____no_output_____"
],
[
"z = q.copy()\ni, j = 2, 3\nz[i], z[j] = z[j], z[i]\nqap_objective_function(z)",
"_____no_output_____"
],
[
"def delta(p, r, s, A, B):\n n = len(p)\n ans = 0\n q = p.copy()\n q[r], q[s] = q[s], q[r]\n for i in range(n):\n for j in range(n):\n ans += A[i][j] * B[p[i], p[j]] - A[i][j] * B[q[i], q[j]]\n return -ans",
"_____no_output_____"
],
[
"def delta2(p, r, s, A, B):\n n = len(p)\n ans = 0\n for k in range(n):\n if k == r or k == s:\n continue\n ans += (A[s, k] - A[r, k]) * (B[p[s], p[k]] - B[p[r], p[k]])\n \n return -2 * ans",
"_____no_output_____"
],
[
"r, s = 4, 8\nu, v = 2, 6\n\nprint(delta2(p, r, s, A, B))\nprint(delta2(q, u, v, A, B))",
"76.0\n54.0\n"
],
[
"for u in range(n):\n for v in range(n):\n a = delta2(p, r, s, A, B)\n b = delta2(q, u, v, A, B)\n c = delta2(p, u, v, A, B) + 2 * (A[r, u] - A[r, v] + A[s, v] - A[s, u]) * (B[q[s], q[u]] - B[q[s], q[v]] + B[q[r], q[v]] - B[q[r], q[u]])\n if u in [r, s] or v in [r, s]:\n 2 + 2\n else:\n if b != c:\n print(u, v)\n print(a)\n print(b)\n print(c)\n print()",
"_____no_output_____"
],
[
"class SA:\n def __init__(self, flow_m, dist_m, QAP_name='', T=10000, radius=1, alpha=1.0, cost_dist=False):\n self.T = T\n self.radius = radius\n self.alpha = alpha\n \n n1, n2 = flow_m.shape\n n3, n4 = dist_m.shape\n assert(n1 == n2 == n3 == n4)\n self.FLOW = flow_m\n self.DIST = dist_m\n self.n = n1\n \n self.last_perm = np.random.permutation(n1)\n self.act_perm = self.last_perm.copy()\n self.r, self.s = 0, 1\n self.act_perm[0], self.act_perm[1] = self.act_perm[1], self.act_perm[0]\n \n self.p_cost = self.qap_objective_function(self.act_perm)\n self.costs = np.zeros(T)\n self.all_perms = [self.act_perm]\n self.all_jumps = []\n self.QAP_name = QAP_name\n self.cost_and_perms_dist = []\n self.cost_dist = cost_dist\n \n self.last_swap_costs = {}\n self.act_swap_costs = {}\n for u in range(self.n):\n for v in range(u, self.n):\n self.last_swap_costs[(u, v)] = self.delta(self.last_perm, u, v)\n self.act_swap_costs[(u, v)] = self.delta(self.act_perm, u, v)\n \n \n def delta(self, p, r, s):\n '''\n Change of Cost after swap(r, s)\n '''\n n = self.n\n A, B = self.FLOW, self.DIST\n ans = 0\n for k in range(n):\n if k == r or k == s:\n continue\n ans += (A[s, k] - A[r, k]) * (B[p[s], p[k]] - B[p[r], p[k]])\n\n return -2 * ans\n\n\n def qap_objective_function(self, p):\n s = 0.0\n for i in range(self.n):\n s += (self.FLOW[i, :] * self.DIST[p, p[i]]).sum()\n return s\n\n\n def random_neighbor(self):\n q = self.act_perm.copy()\n for r in range(self.radius):\n i, j = np.random.choice(self.n, 2, replace=False)\n q[i], q[j] = q[j], q[i]\n return q\n \n \n def run(self):\n for t in tqdm(range(self.T), desc='Simulated Annealing', position=0):\n good_jump, random_jump = 0, 0\n q = self.random_neighbor()\n q_cost = self.qap_objective_function(q)\n \n \n if(q_cost < self.p_cost):\n if self.cost_dist:\n self.cost_and_perms_dist.append((self.p_cost - q_cost))\n \n self.act_perm, self.p_cost = q, q_cost\n self.all_perms.append(self.act_perm)\n good_jump = 1\n \n elif(np.random.rand() < np.exp(-self.alpha * (q_cost - self.p_cost) * t / self.T)):\n self.act_perm, self.p_cost = q, q_cost\n self.all_perms.append(self.act_perm)\n random_jump = 1\n \n self.costs[t] = self.p_cost\n self.all_jumps.append((good_jump, random_jump))\n \n \n def update_swap_costs(self):\n cost = 0\n r, s = self.r, self.s\n A, B = self.FLOW, self.DIST\n q = self.act_perm\n \n self.last_swap_costs = self.act_swap_costs\n self.act_swap_costs = {}\n \n for u in range(self.n):\n for v in range(u, self.n):\n if u in [r, s] or v in [r, s]:\n # calculate (u, v) swap cost in O(N)\n self.act_swap_costs[(u, v)] = self.delta(self.act_perm, u, v)\n else:\n # swap in O(1)\n self.act_swap_costs[(u, v)] = self.last_swap_costs[u, v] \\\n + 2 * (A[r, u] - A[r, v] + A[s, v] - A[s, u]) \\\n * (B[q[s], q[u]] - B[q[s], q[v]] + B[q[r], q[v]] - B[q[r], q[u]])\n \n \n def run_faster(self):\n '''\n We make only single swaps per iter.\n Insted of calculating qap_objective_function in every iteration,\n we calculate cost of swapping every pair every time\n we change a state.\n \n qap_objective_function -> O(n^2) every itereation\n new approach -> O(n^2) only when we change a state\n '''\n for t in tqdm(range(self.T), desc='Simulated Annealing', position=0):\n good_jump, random_jump = 0, 0\n u, v = np.random.choice(self.n, 2, replace=False)\n q = self.act_perm.copy()\n q[u], q[v] = q[v], q[u]\n q_cost = self.act_swap_costs[(min(u, v), max(u, v))] + self.p_cost\n \n# if q_cost != self.qap_objective_function(q):\n# print('IMPLEMENTATION DOES NOT WORK!!!')\n \n if(q_cost < self.p_cost):\n self.act_perm, self.p_cost = q, q_cost\n self.all_perms.append(self.act_perm)\n self.r, self.s = u, v\n self.last_perm = self.act_perm\n self.update_swap_costs()\n good_jump = 1\n \n elif(np.random.rand() < np.exp(-self.alpha * (q_cost - self.p_cost) * t / self.T)):\n self.act_perm, self.p_cost = q, q_cost\n self.all_perms.append(self.act_perm)\n self.r, self.s = u, v\n self.last_perm = self.act_perm\n self.update_swap_costs()\n random_jump = 1\n \n self.costs[t] = self.p_cost\n self.all_jumps.append((good_jump, random_jump))\n \n \n def plot_cost(self):\n plt.figure(figsize=(15, 5))\n plt.plot(self.costs)\n plt.title('Cost function ' + self.QAP_name)\n plt.show()\n \n \n def plot_hist(self, bins):\n plt.figure(figsize=(15, 5))\n plt.hist(self.costs, bins=bins, edgecolor='black')\n plt.title('Cost function histogram ' + self.QAP_name)\n plt.show()\n \n \n def plot_jumps(self):\n x = np.array(self.all_jumps).reshape(-1, 50)\n x = x.sum(axis=1)\n\n f, ax = plt.subplots(2,1, figsize=(15,10))\n ax[0].bar(range(x.shape[0]), x, color='green')\n ax[1].bar(range(x.shape[0]), x, color='red')\n ax[0].set_title('Successes')\n ax[1].set_title('Accepted failures')\n plt.show()\n \n \n def plot_all(self, bins):\n self.plot_cost()\n self.plot_hist(bins=bins)\n self.plot_jumps()",
"_____no_output_____"
]
],
[
[
"### 4.1 Basic implementation",
"_____no_output_____"
]
],
[
[
"%%time\n\nn, A, B = get_nug(12)\nsimulation = SA(flow_m=A, dist_m=B, QAP_name='nug12', T=500000, cost_dist=True)\nsimulation.run()",
"Simulated Annealing: 100%|███████████████████████████████████████████████████| 500000/500000 [01:01<00:00, 8138.42it/s]\n"
],
[
"simulation.costs.min()",
"_____no_output_____"
],
[
"simulation.plot_all(bins=50)",
"_____no_output_____"
]
],
[
[
"### 4.2 Improved implementation based on\nhttps://arxiv.org/pdf/1111.1353.pdf",
"_____no_output_____"
]
],
[
[
"%%time\n\nn, A, B = get_nug(12)\nsimulation = SA(flow_m=A, dist_m=B, QAP_name='nug12', T=500000)\nsimulation.run_faster()",
"Simulated Annealing: 100%|██████████████████████████████████████████████████| 500000/500000 [00:44<00:00, 11117.41it/s]\n"
],
[
"simulation.costs.min()",
"_____no_output_____"
]
],
[
[
"## 5. Parameters tunning",
"_____no_output_____"
],
[
"### 5.1 Radius",
"_____no_output_____"
]
],
[
[
"%%time\nscores_r = []\n\nfor r in range(1, 20):\n print(f'r: {r}')\n n, A, B = get_nug(12)\n simulation = SA(flow_m=A, dist_m=B, QAP_name='nug12', T=200000, radius=r, cost_dist=True)\n simulation.run()\n scores_r.append((r, simulation.costs, simulation.cost_and_perms_dist))",
"r: 1\n"
],
[
"scores_r = np.array(scores_r)\nplt.plot(scores_r[:, 0], list(map(lambda x: min(x), scores_r[:, 1])))\nplt.xlabel('Distance between permutations')\nplt.ylabel('Min cost')\n\n# save scores\n# pd.DataFrame(scores_r).to_csv('scores_for_different_radius.csv')",
"_____no_output_____"
],
[
"# scores_r = pd.read_csv('scores_for_different_radius.csv', names=['r', 'cost'],\n# header=None).reset_index(drop=True).iloc[1:]",
"_____no_output_____"
],
[
"# distances between cost_f(perms) in successes\n\nf, ax = plt.subplots(4, 5, figsize=(17,15))\nfor r in range(19):\n ax[r // 5, r % 5].plot(scores_r[:, 2][r])\n ax[r // 5, r % 5].set_title(f'r: {r}')",
"_____no_output_____"
]
],
[
[
"### 5.2 alpha",
"_____no_output_____"
]
],
[
[
"%%time\n\nscores_a = []\nalphas = [0.01, 0.02, 0.03, 0.05, 0.1, 0.2, 0.3, 0.5, 0.7, 1.01, 1.03, 1.05, 1.1, 1.2, 1.4, 1.5, 1.7, 2, 4, 5, 10]\nfor a in alphas:\n print(f'alpha: {a}')\n n, A, B = get_nug(12)\n simulation = SA(flow_m=A, dist_m=B, QAP_name='nug12', T=200000, alpha=a)\n simulation.run_faster()\n scores_a.append(simulation.costs)",
"alpha: 0.01\n"
],
[
"# save scores\n\n# pd.DataFrame(scores_a).to_csv('scores_for_different_alpha.csv')",
"_____no_output_____"
],
[
"scores_a = np.array(scores_a)\nplt.figure(figsize=(15, 5))\nplt.xticks(np.linspace(0, 10, 20))\nplt.plot(alphas, scores_a.min(axis=1))",
"_____no_output_____"
],
[
"for i, a in enumerate(alphas):\n print(a, ' -> ', scores_a.min(axis=1)[i])",
"0.01 -> 610.0\n0.02 -> 606.0\n0.03 -> 600.0\n0.05 -> 582.0\n0.1 -> 578.0\n0.2 -> 578.0\n0.3 -> 578.0\n0.5 -> 578.0\n0.7 -> 578.0\n1.01 -> 578.0\n1.03 -> 578.0\n1.05 -> 582.0\n1.1 -> 578.0\n1.2 -> 578.0\n1.4 -> 578.0\n1.5 -> 578.0\n1.7 -> 578.0\n2 -> 586.0\n4 -> 578.0\n5 -> 586.0\n10 -> 586.0\n"
]
],
[
[
"## 6. SA on different QAP",
"_____no_output_____"
]
],
[
[
"from urllib.error import HTTPError\n\nnugs = []\nfor i in range(12, 30):\n try:\n x = get_nug(i)\n nugs.append(x)\n print(f'Nug {i} appended!')\n except HTTPError:\n print(f'Nug {i} wasn t found :c')",
"Nug 12 appended!\nNug 13 wasn t found :c\nNug 14 appended!\nNug 15 appended!\nNug 16 wasn t found :c\nNug 17 appended!\nNug 18 appended!\nNug 19 wasn t found :c\nNug 20 appended!\nNug 21 appended!\nNug 22 appended!\nNug 23 wasn t found :c\nNug 24 appended!\nNug 25 appended!\nNug 26 wasn t found :c\nNug 27 appended!\nNug 28 appended!\nNug 29 wasn t found :c\n"
],
[
"%%time\n\nSCORES = []\nfor n, A, B in nugs:\n print(f'Running nug: {n}')\n simulation = SA(flow_m=A, dist_m=B, QAP_name=f'Nug {n}', T=1000000)\n simulation.run_faster()\n \n SCORES.append(simulation.costs)\n save_costs = pd.DataFrame(simulation.costs)\n save_costs.to_csv(f'scores_for_nug{n}.csv')",
"Running nug: 12\n"
],
[
"SCORES = np.array(SCORES)\n\nn, m = SCORES.shape\nn, m",
"_____no_output_____"
],
[
"f, ax = plt.subplots(4, 3, figsize=(17,15))\nfor i in range(n):\n ax[i // 3, i % 3].plot(SCORES[i])\n ax[i // 3, i % 3].set_title(f'nug{nugs[i][0]}')",
"_____no_output_____"
],
[
"print('Min cost:')\nfor i, s in enumerate(SCORES.min(axis=1)):\n print(f'nug{nugs[i][0]}: {s}')",
"Min cost:\nnug12: 578.0\nnug14: 1016.0\nnug15: 1150.0\nnug17: 1736.0\nnug18: 1938.0\nnug20: 2570.0\nnug21: 2438.0\nnug22: 3596.0\nnug24: 3488.0\nnug25: 3758.0\nnug27: 5234.0\nnug28: 5186.0\n"
]
]
] |
[
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"code"
],
[
"markdown",
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code"
]
] |
cbd9b7e23a36f5e35313bf0f9cea306bed8d1773
| 10,404 |
ipynb
|
Jupyter Notebook
|
ipynb/TM_2/4_BB/2_BL/2.4.2.K-FEM.ipynb
|
kassbohm/tm-snippets
|
5e0621ba2470116e54643b740d1b68b9f28bff12
|
[
"MIT"
] | null | null | null |
ipynb/TM_2/4_BB/2_BL/2.4.2.K-FEM.ipynb
|
kassbohm/tm-snippets
|
5e0621ba2470116e54643b740d1b68b9f28bff12
|
[
"MIT"
] | null | null | null |
ipynb/TM_2/4_BB/2_BL/2.4.2.K-FEM.ipynb
|
kassbohm/tm-snippets
|
5e0621ba2470116e54643b740d1b68b9f28bff12
|
[
"MIT"
] | null | null | null | 37.559567 | 335 | 0.357939 |
[
[
[
"from sympy import *\nEI, a, q = var(\"EI, a, q\")\npprint(\"\\nFEM-Solution:\")\n\n# 1: Stiffness Matrices:\n# Element 1\nl = 2*a\nl2 = l*l\nl3 = l*l*l\nK = EI/l3 * Matrix(\n [\n [ 4*l2 , -6*l , 2*l2 , 6*l , 0 , 0 ],\n [ -6*l , 12 , -6*l , -12 , 0 , 0 ],\n [ 2*l2 , -6*l , 4*l2 , 6*l , 0 , 0 ],\n [ 6*l , -12 , 6*l , 12 , 0 , 0 ],\n [ 0 , 0 , 0 , 0 , 0 , 0 ],\n [ 0 , 0 , 0 , 0 , 0 , 0 ],\n ]\n )\n# Element 2\nl = a\nl2 = l*l\nl3 = l*l*l\nK += EI/l3 * Matrix(\n [\n [ 0 , 0 , 0 , 0 , 0 , 0 ],\n [ 0 , 0 , 0 , 0 , 0 , 0 ],\n [ 0 , 0 , 4*l2 , -6*l , 2*l2 , 6*l ],\n [ 0 , 0 , -6*l , 12 , -6*l , -12 ],\n [ 0 , 0 , 2*l2 , -6*l , 4*l2 , 6*l ],\n [ 0 , 0 , 6*l , -12 , 6*l , 12 ],\n ]\n )\n\n# 2: BCs:\np0,w0,p1,w1,p2,w2 = var(\"ψ₀,w₀,ψ₁,w₁,ψ₂,w₂\")\nM0,F0,M1,F1,M2,F2 = var(\"M₀,F₀,M₁,F₁,M₂,F₂\")\n\nMq1, Fq1 = -q/12*a*a, q/2*a\nMq2, Fq2 = -Mq1, Fq1\n\n# 0 1 2\n# qqqqqqqqqqqqqqqq\n# |-------------------------A---------------\nu = Matrix([ 0,0, p1,0, p2,w2 ] )\nf = Matrix([ M0,F0, Mq1,Fq1+F1, Mq2,Fq2 ] )\nunks = [ M0,F0, p1,F1, p2,w2 ]\n\n# 3: Solution:\neq = Eq(K*u, f)\nsol = solve(eq, unks)\npprint(sol)\n\npprint(\"\\nMinimum-Total-Potential-Energy-Principle-Solution:\")\n\n# 1: Ansatz:\na0, a1, a2, a3 = var(\"a0, a1, a2, a3\")\nb0, b1, b2 = var(\"b0, b1, b2\")\n\norder = 2\nif (order == 2):\n # w2 has order 2\n b3, b4 = 0, 0\nelse:\n # w2 has order 4\n b3, b4 = var(\"b3, b4\")\n\nx1, x2 = var(\"x1, x2\")\nw1 = a0 + a1*x1 + a2*x1**2 + a3*x1**3\nw1p = diff(w1, x1)\nw1pp = diff(w1p, x1)\n\nw2 = b0 + b1*x2 + b2*x2**2 + b3*x2**3 + b4*x2**4\nw2p = diff(w2, x2)\nw2pp = diff(w2p, x2)\n\npprint(\"\\nw1 and w2:\")\npprint(w1)\npprint(w2)\n\n\n# 2: Using BCs:\npprint(\"\\nElimination of a0, a1, a2, a3, b0 using BCs:\")\n# w1(0)=0\ne1 = Eq(w1.subs(x1, 0))\n# w1'(0)=0\ne2 = Eq(w1p.subs(x1, 0))\n# w1(2l)=0\ne3 = Eq(w1.subs(x1, 2*a))\n# w2(0)=0\ne4 = Eq(w2.subs(x2, 0))\n# w1p(2a)=w2p(0)\ne5 = Eq(w1p.subs(x1, 2*a), w2p.subs(x2, 0))\n\neqns, unks = [e1, e2, e3, e4, e5], [a0, a1, a2, a3, b0]\nsol = solve(eqns, unks)\npprint(sol)\n\nsub_list=[\n (a0, sol[a0]),\n (a1, sol[a1]),\n (a2, sol[a2]),\n (a3, sol[a3]),\n (b0, sol[b0]),\n ]\n\npprint(\"\\nw1 and w2:\")\nw1 = w1.subs(sub_list)\nw2 = w2.subs(sub_list)\npprint(w1)\npprint(w2)\n\npprint(\"\\nw1'' and w2'':\")\nw1pp = w1pp.subs(sub_list)\nw2pp = w2pp.subs(sub_list)\npprint(w1pp)\npprint(w2pp)\n\n# 3: Using Principle:\npprint(\"\\nU1, U2, Uq:\")\ni1 = w1pp*w1pp\nI1 = integrate(i1, x1)\nI1 = I1.subs(x1,2*a) - I1.subs(x1,0)\nU1 = EI*I1/2\npprint(U1)\n\ni2 = w2pp*w2pp\nI2 = integrate(i2, x2)\nI2 = I2.subs(x2,a) - I2.subs(x2,0)\nU2 = EI*I2/2\npprint(U2)\n\ni2 = q*w2\nI2 = integrate(i2, x2)\nI2 = I2.subs(x2,a) - I2.subs(x2,0)\nUq = I2\npprint(Uq)\n\npprint(\"\\nParameters for U1 + U2 - Uq = Min:\")\nU = U1 + U2 - Uq\n\ne1 = Eq(diff(U, b1))\ne2 = Eq(diff(U, b2))\nif (order == 2):\n eqns = [e1, e2]\n unks = [b1, b2]\n sol = solve(eqns, unks)\n sub_list=[\n (b1, sol[b1]),\n (b2, sol[b2]),\n ]\n w2 = w2.subs(sub_list)\nelse:\n e3 = Eq(diff(U, b3))\n e4 = Eq(diff(U, b4))\n eqns = [e1, e2, e3, e4]\n unks = [b1, b2, b3, b4]\n sol = solve(eqns, unks)\n sub_list=[\n (b1, sol[b1]),\n (b2, sol[b2]),\n (b3, sol[b3]),\n (b4, sol[b4]),\n ]\n w2 = w2.subs(sub_list)\n\npprint(sol)\n\npprint(\"\\nw2:\")\npprint(w2)\n\npprint(\"\\nw2(a):\")\nw2 = w2.subs(x2, a)\npprint(w2)\n\n# FEM-Solution:\n# ⎧ 2 4 3 3 ⎫\n# ⎪ 3⋅a⋅q -11⋅a⋅q -a ⋅q 3⋅a ⋅q -a ⋅q -5⋅a ⋅q ⎪\n# ⎨F₀: ─────, F₁: ────────, M₀: ──────, w₂: ──────, ψ₁: ──────, ψ₂: ────────⎬\n# ⎪ 8 8 4 8⋅EI 4⋅EI 12⋅EI ⎪\n# ⎩ ⎭\n#\n# 2. Minimum-Total-Potential-Energy-Principle-Solution:\n#\n# w1 and w2:\n# 2 3\n# a₀ + a₁⋅x₁ + a₂⋅x₁ + a₃⋅x₁\n# 2\n# b₀ + b₁⋅x₂ + b₂⋅x₂\n#\n# Elimination of a0, a1, a2, a3, b0 using BCs:\n# ⎧ -b₁ b₁ ⎫\n# ⎪a₀: 0, a₁: 0, a₂: ────, a₃: ────, b₀: 0⎪\n# ⎨ 2⋅a 2 ⎬\n# ⎪ 4⋅a ⎪\n# ⎩ ⎭\n#\n# w1 and w2:\n# 2 3\n# b₁⋅x₁ b₁⋅x₁\n# - ────── + ──────\n# 2⋅a 2\n# 4⋅a\n# 2\n# b₁⋅x₂ + b₂⋅x₂\n#\n# w1'' and w2'':\n# b₁ 3⋅b₁⋅x₁\n# - ── + ───────\n# a 2\n# 2⋅a\n# 2⋅b₂\n#\n# U1, U2, Uq:\n# 2\n# EI⋅b₁\n# ──────\n# a\n# 2\n# 2⋅EI⋅a⋅b₂\n# 3 2\n# a ⋅b₂⋅q a ⋅b₁⋅q\n# ─────── + ───────\n# 3 2\n#\n# Parameters for U1 + U2 - Uq = Min:\n# ⎧ 3 2 ⎫\n# ⎪ a ⋅q a ⋅q⎪\n# ⎨b₁: ────, b₂: ─────⎬\n# ⎪ 4⋅EI 12⋅EI⎪\n# ⎩ ⎭\n#\n# w2:\n# 3 2 2\n# a ⋅q⋅x₂ a ⋅q⋅x₂\n# ─────── + ────────\n# 4⋅EI 12⋅EI\n#\n# w2(a):\n# 4\n# a ⋅q\n# ────\n# 3⋅EI\n",
"_____no_output_____"
]
]
] |
[
"code"
] |
[
[
"code"
]
] |
cbd9c04dfec491439daca0abfc1390b6b4a1e418
| 586,902 |
ipynb
|
Jupyter Notebook
|
CV/bandits/bandits.ipynb
|
baumanproject/CV_NLP_WORKS
|
bf7b775ec1034209f4d26636d6a418738399d87a
|
[
"MIT"
] | null | null | null |
CV/bandits/bandits.ipynb
|
baumanproject/CV_NLP_WORKS
|
bf7b775ec1034209f4d26636d6a418738399d87a
|
[
"MIT"
] | null | null | null |
CV/bandits/bandits.ipynb
|
baumanproject/CV_NLP_WORKS
|
bf7b775ec1034209f4d26636d6a418738399d87a
|
[
"MIT"
] | null | null | null | 597.659878 | 245,980 | 0.94356 |
[
[
[
"import enum\nfrom abc import ABCMeta, abstractmethod, abstractproperty\n\nimport numpy as np\nimport pandas as pd\n\nfrom matplotlib import pyplot as plt\n%matplotlib inline",
"_____no_output_____"
]
],
[
[
"# Bayesian bandits\n\nLet us explore a simple task of multi-armed bandits with Bernoulli distributions and several strategies for it.\n\nBandits have $K$ actions. An action leads to a reward $r=1$ with a probability $0 \\le \\theta_k \\le 1$ unknown to an agent but fixed over time. The agent goal is to minimize total suboptimality over $T$ actions:\n\n$$\\rho = T\\theta^* - \\sum_{t=1}^T r_t,$$\n\nand $\\theta^* = \\max_k\\{\\theta_k\\}$.\n\n**Real-life example** - clinical trials: we have $k$ drug types and $T$ patients. After some drug is taken a patient is cured with a probability $\\theta_k$. The goal is to find an optimal drug. The survey on clinical trials - https://arxiv.org/pdf/1507.08025.pdf.\n\n### Hometask\n\nThe task is complementary for the reinforcement learning task, the maximum is 10 points for both.\n\nImplement all the unimplemented agents:\n\n1. [1 балл] $\\varepsilon$-greedy\n2. [1 балл] UCB\n3. [1 балл] Thompson sampling\n4. [2 балла] Custom strategy\n5. [2 балла] $\\varepsilon$-greedy для riverflow\n6. [3 балла] PSRL agent для riverflow",
"_____no_output_____"
]
],
[
[
"class BernoulliBandit:\n def __init__(self, n_actions=5):\n self._probs = np.random.random(n_actions)\n\n @property\n def action_count(self):\n return len(self._probs)\n\n def pull(self, action):\n if np.random.random() > self._probs[action]:\n return 0.0\n return 1.0\n\n def optimal_reward(self):\n \"\"\" Used for regret calculation\n \"\"\"\n return np.max(self._probs)\n\n def step(self):\n \"\"\" Used in nonstationary version\n \"\"\"\n pass\n\n def reset(self):\n \"\"\" Used in nonstationary version\n \"\"\"",
"_____no_output_____"
],
[
"class AbstractAgent(metaclass=ABCMeta): \n def init_actions(self, n_actions):\n self._successes = np.zeros(n_actions)\n self._failures = np.zeros(n_actions)\n self._total_pulls = 0\n\n @abstractmethod\n def get_action(self):\n \"\"\"\n Get current best action\n :rtype: int\n \"\"\"\n pass\n\n def update(self, action, reward):\n \"\"\"\n Observe reward from action and update agent's internal parameters\n :type action: int\n :type reward: int\n \"\"\"\n self._total_pulls += 1\n if reward == 1:\n self._successes[action] += 1\n else:\n self._failures[action] += 1\n\n @property\n def name(self):\n return self.__class__.__name__\n\n\nclass RandomAgent(AbstractAgent): \n def get_action(self):\n return np.random.randint(0, len(self._successes))",
"_____no_output_____"
]
],
[
[
"### Epsilon-greedy agent\n\n> **for** $t = 1,2,...$ **do**\n\n>> **for** $k = 1,...,K$ **do**\n\n>>> $\\hat\\theta_k \\leftarrow \\alpha_k / (\\alpha_k + \\beta_k)$\n\n>> **end for** \n\n>> $x_t \\leftarrow argmax_{k}\\hat\\theta$ with probability $1 - \\varepsilon$ or random action with probability $\\varepsilon$\n\n>> Apply $x_t$ and observe $r_t$\n\n>> $(\\alpha_{x_t}, \\beta_{x_t}) \\leftarrow (\\alpha_{x_t}, \\beta_{x_t}) + (r_t, 1-r_t)$\n\n> **end for**\n\nImplement the algorithm described:",
"_____no_output_____"
]
],
[
[
"class EpsilonGreedyAgent(AbstractAgent):\n def __init__(self, epsilon = 0.01):\n self._epsilon = epsilon\n \n def init_actions(self, n_actions):\n self._a = np.ones(n_actions) * 1e-5\n self._b = np.ones(n_actions) * 1e-5\n self._n_actions = n_actions\n\n def get_action(self):\n if np.random.random() < self._epsilon:\n return np.random.randint(self._n_actions)\n thetas = self._a / (self._a + self._b) \n return np.argmax(thetas)\n\n def update(self, action, reward):\n self._a[action] += reward\n self._b[action] += (1 - reward) \n \n @property\n def name(self):\n return self.__class__.__name__ + \"(epsilon={})\".format(self._epsilon) ",
"_____no_output_____"
]
],
[
[
"### UCB agent\nEpsilon-greedy strategy has no preferences on the random choice stage. Probably it is better to choose the actions we are not sure enough in, but having a potential to become optimal in the future. We could create the measure for both optimality and uncertainty in the same time.\n\nOne of possible sollutions for that is UCB1 algorithm:\n\n> **for** $t = 1,2,...$ **do**\n>> **for** $k = 1,...,K$ **do**\n>>> $w_k \\leftarrow \\alpha_k / (\\alpha_k + \\beta_k) + \\sqrt{2log\\ t \\ / \\ (\\alpha_k + \\beta_k)}$\n\n>> **end for** \n\n>> $x_t \\leftarrow argmax_{k}w$\n\n>> Apply $x_t$ and observe $r_t$\n\n>> $(\\alpha_{x_t}, \\beta_{x_t}) \\leftarrow (\\alpha_{x_t}, \\beta_{x_t}) + (r_t, 1-r_t)$\n\n> **end for**\n\nOther solutions and optimality analysis - https://homes.di.unimi.it/~cesabian/Pubblicazioni/ml-02.pdf.",
"_____no_output_____"
]
],
[
[
"class UCBAgent(AbstractAgent):\n def __init__(self):\n pass\n \n def init_actions(self, n_actions):\n self._a = np.ones(n_actions) * 1e-5\n self._b = np.ones(n_actions) * 1e-5\n self._t = 1\n\n def get_action(self):\n doubleus = self._a / (self._a + self._b) + np.sqrt(2 * np.log(self._t) / (self._a + self._b))\n return np.argmax(doubleus)\n\n def update(self, action, reward):\n self._a[action] += reward\n self._b[action] += (1 - reward)\n self._t += 1",
"_____no_output_____"
]
],
[
[
"### Thompson sampling\n\nUCB1 algorithm does not consider the rewards distribution. If it is known, we can improve it with Thompson sampling.\n\nWe believe that $\\theta_k$ are independent and indentically distributed. As a prior distribution for them we will use beta-distribution with parameters $\\alpha=(\\alpha_1, \\dots, \\alpha_k)$ and $\\beta=(\\beta_1, \\dots, \\beta_k)$. Thus for each parameter $\\theta_k$ an apriori probability function looks like\n$$\np(\\theta_k) =\n\\frac{\\Gamma(\\alpha_k + \\beta_k)}{\\Gamma(\\alpha_k) + \\Gamma(\\beta_k)}\n\\theta_k^{\\alpha_k - 1}(1 - \\theta_k)^{\\beta_k - 1}\n$$\nAfter receiving new evidence the distribution is updated following the Bayes rule.\n\nBeta distribution is comfortable to work with because of conjugation - actions aposterior distribution will be Beta too so we can easily update them:\n\n> **for** $t = 1,2,...$ **do**\n>> **for** $k = 1,...,K$ **do**\n>>> Sample $\\hat\\theta_k \\sim beta(\\alpha_k, \\beta_k)$\n\n>> **end for** \n\n>> $x_t \\leftarrow argmax_{k}\\hat\\theta$\n\n>> Apply $x_t$ and observe $r_t$\n\n>> $(\\alpha_{x_t}, \\beta_{x_t}) \\leftarrow (\\alpha_{x_t}, \\beta_{x_t}) + (r_t, 1-r_t)$\n\n> **end for**\n \nHometask: https://web.stanford.edu/~bvr/pubs/TS_Tutorial.pdf.",
"_____no_output_____"
]
],
[
[
"class ThompsonSamplingAgent(AbstractAgent):\n def __init__(self):\n pass\n \n def init_actions(self, n_actions):\n self._a = np.ones(n_actions) * 1e-5\n self._b = np.ones(n_actions) * 1e-5\n\n def get_action(self):\n thetas = np.random.beta(self._a, self._b)\n return np.argmax(thetas)\n\n def update(self, action, reward):\n self._a[action] += reward\n self._b[action] += (1 - reward)",
"_____no_output_____"
],
[
"def plot_regret(env, agents, n_steps=5000, n_trials=50):\n scores = {\n agent.name : [0.0 for step in range(n_steps)] for agent in agents\n }\n\n for trial in range(n_trials):\n env.reset()\n \n for a in agents:\n a.init_actions(env.action_count)\n\n for i in range(n_steps):\n optimal_reward = env.optimal_reward()\n \n for agent in agents:\n action = agent.get_action()\n reward = env.pull(action)\n agent.update(action, reward)\n scores[agent.name][i] += optimal_reward - reward\n \n env.step() # change bandit's state if it is unstationary\n\n plt.figure(figsize=(17, 8))\n for agent in agents:\n plt.plot(np.cumsum(scores[agent.name]) / n_trials)\n\n plt.legend([agent.name for agent in agents])\n \n plt.ylabel(\"regret\")\n plt.xlabel(\"steps\")\n \n plt.show()",
"_____no_output_____"
],
[
"# Uncomment agents\nagents = [\n EpsilonGreedyAgent(),\n UCBAgent(),\n ThompsonSamplingAgent()\n]\n\nplot_regret(BernoulliBandit(), agents, n_steps=10000, n_trials=10)",
"_____no_output_____"
]
],
[
[
"# Non-stationary bandits\n\nBut what if the probabilities change over time? For example",
"_____no_output_____"
]
],
[
[
"class DriftingBandit(BernoulliBandit):\n def __init__(self, n_actions=5, gamma=0.01):\n \"\"\"\n Idea from https://github.com/iosband/ts_tutorial\n \"\"\"\n super().__init__(n_actions)\n \n self._gamma = gamma\n \n self._successes = None\n self._failures = None\n self._steps = 0\n \n self.reset()\n \n def reset(self):\n self._successes = np.zeros(self.action_count) + 1.0\n self._failures = np.zeros(self.action_count) + 1.0\n self._steps = 0\n \n def step(self):\n action = np.random.randint(self.action_count)\n reward = self.pull(action)\n self._step(action, reward)\n \n def _step(self, action, reward): \n self._successes = self._successes * (1 - self._gamma) + self._gamma\n self._failures = self._failures * (1 - self._gamma) + self._gamma\n self._steps += 1\n\n self._successes[action] += reward\n self._failures[action] += 1.0 - reward\n\n self._probs = np.random.beta(self._successes, self._failures)",
"_____no_output_____"
]
],
[
[
"Letus look at the reward probabilities changing over time:",
"_____no_output_____"
]
],
[
[
"drifting_env = DriftingBandit(n_actions=5)\n\ndrifting_probs = []\nfor i in range(20000):\n drifting_env.step()\n drifting_probs.append(drifting_env._probs)\n\nplt.figure(figsize=(14, 8))\nplt.plot(pd.DataFrame(drifting_probs).rolling(window=20).mean())\n\nplt.xlabel(\"steps\")\nplt.ylabel(\"Success probability\")\nplt.title(\"Reward probabilities over time\")\nplt.legend([\"Action {}\".format(i) for i in range(drifting_env.action_count)])\nplt.show()",
"_____no_output_____"
]
],
[
[
"**Task** - create an agent that will work better then any of the stationary algorithms above.",
"_____no_output_____"
]
],
[
[
"class YourAgent(AbstractAgent):\n def __init__(self, coef=0.995):\n self._coef = coef\n \n def init_actions(self, n_actions):\n self._a = np.ones(n_actions) * 1e-5\n self._b = np.ones(n_actions) * 1e-5\n self._t = 1\n \n def get_action(self):\n doubleus = (self._a / (self._a + self._b) + np.sqrt(2 * np.log(self._t) / (self._a + self._b)))\n return np.argmax(doubleus)\n\n def update(self, action, reward):\n self._a[action] = self._coef * self._a[action] + reward\n self._b[action] = self._coef * self._b[action] + (1 - reward)\n self._t = self._coef * self._t + 1",
"_____no_output_____"
],
[
"drifting_agents = [\n ThompsonSamplingAgent(),\n EpsilonGreedyAgent(),\n UCBAgent(),\n YourAgent()\n]\n\nplot_regret(DriftingBandit(), drifting_agents, n_steps=20000, n_trials=10)",
"_____no_output_____"
]
],
[
[
"## Exploration в MPDP\n\nThe next task, \"river flow\", illustrates the importance of exploration with Markov decision-making processes example.",
"_____no_output_____"
],
[
"<img src=\"river_swim.png\">\n\nIllustration from https://arxiv.org/abs/1306.0940",
"_____no_output_____"
],
[
"Both rewards and transitions are unknown to the agent.\n\nThe optimal strategy is to go right against the flow, but the easiest one is to keep going left and get small rewards each time.",
"_____no_output_____"
]
],
[
[
"class RiverSwimEnv:\n LEFT_REWARD = 5.0 / 1000\n RIGHT_REWARD = 1.0\n \n def __init__(self, intermediate_states_count=4, max_steps=16):\n self._max_steps = max_steps\n self._current_state = None\n self._steps = None\n self._interm_states = intermediate_states_count\n self.reset()\n \n def reset(self):\n self._steps = 0\n self._current_state = 1\n return self._current_state, 0.0, False\n \n @property\n def n_actions(self):\n return 2\n \n @property\n def n_states(self):\n return 2 + self._interm_states\n \n def _get_transition_probs(self, action):\n if action == 0:\n if self._current_state == 0:\n return [0, 1.0, 0]\n else:\n return [1.0, 0, 0]\n \n elif action == 1:\n if self._current_state == 0:\n return [0, .4, .6]\n if self._current_state == self.n_states - 1:\n return [.4, .6, 0]\n else:\n return [.05, .6, .35]\n else:\n raise RuntumeError(\"Unknown action {}. Max action is {}\".format(action, self.n_actions))\n\n def step(self, action):\n \"\"\"\n :param action:\n :type action: int\n :return: observation, reward, is_done\n :rtype: (int, float, bool)\n \"\"\"\n reward = 0.0\n \n if self._steps >= self._max_steps:\n return self._current_state, reward, True\n \n transition = np.random.choice(range(3), p=self._get_transition_probs(action))\n if transition == 0:\n self._current_state -= 1\n elif transition == 1:\n pass\n else:\n self._current_state += 1\n \n if self._current_state == 0:\n reward = self.LEFT_REWARD\n elif self._current_state == self.n_states - 1:\n reward = self.RIGHT_REWARD\n \n self._steps += 1\n return self._current_state, reward, False",
"_____no_output_____"
]
],
[
[
"Lets impement Q-learning agent with an $\\varepsilon$-greedy strategy and look at its performance.",
"_____no_output_____"
]
],
[
[
"class QLearningAgent:\n def __init__(self, n_states, n_actions, lr=0.2, gamma=0.95, epsilon=0.1):\n self._gamma = gamma\n self._epsilon = epsilon\n self._q_matrix = np.zeros((n_states, n_actions))\n self._lr = lr\n\n def get_action(self, state):\n if np.random.random() < self._epsilon:\n return np.random.randint(0, self._q_matrix.shape[1])\n else:\n return np.argmax(self._q_matrix[state])\n \n def get_q_matrix(self):\n \"\"\" Used for policy visualization\n \"\"\"\n \n return self._q_matrix\n \n def start_episode(self):\n \"\"\" Used in PSRL agent\n \"\"\"\n pass\n\n def update(self, state, action, reward, next_state):\n max_val = self._q_matrix[next_state].max()\n self._q_matrix[state][action] += self._lr * \\\n (reward + self._gamma * max_val - self._q_matrix[state][action])",
"_____no_output_____"
],
[
"def train_mdp_agent(agent, env, n_episodes):\n episode_rewards = []\n \n for ep in range(n_episodes):\n state, ep_reward, is_done = env.reset()\n agent.start_episode()\n while not is_done:\n action = agent.get_action(state)\n \n next_state, reward, is_done = env.step(action)\n agent.update(state, action, reward, next_state)\n \n state = next_state\n ep_reward += reward\n \n episode_rewards.append(ep_reward)\n return episode_rewards",
"_____no_output_____"
],
[
"env = RiverSwimEnv()\nagent = QLearningAgent(env.n_states, env.n_actions)\nrews = train_mdp_agent(agent, env, 1000)\nplt.figure(figsize=(15, 8))\n\nplt.plot(pd.DataFrame(np.array(rews)).ewm(alpha=.1).mean())\nplt.xlabel(\"Episode count\")\nplt.ylabel(\"Reward\")\nplt.show()",
"_____no_output_____"
]
],
[
[
"Implement policy decisions visualization:",
"_____no_output_____"
]
],
[
[
"def plot_policy(agent):\n fig = plt.figure(figsize=(15, 8))\n ax = fig.add_subplot(111)\n ax.matshow(agent.get_q_matrix().T)\n ax.set_yticklabels(['', 'left', 'right'])\n plt.xlabel(\"State\")\n plt.ylabel(\"Action\")\n plt.title(\"Values of state-action pairs\")\n plt.show()",
"_____no_output_____"
],
[
"plot_policy(agent)",
"_____no_output_____"
]
],
[
[
"We can see that the agent uses non-optimalstrategy and keeps going left, not knowing the better option.",
"_____no_output_____"
],
[
"## Posterior sampling RL",
"_____no_output_____"
],
[
"Let us implement Thompson Sampling for MPDP!\n\nThe algorithm:\n\n>**for** episode $k = 1,2,...$ **do**\n>> sample $M_k \\sim f(\\bullet\\ |\\ H_k)$\n\n>> compute policy $\\mu_k$ for $M_k$\n\n>> **for** time $t = 1, 2,...$ **do**\n\n>>> take action $a_t$ from $\\mu_k$ \n\n>>> observe $r_t$ and $s_{t+1}$\n>>> update $H_k$\n\n>> **end for**\n\n>**end for**\n\n$M_k$ is modeled as two matrices: transitions and rewards. Transition matrix is sampled from Dirichlet distribution while reward matrix is sampled from Gamma-normal distribution.\n\nThe distibutions are updated following the bayes rule (see https://en.wikipedia.org/wiki/Conjugate_prior for continous distributions).\n\nFollow-up reading - https://arxiv.org/abs/1306.0940.",
"_____no_output_____"
]
],
[
[
"def bellman(val, trans, rewards):\n q = np.zeros(rewards.shape)\n for state in range(rewards.shape[0]):\n for action in range(rewards.shape[1]):\n q[state][action] = rewards[state][action] + val.dot(trans[state,:,action])\n return q.max(axis=1)",
"_____no_output_____"
],
[
"def sample_normal_gamma(mu, lmbd, alpha, beta):\n \"\"\" https://en.wikipedia.org/wiki/Normal-gamma_distribution\n \"\"\"\n tau = np.random.gamma(alpha, beta)\n mu = np.random.normal(mu , 1.0 / np.sqrt(lmbd * tau))\n return mu, tau\n\nclass PsrlAgent:\n def __init__(self, n_states, n_actions, horizon=10):\n self._n_states = n_states\n self._n_actions = n_actions\n self._horizon = horizon\n \n # params for transition sampling - Dirichlet distribution\n self._transition_counts = np.zeros((n_states, n_states, n_actions)) + 1.0\n \n # params for reward sampling - Normal-gamma distribution\n self._mu_matrix = np.zeros((n_states, n_actions)) + 1.0\n self._state_action_counts = np.zeros((n_states, n_actions)) + 1.0 # lambda\n \n self._alpha_matrix = np.zeros((n_states, n_actions)) + 1.0\n self._beta_matrix = np.zeros((n_states, n_actions)) + 1.0\n self._val = np.zeros(n_states)\n \n def _value_iteration(self, transitions, rewards):\n for _ in range(self._horizon):\n self._val = bellman(self._val, transitions, rewards)\n return self._val\n \n def start_episode(self):\n # sample new mdp\n self._sampled_transitions = np.apply_along_axis(np.random.dirichlet, 1, self._transition_counts)\n \n sampled_reward_mus, sampled_reward_stds = sample_normal_gamma(\n self._mu_matrix,\n self._state_action_counts,\n self._alpha_matrix,\n self._beta_matrix\n )\n\n self._sampled_rewards = sampled_reward_mus\n self._current_value_function = self._value_iteration(self._sampled_transitions, self._sampled_rewards)\n\n def get_action(self, state):\n return np.argmax(self._sampled_rewards[state] + \n self._current_value_function.dot(self._sampled_transitions[state]))\n\n def update(self, state, action, reward, next_state):\n \n mu0 = self._mu_matrix[state][action]\n v = self._state_action_counts[state][action]\n a = self._alpha_matrix[state][action]\n b = self._beta_matrix[state][action]\n \n self._mu_matrix[state][action] = (mu0 * v + reward) / (v + 1)\n self._state_action_counts[state][action] = v + 1\n self._alpha_matrix[state][action] = a + 1/2\n self._beta_matrix[state][action] = b + (v / (v + 1)) * (((reward - mu0) ** 2) / 2)\n \n self._transition_counts[state][next_state][action] += 1\n \n def get_q_matrix(self):\n return self._sampled_rewards + self._current_value_function.dot(self._sampled_transitions)",
"_____no_output_____"
],
[
"horizon = 20\nenv = RiverSwimEnv(max_steps=horizon)\nagent = PsrlAgent(env.n_states, env.n_actions, horizon=horizon)\nrews = train_mdp_agent(agent, env, 1000)\n\nplt.figure(figsize=(15, 8))\nplt.plot(pd.DataFrame(np.array(rews)).ewm(alpha=.1).mean())\n\nplt.xlabel(\"Episode count\")\nplt.ylabel(\"Reward\")\nplt.show()",
"_____no_output_____"
],
[
"plot_policy(agent)",
"_____no_output_____"
]
]
] |
[
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown",
"markdown"
],
[
"code",
"code",
"code",
"code"
]
] |
cbd9c5a76e011a3fcac8e4b01a6f0d6933cd7413
| 579,607 |
ipynb
|
Jupyter Notebook
|
miscellaneous/Helmet_recognition_coco(data prepare).ipynb
|
vemodalen-x/ISS_Project
|
85d92d8283e50550a42fc3fe367c61a253ed9730
|
[
"MIT"
] | null | null | null |
miscellaneous/Helmet_recognition_coco(data prepare).ipynb
|
vemodalen-x/ISS_Project
|
85d92d8283e50550a42fc3fe367c61a253ed9730
|
[
"MIT"
] | null | null | null |
miscellaneous/Helmet_recognition_coco(data prepare).ipynb
|
vemodalen-x/ISS_Project
|
85d92d8283e50550a42fc3fe367c61a253ed9730
|
[
"MIT"
] | null | null | null | 579,607 | 579,607 | 0.831218 |
[
[
[
"import os\nfrom google.colab import drive\ndrive.mount('/content/gdrive')",
"Drive already mounted at /content/gdrive; to attempt to forcibly remount, call drive.mount(\"/content/gdrive\", force_remount=True).\n"
],
[
"!mkdir train_local\n!unzip /content/gdrive/MyDrive/NFL_Helmet/nfl-health-and-safety-helmet-assignment.zip -d train_local/",
"\u001b[1;30;43m流式输出内容被截断，只能显示最后 5000 行内容。\u001b[0m\n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57830_000403_Sideline_frame0871.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57830_000734_Endzone_frame60.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57830_000734_Sideline_frame0520.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57830_000734_Sideline_frame89.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57830_000965_Endzone_frame192.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57830_000965_Sideline_frame3.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57830_001765_Sideline_frame0675.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57830_002030_Endzone_frame496.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57830_002030_Sideline_frame191.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57830_002073_Endzone_frame1065.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57830_002900_Endzone_frame166.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57830_002900_Sideline_frame174.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57830_002944_Endzone_frame0746.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57831_000168_Endzone_frame0856.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57831_000400_Endzone_frame0693.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57831_000530_Endzone_frame47.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57831_000530_Sideline_frame159.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57831_001771_Endzone_frame0880.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57831_002303_Sideline_frame0977.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57831_002356_Sideline_frame0637.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57831_002476_Endzone_frame137.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57831_002476_Sideline_frame200.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57831_002605_Sideline_frame0587.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57831_003460_Endzone_frame163.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57831_003460_Sideline_frame593.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57831_003702_Endzone_frame0630.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57831_004235_Endzone_frame0656.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57831_004348_Sideline_frame0509.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57832_000221_Sideline_frame0921.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57832_000381_Endzone_frame191.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57832_000381_Sideline_frame34.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57832_001054_Endzone_frame78.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57832_001054_Sideline_frame48.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57832_001148_Sideline_frame1110.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57832_001281_Endzone_frame157.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57832_001281_Sideline_frame401.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57832_002095_Sideline_frame1315.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57832_002319_Sideline_frame0978.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57832_002481_Endzone_frame0772.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57832_002984_Sideline_frame0997.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57832_003363_Endzone_frame0651.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57832_004201_Endzone_frame0667.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57832_004278_Endzone_frame178.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57832_004278_Sideline_frame311.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57833_000420_Endzone_frame178.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57833_000420_Sideline_frame137.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57833_000442_Endzone_frame161.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57833_000442_Sideline_frame404.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57833_001229_Sideline_frame0787.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57833_001450_Endzone_frame1056.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57833_001904_Sideline_frame1366.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57833_002035_Endzone_frame0963.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57833_002930_Endzone_frame0764.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57833_003392_Endzone_frame1210.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57833_003556_Sideline_frame0783.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57833_003999_Sideline_frame0745.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57833_004038_Endzone_frame0999.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57833_004284_Endzone_frame272.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57833_004284_Sideline_frame38.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57834_000186_Endzone_frame0319.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57834_000305_Sideline_frame0625.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57834_000865_Endzone_frame0477.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57834_001495_Endzone_frame0778.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57834_001520_Sideline_frame0590.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57834_001542_Endzone_frame1033.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57834_002150_Sideline_frame0909.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57834_002800_Sideline_frame0833.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57834_002954_Endzone_frame1044.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57835_000036_Sideline_frame1526.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57835_000058_Endzone_frame0922.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57835_000416_Sideline_frame1019.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57835_001270_Sideline_frame1213.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57835_001636_Endzone_frame0903.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57835_001765_Endzone_frame0566.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57835_003370_Endzone_frame1096.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57835_004502_Sideline_frame0892.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57836_000075_Sideline_frame0716.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57836_000264_Endzone_frame0556.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57836_000815_Endzone_frame0781.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57836_001102_Endzone_frame0497.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57836_001124_Endzone_frame374.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57836_001124_Sideline_frame219.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57836_001480_Endzone_frame0611.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57836_001662_Endzone_frame0673.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57836_002246_Endzone_frame16.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57836_002246_Sideline_frame0830.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57836_002246_Sideline_frame50.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57836_002303_Endzone_frame567.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57836_002303_Sideline_frame107.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57836_003188_Endzone_frame117.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57836_003188_Sideline_frame189.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57836_003379_Sideline_frame0435.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57837_000160_Endzone_frame0560.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57837_001920_Endzone_frame31.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57837_001920_Sideline_frame126.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57837_002359_Endzone_frame77.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57837_002359_Sideline_frame283.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57837_002514_Sideline_frame0728.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57837_002735_Endzone_frame1108.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57837_002822_Sideline_frame0559.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57837_002925_Sideline_frame0939.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57837_003066_Endzone_frame230.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57837_003066_Sideline_frame304.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57837_003130_Sideline_frame0865.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57837_003294_Endzone_frame1096.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57837_003479_Endzone_frame0799.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57837_003682_Sideline_frame0856.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57838_000193_Endzone_frame0778.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57838_000388_Endzone_frame0946.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57838_001437_Endzone_frame0599.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57838_001572_Endzone_frame147.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57838_001572_Sideline_frame421.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57838_001763_Endzone_frame0853.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57838_002081_Sideline_frame0660.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57838_002242_Endzone_frame360.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57838_002242_Sideline_frame102.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57838_002938_Sideline_frame0770.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57838_003242_Sideline_frame0664.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57838_003843_Endzone_frame0.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57838_003843_Sideline_frame368.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57838_003885_Endzone_frame203.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57838_003885_Sideline_frame265.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57838_004547_Sideline_frame1059.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57838_004795_Sideline_frame0537.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57839_000135_Sideline_frame0883.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57839_000473_Sideline_frame0860.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57839_000572_Endzone_frame547.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57839_000572_Sideline_frame261.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57839_000695_Endzone_frame1332.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57839_001204_Endzone_frame0907.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57839_001233_Sideline_frame0705.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57839_001302_Endzone_frame0948.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57839_001664_Sideline_frame0770.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57839_001685_Endzone_frame1163.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57839_002739_Sideline_frame1008.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57839_003183_Endzone_frame388.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57839_003183_Sideline_frame446.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57839_003645_Endzone_frame144.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57839_003645_Sideline_frame541.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57839_003812_Endzone_frame351.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57839_003812_Sideline_frame422.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57840_000361_Endzone_frame0486.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57840_000619_Sideline_frame0878.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57840_000999_Endzone_frame0591.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57840_001248_Endzone_frame0705.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57840_001441_Endzone_frame403.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57840_001441_Sideline_frame442.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57840_001976_Sideline_frame0840.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57840_002720_Endzone_frame215.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57840_002720_Sideline_frame89.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57840_002972_Endzone_frame282.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57840_002972_Sideline_frame459.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57840_003320_Endzone_frame0475.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57840_003865_Sideline_frame0867.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57840_003915_Sideline_frame1186.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57840_004138_Endzone_frame262.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57840_004138_Sideline_frame533.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57840_004381_Sideline_frame0814.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57841_000578_Endzone_frame28.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57841_000578_Sideline_frame311.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57841_000670_Endzone_frame0707.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57841_000756_Endzone_frame0722.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57841_001360_Endzone_frame0615.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57841_001462_Endzone_frame0650.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57841_001487_Endzone_frame0966.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57841_001553_Sideline_frame0766.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57841_002420_Endzone_frame217.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57841_002420_Sideline_frame431.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57841_002911_Sideline_frame0506.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57841_003434_Endzone_frame221.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57841_003434_Sideline_frame199.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57841_003938_Sideline_frame0938.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57842_000069_Sideline_frame0782.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57842_000438_Endzone_frame124.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57842_000438_Sideline_frame175.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57842_000981_Sideline_frame1335.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57842_001136_Endzone_frame393.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57842_001136_Sideline_frame359.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57842_001715_Endzone_frame0596.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57842_002734_Endzone_frame0902.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57842_003103_Sideline_frame0898.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57842_003136_Endzone_frame454.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57842_003136_Sideline_frame390.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57842_003259_Sideline_frame0736.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57842_003455_Endzone_frame0527.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57842_003616_Endzone_frame1627.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57842_003734_Sideline_frame0837.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57842_003821_Endzone_frame390.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57842_003821_Sideline_frame142.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57843_000608_Endzone_frame257.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57843_000608_Sideline_frame413.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57843_001020_Endzone_frame0657.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57843_002655_Sideline_frame0921.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57843_002729_Sideline_frame1034.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57843_002837_Sideline_frame1043.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57843_003018_Endzone_frame0604.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57843_003602_Sideline_frame311.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57843_003757_Sideline_frame0917.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57843_003842_Endzone_frame0541.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57843_004469_Endzone_frame98.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57843_004469_Sideline_frame14.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57843_004591_Endzone_frame201.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57843_004591_Sideline_frame345.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57843_004639_Endzone_frame0396.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57844_000460_Sideline_frame0763.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57844_000507_Endzone_frame216.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57844_000507_Sideline_frame65.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57844_000826_Sideline_frame0803.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57844_001322_Sideline_frame0635.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57844_001599_Endzone_frame450.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57844_001599_Sideline_frame7.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57844_001621_Sideline_frame134.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57844_002117_Endzone_frame132.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57844_002117_Sideline_frame342.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57844_002159_Sideline_frame1167.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57844_002751_Endzone_frame1055.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57844_002825_Endzone_frame0613.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57844_003594_Sideline_frame1110.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57844_003933_Endzone_frame0683.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57844_004173_Endzone_frame0559.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57845_000953_Endzone_frame114.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57845_000953_Sideline_frame223.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57845_001269_Endzone_frame0525.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57845_001523_Sideline_frame0816.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57845_001548_Sideline_frame0524.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57845_002192_Endzone_frame254.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57845_002192_Sideline_frame555.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57845_002746_Endzone_frame0795.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57845_003013_Sideline_frame1016.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57845_003613_Endzone_frame186.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57845_003613_Sideline_frame310.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57845_003641_Endzone_frame0779.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57845_003889_Sideline_frame1546.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57845_004007_Sideline_frame0675.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57845_004201_Endzone_frame0870.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57845_004310_Endzone_frame136.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57845_004310_Sideline_frame121.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57846_000457_Endzone_frame1720.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57846_000482_Endzone_frame134.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57846_000482_Sideline_frame42.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57846_001230_Endzone_frame1516.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57846_001254_Endzone_frame391.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57846_001254_Sideline_frame327.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57846_001361_Sideline_frame1024.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57846_001791_Sideline_frame1225.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57846_001992_Endzone_frame230.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57846_001992_Sideline_frame205.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57846_002882_Endzone_frame515.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57846_002882_Sideline_frame453.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57846_003073_Sideline_frame0937.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57846_003111_Sideline_frame1026.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57846_003187_Endzone_frame1115.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57846_004213_Endzone_frame0711.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57846_004292_Sideline_frame0831.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57847_000472_Sideline_frame0711.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57847_001039_Endzone_frame0732.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57847_002157_Endzone_frame0557.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57847_002593_Sideline_frame1693.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57847_002745_Sideline_frame1313.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57847_003031_Endzone_frame0793.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57847_003167_Sideline_frame0970.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57847_003461_Endzone_frame1015.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57847_004097_Sideline_frame0742.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57848_000428_Endzone_frame0822.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57848_000738_Sideline_frame0669.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57848_000983_Endzone_frame0507.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57848_001153_Endzone_frame144.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57848_001153_Sideline_frame290.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57848_001505_Endzone_frame175.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57848_001505_Sideline_frame197.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57848_001598_Sideline_frame1173.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57848_001665_Endzone_frame0849.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57848_001826_Endzone_frame357.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57848_001826_Sideline_frame472.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57848_002061_Sideline_frame0529.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57848_002515_Endzone_frame107.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57848_002515_Sideline_frame0794.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57848_002515_Sideline_frame11.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57848_003040_Sideline_frame1116.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57848_003716_Endzone_frame0604.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57849_000179_Sideline_frame0894.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57849_000200_Sideline_frame1137.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57849_000998_Sideline_frame0580.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57849_002130_Sideline_frame0619.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57849_002749_Endzone_frame1387.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57849_003301_Endzone_frame0324.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57849_003631_Endzone_frame0514.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57849_004432_Sideline_frame1130.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57849_004601_Endzone_frame0800.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57850_000268_Endzone_frame0636.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57850_001305_Sideline_frame0864.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57850_001681_Endzone_frame0730.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57850_001757_Endzone_frame0811.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57850_002530_Sideline_frame0954.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57850_002638_Endzone_frame0868.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57850_003329_Sideline_frame0664.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57850_003796_Endzone_frame0742.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57850_004843_Sideline_frame0279.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57851_000379_Sideline_frame0897.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57851_001458_Sideline_frame1205.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57851_001480_Sideline_frame0810.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57851_001942_Endzone_frame484.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57851_001942_Sideline_frame205.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57851_002468_Endzone_frame1183.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57851_002564_Endzone_frame0807.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57851_003068_Endzone_frame54.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57851_003068_Sideline_frame321.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57851_003245_Sideline_frame1131.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57851_003303_Endzone_frame0892.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57851_003521_Endzone_frame209.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57851_003521_Sideline_frame0806.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57851_003521_Sideline_frame355.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57851_003792_Endzone_frame179.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57851_003792_Sideline_frame459.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57851_003816_Endzone_frame0672.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57852_000072_Endzone_frame0496.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57852_000114_Endzone_frame65.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57852_000114_Sideline_frame199.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57852_000596_Endzone_frame0587.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57852_000667_Sideline_frame0611.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57852_002070_Endzone_frame377.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57852_002070_Sideline_frame273.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57852_002511_Endzone_frame333.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57852_002511_Sideline_frame82.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57852_002630_Sideline_frame0680.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57852_002927_Endzone_frame0788.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57852_003541_Endzone_frame319.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57852_003541_Sideline_frame20.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57852_003781_Endzone_frame0524.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57852_003972_Endzone_frame0692.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57852_004190_Sideline_frame0680.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57853_000512_Sideline_frame0742.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57853_000582_Endzone_frame37.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57853_000582_Sideline_frame41.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57853_000823_Endzone_frame1008.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57853_001195_Endzone_frame1007.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57853_001454_Sideline_frame1047.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57853_001717_Endzone_frame76.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57853_001717_Sideline_frame127.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57853_002396_Endzone_frame425.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57853_002396_Sideline_frame403.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57853_002417_Sideline_frame0485.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57853_002744_Sideline_frame0778.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57853_003342_Sideline_frame1272.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57853_003496_Endzone_frame0971.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57853_004081_Endzone_frame0999.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57854_000409_Endzone_frame0634.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57854_001255_Sideline_frame0703.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57854_001492_Endzone_frame138.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57854_001492_Sideline_frame265.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57854_001622_Endzone_frame1060.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57854_001643_Endzone_frame0923.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57854_001703_Sideline_frame0843.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57854_002197_Endzone_frame1274.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57854_002380_Sideline_frame1262.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57854_002506_Endzone_frame109.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57854_002506_Sideline_frame213.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57854_002670_Endzone_frame0796.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57854_003178_Endzone_frame60.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57854_003178_Sideline_frame201.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57854_003520_Endzone_frame177.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57854_003520_Sideline_frame134.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57854_003643_Sideline_frame1168.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57855_000602_Sideline_frame0466.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57855_001290_Endzone_frame396.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57855_001290_Sideline_frame221.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57855_001426_Endzone_frame459.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57855_001426_Sideline_frame278.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57855_001574_Endzone_frame0513.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57855_001908_Endzone_frame1538.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57855_002334_Endzone_frame437.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57855_002334_Sideline_frame98.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57855_002546_Sideline_frame0759.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57855_003021_Sideline_frame0696.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57855_003492_Endzone_frame0764.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57855_003717_Sideline_frame1136.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57855_003787_Endzone_frame32.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57855_003787_Sideline_frame199.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57855_003907_Endzone_frame0772.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57855_004245_Sideline_frame1218.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57856_000536_Endzone_frame0547.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57856_000700_Sideline_frame0957.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57856_000891_Endzone_frame260.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57856_000891_Sideline_frame403.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57856_001638_Endzone_frame1073.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57856_002394_Endzone_frame244.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57856_002394_Sideline_frame267.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57856_002462_Endzone_frame231.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57856_002462_Sideline_frame399.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57856_002908_Sideline_frame0848.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57856_003041_Endzone_frame1496.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57856_003483_Sideline_frame0762.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57856_003716_Endzone_frame0704.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57856_004065_Sideline_frame0865.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57856_004275_Sideline_frame0786.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57857_000103_Endzone_frame2110.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57857_000708_Endzone_frame0854.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57857_000880_Endzone_frame0768.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57857_000947_Endzone_frame1113.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57857_001101_Endzone_frame100.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57857_001101_Sideline_frame123.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57857_002070_Sideline_frame0619.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57857_003460_Sideline_frame1323.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57857_003482_Sideline_frame0791.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57857_003854_Endzone_frame0637.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57857_003854_Endzone_frame401.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57857_003854_Sideline_frame331.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57857_004092_Sideline_frame1048.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57857_004288_Endzone_frame132.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57857_004288_Sideline_frame247.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57857_004335_Endzone_frame205.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57857_004335_Sideline_frame110.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57858_000641_Endzone_frame1345.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57858_001342_Endzone_frame0611.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57858_001595_Endzone_frame278.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57858_001595_Sideline_frame100.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57858_001786_Sideline_frame0713.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57858_002102_Endzone_frame0414.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57858_002200_Sideline_frame0757.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57858_002639_Endzone_frame0505.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57858_002751_Endzone_frame0920.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57858_003016_Sideline_frame1198.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57858_003061_Endzone_frame132.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57858_003061_Sideline_frame267.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57858_004258_Endzone_frame49.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57858_004258_Sideline_frame0656.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57858_004258_Sideline_frame144.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57858_004416_Endzone_frame109.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57858_004416_Sideline_frame159.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57859_000067_Sideline_frame0676.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57859_000252_Endzone_frame295.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57859_000252_Sideline_frame291.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57859_000636_Sideline_frame0990.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57859_000930_Endzone_frame520.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57859_000930_Sideline_frame538.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57859_001589_Endzone_frame0575.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57859_001778_Endzone_frame0990.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57859_002013_Endzone_frame0709.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57859_002420_Sideline_frame1132.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57859_002567_Endzone_frame137.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57859_002567_Sideline_frame347.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57859_003221_Endzone_frame5.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57859_003221_Sideline_frame217.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57859_003364_Endzone_frame1100.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57859_004648_Sideline_frame0925.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57860_000037_Endzone_frame1217.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57860_000200_Endzone_frame1003.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57860_000433_Endzone_frame0742.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57860_000848_Sideline_frame1113.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57860_000962_Endzone_frame1067.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57860_001311_Endzone_frame212.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57860_001311_Sideline_frame190.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57860_001591_Endzone_frame299.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57860_001591_Sideline_frame346.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57860_001869_Endzone_frame333.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57860_001869_Sideline_frame247.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57860_002557_Sideline_frame0992.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57860_002578_Endzone_frame1532.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57860_002643_Endzone_frame60.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57860_002643_Sideline_frame122.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57860_003211_Sideline_frame1016.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57860_003813_Sideline_frame1071.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57861_000263_Endzone_frame173.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57861_000263_Sideline_frame553.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57861_000746_Endzone_frame0604.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57861_002265_Endzone_frame0777.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57861_002452_Endzone_frame120.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57861_002452_Sideline_frame264.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57861_002587_Endzone_frame449.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57861_002587_Sideline_frame13.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57861_002813_Sideline_frame0985.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57861_002930_Endzone_frame0985.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57861_003088_Endzone_frame0710.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57861_003449_Sideline_frame0515.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57861_003470_Sideline_frame0956.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57861_003562_Endzone_frame1068.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57861_003745_Sideline_frame0564.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57861_004079_Endzone_frame329.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57861_004079_Sideline_frame279.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57862_000058_Endzone_frame99.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57862_000058_Sideline_frame35.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57862_000244_Sideline_frame0657.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57862_000285_Sideline_frame0773.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57862_000713_Endzone_frame9.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57862_000713_Sideline_frame222.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57862_001243_Endzone_frame0904.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57862_001243_Endzone_frame74.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57862_001243_Sideline_frame7.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57862_001506_Endzone_frame0760.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57862_002634_Endzone_frame0745.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57862_003033_Sideline_frame0711.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57862_003407_Endzone_frame0465.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57862_003625_Sideline_frame0709.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57862_003668_Endzone_frame0569.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57863_000036_Sideline_frame0838.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57863_000405_Endzone_frame0783.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57863_000785_Endzone_frame273.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57863_000785_Sideline_frame199.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57863_001027_Sideline_frame0721.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57863_001785_Endzone_frame133.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57863_001785_Sideline_frame109.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57863_001885_Endzone_frame1066.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57863_002060_Endzone_frame0968.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57863_002142_Endzone_frame44.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57863_002142_Sideline_frame59.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57863_002717_Endzone_frame234.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57863_002717_Sideline_frame168.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57863_003140_Endzone_frame0359.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57863_003745_Sideline_frame1157.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57863_003792_Endzone_frame0909.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57864_000277_Sideline_frame0917.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57864_000348_Endzone_frame363.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57864_000348_Sideline_frame512.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57864_001110_Sideline_frame0841.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57864_001643_Endzone_frame1110.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57864_002408_Sideline_frame0700.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57864_002547_Endzone_frame0594.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57864_002907_Endzone_frame18.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57864_002907_Sideline_frame260.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57864_002985_Endzone_frame211.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57864_002985_Sideline_frame201.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57864_003031_Sideline_frame1137.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57864_003580_Endzone_frame0973.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57864_003649_Endzone_frame1065.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57864_003759_Sideline_frame0841.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57864_003781_Endzone_frame379.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57864_003781_Sideline_frame365.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57865_000051_Endzone_frame117.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57865_000051_Sideline_frame25.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57865_000278_Endzone_frame0841.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57865_000547_Sideline_frame1714.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57865_000710_Endzone_frame275.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57865_000710_Sideline_frame7.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57865_000751_Endzone_frame0743.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57865_000961_Endzone_frame280.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57865_000961_Sideline_frame118.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57865_001092_Sideline_frame0771.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57865_001326_Endzone_frame0916.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57865_001454_Sideline_frame1207.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57865_001912_Endzone_frame251.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57865_001912_Sideline_frame111.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57865_002788_Sideline_frame1203.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57865_002980_Sideline_frame1382.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57865_003825_Endzone_frame0629.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57866_000749_Endzone_frame190.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57866_000749_Sideline_frame5.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57866_001035_Sideline_frame0976.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57866_001078_Endzone_frame134.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57866_001078_Sideline_frame57.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57866_001101_Sideline_frame1002.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57866_001566_Endzone_frame0502.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57866_002064_Endzone_frame115.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57866_002064_Sideline_frame238.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57866_002590_Sideline_frame1599.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57866_002678_Sideline_frame1073.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57866_002983_Endzone_frame0942.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57866_003916_Endzone_frame0633.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57866_003987_Endzone_frame236.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57866_003987_Sideline_frame208.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57866_004224_Endzone_frame1366.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57866_004271_Sideline_frame0865.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57867_000132_Endzone_frame106.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57867_000132_Sideline_frame378.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57867_000392_Sideline_frame0540.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57867_000452_Endzone_frame0698.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57867_000661_Sideline_frame0992.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57867_000997_Endzone_frame0576.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57867_001019_Sideline_frame0813.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57867_001631_Endzone_frame1083.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57867_001696_Endzone_frame1273.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57867_001860_Endzone_frame265.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57867_001860_Sideline_frame169.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57867_002093_Endzone_frame154.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57867_002093_Sideline_frame230.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57867_003433_Sideline_frame0598.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57867_004039_Endzone_frame3.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57867_004039_Sideline_frame90.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57867_004300_Sideline_frame0680.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57868_000057_Sideline_frame0494.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57868_000368_Endzone_frame15.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57868_000368_Sideline_frame414.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57868_000420_Sideline_frame0997.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57868_001209_Sideline_frame0917.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57868_001465_Endzone_frame0565.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57868_001636_Endzone_frame234.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57868_001636_Sideline_frame60.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57868_001684_Endzone_frame0805.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57868_002672_Endzone_frame2606.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57868_002713_Sideline_frame1108.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57868_002755_Endzone_frame146.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57868_002755_Sideline_frame286.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57868_003411_Endzone_frame131.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57868_003411_Sideline_frame106.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57868_003636_Sideline_frame0488.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57868_003657_Endzone_frame0868.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57869_000393_Endzone_frame249.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57869_000393_Sideline_frame5.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57869_000541_Endzone_frame0520.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57869_000649_Endzone_frame234.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57869_000649_Sideline_frame364.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57869_001140_Endzone_frame121.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57869_001140_Sideline_frame1604.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57869_001140_Sideline_frame64.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57869_001885_Sideline_frame0840.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57869_002323_Sideline_frame1547.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57869_002392_Sideline_frame0927.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57869_002669_Endzone_frame0588.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57869_002968_Endzone_frame379.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57869_002968_Sideline_frame397.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57869_003030_Endzone_frame0974.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57869_003321_Endzone_frame1010.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57869_003452_Sideline_frame0767.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57870_000298_Endzone_frame147.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57870_000298_Sideline_frame185.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57870_000367_Endzone_frame357.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57870_000367_Sideline_frame39.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57870_001167_Endzone_frame0899.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57870_001321_Endzone_frame0693.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57870_001345_Sideline_frame0905.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57870_001511_Endzone_frame0938.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57870_002605_Endzone_frame72.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57870_002605_Sideline_frame426.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57870_002879_Sideline_frame0587.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57870_003060_Endzone_frame1223.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57870_003300_Sideline_frame0581.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57870_003474_Sideline_frame0707.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57870_004246_Endzone_frame204.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57870_004246_Sideline_frame495.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57870_004466_Sideline_frame0765.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57871_000794_Endzone_frame0854.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57871_000920_Sideline_frame0656.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57871_001271_Endzone_frame331.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57871_001271_Sideline_frame302.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57871_001337_Endzone_frame1109.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57871_001486_Endzone_frame1001.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57871_001610_Sideline_frame0729.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57871_002144_Sideline_frame1596.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57871_002424_Endzone_frame206.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57871_002424_Sideline_frame416.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57871_002448_Sideline_frame0439.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57871_002583_Endzone_frame14.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57871_002583_Sideline_frame415.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57871_003564_Endzone_frame0919.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57871_003792_Endzone_frame152.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57871_003792_Sideline_frame301.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57872_000904_Sideline_frame1130.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57872_000933_Endzone_frame275.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57872_000933_Sideline_frame406.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57872_000962_Endzone_frame0632.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57872_000984_Endzone_frame8.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57872_000984_Sideline_frame170.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57872_001046_Endzone_frame381.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57872_001046_Sideline_frame224.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57872_001742_Endzone_frame0822.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57872_001907_Sideline_frame0878.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57872_002041_Endzone_frame185.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57872_002041_Sideline_frame60.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57872_002086_Endzone_frame0977.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57872_002171_Sideline_frame1061.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57872_002509_Endzone_frame0694.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57872_002937_Sideline_frame0808.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57872_003643_Endzone_frame0533.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57873_000268_Endzone_frame0806.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57873_000399_Endzone_frame1010.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57873_000664_Sideline_frame1133.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57873_000752_Endzone_frame92.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57873_000752_Sideline_frame530.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57873_001154_Endzone_frame0965.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57873_001178_Sideline_frame1074.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57873_001204_Endzone_frame0573.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57873_001907_Endzone_frame164.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57873_001907_Sideline_frame108.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57873_002819_Sideline_frame0642.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57873_002888_Sideline_frame1133.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57873_003005_Endzone_frame356.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57873_003005_Sideline_frame171.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57873_003110_Endzone_frame0437.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57873_003339_Endzone_frame84.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57873_003339_Sideline_frame24.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57874_000138_Sideline_frame0707.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57874_000204_Endzone_frame351.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57874_000204_Sideline_frame437.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57874_000516_Endzone_frame0619.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57874_001471_Endzone_frame0755.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57874_001635_Sideline_frame0874.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57874_001774_Endzone_frame288.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57874_001774_Sideline_frame158.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57874_001856_Endzone_frame1067.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57874_001931_Endzone_frame350.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57874_001931_Sideline_frame79.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57874_001952_Sideline_frame0693.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57874_001995_Endzone_frame0759.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57874_002211_Endzone_frame391.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57874_002211_Sideline_frame1.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57874_002561_Endzone_frame0618.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57874_003702_Sideline_frame1044.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57875_000223_Endzone_frame143.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57875_000223_Sideline_frame188.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57875_000392_Sideline_frame1101.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57875_000560_Endzone_frame0945.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57875_001272_Sideline_frame1190.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57875_001734_Endzone_frame256.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57875_001734_Sideline_frame233.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57875_001808_Endzone_frame454.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57875_001808_Sideline_frame556.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57875_001993_Endzone_frame1037.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57875_002607_Endzone_frame529.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57875_002607_Sideline_frame257.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57875_003103_Sideline_frame0855.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57875_003642_Endzone_frame0535.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57875_003695_Sideline_frame1197.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57875_003922_Sideline_frame0813.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57875_004182_Endzone_frame0633.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57876_001121_Endzone_frame278.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57876_001121_Sideline_frame368.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57876_001334_Sideline_frame0841.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57876_001353_Endzone_frame0774.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57876_001423_Endzone_frame0651.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57876_001627_Endzone_frame218.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57876_001627_Sideline_frame364.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57876_001916_Sideline_frame0698.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57876_002428_Endzone_frame0763.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57876_002450_Sideline_frame0744.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57876_003397_Sideline_frame0747.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57876_003520_Endzone_frame0437.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57876_003707_Sideline_frame0719.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57876_003782_Endzone_frame222.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57876_003782_Sideline_frame288.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57877_000688_Endzone_frame40.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57877_000688_Sideline_frame408.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57877_000750_Sideline_frame0874.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57877_001001_Endzone_frame0741.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57877_001169_Endzone_frame1328.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57877_001423_Sideline_frame0987.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57877_001492_Endzone_frame1348.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57877_002023_Endzone_frame254.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57877_002023_Sideline_frame161.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57877_002073_Sideline_frame0466.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57877_002141_Endzone_frame62.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57877_002141_Sideline_frame397.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57877_002489_Sideline_frame1491.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57877_003291_Endzone_frame0983.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57877_003458_Endzone_frame296.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57877_003458_Sideline_frame32.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57878_000159_Endzone_frame0679.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57878_000359_Sideline_frame0893.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57878_000583_Sideline_frame0619.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57878_000748_Endzone_frame310.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57878_000748_Sideline_frame269.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57878_001177_Endzone_frame1099.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57878_001463_Endzone_frame351.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57878_001463_Sideline_frame442.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57878_001705_Sideline_frame1162.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57878_002280_Endzone_frame1371.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57878_002902_Endzone_frame181.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57878_002902_Sideline_frame353.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57878_003104_Endzone_frame0811.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57878_004191_Endzone_frame346.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57878_004191_Sideline_frame119.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57878_004507_Sideline_frame0913.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57879_000140_Endzone_frame0996.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57879_000376_Endzone_frame359.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57879_000376_Sideline_frame137.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57879_000467_Sideline_frame0723.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57879_000814_Endzone_frame0887.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57879_001742_Endzone_frame0635.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57879_001826_Sideline_frame0582.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57879_002450_Endzone_frame23.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57879_002450_Sideline_frame160.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57879_003000_Endzone_frame0925.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57879_003151_Sideline_frame1110.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57879_003416_Sideline_frame0667.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57879_004512_Endzone_frame0763.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57880_000313_Endzone_frame0963.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57880_000354_Endzone_frame134.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57880_000354_Sideline_frame372.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57880_000616_Sideline_frame0961.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57880_000685_Sideline_frame1042.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57880_000824_Endzone_frame26.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57880_000824_Sideline_frame292.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57880_001273_Endzone_frame0613.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57880_001439_Sideline_frame1035.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57880_001495_Endzone_frame1047.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57880_001564_Endzone_frame0810.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57880_001623_Sideline_frame0664.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57880_001839_Endzone_frame46.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57880_001839_Sideline_frame188.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57880_002553_Endzone_frame151.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57880_002553_Sideline_frame441.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57880_003079_Sideline_frame0906.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57881_000385_Endzone_frame1001.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57881_000504_Endzone_frame0528.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57881_001212_Endzone_frame0565.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57881_001300_Sideline_frame0829.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57881_001393_Endzone_frame271.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57881_001393_Sideline_frame263.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57881_001506_Sideline_frame1552.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57881_001748_Sideline_frame0878.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57881_002465_Sideline_frame0637.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57881_002581_Endzone_frame0454.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57881_003159_Endzone_frame110.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57881_003159_Sideline_frame52.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57881_003595_Endzone_frame285.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57881_003595_Sideline_frame212.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57881_003779_Endzone_frame0713.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57882_000240_Endzone_frame0716.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57882_000546_Sideline_frame1315.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57882_000958_Endzone_frame1102.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57882_001008_Endzone_frame35.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57882_001008_Sideline_frame242.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57882_001482_Endzone_frame0554.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57882_001915_Endzone_frame197.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57882_001915_Sideline_frame97.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57882_001957_Endzone_frame0463.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57882_002352_Endzone_frame62.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57882_002352_Sideline_frame106.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57882_002508_Endzone_frame44.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57882_002508_Sideline_frame158.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57882_003328_Sideline_frame0980.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57882_003387_Sideline_frame0370.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57882_003901_Endzone_frame0780.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57882_004052_Sideline_frame0829.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57883_000234_Endzone_frame1012.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57883_000452_Sideline_frame0700.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57883_000543_Endzone_frame65.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57883_000543_Sideline_frame438.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57883_000883_Endzone_frame1662.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57883_001097_Endzone_frame19.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57883_001097_Sideline_frame19.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57883_001138_Endzone_frame230.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57883_001138_Sideline_frame37.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57883_001160_Sideline_frame0687.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57883_002580_Endzone_frame271.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57883_002580_Sideline_frame238.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57883_002751_Endzone_frame0809.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57883_002908_Endzone_frame0703.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57883_003015_Sideline_frame0914.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57883_003060_Sideline_frame0804.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57883_003152_Endzone_frame0581.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57884_000277_Endzone_frame204.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57884_000277_Sideline_frame160.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57884_000298_Sideline_frame1143.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57884_000430_Endzone_frame0555.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57884_001286_Sideline_frame0769.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57884_002225_Endzone_frame0632.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57884_002370_Sideline_frame1480.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57884_002389_Endzone_frame0866.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57884_002639_Endzone_frame65.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57884_002639_Sideline_frame64.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57884_002831_Sideline_frame0897.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57884_002855_Endzone_frame25.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57884_002855_Sideline_frame36.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57884_003311_Endzone_frame0507.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57884_003644_Endzone_frame169.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57884_003644_Sideline_frame471.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57884_004074_Endzone_frame0944.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57885_000091_Endzone_frame0757.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57885_000357_Sideline_frame0986.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57885_000456_Sideline_frame0415.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57885_000519_Sideline_frame0602.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57885_001364_Endzone_frame113.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57885_001364_Sideline_frame307.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57885_002279_Sideline_frame0726.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57885_002936_Endzone_frame307.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57885_002936_Sideline_frame154.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57885_002966_Endzone_frame170.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57885_002966_Sideline_frame23.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57885_003058_Endzone_frame0601.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57885_003400_Endzone_frame0885.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57885_003661_Endzone_frame0692.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57885_003850_Sideline_frame0543.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57885_003885_Endzone_frame380.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57885_003885_Sideline_frame501.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57886_000373_Endzone_frame1161.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57886_000630_Endzone_frame0912.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57886_001107_Endzone_frame1124.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57886_001131_Sideline_frame0862.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57886_001177_Sideline_frame0982.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57886_001997_Endzone_frame43.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57886_001997_Sideline_frame573.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57886_002019_Endzone_frame510.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57886_002019_Sideline_frame216.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57886_002048_Endzone_frame429.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57886_002048_Sideline_frame38.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57886_002072_Endzone_frame0495.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57886_002211_Endzone_frame0951.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57886_002420_Sideline_frame1480.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57886_002497_Sideline_frame0969.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57887_000051_Sideline_frame0907.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57887_000810_Sideline_frame1004.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57887_000958_Endzone_frame341.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57887_000958_Sideline_frame196.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57887_001028_Sideline_frame0634.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57887_001486_Sideline_frame345.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57887_002233_Endzone_frame271.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57887_002233_Sideline_frame79.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57887_002362_Endzone_frame0816.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57887_002629_Endzone_frame0934.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57887_002651_Endzone_frame0582.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57887_002685_Endzone_frame0915.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57887_003003_Endzone_frame357.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57887_003003_Sideline_frame442.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57887_003241_Endzone_frame1078.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57887_003760_Sideline_frame0885.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57888_000104_Endzone_frame0557.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57888_001500_Endzone_frame1077.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57888_002122_Endzone_frame362.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57888_002122_Sideline_frame128.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57888_002551_Endzone_frame0597.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57888_002842_Sideline_frame1100.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57888_003122_Endzone_frame0623.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57888_003214_Sideline_frame0833.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57888_004072_Endzone_frame61.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57888_004072_Sideline_frame216.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57888_004107_Endzone_frame303.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57888_004107_Sideline_frame329.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57888_004149_Endzone_frame25.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57888_004149_Sideline_frame318.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57888_004332_Sideline_frame0833.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57888_004517_Endzone_frame0794.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57889_000120_Endzone_frame0794.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57889_000598_Endzone_frame493.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57889_000598_Sideline_frame467.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57889_001236_Endzone_frame1252.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57889_001319_Sideline_frame0771.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57889_001517_Endzone_frame559.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57889_001517_Sideline_frame516.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57889_001746_Endzone_frame0762.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57889_002316_Sideline_frame1114.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57889_002551_Endzone_frame469.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57889_002551_Sideline_frame207.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57889_002711_Sideline_frame0597.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57889_003437_Sideline_frame0598.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57889_003554_Sideline_frame1239.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57890_000247_Endzone_frame0624.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57890_000330_Sideline_frame0685.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57890_000485_Endzone_frame83.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57890_000485_Sideline_frame93.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57890_000622_Endzone_frame0866.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57890_000680_Endzone_frame347.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57890_000680_Sideline_frame448.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57890_000897_Endzone_frame0698.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57890_001228_Sideline_frame0843.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57890_001252_Endzone_frame202.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57890_001252_Sideline_frame62.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57890_002082_Sideline_frame1140.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57890_002729_Sideline_frame1070.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57890_003130_Endzone_frame0595.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57890_003483_Endzone_frame258.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57890_003483_Sideline_frame266.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57891_000122_Sideline_frame0680.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57891_000318_Endzone_frame0967.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57891_001193_Sideline_frame1090.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57891_001355_Sideline_frame0558.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57891_001380_Sideline_frame0528.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57891_001762_Endzone_frame151.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57891_001762_Sideline_frame0.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57891_001800_Endzone_frame0770.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57891_002141_Sideline_frame0605.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57891_002358_Endzone_frame105.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57891_002358_Sideline_frame12.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57891_002457_Endzone_frame240.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57891_002457_Sideline_frame179.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57891_002549_Endzone_frame535.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57891_002549_Sideline_frame323.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57891_002980_Endzone_frame1189.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57891_003743_Endzone_frame0632.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57892_000051_Endzone_frame0428.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57892_000367_Sideline_frame0676.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57892_000951_Sideline_frame0728.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57892_001134_Endzone_frame0876.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57892_001216_Sideline_frame1227.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57892_001297_Endzone_frame382.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57892_001297_Sideline_frame66.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57892_001771_Endzone_frame1425.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57892_002031_Endzone_frame140.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57892_002031_Sideline_frame160.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57892_002499_Endzone_frame37.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57892_002499_Sideline_frame57.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57892_002520_Endzone_frame0674.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57892_002732_Sideline_frame0431.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57892_003092_Sideline_frame1241.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57892_003149_Endzone_frame308.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57892_003149_Sideline_frame578.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57893_000772_Sideline_frame0641.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57893_000812_Endzone_frame170.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57893_000812_Sideline_frame240.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57893_001136_Endzone_frame0635.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57893_001315_Endzone_frame82.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57893_001315_Sideline_frame172.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57893_001370_Endzone_frame0810.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57893_001479_Endzone_frame0775.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57893_002454_Endzone_frame144.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57893_002454_Sideline_frame106.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57893_002541_Endzone_frame330.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57893_002541_Sideline_frame241.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57893_002644_Sideline_frame0801.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57893_002789_Endzone_frame1012.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57893_002872_Sideline_frame1400.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57893_003926_Sideline_frame0754.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57893_004224_Sideline_frame0289.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57894_000909_Endzone_frame0838.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57894_001500_Sideline_frame0662.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57894_001645_Sideline_frame0627.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57894_001711_Endzone_frame0516.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57894_001733_Endzone_frame0488.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57894_002326_Sideline_frame0868.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57894_002844_Endzone_frame129.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57894_002844_Sideline_frame1162.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57894_002844_Sideline_frame64.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57894_003101_Endzone_frame438.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57894_003101_Sideline_frame292.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57894_003267_Endzone_frame1065.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57894_003333_Endzone_frame75.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57894_003333_Sideline_frame25.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57894_003526_Endzone_frame183.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57894_003526_Sideline_frame114.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57894_003937_Endzone_frame0955.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57895_000469_Endzone_frame0714.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57895_000807_Endzone_frame386.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57895_000807_Sideline_frame34.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57895_000949_Endzone_frame1019.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57895_000973_Sideline_frame0490.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57895_001001_Endzone_frame50.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57895_001001_Sideline_frame348.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57895_001504_Endzone_frame0566.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57895_001916_Endzone_frame0762.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57895_001940_Endzone_frame0935.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57895_002206_Endzone_frame179.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57895_002206_Sideline_frame45.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57895_002913_Sideline_frame1099.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57895_003094_Endzone_frame49.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57895_003094_Sideline_frame402.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57895_003450_Sideline_frame0856.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57896_000036_Endzone_frame0976.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57896_000079_Endzone_frame325.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57896_000079_Sideline_frame410.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57896_001089_Endzone_frame243.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57896_001089_Sideline_frame337.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57896_001311_Endzone_frame1024.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57896_001851_Endzone_frame420.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57896_001851_Sideline_frame384.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57896_001873_Endzone_frame0869.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57896_002102_Sideline_frame1607.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57896_002137_Sideline_frame0639.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57896_002291_Sideline_frame0589.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57896_002713_Endzone_frame29.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57896_002713_Sideline_frame24.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57896_002962_Sideline_frame0548.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57896_003350_Endzone_frame0860.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57896_003395_Sideline_frame1067.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57897_000289_Endzone_frame0558.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57897_000551_Endzone_frame0879.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57897_000926_Sideline_frame1113.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57897_001114_Endzone_frame0375.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57897_001214_Endzone_frame159.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57897_001214_Sideline_frame157.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57897_001515_Endzone_frame0923.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57897_002345_Endzone_frame488.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57897_002345_Sideline_frame238.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57897_002805_Endzone_frame31.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57897_002805_Sideline_frame344.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57897_003047_Endzone_frame86.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57897_003047_Sideline_frame489.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57897_003251_Sideline_frame0694.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57897_003424_Sideline_frame1016.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57897_003819_Endzone_frame0591.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57897_003882_Sideline_frame0603.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57898_000128_Endzone_frame5.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57898_000128_Sideline_frame193.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57898_000238_Sideline_frame0697.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57898_000653_Sideline_frame0728.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57898_000956_Endzone_frame0401.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57898_001109_Sideline_frame1126.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57898_001295_Endzone_frame0586.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57898_001753_Sideline_frame0681.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57898_001952_Endzone_frame88.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57898_001952_Sideline_frame404.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57898_002084_Endzone_frame0705.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57898_002227_Endzone_frame188.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57898_002227_Sideline_frame397.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57898_003221_Sideline_frame0716.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57898_003578_Endzone_frame1059.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57899_000362_Endzone_frame80.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57899_000362_Sideline_frame245.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57899_001092_Endzone_frame113.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57899_001092_Sideline_frame307.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57899_001200_Endzone_frame1082.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57899_001645_Sideline_frame0422.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57899_001745_Sideline_frame0632.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57899_001929_Sideline_frame0602.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57899_002782_Endzone_frame0977.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57899_002827_Endzone_frame520.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57899_002827_Sideline_frame461.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57899_003189_Endzone_frame0908.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57899_003211_Endzone_frame36.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57899_003211_Sideline_frame152.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57899_003587_Sideline_frame0667.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57899_004390_Endzone_frame0823.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57900_000443_Endzone_frame618.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57900_000443_Sideline_frame115.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57900_003349_Endzone_frame512.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57900_003349_Sideline_frame396.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57900_004532_Endzone_frame255.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57900_004532_Sideline_frame137.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57904_000337_Endzone_frame302.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57904_000337_Sideline_frame345.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57904_003413_Endzone_frame231.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57904_003413_Sideline_frame214.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57904_003476_Endzone_frame137.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57904_003476_Sideline_frame355.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57904_003507_Endzone_frame229.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57904_003507_Sideline_frame92.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57905_000510_Endzone_frame431.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57905_000510_Sideline_frame79.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57905_000570_Endzone_frame279.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57905_000570_Sideline_frame536.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57905_000593_Endzone_frame233.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57905_000593_Sideline_frame376.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57905_003882_Endzone_frame79.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57905_003882_Sideline_frame95.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57906_001930_Endzone_frame40.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57906_001930_Sideline_frame106.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57906_002077_Endzone_frame465.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57906_002077_Sideline_frame154.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57906_003724_Endzone_frame266.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57906_003724_Sideline_frame155.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57907_001496_Endzone_frame189.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57907_001496_Sideline_frame58.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57907_002016_Endzone_frame25.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57907_002016_Sideline_frame79.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57907_002300_Endzone_frame167.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57907_002300_Sideline_frame389.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57907_003078_Endzone_frame389.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57907_003078_Sideline_frame31.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57908_001357_Endzone_frame361.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57908_001357_Sideline_frame68.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57908_001947_Endzone_frame285.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57908_001947_Sideline_frame154.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57908_002232_Endzone_frame255.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57908_002232_Sideline_frame288.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57908_002382_Endzone_frame171.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57908_002382_Sideline_frame334.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57909_000290_Endzone_frame209.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57909_000290_Sideline_frame106.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57909_002209_Endzone_frame54.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57909_002209_Sideline_frame529.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57909_002800_Endzone_frame315.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57909_002800_Sideline_frame240.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57910_000860_Endzone_frame178.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57910_000860_Sideline_frame95.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57910_002503_Endzone_frame447.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57910_002503_Sideline_frame236.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57910_003591_Endzone_frame34.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57910_003591_Sideline_frame326.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57910_004148_Endzone_frame250.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57910_004148_Sideline_frame40.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57913_000083_Endzone_frame91.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57913_000083_Sideline_frame193.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57913_001538_Endzone_frame188.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57913_001538_Sideline_frame33.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57913_002056_Endzone_frame39.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57913_002056_Sideline_frame131.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57913_003545_Endzone_frame458.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57913_003545_Sideline_frame223.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57914_000093_Endzone_frame561.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57914_000093_Sideline_frame340.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57914_002821_Endzone_frame246.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57914_002821_Sideline_frame148.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57914_003622_Endzone_frame312.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57914_003622_Sideline_frame115.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57914_003818_Endzone_frame25.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57914_003818_Sideline_frame700.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57915_002103_Endzone_frame491.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57915_002103_Sideline_frame497.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57915_002216_Endzone_frame95.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57915_002216_Sideline_frame312.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57915_002673_Endzone_frame87.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57915_002673_Sideline_frame303.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57915_002800_Endzone_frame257.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57915_002800_Sideline_frame565.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57916_000332_Sideline_frame0872.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57916_000551_Endzone_frame0751.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57916_000939_Sideline_frame0177.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57916_001173_Endzone_frame59.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57916_001173_Sideline_frame58.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57916_001419_Sideline_frame0806.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57916_001991_Sideline_frame0840.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57916_002148_Endzone_frame384.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57916_002148_Sideline_frame108.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57916_002526_Endzone_frame0834.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57916_002881_Sideline_frame0927.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57916_003073_Endzone_frame161.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57916_003073_Sideline_frame245.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57916_003341_Endzone_frame160.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57916_003341_Sideline_frame94.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57916_003686_Endzone_frame0731.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57916_003736_Endzone_frame0572.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57917_000072_Endzone_frame0742.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57917_000866_Sideline_frame0830.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57917_001126_Sideline_frame0647.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57917_001959_Endzone_frame215.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57917_001959_Sideline_frame324.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57917_002248_Sideline_frame1242.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57917_002971_Endzone_frame181.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57917_002971_Sideline_frame236.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57917_003114_Endzone_frame0267.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57917_003635_Sideline_frame0689.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57917_003687_Endzone_frame226.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57917_003687_Sideline_frame406.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57917_004037_Endzone_frame0358.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57917_004060_Endzone_frame0832.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57917_004107_Endzone_frame210.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57917_004107_Sideline_frame318.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57917_004183_Sideline_frame0788.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57918_000125_Sideline_frame1007.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57918_000279_Sideline_frame0691.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57918_000367_Endzone_frame183.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57918_000367_Sideline_frame357.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57918_000606_Sideline_frame1028.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57918_000868_Sideline_frame0982.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57918_001452_Endzone_frame40.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57918_001452_Sideline_frame8.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57918_001619_Endzone_frame0531.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57918_002035_Endzone_frame0919.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57918_002523_Endzone_frame1081.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57918_002752_Endzone_frame0983.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57918_003963_Endzone_frame295.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57918_003963_Sideline_frame39.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57918_004065_Endzone_frame0849.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57918_004086_Endzone_frame158.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57918_004086_Sideline_frame263.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57919_000082_Sideline_frame1075.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57919_000262_Sideline_frame1483.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57919_000437_Endzone_frame0904.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57919_000520_Sideline_frame0679.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57919_000678_Endzone_frame19.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57919_000678_Sideline_frame34.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57919_001279_Endzone_frame0698.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57919_002198_Endzone_frame204.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57919_002198_Sideline_frame68.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57919_002711_Endzone_frame289.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57919_002711_Sideline_frame41.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57919_002834_Endzone_frame0782.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57919_003317_Endzone_frame0816.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57919_003494_Endzone_frame0814.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57919_003741_Sideline_frame1397.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57919_003765_Endzone_frame34.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57919_003765_Sideline_frame210.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57920_000130_Endzone_frame1090.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57920_000236_Endzone_frame0835.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57920_000487_Endzone_frame1152.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57920_000696_Endzone_frame0614.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57920_001165_Endzone_frame0765.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57920_001491_Endzone_frame28.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57920_001491_Sideline_frame534.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57920_002334_Sideline_frame0884.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57920_002382_Endzone_frame12.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57920_002382_Sideline_frame80.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57920_002811_Endzone_frame431.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57920_002811_Sideline_frame269.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57920_003467_Sideline_frame0481.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57920_003584_Sideline_frame0645.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57920_003937_Endzone_frame203.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57920_003937_Sideline_frame224.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57921_000141_Sideline_frame0712.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57921_000351_Sideline_frame0994.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57921_000937_Sideline_frame1103.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57921_001304_Sideline_frame0449.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57921_001528_Endzone_frame478.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57921_001528_Sideline_frame125.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57921_001944_Endzone_frame18.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57921_001944_Sideline_frame224.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57921_002619_Endzone_frame0773.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57921_002745_Endzone_frame1606.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57921_003088_Sideline_frame1198.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57921_003142_Endzone_frame277.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57921_003142_Sideline_frame176.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57921_003299_Endzone_frame1034.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57921_003349_Endzone_frame0637.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57921_003707_Endzone_frame224.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57921_003707_Sideline_frame46.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57922_000115_Endzone_frame1238.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57922_000679_Endzone_frame0968.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57922_001174_Endzone_frame0603.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57922_001302_Endzone_frame66.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57922_001302_Sideline_frame59.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57922_001469_Sideline_frame1009.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57922_001500_Sideline_frame1201.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57922_002236_Endzone_frame1169.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57922_002944_Endzone_frame216.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57922_002944_Sideline_frame273.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57922_003191_Endzone_frame134.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57922_003191_Sideline_frame184.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57922_003229_Sideline_frame1085.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57922_003591_Sideline_frame1957.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57922_003779_Endzone_frame0977.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57922_003889_Endzone_frame195.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57922_003889_Sideline_frame19.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57923_000533_Sideline_frame1057.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57923_000729_Sideline_frame1303.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57923_001743_Endzone_frame265.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57923_001743_Sideline_frame258.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57923_002198_Endzone_frame0849.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57923_002258_Endzone_frame385.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57923_002258_Sideline_frame337.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57923_002350_Endzone_frame0870.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57923_002924_Endzone_frame257.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57923_002924_Sideline_frame470.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57923_002972_Sideline_frame0957.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57923_004088_Sideline_frame0630.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57923_004227_Endzone_frame603.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57923_004227_Sideline_frame46.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57923_004245_Endzone_frame0792.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57923_004269_Endzone_frame0550.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57923_004305_Sideline_frame1039.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57924_000051_Sideline_frame1131.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57924_000545_Endzone_frame0568.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57924_001105_Sideline_frame106.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57924_001397_Endzone_frame305.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57924_001397_Sideline_frame160.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57924_001420_Endzone_frame0959.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57924_001779_Sideline_frame0979.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57924_002127_Endzone_frame0546.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57924_002422_Endzone_frame208.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57924_002422_Sideline_frame12.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57924_002544_Sideline_frame0749.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57924_002703_Endzone_frame0640.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57924_003058_Endzone_frame0759.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57924_003089_Endzone_frame278.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57924_003089_Sideline_frame150.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57924_003875_Sideline_frame1206.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57925_000365_Sideline_frame0778.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57925_000918_Sideline_frame0855.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57925_001066_Endzone_frame0990.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57925_001537_Sideline_frame0524.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57925_001738_Endzone_frame60.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57925_001738_Sideline_frame139.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57925_001773_Endzone_frame0986.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57925_001896_Endzone_frame0580.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57925_002084_Endzone_frame388.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57925_002084_Sideline_frame353.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57925_002136_Endzone_frame18.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57925_002136_Sideline_frame218.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57925_002243_Endzone_frame406.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57925_002243_Sideline_frame478.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57925_002774_Sideline_frame0899.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57925_003350_Endzone_frame1135.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57925_003958_Endzone_frame0402.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57926_000079_Endzone_frame0929.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57926_001029_Endzone_frame71.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57926_001029_Sideline_frame415.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57926_001119_Endzone_frame160.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57926_001119_Sideline_frame106.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57926_001550_Sideline_frame0496.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57926_001630_Endzone_frame1431.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57926_001860_Endzone_frame109.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57926_001860_Sideline_frame177.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57926_002523_Sideline_frame0446.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57926_002717_Endzone_frame0666.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57926_003257_Endzone_frame0918.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57926_003366_Sideline_frame1851.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57926_003465_Endzone_frame0717.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57927_000354_Endzone_frame0721.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57927_000412_Sideline_frame0939.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57927_000714_Endzone_frame145.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57927_000714_Sideline_frame373.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57927_001476_Endzone_frame258.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57927_001476_Sideline_frame493.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57927_001662_Sideline_frame1331.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57927_002521_Endzone_frame1103.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57927_003124_Sideline_frame1273.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57927_003231_Endzone_frame270.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57927_003231_Sideline_frame544.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57927_003252_Sideline_frame1651.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57927_003573_Sideline_frame1692.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57927_003826_Endzone_frame1071.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57927_003826_Endzone_frame162.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57927_003826_Sideline_frame311.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57928_000243_Sideline_frame1281.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57928_000390_Endzone_frame264.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57928_000390_Sideline_frame381.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57928_001095_Sideline_frame1198.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57928_001630_Endzone_frame249.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57928_001630_Sideline_frame113.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57928_002004_Sideline_frame0820.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57928_003139_Endzone_frame254.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57928_003139_Sideline_frame64.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57928_003220_Sideline_frame0954.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57928_003575_Endzone_frame118.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57928_003575_Sideline_frame380.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57928_003838_Endzone_frame0594.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57928_004122_Endzone_frame0510.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57928_004142_Endzone_frame0592.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57928_004277_Sideline_frame1255.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57929_000909_Endzone_frame406.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57929_000909_Sideline_frame190.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57929_001639_Endzone_frame0533.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57929_001842_Sideline_frame0757.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57929_001930_Endzone_frame1135.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57929_001968_Endzone_frame37.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57929_001968_Sideline_frame228.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57929_002399_Endzone_frame0621.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57929_003104_Endzone_frame0939.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57929_003313_Endzone_frame106.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57929_003313_Sideline_frame103.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57929_003365_Sideline_frame1051.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57929_003479_Sideline_frame0967.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57929_004170_Endzone_frame0738.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57930_000147_Sideline_frame0523.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57930_000234_Endzone_frame0580.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57930_000388_Endzone_frame221.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57930_000388_Sideline_frame225.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57930_000444_Sideline_frame0832.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57930_001294_Endzone_frame0581.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57930_002017_Sideline_frame0422.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57930_002039_Endzone_frame0823.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57930_002065_Sideline_frame1043.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57930_004053_Endzone_frame0522.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57930_004116_Endzone_frame0576.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57930_004167_Endzone_frame116.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57930_004167_Sideline_frame111.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57930_004418_Endzone_frame165.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57930_004418_Sideline_frame37.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57931_000180_Endzone_frame1299.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57931_000201_Sideline_frame1408.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57931_000347_Endzone_frame1067.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57931_001664_Sideline_frame0930.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57931_001803_Endzone_frame0717.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57931_002010_Endzone_frame127.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57931_002010_Sideline_frame217.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57931_002651_Endzone_frame1295.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57931_002651_Endzone_frame545.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57931_002651_Sideline_frame629.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57931_002956_Sideline_frame0674.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57931_003255_Endzone_frame652.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57931_003255_Sideline_frame160.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57931_003542_Endzone_frame0763.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57931_004024_Endzone_frame159.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57931_004024_Sideline_frame398.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57931_004048_Sideline_frame0534.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57932_000229_Endzone_frame0547.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57932_000229_Endzone_frame89.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57932_000229_Sideline_frame307.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57932_000726_Sideline_frame0770.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57932_001165_Endzone_frame376.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57932_001165_Sideline_frame254.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57932_002365_Sideline_frame0836.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57932_002798_Sideline_frame1098.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57932_002871_Endzone_frame1150.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57932_003054_Endzone_frame0857.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57932_003123_Endzone_frame0872.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57932_003668_Sideline_frame1315.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57932_003752_Endzone_frame252.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57932_003752_Sideline_frame82.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57933_000572_Sideline_frame0776.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57933_000960_Endzone_frame1555.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57933_001040_Endzone_frame0941.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57933_001129_Endzone_frame0548.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57933_001858_Sideline_frame0897.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57933_002494_Sideline_frame1114.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57933_003162_Endzone_frame0753.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57933_003678_Sideline_frame1066.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57933_004055_Endzone_frame118.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57933_004055_Sideline_frame127.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57933_004097_Endzone_frame246.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57933_004097_Sideline_frame302.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57933_004594_Endzone_frame76.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57933_004594_Sideline_frame53.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57933_004653_Sideline_frame0733.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57934_000222_Sideline_frame0792.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57934_000322_Sideline_frame0982.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57934_000365_Endzone_frame205.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57934_000365_Sideline_frame425.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57934_000387_Endzone_frame1011.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57934_000406_Endzone_frame0794.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57934_000838_Endzone_frame0979.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57934_001488_Sideline_frame0753.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57934_002386_Sideline_frame1040.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57934_002877_Endzone_frame1882.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57934_003035_Endzone_frame44.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57934_003035_Sideline_frame87.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57934_004175_Endzone_frame335.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57934_004175_Sideline_frame1057.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57934_004175_Sideline_frame164.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57935_000324_Sideline_frame0440.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57935_001282_Sideline_frame0666.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57935_001306_Endzone_frame260.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57935_001306_Sideline_frame113.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57935_001431_Endzone_frame0445.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57935_001879_Endzone_frame1451.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57935_002250_Endzone_frame0885.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57935_002271_Sideline_frame0633.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57935_003073_Sideline_frame0518.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57935_003316_Sideline_frame0906.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57935_003534_Endzone_frame0845.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57936_001196_Endzone_frame0724.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57936_001196_Endzone_frame271.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57936_001196_Sideline_frame27.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57936_001237_Endzone_frame176.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57936_001237_Sideline_frame54.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57936_001622_Endzone_frame0951.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57936_001645_Endzone_frame1324.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57936_001960_Endzone_frame243.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57936_001960_Sideline_frame283.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57936_002097_Endzone_frame90.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57936_002097_Sideline_frame73.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57936_002159_Endzone_frame0668.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57936_002253_Sideline_frame0559.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57936_002298_Sideline_frame0824.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57936_002908_Endzone_frame1320.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57936_003367_Sideline_frame0595.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57936_003525_Sideline_frame0717.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57937_000107_Sideline_frame0801.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57937_000969_Endzone_frame0428.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57937_001145_Sideline_frame0711.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57937_002196_Sideline_frame0939.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57937_002362_Endzone_frame0522.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57937_002383_Endzone_frame372.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57937_002383_Sideline_frame147.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57937_003316_Sideline_frame0663.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57937_003543_Endzone_frame0816.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57937_003779_Endzone_frame79.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57937_003779_Sideline_frame52.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57937_003888_Endzone_frame0604.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57937_003909_Sideline_frame0752.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57937_004038_Endzone_frame87.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57937_004038_Sideline_frame343.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57938_000036_Endzone_frame0973.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57938_000284_Endzone_frame97.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57938_000284_Sideline_frame403.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57938_000556_Endzone_frame216.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57938_000556_Sideline_frame530.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57938_000654_Endzone_frame210.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57938_000654_Sideline_frame249.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57938_000687_Sideline_frame1360.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57938_001374_Endzone_frame0688.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57938_001562_Sideline_frame0531.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57938_001676_Sideline_frame0830.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57938_001971_Endzone_frame0634.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57938_002090_Sideline_frame0548.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57938_003103_Endzone_frame0716.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57938_003234_Sideline_frame1197.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57939_000459_Endzone_frame164.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57939_000459_Sideline_frame1017.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57939_000459_Sideline_frame197.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57939_000629_Sideline_frame0665.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57939_000747_Sideline_frame1474.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57939_000814_Endzone_frame1357.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57939_000951_Endzone_frame81.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57939_000951_Sideline_frame70.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57939_001387_Endzone_frame1210.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57939_001683_Endzone_frame1485.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57939_002203_Sideline_frame0669.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57939_003179_Endzone_frame303.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57939_003179_Sideline_frame87.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57939_003469_Sideline_frame1439.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57939_003699_Endzone_frame0674.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57939_003815_Endzone_frame217.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57939_003815_Sideline_frame31.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57940_000127_Endzone_frame347.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57940_000127_Sideline_frame474.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57940_000501_Sideline_frame0770.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57940_000620_Endzone_frame0722.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57940_000644_Endzone_frame88.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57940_000644_Sideline_frame348.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57940_001316_Endzone_frame0818.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57940_001356_Sideline_frame1385.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57940_001698_Endzone_frame0817.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57940_002060_Endzone_frame0677.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57940_002513_Sideline_frame1389.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57940_002775_Endzone_frame1066.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57940_003225_Endzone_frame196.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57940_003225_Sideline_frame215.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57940_003487_Sideline_frame0790.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57940_003719_Endzone_frame118.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57940_003719_Sideline_frame144.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57941_000497_Endzone_frame0833.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57941_000574_Endzone_frame0708.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57941_000711_Sideline_frame1004.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57941_000988_Endzone_frame158.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57941_000988_Sideline_frame335.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57941_001034_Endzone_frame1043.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57941_001229_Sideline_frame0481.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57941_001409_Sideline_frame0946.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57941_002499_Endzone_frame145.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57941_002499_Sideline_frame33.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57941_002981_Sideline_frame1244.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57941_004129_Endzone_frame0488.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57941_004169_Endzone_frame1067.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57941_004297_Endzone_frame210.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57941_004297_Sideline_frame157.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57942_000126_Endzone_frame0946.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57942_000281_Endzone_frame186.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57942_000281_Sideline_frame224.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57942_000626_Endzone_frame53.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57942_000626_Sideline_frame249.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57942_000905_Endzone_frame0994.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57942_000957_Sideline_frame0795.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57942_001110_Endzone_frame0826.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57942_001131_Endzone_frame1497.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57942_001434_Sideline_frame0767.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57942_001768_Endzone_frame277.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57942_001768_Sideline_frame377.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57942_002555_Endzone_frame0779.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57942_002807_Sideline_frame0861.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57942_004235_Endzone_frame63.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57942_004235_Sideline_frame201.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57942_004271_Sideline_frame0801.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57943_001328_Sideline_frame0845.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57943_001424_Endzone_frame80.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57943_001424_Sideline_frame288.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57943_001531_Endzone_frame0982.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57943_001672_Sideline_frame1289.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57943_001982_Endzone_frame59.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57943_001982_Sideline_frame348.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57943_002142_Endzone_frame0871.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57943_002257_Endzone_frame0917.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57943_002648_Endzone_frame510.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57943_002648_Sideline_frame1610.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57943_002648_Sideline_frame532.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57943_002823_Endzone_frame0701.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57943_003025_Sideline_frame1000.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57943_003086_Endzone_frame0.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57943_003086_Sideline_frame458.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57943_003346_Endzone_frame1276.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57944_000720_Endzone_frame0542.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57944_000890_Sideline_frame0633.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57944_001376_Endzone_frame311.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57944_001376_Sideline_frame66.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57944_001608_Endzone_frame0757.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57944_001833_Endzone_frame0866.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57944_002576_Sideline_frame1177.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57944_002598_Endzone_frame149.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57944_002598_Sideline_frame147.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57944_002804_Sideline_frame0667.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57944_003061_Sideline_frame0776.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57944_003395_Endzone_frame165.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57944_003395_Sideline_frame65.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57944_003757_Sideline_frame0564.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57944_004107_Endzone_frame1458.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57945_000051_Endzone_frame0529.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57945_000454_Endzone_frame1333.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57945_001542_Endzone_frame255.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57945_001542_Sideline_frame430.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57945_001840_Sideline_frame0933.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57945_002064_Sideline_frame0708.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57945_002277_Endzone_frame0680.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57945_002649_Sideline_frame1156.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57945_002776_Sideline_frame0776.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57945_003594_Endzone_frame0800.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57945_003776_Endzone_frame251.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57945_003776_Sideline_frame314.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57945_003860_Endzone_frame240.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57945_003860_Sideline_frame34.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57945_004002_Endzone_frame0561.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57945_004171_Endzone_frame132.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57945_004171_Sideline_frame248.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57946_000380_Endzone_frame595.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57946_000380_Sideline_frame605.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57946_000565_Sideline_frame1362.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57946_000725_Endzone_frame314.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57946_000725_Sideline_frame283.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57946_001177_Sideline_frame0844.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57946_001643_Sideline_frame0769.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57946_002162_Sideline_frame1465.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57946_002417_Endzone_frame0841.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57946_002483_Endzone_frame4.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57946_002483_Sideline_frame492.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57946_002748_Endzone_frame1048.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57946_002990_Endzone_frame1576.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57946_003384_Endzone_frame0783.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57946_003510_Endzone_frame63.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57946_003510_Sideline_frame235.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57946_003625_Endzone_frame0445.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57947_000248_Sideline_frame0545.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57947_000540_Sideline_frame0627.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57947_000653_Endzone_frame0564.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57947_000844_Endzone_frame124.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57947_000844_Sideline_frame115.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57947_001262_Endzone_frame0796.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57947_001577_Endzone_frame266.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57947_001577_Sideline_frame106.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57947_002165_Endzone_frame181.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57947_002165_Sideline_frame83.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57947_002413_Endzone_frame1285.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57947_002932_Endzone_frame1084.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57947_002956_Sideline_frame0694.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57947_003270_Endzone_frame0464.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57947_003517_Sideline_frame0696.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57947_003626_Endzone_frame628.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57947_003626_Sideline_frame391.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57948_000191_Endzone_frame495.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57948_000191_Sideline_frame57.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57948_000564_Endzone_frame402.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57948_000564_Sideline_frame87.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57948_001079_Sideline_frame0995.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57948_001559_Endzone_frame334.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57948_001559_Sideline_frame90.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57948_001714_Sideline_frame0813.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57948_001838_Sideline_frame0866.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57948_002112_Endzone_frame0726.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57948_002205_Endzone_frame0633.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57948_002221_Endzone_frame0931.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57948_002582_Sideline_frame1017.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57948_003479_Endzone_frame1150.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57948_003549_Endzone_frame1176.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57948_003822_Endzone_frame248.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57948_003822_Sideline_frame261.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57949_000602_Endzone_frame0875.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57949_000845_Sideline_frame1453.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57949_001177_Endzone_frame75.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57949_001177_Sideline_frame303.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57949_001374_Sideline_frame0689.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57949_002043_Endzone_frame0941.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57949_002129_Sideline_frame0763.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57949_002195_Endzone_frame14.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57949_002195_Sideline_frame33.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57949_002698_Sideline_frame0755.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57949_002722_Endzone_frame448.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57949_002722_Sideline_frame584.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57949_003020_Sideline_frame0688.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57949_003042_Endzone_frame1014.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57949_003407_Endzone_frame279.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57949_003407_Sideline_frame184.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57949_003758_Endzone_frame1256.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57950_000278_Endzone_frame0647.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57950_000529_Sideline_frame0633.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57950_000610_Endzone_frame215.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57950_000610_Sideline_frame1066.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57950_000610_Sideline_frame54.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57950_001114_Endzone_frame0947.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57950_001540_Endzone_frame52.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57950_001540_Sideline_frame211.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57950_001844_Sideline_frame0942.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57950_001938_Sideline_frame1558.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57950_002316_Endzone_frame0932.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57950_003106_Sideline_frame1029.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57950_003253_Endzone_frame158.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57950_003253_Sideline_frame150.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57950_003289_Endzone_frame0702.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57950_003825_Endzone_frame22.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57950_003825_Sideline_frame79.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57951_000420_Endzone_frame0631.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57951_000705_Sideline_frame0984.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57951_000815_Endzone_frame154.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57951_000815_Sideline_frame176.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57951_000947_Endzone_frame289.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57951_000947_Sideline_frame242.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57951_001081_Endzone_frame172.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57951_001081_Sideline_frame478.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57951_002398_Endzone_frame0631.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57951_002633_Endzone_frame326.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57951_002633_Sideline_frame580.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57951_002714_Endzone_frame0536.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57951_003181_Sideline_frame1523.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57951_003221_Sideline_frame1209.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57951_004202_Endzone_frame0535.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57951_004385_Sideline_frame0827.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57952_000036_Endzone_frame0694.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57952_000036_Endzone_frame106.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57952_000036_Sideline_frame135.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57952_000060_Sideline_frame0653.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57952_000415_Sideline_frame0931.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57952_000966_Endzone_frame258.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57952_000966_Sideline_frame252.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57952_001915_Endzone_frame0540.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57952_002115_Endzone_frame234.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57952_002115_Sideline_frame83.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57952_003784_Sideline_frame0404.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57952_003806_Sideline_frame1168.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57952_003882_Endzone_frame0522.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57952_004612_Endzone_frame219.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57952_004612_Sideline_frame274.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57952_004700_Sideline_frame0684.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57953_000061_Endzone_frame0673.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57953_000106_Endzone_frame8.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57953_000106_Sideline_frame223.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57953_000541_Endzone_frame0914.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57953_000985_Sideline_frame1060.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57953_001004_Endzone_frame0830.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57953_001617_Sideline_frame0962.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57953_001906_Endzone_frame108.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57953_001906_Sideline_frame48.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57953_001963_Endzone_frame335.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57953_001963_Sideline_frame10.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57953_002457_Endzone_frame81.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57953_002457_Sideline_frame67.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57953_002990_Sideline_frame0743.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57953_003306_Sideline_frame0509.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57953_003571_Sideline_frame1219.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57953_003656_Endzone_frame0558.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57954_000360_Sideline_frame1019.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57954_000429_Endzone_frame608.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57954_000429_Sideline_frame460.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57954_000914_Endzone_frame25.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57954_000914_Sideline_frame73.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57954_001452_Endzone_frame0577.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57954_001564_Endzone_frame339.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57954_001564_Sideline_frame157.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57954_001629_Sideline_frame1140.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57954_002289_Sideline_frame0904.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57954_002914_Sideline_frame0719.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57954_003182_Endzone_frame0878.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57954_003471_Sideline_frame0507.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57954_003871_Endzone_frame0869.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57954_003986_Endzone_frame1092.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57954_004251_Endzone_frame115.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57954_004251_Sideline_frame359.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57955_000985_Endzone_frame0394.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57955_001132_Endzone_frame0781.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57955_001332_Endzone_frame1119.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57955_001808_Endzone_frame302.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57955_001808_Sideline_frame1434.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57955_001808_Sideline_frame8.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57955_002613_Sideline_frame1308.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57955_002762_Endzone_frame390.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57955_002762_Sideline_frame417.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57955_003020_Endzone_frame80.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57955_003020_Sideline_frame243.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57955_003107_Sideline_frame0665.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57955_003361_Sideline_frame0700.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57955_003490_Endzone_frame0588.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57955_003723_Endzone_frame191.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57955_003723_Sideline_frame0818.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57955_003723_Sideline_frame87.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57956_000536_Sideline_frame0424.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57956_000661_Endzone_frame232.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57956_000661_Sideline_frame111.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57956_000770_Sideline_frame1018.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57956_001367_Endzone_frame0763.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57956_001388_Endzone_frame272.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57956_001388_Sideline_frame34.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57956_001439_Endzone_frame494.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57956_001439_Sideline_frame103.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57956_001626_Endzone_frame0602.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57956_002238_Endzone_frame0877.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57956_002262_Endzone_frame26.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57956_002262_Sideline_frame384.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57956_002597_Sideline_frame0698.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57956_003632_Endzone_frame1182.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57956_003722_Endzone_frame1268.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57956_003779_Sideline_frame0771.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57957_000447_Sideline_frame0922.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57957_000514_Sideline_frame0821.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57957_000810_Sideline_frame0563.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57957_001152_Sideline_frame0905.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57957_001471_Sideline_frame0943.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57957_001721_Endzone_frame148.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57957_001721_Sideline_frame537.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57957_001746_Endzone_frame1407.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57957_002176_Endzone_frame0548.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57957_003347_Endzone_frame255.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57957_003347_Sideline_frame101.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57957_003756_Endzone_frame1047.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57957_003789_Endzone_frame0882.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57957_003880_Endzone_frame320.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57957_003880_Sideline_frame84.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57958_000893_Endzone_frame0910.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57958_000997_Endzone_frame279.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57958_000997_Sideline_frame443.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57958_001727_Endzone_frame0882.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57958_001814_Sideline_frame1287.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57958_002454_Endzone_frame74.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57958_002454_Sideline_frame569.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57958_002856_Sideline_frame1087.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57958_002900_Endzone_frame0918.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57958_003069_Sideline_frame0757.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57958_003948_Endzone_frame289.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57958_003948_Sideline_frame45.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57958_004213_Sideline_frame1273.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57958_004446_Endzone_frame120.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57958_004446_Sideline_frame333.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57958_004547_Sideline_frame1134.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57958_004772_Endzone_frame0866.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57959_000655_Endzone_frame0683.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57959_000996_Endzone_frame186.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57959_000996_Sideline_frame96.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57959_001529_Sideline_frame0647.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57959_001562_Endzone_frame0785.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57959_001586_Endzone_frame0827.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57959_002323_Endzone_frame1774.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57959_002979_Endzone_frame163.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57959_002979_Sideline_frame208.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57959_003369_Sideline_frame1189.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57959_003529_Endzone_frame347.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57959_003529_Sideline_frame305.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57959_004063_Sideline_frame0483.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57959_004139_Sideline_frame1588.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57959_004244_Endzone_frame1144.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57960_000428_Sideline_frame0688.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57960_001459_Endzone_frame0469.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57960_001484_Endzone_frame0612.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57960_001530_Sideline_frame1036.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57960_001907_Endzone_frame329.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57960_001907_Sideline_frame86.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57960_002282_Endzone_frame255.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57960_002282_Sideline_frame532.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57960_002937_Sideline_frame2006.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57960_003016_Endzone_frame293.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57960_003016_Sideline_frame328.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57960_003195_Endzone_frame84.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57960_003195_Sideline_frame371.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57960_003380_Endzone_frame0638.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57960_003834_Endzone_frame0506.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57960_004007_Sideline_frame0724.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57960_004109_Sideline_frame0948.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57961_000657_Endzone_frame1300.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57961_001107_Sideline_frame1108.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57961_001225_Sideline_frame0756.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57961_001328_Sideline_frame1035.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57961_001352_Endzone_frame553.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57961_001352_Sideline_frame137.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57961_001593_Endzone_frame357.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57961_001593_Sideline_frame367.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57961_001785_Sideline_frame0939.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57961_001871_Endzone_frame363.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57961_001871_Sideline_frame376.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57961_001997_Endzone_frame0971.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57961_002323_Endzone_frame0711.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57961_002400_Endzone_frame125.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57961_002400_Sideline_frame53.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57961_002841_Endzone_frame0741.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57961_004100_Endzone_frame0232.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57962_000212_Endzone_frame1049.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57962_000426_Sideline_frame0473.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57962_000472_Sideline_frame0972.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57962_000637_Endzone_frame0716.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57962_000681_Endzone_frame353.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57962_000681_Sideline_frame92.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57962_000698_Endzone_frame0918.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57962_000722_Endzone_frame0506.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57962_000873_Sideline_frame1267.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57962_001156_Endzone_frame255.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57962_001156_Sideline_frame499.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57962_001644_Sideline_frame1023.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57962_001694_Sideline_frame1155.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57962_001844_Endzone_frame215.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57962_001844_Sideline_frame6.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57962_002354_Endzone_frame224.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57962_002354_Sideline_frame89.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57963_000256_Sideline_frame1188.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57963_000651_Endzone_frame130.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57963_000651_Sideline_frame12.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57963_001260_Sideline_frame1365.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57963_002522_Endzone_frame1088.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57963_002589_Endzone_frame0824.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57963_003133_Endzone_frame155.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57963_003133_Sideline_frame240.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57963_003170_Sideline_frame0791.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57963_004317_Endzone_frame0625.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57963_004441_Endzone_frame221.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57963_004441_Sideline_frame291.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57963_004466_Endzone_frame0545.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57963_004573_Endzone_frame0447.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57963_004671_Sideline_frame0741.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57964_000084_Endzone_frame293.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57964_000084_Sideline_frame60.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57964_000330_Endzone_frame0633.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57964_000497_Endzone_frame199.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57964_000497_Sideline_frame159.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57964_000831_Sideline_frame1628.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57964_000993_Endzone_frame0882.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57964_001225_Endzone_frame0866.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57964_001293_Endzone_frame0969.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57964_001517_Endzone_frame89.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57964_001517_Sideline_frame246.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57964_001655_Endzone_frame0818.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57964_003164_Sideline_frame1007.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57964_003465_Endzone_frame381.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57964_003465_Sideline_frame308.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57964_003648_Sideline_frame0398.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57964_004567_Sideline_frame1135.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57965_000060_Sideline_frame0955.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57965_000172_Endzone_frame83.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57965_000172_Sideline_frame57.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57965_000564_Sideline_frame0636.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57965_001111_Endzone_frame0498.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57965_001749_Endzone_frame0865.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57965_001836_Endzone_frame314.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57965_001836_Sideline_frame453.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57965_001924_Endzone_frame0565.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57965_002511_Sideline_frame1050.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57965_002778_Endzone_frame1281.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57965_003194_Endzone_frame0877.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57965_003344_Sideline_frame0702.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57965_003826_Endzone_frame199.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57965_003826_Sideline_frame63.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57965_004028_Endzone_frame286.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57965_004028_Sideline_frame194.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57966_000061_Sideline_frame0819.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57966_000425_Endzone_frame0430.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57966_000441_Endzone_frame60.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57966_000441_Sideline_frame54.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57966_000543_Sideline_frame0818.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57966_000674_Endzone_frame0495.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57966_002275_Sideline_frame1225.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57966_002817_Endzone_frame1079.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57966_002877_Endzone_frame0602.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57966_002949_Endzone_frame18.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57966_002949_Sideline_frame346.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57966_003006_Sideline_frame0828.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57966_003460_Endzone_frame270.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57966_003460_Sideline_frame210.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57966_003609_Endzone_frame0700.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57966_004268_Endzone_frame25.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57966_004268_Sideline_frame293.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57967_000365_Sideline_frame0899.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57967_000479_Endzone_frame1330.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57967_000684_Endzone_frame131.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57967_000684_Sideline_frame104.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57967_000722_Endzone_frame0568.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57967_001277_Endzone_frame0695.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57967_001433_Endzone_frame374.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57967_001433_Sideline_frame243.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57967_002407_Sideline_frame0823.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57967_002521_Sideline_frame0775.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57967_002717_Endzone_frame254.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57967_002717_Sideline_frame297.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57967_002812_Endzone_frame455.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57967_002812_Sideline_frame494.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57967_002875_Endzone_frame0820.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57967_003630_Sideline_frame1055.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57967_003740_Endzone_frame0941.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57968_000295_Sideline_frame1241.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57968_000733_Endzone_frame0644.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57968_000929_Endzone_frame0614.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57968_001303_Sideline_frame1101.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57968_001428_Endzone_frame0571.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57968_001449_Sideline_frame0674.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57968_001980_Endzone_frame4.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57968_001980_Sideline_frame51.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57968_002054_Endzone_frame0683.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57968_002349_Sideline_frame0797.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57968_002505_Endzone_frame170.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57968_002505_Sideline_frame338.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57968_002637_Endzone_frame272.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57968_002637_Sideline_frame262.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57968_003001_Endzone_frame243.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57968_003001_Sideline_frame176.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57968_004168_Sideline_frame0570.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57969_000044_Endzone_frame0699.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57969_000530_Sideline_frame1791.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57969_000592_Endzone_frame222.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57969_000592_Sideline_frame103.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57969_001089_Endzone_frame191.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57969_001089_Sideline_frame698.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57969_002166_Sideline_frame1096.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57969_002254_Endzone_frame348.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57969_002254_Sideline_frame108.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57969_002539_Sideline_frame1580.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57969_002554_Endzone_frame0935.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57969_002682_Endzone_frame0960.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57969_002704_Sideline_frame2065.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57969_002976_Endzone_frame0624.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57969_003918_Endzone_frame238.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57969_003918_Sideline_frame155.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57969_004169_Endzone_frame1402.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57970_000320_Endzone_frame0815.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57970_000401_Sideline_frame0969.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57970_000935_Sideline_frame1240.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57970_001355_Endzone_frame0996.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57970_001458_Endzone_frame462.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57970_001458_Sideline_frame194.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57970_001546_Endzone_frame87.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57970_001546_Sideline_frame160.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57970_001620_Sideline_frame0997.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57970_002402_Endzone_frame106.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57970_002402_Sideline_frame21.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57970_002912_Endzone_frame0747.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57970_003070_Endzone_frame1013.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57970_003099_Sideline_frame1299.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57971_000496_Endzone_frame0880.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57971_000589_Endzone_frame223.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57971_000589_Sideline_frame526.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57971_001800_Endzone_frame0531.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57971_002014_Endzone_frame1096.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57971_002130_Sideline_frame0736.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57971_002690_Sideline_frame1285.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57971_003214_Endzone_frame0713.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57971_003338_Endzone_frame173.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57971_003338_Sideline_frame281.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57971_003764_Sideline_frame0599.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57971_003911_Sideline_frame0639.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57971_004196_Endzone_frame4.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57971_004196_Sideline_frame1028.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57971_004196_Sideline_frame47.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57972_000165_Endzone_frame0950.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57972_000189_Endzone_frame610.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57972_000189_Sideline_frame477.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57972_000434_Sideline_frame0651.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57972_000508_Endzone_frame0732.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57972_000870_Sideline_frame0616.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57972_001431_Endzone_frame179.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57972_001431_Sideline_frame194.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57972_001455_Endzone_frame0834.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57972_001674_Sideline_frame0552.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57972_001771_Sideline_frame0634.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57972_001899_Endzone_frame1206.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57972_002009_Endzone_frame106.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57972_002009_Sideline_frame63.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57972_003224_Endzone_frame336.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57972_003224_Sideline_frame223.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57972_003384_Sideline_frame0926.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57973_000240_Sideline_frame0438.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57973_000269_Endzone_frame0912.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57973_000269_Endzone_frame368.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57973_000269_Sideline_frame75.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57973_000709_Endzone_frame0595.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57973_000855_Sideline_frame0472.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57973_000935_Endzone_frame409.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57973_000935_Sideline_frame306.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57973_001618_Endzone_frame1249.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57973_002478_Endzone_frame277.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57973_002478_Sideline_frame182.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57973_003030_Endzone_frame15.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57973_003030_Sideline_frame50.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57973_003894_Sideline_frame1159.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57973_004168_Sideline_frame0881.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57973_004313_Endzone_frame1051.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57974_000188_Sideline_frame0820.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57974_000495_Sideline_frame0980.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57974_000749_Endzone_frame0763.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57974_001096_Endzone_frame220.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57974_001096_Sideline_frame53.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57974_001492_Endzone_frame0839.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57974_002057_Endzone_frame85.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57974_002057_Sideline_frame35.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57974_002280_Endzone_frame1161.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57974_002571_Endzone_frame72.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57974_002571_Sideline_frame500.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57974_002705_Sideline_frame1063.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57974_002798_Endzone_frame0462.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57974_002819_Endzone_frame383.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57974_002819_Sideline_frame39.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57974_003672_Sideline_frame1191.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57974_003733_Endzone_frame0720.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57975_000264_Sideline_frame0962.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57975_000335_Endzone_frame1334.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57975_000983_Endzone_frame1106.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57975_001629_Sideline_frame1149.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57975_001650_Endzone_frame156.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57975_001650_Sideline_frame373.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57975_002068_Endzone_frame359.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57975_002068_Sideline_frame394.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57975_002425_Sideline_frame1269.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57975_002795_Endzone_frame0737.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57975_003021_Endzone_frame1106.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57975_003198_Sideline_frame0804.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57975_004085_Endzone_frame94.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57975_004085_Sideline_frame497.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57975_004576_Endzone_frame1179.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57976_000596_Sideline_frame1082.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57976_000661_Sideline_frame1407.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57976_000768_Endzone_frame0798.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57976_000847_Endzone_frame1468.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57976_001389_Endzone_frame1212.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57976_001787_Sideline_frame1053.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57976_001877_Endzone_frame63.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57976_001877_Sideline_frame285.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57976_002449_Sideline_frame0866.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57976_002468_Endzone_frame491.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57976_002468_Sideline_frame278.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57976_003113_Endzone_frame0595.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57976_003269_Endzone_frame101.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57976_003269_Sideline_frame360.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57976_003929_Endzone_frame207.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57976_003929_Sideline_frame428.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57976_004454_Endzone_frame0743.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57977_000427_Endzone_frame147.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57977_000427_Sideline_frame35.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57977_000470_Sideline_frame0721.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57977_000489_Endzone_frame1287.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57977_001204_Endzone_frame0744.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57977_001736_Endzone_frame53.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57977_001736_Sideline_frame33.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57977_002336_Endzone_frame1183.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57977_002491_Endzone_frame0677.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57977_002820_Endzone_frame340.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57977_002820_Sideline_frame260.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57977_003322_Endzone_frame0767.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57977_003348_Sideline_frame1420.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57977_003863_Sideline_frame0620.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57977_003982_Endzone_frame157.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57977_003982_Sideline_frame210.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57978_000079_Endzone_frame115.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57978_000079_Sideline_frame85.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57978_000224_Sideline_frame0985.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57978_000332_Endzone_frame0895.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57978_000455_Endzone_frame2501.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57978_001565_Endzone_frame49.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57978_001565_Sideline_frame343.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57978_001586_Sideline_frame1010.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57978_001962_Endzone_frame320.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57978_001962_Sideline_frame345.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57978_002139_Endzone_frame0928.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57978_002470_Sideline_frame1524.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57978_002976_Endzone_frame0780.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57978_003584_Endzone_frame503.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57978_003584_Sideline_frame480.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57978_004039_Sideline_frame0270.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57980_000090_Endzone_frame0639.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57980_000111_Endzone_frame327.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57980_000111_Sideline_frame187.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57980_000220_Endzone_frame0749.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57980_000953_Endzone_frame0854.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57980_001358_Endzone_frame253.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57980_001358_Sideline_frame191.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57980_001652_Endzone_frame279.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57980_001652_Sideline_frame9.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57980_001917_Sideline_frame0810.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57980_002245_Endzone_frame0690.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57980_002823_Sideline_frame1124.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57980_003761_Sideline_frame0742.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57980_003842_Endzone_frame0480.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57980_004128_Sideline_frame0528.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57980_004314_Endzone_frame99.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57980_004314_Sideline_frame130.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57981_000358_Sideline_frame1490.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57981_000647_Endzone_frame0634.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57981_000869_Sideline_frame0680.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57981_000991_Endzone_frame65.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57981_000991_Sideline_frame313.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57981_001146_Endzone_frame1257.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57981_001252_Sideline_frame0987.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57981_001421_Endzone_frame0711.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57981_001702_Endzone_frame511.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57981_001702_Sideline_frame318.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57981_001990_Endzone_frame233.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57981_001990_Sideline_frame187.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57981_002432_Endzone_frame0785.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57981_002590_Endzone_frame1105.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57981_003735_Sideline_frame2066.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57982_000134_Endzone_frame0541.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57982_000320_Endzone_frame23.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57982_000320_Sideline_frame274.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57982_000598_Endzone_frame422.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57982_000598_Sideline_frame145.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57982_000707_Endzone_frame324.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57982_000707_Sideline_frame322.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57982_001001_Endzone_frame0845.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57982_001507_Endzone_frame0545.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57982_001553_Endzone_frame0721.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57982_001914_Sideline_frame0736.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57982_002152_Endzone_frame217.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57982_002152_Sideline_frame359.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57982_002800_Sideline_frame1058.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57982_002905_Sideline_frame0773.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57982_003348_Endzone_frame0716.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57982_003394_Sideline_frame0552.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57983_000036_Sideline_frame1479.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57983_000210_Endzone_frame0745.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57983_000232_Endzone_frame0675.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57983_001603_Sideline_frame0787.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57983_001648_Endzone_frame165.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57983_001648_Sideline_frame299.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57983_001729_Endzone_frame301.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57983_001729_Sideline_frame559.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57983_003011_Sideline_frame1011.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57983_003209_Endzone_frame0545.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57983_003513_Endzone_frame0960.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57983_004001_Endzone_frame301.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57983_004001_Sideline_frame240.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57983_004209_Endzone_frame0616.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57983_004292_Sideline_frame1370.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57983_004395_Endzone_frame69.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57983_004395_Sideline_frame59.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57984_001874_Endzone_frame17.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57984_001874_Sideline_frame18.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57984_002044_Endzone_frame0790.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57984_002493_Sideline_frame0885.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57984_002679_Endzone_frame1317.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57984_002764_Sideline_frame0849.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57984_002884_Sideline_frame0936.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57984_003083_Sideline_frame0831.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57984_003402_Endzone_frame1714.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57984_003450_Endzone_frame0702.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57984_003515_Endzone_frame109.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57984_003515_Sideline_frame204.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57984_003663_Endzone_frame296.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57984_003663_Sideline_frame267.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57984_004151_Endzone_frame238.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57984_004151_Sideline_frame91.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57984_004223_Sideline_frame0995.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57985_000051_Endzone_frame49.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57985_000051_Sideline_frame70.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57985_000122_Sideline_frame0647.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57985_000670_Endzone_frame1034.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57985_000670_Endzone_frame22.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57985_000670_Sideline_frame568.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57985_000702_Endzone_frame0555.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57985_000789_Sideline_frame0821.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57985_002665_Sideline_frame1192.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57985_003175_Sideline_frame1329.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57985_003429_Endzone_frame266.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57985_003429_Sideline_frame478.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57985_003462_Sideline_frame0835.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57985_003505_Endzone_frame183.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57985_003505_Sideline_frame39.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57985_003692_Endzone_frame0782.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57985_003872_Endzone_frame0593.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57986_000635_Sideline_frame0976.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57986_000993_Endzone_frame209.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57986_000993_Sideline_frame62.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57986_002200_Sideline_frame0605.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57986_002432_Endzone_frame0563.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57986_002495_Endzone_frame0987.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57986_002628_Sideline_frame1165.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57986_002990_Sideline_frame1037.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57986_003630_Sideline_frame0820.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57986_003952_Endzone_frame0864.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57986_003974_Endzone_frame0681.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57986_004373_Endzone_frame190.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57986_004373_Sideline_frame340.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57986_004635_Endzone_frame73.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57986_004635_Sideline_frame93.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57987_000278_Sideline_frame0916.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57987_000342_Endzone_frame17.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57987_000342_Sideline_frame1227.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57987_000342_Sideline_frame42.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57987_000604_Endzone_frame293.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57987_000604_Sideline_frame161.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57987_000772_Endzone_frame330.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57987_000772_Sideline_frame385.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57987_001421_Endzone_frame0826.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57987_002423_Endzone_frame0778.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57987_002647_Endzone_frame0549.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57987_002741_Endzone_frame325.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57987_002741_Sideline_frame316.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57987_003113_Endzone_frame1027.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57987_003168_Endzone_frame0809.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57987_003442_Sideline_frame0789.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57987_003520_Sideline_frame0639.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57988_000253_Endzone_frame0894.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57988_000489_Endzone_frame0580.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57988_000736_Endzone_frame278.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57988_000736_Sideline_frame42.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57988_001226_Endzone_frame307.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57988_001226_Sideline_frame70.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57988_001347_Sideline_frame0944.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57988_001697_Sideline_frame1022.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57988_002938_Sideline_frame0655.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57988_003287_Endzone_frame127.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57988_003287_Sideline_frame36.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57988_003646_Endzone_frame0689.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57988_003757_Endzone_frame0899.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57988_004000_Sideline_frame2022.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57989_000036_Sideline_frame0952.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57989_000144_Endzone_frame1048.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57989_000498_Sideline_frame0873.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57989_000542_Endzone_frame273.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57989_000542_Sideline_frame245.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57989_000730_Sideline_frame0857.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57989_001264_Endzone_frame316.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57989_001264_Sideline_frame62.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57989_001370_Sideline_frame0987.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57989_001475_Sideline_frame0751.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57989_001655_Endzone_frame0735.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57989_001890_Endzone_frame218.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57989_001890_Sideline_frame133.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57989_001931_Endzone_frame1212.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57989_002629_Endzone_frame567.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57989_002629_Sideline_frame476.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57989_003138_Endzone_frame1336.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57990_000233_Endzone_frame70.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57990_000233_Sideline_frame353.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57990_000360_Endzone_frame152.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57990_000360_Sideline_frame234.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57990_000557_Endzone_frame330.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57990_000557_Sideline_frame28.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57990_000831_Sideline_frame1085.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57990_000850_Endzone_frame0892.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57990_001643_Endzone_frame235.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57990_001643_Sideline_frame257.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57990_002013_Endzone_frame1475.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57990_002492_Endzone_frame0798.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57990_002662_Sideline_frame1056.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57990_003016_Endzone_frame0751.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57990_003457_Sideline_frame1056.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57990_003598_Endzone_frame0548.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57990_003677_Sideline_frame1551.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57991_000221_Endzone_frame326.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57991_000221_Sideline_frame59.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57991_000830_Endzone_frame1067.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57991_001202_Sideline_frame0733.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57991_001661_Endzone_frame275.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57991_001661_Sideline_frame58.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57991_001897_Sideline_frame1437.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57991_002726_Endzone_frame0481.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57991_003101_Endzone_frame1652.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57991_003449_Endzone_frame1033.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57991_003523_Sideline_frame0640.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57991_003744_Endzone_frame100.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57991_003744_Sideline_frame339.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57991_003837_Endzone_frame159.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57991_003837_Sideline_frame84.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57991_004087_Endzone_frame0423.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/57991_004108_Sideline_frame0442.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57993_000059_Endzone_frame153.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57993_000059_Sideline_frame140.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57993_001396_Endzone_frame331.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57993_001396_Sideline_frame144.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57993_001592_Endzone_frame299.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57993_001592_Sideline_frame202.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57993_004103_Endzone_frame129.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57993_004103_Sideline_frame273.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57995_000534_Endzone_frame474.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57995_000534_Sideline_frame5.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57995_000772_Endzone_frame243.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57995_000772_Sideline_frame43.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57995_000815_Endzone_frame492.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57995_000815_Sideline_frame454.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57995_002799_Endzone_frame3.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57995_002799_Sideline_frame313.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57997_003050_Endzone_frame555.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57997_003050_Sideline_frame687.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57997_003154_Endzone_frame370.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57997_003154_Sideline_frame5.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57997_003733_Endzone_frame638.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57997_003733_Sideline_frame208.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57998_001611_Endzone_frame431.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57998_001611_Sideline_frame466.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57998_003205_Endzone_frame130.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57998_003205_Sideline_frame215.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57998_004427_Endzone_frame281.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/57998_004427_Sideline_frame9.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58000_000589_Endzone_frame402.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58000_000589_Sideline_frame121.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58000_001306_Endzone_frame325.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58000_001306_Sideline_frame83.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58000_002076_Endzone_frame210.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58000_002076_Sideline_frame346.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58000_003144_Endzone_frame8.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58000_003144_Sideline_frame174.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58002_000537_Endzone_frame100.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58002_000537_Sideline_frame424.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58002_000769_Endzone_frame95.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58002_000769_Sideline_frame323.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58002_000843_Endzone_frame142.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58002_000843_Sideline_frame244.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58002_001798_Endzone_frame87.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58004_000601_Endzone_frame273.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58004_000601_Sideline_frame171.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58004_001269_Endzone_frame23.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58004_001269_Sideline_frame470.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58004_001473_Endzone_frame284.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58004_001473_Sideline_frame382.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58004_003208_Endzone_frame266.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58004_003208_Sideline_frame88.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58005_002086_Endzone_frame182.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58005_002086_Sideline_frame46.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58005_003391_Endzone_frame82.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58005_003391_Sideline_frame134.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58005_004289_Endzone_frame144.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58005_004289_Sideline_frame265.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58006_000705_Endzone_frame0503.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58006_001332_Sideline_frame1262.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58006_001448_Endzone_frame301.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58006_001448_Sideline_frame33.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58006_001463_Endzone_frame0541.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58006_001657_Endzone_frame0584.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58006_001739_Endzone_frame518.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58006_001739_Sideline_frame333.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58006_001947_Endzone_frame0547.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58006_001947_Endzone_frame59.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58006_001947_Sideline_frame299.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58006_002352_Endzone_frame204.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58006_002352_Sideline_frame78.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58006_002512_Sideline_frame0706.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58006_002674_Sideline_frame0710.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58006_002770_Sideline_frame0724.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58006_002947_Endzone_frame0856.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58007_000282_Endzone_frame540.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58007_000282_Sideline_frame541.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58007_000425_Endzone_frame0856.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58007_000535_Endzone_frame0508.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58007_000640_Endzone_frame31.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58007_000640_Sideline_frame8.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58007_000690_Sideline_frame0726.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58007_001307_Endzone_frame0756.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58007_001307_Endzone_frame282.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58007_001307_Sideline_frame135.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58007_001463_Sideline_frame0558.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58007_001742_Sideline_frame0724.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58007_002136_Endzone_frame0664.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58007_003184_Sideline_frame1025.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58007_003619_Endzone_frame349.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58007_003619_Sideline_frame13.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58007_003967_Sideline_frame0859.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58008_000253_Endzone_frame0669.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58008_000300_Endzone_frame0494.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58008_000353_Sideline_frame0737.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58008_000393_Endzone_frame0823.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58008_000768_Endzone_frame195.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58008_000768_Sideline_frame290.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58008_000979_Endzone_frame24.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58008_000979_Sideline_frame3.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58008_001454_Sideline_frame0739.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58008_001524_Endzone_frame0543.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58008_002716_Sideline_frame0819.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58008_003193_Endzone_frame0628.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58008_003496_Endzone_frame84.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58008_003496_Sideline_frame142.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58008_003722_Sideline_frame0930.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58008_003967_Endzone_frame82.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58008_003967_Sideline_frame210.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58009_000125_Sideline_frame2063.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58009_000391_Endzone_frame355.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58009_000391_Sideline_frame662.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58009_000455_Sideline_frame0877.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58009_000604_Endzone_frame1047.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58009_000939_Sideline_frame0975.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58009_001105_Endzone_frame0952.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58009_001162_Endzone_frame0497.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58009_001288_Endzone_frame1038.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58009_001718_Endzone_frame1804.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58009_002341_Sideline_frame0683.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58009_002840_Endzone_frame96.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58009_002840_Sideline_frame39.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58009_003235_Endzone_frame117.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58009_003235_Sideline_frame346.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58010_000787_Endzone_frame440.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58010_000787_Sideline_frame647.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58010_001062_Endzone_frame0478.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58010_001062_Endzone_frame74.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58010_001062_Sideline_frame276.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58010_001134_Endzone_frame0451.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58010_001312_Sideline_frame0783.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58010_001578_Sideline_frame0840.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58010_001728_Sideline_frame0791.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58010_003625_Sideline_frame1011.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58010_003749_Endzone_frame133.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58010_003749_Sideline_frame278.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58010_003813_Endzone_frame0755.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58011_000815_Sideline_frame0831.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58011_001209_Sideline_frame0700.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58011_001351_Endzone_frame0771.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58011_001750_Endzone_frame293.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58011_001750_Sideline_frame148.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58011_001800_Endzone_frame296.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58011_001800_Sideline_frame262.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58011_002099_Sideline_frame1005.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58011_002234_Endzone_frame129.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58011_002234_Sideline_frame191.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58011_002716_Endzone_frame0753.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58011_002758_Endzone_frame0774.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58011_003321_Sideline_frame0635.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58011_003895_Sideline_frame0957.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58011_004064_Endzone_frame0664.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58012_000278_Endzone_frame0950.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58012_000369_Endzone_frame46.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58012_000369_Sideline_frame4.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58012_000511_Endzone_frame0426.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58012_000942_Endzone_frame119.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58012_000942_Sideline_frame393.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58012_002128_Sideline_frame0787.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58012_002337_Sideline_frame0783.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58012_002455_Endzone_frame0566.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58012_002477_Endzone_frame247.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58012_002477_Sideline_frame231.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58012_002652_Endzone_frame347.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58012_002652_Sideline_frame423.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58012_002982_Sideline_frame0948.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58012_003115_Sideline_frame0877.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58012_003555_Sideline_frame0628.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58012_003957_Endzone_frame1020.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58013_000568_Endzone_frame23.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58013_000568_Sideline_frame118.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58013_001916_Sideline_frame1096.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58013_001997_Endzone_frame0529.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58013_002089_Endzone_frame241.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58013_002089_Sideline_frame1044.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58013_002089_Sideline_frame43.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58013_002149_Endzone_frame1248.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58013_002430_Sideline_frame0846.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58013_003144_Endzone_frame202.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58013_003144_Sideline_frame307.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58013_003278_Endzone_frame0675.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58013_003447_Sideline_frame0640.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58013_003589_Endzone_frame0718.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58013_004304_Sideline_frame1172.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58014_000051_Endzone_frame309.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58014_000051_Sideline_frame45.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58014_000197_Endzone_frame1200.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58014_000353_Sideline_frame1539.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58014_000405_Sideline_frame1071.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58014_000484_Endzone_frame0737.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58014_000563_Sideline_frame0612.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58014_000612_Endzone_frame1713.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58014_000828_Endzone_frame0653.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58014_001394_Endzone_frame0800.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58014_001703_Sideline_frame0944.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58014_002191_Endzone_frame221.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58014_002191_Sideline_frame67.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58014_002746_Endzone_frame94.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58014_002746_Sideline_frame159.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58014_004050_Endzone_frame169.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58014_004050_Sideline_frame205.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58015_000046_Sideline_frame0997.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58015_000070_Endzone_frame0483.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58015_000231_Endzone_frame305.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58015_000231_Sideline_frame330.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58015_000339_Sideline_frame0816.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58015_000358_Endzone_frame261.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58015_000358_Sideline_frame177.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58015_000847_Endzone_frame0993.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58015_001449_Endzone_frame0978.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58015_001672_Sideline_frame1076.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58015_002428_Endzone_frame0833.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58015_002773_Sideline_frame0685.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58015_002947_Endzone_frame280.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58015_002947_Sideline_frame291.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58015_003253_Endzone_frame1159.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58015_004451_Endzone_frame135.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58015_004451_Sideline_frame117.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58016_000055_Sideline_frame1038.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58016_000100_Endzone_frame1127.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58016_000483_Endzone_frame0789.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58016_000661_Sideline_frame0656.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58016_001673_Endzone_frame0876.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58016_001753_Endzone_frame0898.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58016_001774_Endzone_frame69.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58016_001774_Sideline_frame105.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58016_002130_Endzone_frame48.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58016_002130_Sideline_frame142.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58016_002385_Endzone_frame0722.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58016_002385_Endzone_frame339.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58016_002385_Sideline_frame151.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58016_002807_Endzone_frame265.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58016_002807_Sideline_frame164.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58016_003370_Sideline_frame0551.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58017_000446_Endzone_frame1185.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58017_000627_Sideline_frame1025.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58017_001124_Sideline_frame0963.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58017_001822_Sideline_frame1091.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58017_001863_Endzone_frame156.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58017_001863_Sideline_frame418.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58017_002742_Endzone_frame1062.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58017_002789_Endzone_frame29.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58017_002789_Sideline_frame309.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58017_002932_Sideline_frame0768.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58017_003032_Endzone_frame1374.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58017_003094_Endzone_frame0913.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58017_003141_Endzone_frame1033.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58017_003141_Endzone_frame393.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58017_003141_Sideline_frame53.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58018_000069_Sideline_frame1053.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58018_000472_Sideline_frame1813.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58018_001114_Endzone_frame67.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58018_001114_Sideline_frame138.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58018_001687_Endzone_frame455.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58018_001687_Sideline_frame239.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58018_001819_Endzone_frame182.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58018_001819_Sideline_frame625.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58018_001927_Endzone_frame0830.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58018_002343_Sideline_frame0820.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58018_002385_Sideline_frame0985.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58018_002494_Endzone_frame0932.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58018_002727_Endzone_frame0642.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58018_003169_Endzone_frame67.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58018_003169_Sideline_frame525.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58018_003967_Sideline_frame0899.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58018_004077_Endzone_frame1174.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58019_000150_Endzone_frame0604.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58019_000206_Endzone_frame99.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58019_000206_Sideline_frame521.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58019_000395_Endzone_frame308.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58019_000395_Sideline_frame297.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58019_000438_Endzone_frame0672.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58019_000702_Endzone_frame0517.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58019_001137_Sideline_frame0606.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58019_001955_Endzone_frame61.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58019_001955_Sideline_frame319.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58019_002508_Sideline_frame0861.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58019_002967_Endzone_frame1106.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58019_003252_Sideline_frame0496.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58019_003438_Sideline_frame1288.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58019_003868_Sideline_frame0995.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58020_000184_Endzone_frame29.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58020_000184_Sideline_frame295.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58020_000254_Sideline_frame0808.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58020_000368_Endzone_frame0775.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58020_000894_Sideline_frame0989.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58020_001097_Sideline_frame1264.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58020_001359_Endzone_frame464.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58020_001359_Sideline_frame476.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58020_001662_Sideline_frame0609.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58020_002398_Sideline_frame1236.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58020_003104_Endzone_frame87.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58020_003104_Sideline_frame229.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58020_003432_Endzone_frame0881.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58020_003893_Endzone_frame64.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58020_003893_Sideline_frame69.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58021_000036_Endzone_frame1979.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58021_000908_Sideline_frame0673.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58021_000929_Sideline_frame0602.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58021_001157_Endzone_frame0476.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58021_001518_Sideline_frame0722.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58021_002003_Endzone_frame415.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58021_002003_Sideline_frame173.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58021_002355_Endzone_frame2022.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58021_002611_Endzone_frame0871.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58021_002928_Sideline_frame2175.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58021_003171_Endzone_frame58.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58021_003171_Sideline_frame294.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58021_003192_Endzone_frame0899.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58022_000248_Endzone_frame1043.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58022_000440_Endzone_frame116.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58022_000440_Sideline_frame276.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58022_000706_Endzone_frame189.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58022_000706_Sideline_frame12.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58022_000978_Sideline_frame0566.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58022_001182_Sideline_frame0707.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58022_001622_Endzone_frame0751.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58022_001622_Endzone_frame83.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58022_001622_Sideline_frame25.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58022_001680_Endzone_frame297.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58022_001680_Sideline_frame306.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58022_002136_Sideline_frame0778.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58022_002313_Sideline_frame1086.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58022_002575_Endzone_frame0861.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58022_002803_Endzone_frame0674.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58022_002961_Endzone_frame0564.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58023_000157_Endzone_frame180.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58023_000157_Sideline_frame551.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58023_000280_Endzone_frame112.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58023_000280_Sideline_frame246.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58023_000301_Endzone_frame275.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58023_000301_Sideline_frame226.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58023_000669_Sideline_frame0594.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58023_000861_Endzone_frame0408.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58023_000880_Endzone_frame142.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58023_000880_Sideline_frame359.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58023_000982_Sideline_frame1194.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58023_001738_Sideline_frame1007.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58023_001834_Endzone_frame0598.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58023_001925_Sideline_frame0931.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58023_002768_Sideline_frame0809.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58023_002952_Endzone_frame0937.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58023_003421_Endzone_frame0499.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58024_000144_Endzone_frame0968.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58024_000951_Endzone_frame192.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58024_000951_Sideline_frame0318.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58024_000951_Sideline_frame27.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58024_001072_Endzone_frame287.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58024_001072_Sideline_frame155.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58024_002363_Endzone_frame0953.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58024_002509_Sideline_frame1007.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58024_003061_Endzone_frame0546.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58024_003149_Endzone_frame1260.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58024_003308_Endzone_frame0968.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58024_003948_Endzone_frame255.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58024_003948_Sideline_frame311.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58024_004056_Sideline_frame0856.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58024_004387_Sideline_frame0388.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58025_001014_Sideline_frame1040.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58025_001380_Endzone_frame324.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58025_001380_Sideline_frame390.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58025_002305_Endzone_frame408.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58025_002305_Sideline_frame132.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58025_002992_Endzone_frame0597.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58025_003099_Endzone_frame498.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58025_003099_Sideline_frame82.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58025_003682_Endzone_frame0467.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58025_003799_Sideline_frame1041.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58025_004320_Sideline_frame0909.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58025_004532_Sideline_frame0651.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58026_000051_Endzone_frame0964.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58026_000294_Endzone_frame81.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58026_000294_Sideline_frame330.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58026_000973_Sideline_frame0878.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58026_001343_Endzone_frame1121.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58026_002322_Sideline_frame0546.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58026_002368_Sideline_frame479.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58026_002512_Sideline_frame1064.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58026_003046_Endzone_frame1107.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58026_003198_Endzone_frame1393.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58026_003576_Endzone_frame1023.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58026_003812_Endzone_frame95.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58026_003812_Sideline_frame403.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58026_003873_Sideline_frame0824.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58026_004141_Endzone_frame28.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58026_004141_Sideline_frame187.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58027_000172_Endzone_frame123.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58027_000172_Sideline_frame9.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58027_000284_Endzone_frame361.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58027_000284_Sideline_frame383.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58027_000649_Endzone_frame86.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58027_000649_Sideline_frame6.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58027_000946_Endzone_frame361.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58027_000946_Sideline_frame279.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58027_002340_Endzone_frame0332.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58027_002363_Sideline_frame0823.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58027_002612_Sideline_frame1192.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58027_002707_Endzone_frame0720.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58027_002798_Sideline_frame1180.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58027_003326_Endzone_frame1310.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58027_003458_Sideline_frame0846.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58027_003504_Endzone_frame0900.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58027_003671_Sideline_frame1039.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58028_000193_Sideline_frame0967.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58028_001262_Endzone_frame488.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58028_001262_Sideline_frame204.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58028_001292_Sideline_frame0706.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58028_001880_Endzone_frame0778.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58028_002011_Endzone_frame1325.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58028_002180_Endzone_frame0513.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58028_002264_Sideline_frame0790.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58028_002803_Endzone_frame411.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58028_002803_Sideline_frame267.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58028_003002_Endzone_frame0721.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58028_003832_Sideline_frame0826.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58028_004088_Endzone_frame97.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58028_004088_Sideline_frame66.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58028_004126_Endzone_frame0797.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58029_000441_Sideline_frame0982.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58029_001290_Sideline_frame0851.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58029_001398_Endzone_frame449.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58029_001398_Sideline_frame343.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58029_001463_Endzone_frame0980.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58029_002412_Sideline_frame0687.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58029_002697_Endzone_frame0693.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58029_002940_Endzone_frame1248.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58029_003465_Endzone_frame0760.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58029_003616_Sideline_frame0859.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58029_003722_Endzone_frame307.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58029_003722_Sideline_frame17.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58029_003901_Endzone_frame382.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58029_003901_Sideline_frame278.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58029_004233_Sideline_frame0957.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58030_000857_Endzone_frame0848.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58030_000987_Sideline_frame0998.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58030_001454_Endzone_frame0783.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58030_001560_Sideline_frame0442.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58030_003363_Sideline_frame6.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58030_003811_Endzone_frame301.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58030_003811_Sideline_frame338.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58030_003987_Sideline_frame0657.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58030_004125_Endzone_frame1206.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58030_004149_Endzone_frame0688.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58030_004461_Endzone_frame56.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58030_004461_Sideline_frame206.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58030_004806_Sideline_frame0607.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58030_004954_Endzone_frame277.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58030_004954_Sideline_frame64.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58031_000356_Sideline_frame0851.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58031_001349_Endzone_frame147.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58031_001349_Sideline_frame98.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58031_001866_Sideline_frame0543.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58031_002053_Endzone_frame299.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58031_002053_Sideline_frame77.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58031_002208_Endzone_frame1236.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58031_002331_Endzone_frame0603.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58031_002363_Sideline_frame0954.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58031_002661_Endzone_frame0523.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58031_002864_Sideline_frame0675.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58031_003315_Endzone_frame180.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58031_003315_Sideline_frame420.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58031_003518_Endzone_frame0697.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58031_003654_Endzone_frame194.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58031_003654_Sideline_frame477.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58032_000393_Endzone_frame1063.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58032_000480_Endzone_frame0528.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58032_000571_Sideline_frame0931.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58032_000858_Sideline_frame1261.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58032_001222_Endzone_frame0864.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58032_001278_Endzone_frame142.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58032_001278_Sideline_frame14.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58032_001464_Endzone_frame1360.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58032_002334_Endzone_frame0966.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58032_002785_Endzone_frame210.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58032_002785_Sideline_frame21.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58032_002852_Sideline_frame0775.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58032_002873_Endzone_frame86.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58032_002873_Sideline_frame68.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58033_000685_Sideline_frame1021.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58033_000787_Endzone_frame0835.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58033_000829_Endzone_frame0558.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58033_001507_Endzone_frame400.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58033_001507_Sideline_frame484.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58033_001663_Endzone_frame1055.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58033_002330_Endzone_frame424.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58033_002330_Sideline_frame11.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58033_002368_Sideline_frame0365.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58033_002772_Endzone_frame220.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58033_002772_Sideline_frame24.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58033_002991_Endzone_frame0514.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58033_004131_Sideline_frame0946.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58033_004359_Endzone_frame0123.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58034_000401_Endzone_frame455.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58034_000401_Sideline_frame417.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58034_001003_Sideline_frame1283.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58034_001212_Sideline_frame1182.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58034_001633_Endzone_frame1067.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58034_001710_Endzone_frame1129.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58034_002065_Sideline_frame1007.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58034_002087_Endzone_frame3.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58034_002087_Sideline_frame322.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58034_002299_Endzone_frame0889.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58034_002376_Endzone_frame415.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58034_002376_Sideline_frame59.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58034_002948_Sideline_frame1070.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58034_003201_Endzone_frame26.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58034_003201_Sideline_frame1.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58034_004471_Endzone_frame1222.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58035_000101_Endzone_frame1259.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58035_000101_Endzone_frame54.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58035_000101_Sideline_frame253.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58035_000141_Sideline_frame1874.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58035_000188_Endzone_frame0873.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58035_000357_Sideline_frame0949.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58035_000539_Endzone_frame452.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58035_000539_Sideline_frame303.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58035_000650_Endzone_frame0822.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58035_002991_Sideline_frame1718.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58035_003015_Sideline_frame0821.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58035_003034_Sideline_frame1389.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58035_003206_Endzone_frame337.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58035_003206_Sideline_frame245.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58035_003640_Endzone_frame298.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58035_003640_Sideline_frame75.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58036_000059_Sideline_frame0751.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58036_000171_Endzone_frame0608.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58036_000404_Endzone_frame1522.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58036_000707_Endzone_frame0670.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58036_000769_Endzone_frame353.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58036_000769_Sideline_frame370.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58036_000948_Sideline_frame1017.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58036_001088_Endzone_frame240.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58036_001088_Sideline_frame290.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58036_001908_Sideline_frame0775.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58036_002475_Endzone_frame0714.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58036_002615_Endzone_frame0531.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58036_003522_Endzone_frame153.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58036_003522_Sideline_frame99.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58036_003955_Sideline_frame0890.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58036_004662_Endzone_frame321.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58036_004662_Sideline_frame89.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58037_000173_Endzone_frame194.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58037_000173_Sideline_frame166.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58037_000254_Sideline_frame1147.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58037_000276_Endzone_frame1627.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58037_000769_Endzone_frame0587.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58037_001195_Endzone_frame0792.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58037_001432_Endzone_frame388.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58037_001432_Sideline_frame386.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58037_001982_Endzone_frame0579.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58037_002295_Sideline_frame0683.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58037_002551_Endzone_frame215.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58037_002551_Sideline_frame68.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58037_002951_Endzone_frame0827.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58037_003090_Sideline_frame1000.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58037_003354_Sideline_frame0947.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58038_000081_Sideline_frame1017.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58038_000423_Endzone_frame1113.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58038_000661_Endzone_frame304.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58038_000661_Sideline_frame164.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58038_000687_Endzone_frame0557.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58038_001218_Sideline_frame1398.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58038_001767_Endzone_frame80.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58038_001767_Sideline_frame544.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58038_002309_Endzone_frame0990.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58038_003257_Endzone_frame1022.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58038_003419_Endzone_frame0630.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58038_003646_Endzone_frame463.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58038_003646_Sideline_frame417.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58038_003886_Sideline_frame0907.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58038_003983_Endzone_frame108.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58038_003983_Sideline_frame371.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58038_004141_Sideline_frame0644.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58039_000278_Endzone_frame16.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58039_000278_Sideline_frame429.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58039_000411_Endzone_frame184.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58039_000411_Sideline_frame412.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58039_000521_Endzone_frame56.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58039_000521_Sideline_frame29.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58039_001404_Endzone_frame0997.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58039_001534_Endzone_frame0806.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58039_001739_Sideline_frame0632.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58039_002493_Sideline_frame1021.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58039_002548_Sideline_frame1104.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58039_002612_Endzone_frame0476.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58039_003199_Endzone_frame1038.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58039_003411_Sideline_frame0921.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58039_003976_Endzone_frame183.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58039_003976_Sideline_frame308.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58039_004163_Sideline_frame0517.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58040_000368_Endzone_frame0714.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58040_000387_Endzone_frame401.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58040_000387_Sideline_frame84.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58040_000501_Sideline_frame0699.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58040_000657_Endzone_frame0997.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58040_001314_Endzone_frame123.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58040_001314_Sideline_frame304.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58040_001527_Endzone_frame0435.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58040_001667_Sideline_frame1084.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58040_001834_Sideline_frame0764.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58040_002352_Endzone_frame0624.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58040_002490_Endzone_frame16.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58040_002490_Sideline_frame61.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58040_003235_Sideline_frame0605.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58040_003690_Sideline_frame1217.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58041_000108_Sideline_frame0825.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58041_000510_Endzone_frame547.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58041_000510_Sideline_frame315.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58041_000583_Endzone_frame1078.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58041_000741_Endzone_frame0994.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58041_000772_Endzone_frame1215.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58041_001417_Sideline_frame1640.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58041_001913_Endzone_frame219.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58041_001913_Sideline_frame486.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58041_002353_Endzone_frame1089.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58041_002675_Endzone_frame110.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58041_002675_Sideline_frame296.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58041_003102_Endzone_frame604.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58041_003102_Sideline_frame28.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58041_003581_Endzone_frame0977.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58041_003834_Sideline_frame1043.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58042_000646_Endzone_frame0455.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58042_000863_Sideline_frame0690.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58042_000908_Sideline_frame0832.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58042_001358_Endzone_frame0610.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58042_001934_Endzone_frame165.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58042_001934_Sideline_frame38.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58042_002046_Endzone_frame89.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58042_002046_Sideline_frame254.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58042_002070_Endzone_frame0407.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58042_002880_Sideline_frame1169.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58042_003380_Sideline_frame1232.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58042_003883_Endzone_frame0911.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58042_004069_Endzone_frame116.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58042_004069_Sideline_frame95.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58042_004091_Endzone_frame224.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58042_004091_Sideline_frame120.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58042_004366_Sideline_frame0704.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58043_000466_Endzone_frame1251.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58043_001087_Endzone_frame376.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58043_001087_Sideline_frame251.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58043_001262_Endzone_frame1049.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58043_001642_Endzone_frame0642.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58043_002190_Sideline_frame0899.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58043_002346_Endzone_frame0942.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58043_002685_Endzone_frame243.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58043_002685_Sideline_frame139.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58043_002831_Endzone_frame0545.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58043_003147_Endzone_frame67.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58043_003147_Sideline_frame11.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58043_003292_Sideline_frame0503.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58043_003399_Sideline_frame0585.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58043_004209_Sideline_frame1013.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58044_000060_Endzone_frame0933.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58044_000708_Endzone_frame46.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58044_000708_Sideline_frame286.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58044_000998_Sideline_frame0875.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58044_001071_Endzone_frame86.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58044_001071_Sideline_frame46.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58044_001092_Endzone_frame1283.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58044_001740_Sideline_frame0938.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58044_002450_Endzone_frame0643.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58044_002474_Sideline_frame0616.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58044_002791_Endzone_frame1227.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58044_003195_Endzone_frame295.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58044_003195_Sideline_frame274.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58045_000306_Sideline_frame0964.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58045_000516_Endzone_frame0558.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58045_000584_Endzone_frame0669.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58045_000663_Sideline_frame0721.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58045_000959_Sideline_frame1469.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58045_001100_Endzone_frame0939.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58045_001301_Endzone_frame156.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58045_001301_Sideline_frame656.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58045_001673_Endzone_frame293.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58045_001673_Sideline_frame247.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58045_002955_Endzone_frame406.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58045_002955_Sideline_frame136.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58045_003397_Endzone_frame175.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58045_003397_Sideline_frame1138.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58045_003397_Sideline_frame192.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58045_003506_Endzone_frame0601.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58045_004047_Endzone_frame0358.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58046_000623_Endzone_frame1043.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58046_000882_Endzone_frame0971.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58046_000920_Sideline_frame0797.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58046_000967_Endzone_frame0749.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58046_001676_Sideline_frame0922.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58046_001700_Endzone_frame0662.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58046_001766_Endzone_frame1415.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58046_002442_Sideline_frame0696.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58046_002488_Endzone_frame473.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58046_002488_Sideline_frame228.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58046_002891_Endzone_frame192.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58046_002891_Sideline_frame125.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58046_002938_Sideline_frame1135.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58046_004157_Endzone_frame247.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58046_004157_Sideline_frame388.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58047_000072_Endzone_frame179.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58047_000072_Sideline_frame220.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58047_000736_Endzone_frame0642.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58047_000840_Sideline_frame0808.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58047_001421_Endzone_frame0900.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58047_001796_Endzone_frame1329.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58047_002281_Sideline_frame1871.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58047_002321_Endzone_frame93.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58047_002321_Sideline_frame339.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58047_003057_Sideline_frame0644.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58047_003427_Endzone_frame48.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58047_003427_Sideline_frame177.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58047_003493_Endzone_frame0545.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58047_003682_Endzone_frame180.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58047_003682_Sideline_frame343.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58047_003705_Endzone_frame0577.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58047_004264_Sideline_frame0611.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58048_001131_Endzone_frame212.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58048_001131_Sideline_frame497.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58048_001646_Endzone_frame492.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58048_001646_Sideline_frame48.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58048_002298_Endzone_frame271.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58048_002298_Sideline_frame36.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58049_000327_Endzone_frame150.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58049_000327_Sideline_frame78.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58049_000914_Sideline_frame0694.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58049_000935_Endzone_frame69.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58049_000935_Sideline_frame276.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58049_001121_Endzone_frame0522.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58049_001488_Endzone_frame216.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58049_001488_Sideline_frame216.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58049_001537_Sideline_frame1267.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58049_002655_Sideline_frame0987.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58049_003045_Sideline_frame1388.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58049_003686_Endzone_frame0780.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58049_003707_Sideline_frame0884.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58049_003891_Endzone_frame0804.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58049_003919_Endzone_frame145.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58049_003919_Sideline_frame217.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58049_004143_Endzone_frame0836.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58050_000118_Sideline_frame1301.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58050_000256_Endzone_frame198.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58050_000256_Sideline_frame37.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58050_000762_Endzone_frame0684.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58050_000832_Sideline_frame154.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58050_001469_Sideline_frame0535.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58050_001593_Sideline_frame0943.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58050_001744_Endzone_frame222.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58050_001744_Sideline_frame322.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58050_002781_Endzone_frame1050.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58050_002845_Endzone_frame0886.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58050_002990_Sideline_frame0630.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58050_003168_Endzone_frame0562.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58050_003209_Sideline_frame1059.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58051_000168_Endzone_frame0583.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58051_000338_Endzone_frame0709.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58051_000524_Endzone_frame1235.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58051_000557_Endzone_frame69.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58051_000557_Sideline_frame311.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58051_001996_Endzone_frame0740.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58051_002137_Sideline_frame1361.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58051_002161_Endzone_frame57.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58051_002161_Sideline_frame0936.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58051_002161_Sideline_frame95.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58051_002185_Endzone_frame150.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58051_002185_Sideline_frame32.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58051_002652_Endzone_frame384.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58051_002652_Sideline_frame182.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58051_003298_Sideline_frame0712.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58051_003758_Endzone_frame1137.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58051_004052_Sideline_frame1063.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58052_000140_Endzone_frame0866.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58052_000545_Endzone_frame134.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58052_000545_Sideline_frame81.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58052_000878_Endzone_frame308.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58052_000878_Sideline_frame297.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58052_002814_Sideline_frame1492.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58052_003100_Sideline_frame1106.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58052_003120_Endzone_frame1146.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58052_003200_Endzone_frame0969.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58052_003286_Endzone_frame0853.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58052_003457_Sideline_frame0966.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58052_003602_Endzone_frame382.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58052_003602_Sideline_frame116.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58052_004035_Sideline_frame0282.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58053_000129_Sideline_frame0658.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58053_000491_Endzone_frame1073.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58053_000678_Sideline_frame1150.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58053_001028_Endzone_frame113.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58053_001028_Sideline_frame269.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58053_001044_Endzone_frame1100.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58053_001155_Endzone_frame1378.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58053_001572_Sideline_frame1177.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58053_002097_Sideline_frame0827.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58053_002507_Endzone_frame300.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58053_002507_Sideline_frame416.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58053_002945_Endzone_frame152.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58053_002945_Sideline_frame231.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58053_003639_Endzone_frame0604.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58053_003948_Endzone_frame0960.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58054_000257_Endzone_frame0646.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58054_001081_Endzone_frame0863.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58054_001984_Sideline_frame0933.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58054_002008_Endzone_frame137.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58054_002008_Sideline_frame44.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58054_002037_Endzone_frame1158.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58054_002407_Sideline_frame80.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58054_002438_Endzone_frame286.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58054_002438_Sideline_frame159.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58054_003133_Sideline_frame0672.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58054_003162_Endzone_frame1103.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58054_003482_Sideline_frame1297.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58054_003566_Endzone_frame0627.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58054_003779_Endzone_frame494.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58054_003779_Sideline_frame204.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58054_004019_Sideline_frame1239.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58055_000450_Sideline_frame1111.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58055_000943_Sideline_frame1030.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58055_000976_Endzone_frame152.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58055_000976_Sideline_frame0471.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58055_000976_Sideline_frame131.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58055_001469_Endzone_frame0561.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58055_001677_Sideline_frame0742.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58055_002230_Sideline_frame0819.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58055_002692_Endzone_frame207.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58055_002692_Sideline_frame135.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58055_002717_Sideline_frame0.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58055_002924_Endzone_frame0518.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58055_003792_Endzone_frame0570.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58055_004176_Endzone_frame0374.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58056_001141_Endzone_frame344.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58056_001141_Sideline_frame180.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58056_001277_Endzone_frame1387.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58056_001401_Sideline_frame0949.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58056_001862_Endzone_frame158.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58056_001862_Sideline_frame131.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58056_002161_Sideline_frame0699.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58056_002217_Endzone_frame0736.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58056_002383_Endzone_frame0.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58056_002383_Endzone_frame0656.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58056_002383_Sideline_frame57.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58056_002470_Sideline_frame0925.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58056_003387_Endzone_frame71.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58056_003387_Sideline_frame374.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58056_003520_Sideline_frame0875.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58056_003815_Endzone_frame1383.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58056_004130_Sideline_frame0935.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58057_000893_Sideline_frame0907.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58057_001613_Endzone_frame300.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58057_001613_Sideline_frame367.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58057_001641_Endzone_frame0599.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58057_002043_Sideline_frame1411.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58057_002071_Sideline_frame0784.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58057_002097_Endzone_frame219.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58057_002097_Sideline_frame140.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58057_002466_Endzone_frame0656.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58057_002573_Endzone_frame47.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58057_002573_Sideline_frame417.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58057_003060_Endzone_frame0480.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58057_003255_Sideline_frame0852.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58057_003483_Sideline_frame0655.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58057_003674_Endzone_frame55.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58057_003674_Sideline_frame250.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58057_004714_Endzone_frame0473.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58058_000987_Endzone_frame0759.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58058_001067_Endzone_frame0833.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58058_001157_Endzone_frame160.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58058_001157_Sideline_frame1389.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58058_001157_Sideline_frame688.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58058_001532_Sideline_frame1281.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58058_001903_Endzone_frame41.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58058_001903_Sideline_frame119.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58058_002130_Endzone_frame0935.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58058_002302_Endzone_frame315.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58058_002302_Sideline_frame136.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58058_002642_Sideline_frame1040.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58058_002875_Endzone_frame542.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58058_002875_Sideline_frame502.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58058_003724_Sideline_frame0688.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58058_003911_Endzone_frame0688.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58059_000539_Endzone_frame0588.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58059_000920_Endzone_frame358.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58059_000920_Sideline_frame262.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58059_001277_Endzone_frame0786.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58059_001301_Endzone_frame1198.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58059_001613_Sideline_frame1211.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58059_002125_Endzone_frame0574.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58059_002429_Endzone_frame260.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58059_002429_Sideline_frame57.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58059_002504_Sideline_frame1431.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58059_002724_Endzone_frame0439.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58059_003449_Sideline_frame1328.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58059_004166_Endzone_frame246.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58059_004166_Sideline_frame267.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58060_000420_Endzone_frame218.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58060_000420_Sideline_frame200.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58060_000442_Endzone_frame115.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58060_000442_Sideline_frame400.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58060_000847_Sideline_frame1327.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58060_001026_Sideline_frame1131.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58060_001068_Endzone_frame0688.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58060_001134_Endzone_frame0916.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58060_001218_Sideline_frame1021.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58060_002015_Endzone_frame0924.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58060_002267_Sideline_frame0829.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58060_002393_Sideline_frame1102.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58060_002645_Endzone_frame37.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58060_002645_Sideline_frame84.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58060_003676_Endzone_frame72.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58060_003676_Sideline_frame79.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58060_004019_Endzone_frame0613.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58061_000322_Endzone_frame1420.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58061_000458_Endzone_frame1440.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58061_000929_Endzone_frame228.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58061_000929_Sideline_frame443.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58061_001253_Sideline_frame1106.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58061_001423_Sideline_frame0969.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58061_001516_Endzone_frame112.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58061_001516_Sideline_frame90.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58061_001608_Endzone_frame192.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58061_001608_Sideline_frame142.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58061_001918_Sideline_frame0836.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58061_002706_Endzone_frame0802.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58061_002917_Endzone_frame0817.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58061_003384_Endzone_frame393.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58061_003384_Sideline_frame176.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58061_003568_Sideline_frame0800.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58062_000057_Endzone_frame130.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58062_000057_Sideline_frame443.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58062_000375_Endzone_frame0522.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58062_000510_Endzone_frame272.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58062_000510_Sideline_frame187.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58062_000843_Sideline_frame0833.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58062_000976_Sideline_frame0753.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58062_001042_Sideline_frame0526.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58062_001140_Endzone_frame0940.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58062_001474_Endzone_frame0788.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58062_002355_Sideline_frame1266.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58062_002708_Endzone_frame190.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58062_002708_Sideline_frame331.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58062_002829_Sideline_frame0593.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58062_003458_Endzone_frame217.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58062_003458_Sideline_frame120.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58062_003724_Endzone_frame0799.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58063_000167_Sideline_frame0718.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58063_000905_Endzone_frame312.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58063_000905_Sideline_frame487.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58063_001116_Endzone_frame541.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58063_001116_Sideline_frame293.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58063_001485_Endzone_frame2029.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58063_001552_Sideline_frame0701.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58063_002296_Endzone_frame0593.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58063_002366_Endzone_frame382.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58063_002366_Sideline_frame371.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58063_002868_Endzone_frame1558.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58063_003102_Endzone_frame2232.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58063_003212_Endzone_frame495.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58063_003212_Sideline_frame45.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58063_003238_Sideline_frame0951.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58063_004697_Endzone_frame0534.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58064_000445_Endzone_frame0405.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58064_000734_Sideline_frame1182.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58064_001371_Endzone_frame0825.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58064_001626_Endzone_frame0501.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58064_001957_Endzone_frame18.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58064_001957_Sideline_frame79.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58064_002329_Sideline_frame1396.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58064_002558_Sideline_frame1234.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58064_002800_Endzone_frame235.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58064_002800_Sideline_frame349.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58064_002917_Sideline_frame1346.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58064_002964_Endzone_frame0457.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58064_003043_Endzone_frame354.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58064_003043_Sideline_frame90.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58064_003388_Sideline_frame1557.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58064_003475_Endzone_frame38.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58064_003475_Sideline_frame336.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58065_000070_Sideline_frame0633.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58065_001632_Sideline_frame1663.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58065_002516_Endzone_frame0746.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58065_002540_Sideline_frame0908.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58065_002733_Endzone_frame0649.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58065_002965_Sideline_frame0829.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58065_003038_Endzone_frame80.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58065_003038_Sideline_frame35.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58065_003182_Endzone_frame1098.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58065_003998_Endzone_frame339.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58065_003998_Sideline_frame22.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58065_004123_Sideline_frame231.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58065_004277_Endzone_frame268.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58065_004277_Sideline_frame159.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58065_004364_Endzone_frame0905.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58065_004495_Endzone_frame0871.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58066_000119_Endzone_frame0991.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58066_000295_Endzone_frame0516.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58066_000316_Sideline_frame1120.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58066_000585_Endzone_frame0851.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58066_000675_Endzone_frame165.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58066_000675_Sideline_frame107.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58066_001473_Endzone_frame1082.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58066_001556_Endzone_frame11.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58066_001556_Sideline_frame204.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58066_001916_Endzone_frame222.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58066_001916_Sideline_frame55.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58066_002211_Sideline_frame0746.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58066_002295_Endzone_frame478.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58066_002295_Sideline_frame133.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58066_003300_Endzone_frame0790.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58066_003403_Sideline_frame1688.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58066_003649_Sideline_frame0816.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58067_000410_Endzone_frame0751.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58067_000582_Endzone_frame0538.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58067_000603_Sideline_frame1488.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58067_001287_Endzone_frame0644.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58067_001472_Sideline_frame0905.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58067_001734_Sideline_frame0945.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58067_002079_Endzone_frame351.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58067_002079_Sideline_frame59.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58067_002390_Sideline_frame1031.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58067_003042_Sideline_frame1286.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58067_003138_Endzone_frame1046.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58067_003980_Endzone_frame461.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58067_003980_Sideline_frame385.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58067_004301_Endzone_frame110.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58067_004301_Sideline_frame48.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58068_000124_Endzone_frame0625.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58068_000250_Endzone_frame0647.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58068_000330_Endzone_frame0534.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58068_000401_Endzone_frame154.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58068_000401_Sideline_frame239.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58068_000659_Sideline_frame0652.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58068_001253_Endzone_frame54.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58068_001253_Sideline_frame0684.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58068_001253_Sideline_frame187.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58068_001771_Endzone_frame267.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58068_001771_Sideline_frame135.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58068_002460_Sideline_frame0697.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58068_003185_Endzone_frame0612.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58068_003879_Endzone_frame345.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58068_003879_Sideline_frame338.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58068_004181_Endzone_frame0826.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58068_004574_Sideline_frame0493.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58069_000120_Endzone_frame0964.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58069_000307_Sideline_frame1290.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58069_000367_Sideline_frame1262.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58069_000815_Endzone_frame1075.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58069_001716_Endzone_frame0802.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58069_001738_Endzone_frame390.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58069_001738_Sideline_frame238.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58069_002202_Endzone_frame192.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58069_002202_Sideline_frame36.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58069_002397_Endzone_frame1231.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58069_002495_Sideline_frame1039.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58069_002694_Sideline_frame0591.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58069_003406_Endzone_frame334.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58069_003406_Sideline_frame346.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58069_003451_Endzone_frame481.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58069_003451_Sideline_frame79.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58069_003522_Endzone_frame1176.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58070_000163_Endzone_frame82.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58070_000163_Sideline_frame41.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58070_000502_Endzone_frame0864.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58070_000532_Endzone_frame0693.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58070_001304_Sideline_frame1051.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58070_001620_Endzone_frame41.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58070_001620_Sideline_frame86.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58070_001748_Endzone_frame171.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58070_001748_Sideline_frame98.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58070_002278_Sideline_frame0578.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58070_003606_Sideline_frame0746.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58070_003822_Endzone_frame0692.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58070_003958_Sideline_frame0622.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58070_003958_Sideline_frame215.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58070_003977_Endzone_frame0772.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58071_000060_Endzone_frame350.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58071_000060_Sideline_frame102.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58071_000294_Sideline_frame0861.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58071_000565_Endzone_frame0735.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58071_000587_Sideline_frame1067.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58071_001211_Sideline_frame0690.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58071_001328_Endzone_frame0897.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58071_001836_Sideline_frame0906.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58071_002667_Endzone_frame338.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58071_002667_Sideline_frame240.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58071_003109_Endzone_frame0673.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58071_003399_Endzone_frame151.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58071_003399_Sideline_frame244.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58071_003427_Endzone_frame0958.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58071_003977_Endzone_frame0550.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58072_000072_Endzone_frame86.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58072_000072_Sideline_frame38.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58072_000118_Sideline_frame1741.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58072_000248_Endzone_frame0733.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58072_000423_Endzone_frame417.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58072_000423_Sideline_frame496.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58072_000829_Sideline_frame0810.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58072_001273_Endzone_frame423.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58072_001273_Sideline_frame233.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58072_001404_Sideline_frame0920.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58072_002041_Endzone_frame1240.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58072_002195_Endzone_frame1289.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58072_002435_Endzone_frame0579.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58072_002718_Endzone_frame1352.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58073_000278_Sideline_frame0531.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58073_001010_Endzone_frame30.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58073_001010_Sideline_frame201.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58073_001086_Sideline_frame0405.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58073_001395_Endzone_frame29.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58073_001395_Sideline_frame44.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58073_001556_Endzone_frame0543.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58073_001654_Endzone_frame0694.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58073_002582_Sideline_frame0483.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58073_002868_Endzone_frame0581.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58073_002925_Endzone_frame0706.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58073_003013_Sideline_frame0932.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58073_003087_Endzone_frame94.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58073_003087_Sideline_frame30.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58073_003389_Endzone_frame197.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58073_003389_Sideline_frame51.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58073_003412_Endzone_frame0824.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58074_000160_Endzone_frame88.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58074_000160_Sideline_frame266.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58074_000181_Sideline_frame1006.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58074_000760_Endzone_frame1053.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58074_001199_Endzone_frame55.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58074_001199_Sideline_frame135.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58074_001937_Endzone_frame127.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58074_001937_Sideline_frame203.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58074_002134_Sideline_frame0756.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58074_002443_Sideline_frame1170.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58074_002465_Endzone_frame0753.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58074_002560_Endzone_frame0670.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58074_002860_Sideline_frame0807.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58074_003100_Sideline_frame0947.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58074_003427_Endzone_frame0591.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58075_000529_Endzone_frame1470.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58075_001381_Sideline_frame1204.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58075_001956_Sideline_frame0741.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58075_002093_Endzone_frame27.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58075_002093_Sideline_frame85.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58075_002743_Endzone_frame0657.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58075_002823_Endzone_frame0832.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58075_002919_Sideline_frame0579.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58075_003050_Sideline_frame1297.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58075_003110_Endzone_frame352.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58075_003110_Sideline_frame330.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58075_003327_Endzone_frame102.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58075_003327_Sideline_frame31.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58075_003646_Endzone_frame158.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58075_003646_Sideline_frame283.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58075_003681_Endzone_frame0472.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58075_004003_Sideline_frame0981.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58076_000164_Endzone_frame0621.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58076_000306_Sideline_frame0603.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58076_000593_Endzone_frame0550.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58076_000823_Endzone_frame214.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58076_000823_Sideline_frame377.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58076_000869_Sideline_frame0622.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58076_001171_Endzone_frame153.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58076_001171_Sideline_frame64.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58076_001416_Endzone_frame361.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58076_001416_Sideline_frame24.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58076_001491_Endzone_frame0871.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58076_002384_Endzone_frame0601.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58076_002484_Endzone_frame153.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58076_002484_Sideline_frame18.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58076_002611_Endzone_frame1084.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58076_002817_Sideline_frame0841.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58076_004307_Sideline_frame0820.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58077_000463_Endzone_frame1374.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58077_000995_Sideline_frame0951.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58077_001133_Endzone_frame1346.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58077_001274_Endzone_frame156.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58077_001274_Sideline_frame336.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58077_001741_Endzone_frame88.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58077_001741_Sideline_frame142.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58077_002304_Sideline_frame0785.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58077_002588_Endzone_frame1150.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58077_002617_Endzone_frame165.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58077_002617_Sideline_frame63.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58077_002703_Sideline_frame0957.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58077_002775_Endzone_frame1787.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58077_002881_Endzone_frame0845.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58077_003518_Endzone_frame295.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58077_003518_Sideline_frame111.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58077_004194_Sideline_frame0951.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58078_000051_Endzone_frame1416.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58078_000152_Endzone_frame180.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58078_000152_Sideline_frame131.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58078_000267_Endzone_frame1814.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58078_000792_Endzone_frame1745.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58078_001258_Sideline_frame0582.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58078_001386_Sideline_frame0877.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58078_001428_Endzone_frame0715.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58078_002546_Sideline_frame0881.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58078_003209_Endzone_frame399.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58078_003209_Sideline_frame347.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58078_003436_Endzone_frame0922.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58078_003436_Endzone_frame194.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58078_003436_Sideline_frame181.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58078_004204_Sideline_frame2116.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58078_004220_Endzone_frame22.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58078_004220_Sideline_frame108.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58079_000107_Sideline_frame1305.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58079_000161_Endzone_frame0629.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58079_000243_Endzone_frame82.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58079_000243_Sideline_frame313.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58079_000530_Endzone_frame1242.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58079_000609_Endzone_frame169.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58079_000609_Sideline_frame0466.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58079_000609_Sideline_frame21.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58079_001078_Endzone_frame331.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58079_001078_Sideline_frame319.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58079_001228_Sideline_frame0822.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58079_002156_Sideline_frame0753.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58079_002526_Sideline_frame0746.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58079_002680_Endzone_frame0559.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58079_002951_Endzone_frame0533.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58079_003376_Endzone_frame117.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58079_003376_Sideline_frame156.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58080_000075_Sideline_frame0890.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58080_000609_Sideline_frame0894.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58080_000752_Endzone_frame236.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58080_000752_Sideline_frame193.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58080_000869_Endzone_frame22.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58080_000869_Sideline_frame81.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58080_001982_Endzone_frame310.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58080_001982_Sideline_frame70.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58080_002119_Endzone_frame0836.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58080_002305_Endzone_frame0673.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58080_002582_Endzone_frame0720.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58080_002708_Endzone_frame1057.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58080_003493_Endzone_frame0819.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58080_003512_Endzone_frame101.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58080_003512_Sideline_frame105.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58080_004093_Sideline_frame0470.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58080_004234_Sideline_frame0528.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58081_000051_Endzone_frame1012.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58081_000073_Sideline_frame0964.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58081_000529_Sideline_frame1084.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58081_000850_Endzone_frame1084.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58081_001002_Endzone_frame296.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58081_001002_Sideline_frame374.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58081_001347_Endzone_frame0611.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58081_001390_Endzone_frame0.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58081_001390_Sideline_frame329.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58081_003012_Sideline_frame1200.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58081_003300_Sideline_frame0633.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58081_003631_Endzone_frame0875.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58081_004120_Endzone_frame0708.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58081_004120_Endzone_frame162.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58081_004120_Sideline_frame109.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58082_000215_Endzone_frame0640.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58082_000392_Endzone_frame439.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58082_000392_Sideline_frame8.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58082_000617_Sideline_frame0989.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58082_000830_Endzone_frame252.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58082_000830_Sideline_frame198.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58082_001037_Sideline_frame1061.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58082_001916_Sideline_frame0977.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58082_002360_Endzone_frame1119.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58082_002648_Endzone_frame1308.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58082_002670_Sideline_frame0836.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58082_002844_Endzone_frame124.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58082_002844_Sideline_frame125.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58082_003170_Endzone_frame322.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58082_003170_Sideline_frame51.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58082_003404_Endzone_frame1397.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58082_004170_Sideline_frame0369.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58083_000036_Endzone_frame395.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58083_000036_Sideline_frame62.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58083_000706_Sideline_frame0431.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58083_000888_Endzone_frame0684.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58083_001215_Sideline_frame1076.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58083_001536_Endzone_frame0574.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58083_002231_Endzone_frame0766.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58083_003746_Endzone_frame175.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58083_003746_Sideline_frame1097.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58083_003746_Sideline_frame9.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58083_003859_Endzone_frame0799.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58083_003955_Sideline_frame1008.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58083_004054_Endzone_frame262.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58083_004054_Sideline_frame90.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58083_004091_Endzone_frame372.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58083_004091_Sideline_frame450.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58083_004483_Endzone_frame0352.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58084_000686_Endzone_frame0737.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58084_000728_Sideline_frame0989.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58084_001927_Sideline_frame0810.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58084_002521_Sideline_frame1142.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58084_002587_Endzone_frame283.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58084_002587_Sideline_frame219.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58084_002696_Endzone_frame45.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58084_002696_Sideline_frame71.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58084_002894_Endzone_frame77.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58084_002894_Sideline_frame130.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58084_002969_Endzone_frame0566.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58084_003008_Sideline_frame1285.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58084_003069_Endzone_frame0763.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58084_003454_Endzone_frame62.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58084_003454_Sideline_frame212.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58084_003552_Sideline_frame0812.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58084_003713_Endzone_frame1304.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58085_000240_Endzone_frame219.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58085_000240_Sideline_frame381.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58085_000567_Endzone_frame196.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58085_000567_Sideline_frame98.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58085_000975_Endzone_frame0918.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58085_001539_Endzone_frame95.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58085_001539_Sideline_frame115.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58085_001572_Sideline_frame0894.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58085_002448_Endzone_frame0649.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58085_003070_Sideline_frame0914.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58085_003224_Sideline_frame1271.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58085_003616_Endzone_frame1048.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58085_004291_Sideline_frame0750.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58086_000100_Endzone_frame0698.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58086_000767_Endzone_frame1468.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58086_001282_Endzone_frame0761.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58086_001646_Sideline_frame1027.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58086_001725_Endzone_frame129.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58086_001725_Sideline_frame33.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58086_002092_Sideline_frame1394.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58086_002232_Sideline_frame1030.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58086_002277_Endzone_frame28.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58086_002277_Sideline_frame0939.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58086_002277_Sideline_frame94.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58086_002836_Endzone_frame1038.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58086_003403_Sideline_frame0565.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58086_003747_Endzone_frame252.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58086_003747_Sideline_frame326.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58086_003890_Endzone_frame353.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58086_003890_Sideline_frame308.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58087_000308_Sideline_frame1787.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58087_000668_Endzone_frame80.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58087_000668_Sideline_frame164.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58087_001016_Endzone_frame62.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58087_001016_Sideline_frame216.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58087_002103_Endzone_frame0540.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58087_002357_Endzone_frame104.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58087_002357_Sideline_frame425.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58087_002718_Endzone_frame0600.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58087_003036_Sideline_frame0508.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58087_003150_Endzone_frame0698.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58087_003199_Endzone_frame152.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58087_003199_Sideline_frame118.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58087_003434_Sideline_frame0948.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58087_003663_Endzone_frame0589.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58087_003986_Endzone_frame0462.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58087_004325_Sideline_frame0575.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58088_000420_Endzone_frame0950.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58088_000487_Endzone_frame0470.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58088_000680_Endzone_frame0625.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58088_000750_Sideline_frame1094.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58088_000796_Endzone_frame0871.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58088_000990_Endzone_frame409.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58088_000990_Sideline_frame366.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58088_001139_Endzone_frame308.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58088_001139_Sideline_frame397.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58088_001457_Sideline_frame1257.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58088_001516_Sideline_frame0553.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58088_002530_Endzone_frame0638.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58088_002963_Endzone_frame214.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58088_002963_Sideline_frame135.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58088_003131_Sideline_frame1130.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58088_003294_Endzone_frame73.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58088_003294_Sideline_frame17.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58089_000122_Endzone_frame0875.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58089_000423_Endzone_frame0587.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58089_000678_Endzone_frame299.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58089_000678_Sideline_frame456.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58089_001114_Sideline_frame0717.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58089_001223_Endzone_frame322.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58089_001223_Sideline_frame352.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58089_001925_Endzone_frame0804.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58089_002301_Endzone_frame69.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58089_002301_Sideline_frame348.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58089_002393_Sideline_frame1251.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58089_002442_Sideline_frame1167.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58089_002798_Endzone_frame1073.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58089_002883_Endzone_frame269.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58089_002883_Sideline_frame335.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58089_003437_Sideline_frame0738.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58089_003563_Endzone_frame1063.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58090_000524_Sideline_frame0709.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58090_000831_Sideline_frame0594.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58090_000943_Endzone_frame344.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58090_000943_Sideline_frame241.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58090_001122_Sideline_frame0660.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58090_001557_Endzone_frame1629.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58090_002273_Endzone_frame0559.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58090_002371_Endzone_frame0991.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58090_003287_Endzone_frame203.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58090_003287_Sideline_frame340.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58090_003308_Endzone_frame1393.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58090_003795_Endzone_frame304.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58090_003795_Sideline_frame173.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58090_003860_Endzone_frame0629.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58090_004119_Sideline_frame0604.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58090_004296_Endzone_frame205.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58090_004296_Sideline_frame156.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58091_000148_Endzone_frame262.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58091_000148_Sideline_frame162.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58091_000384_Endzone_frame0460.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58091_000758_Endzone_frame21.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58091_000758_Sideline_frame530.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58091_000865_Endzone_frame354.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58091_000865_Sideline_frame388.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58091_000913_Sideline_frame0930.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58091_001267_Endzone_frame0818.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58091_002053_Endzone_frame122.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58091_002053_Sideline_frame348.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58091_002151_Endzone_frame0652.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58091_002765_Sideline_frame0921.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58091_002987_Sideline_frame0556.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58091_003239_Endzone_frame0816.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58091_003445_Sideline_frame0440.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58093_001894_Endzone_frame394.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58093_001894_Sideline_frame195.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58093_002247_Endzone_frame477.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58093_002247_Sideline_frame413.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58093_002653_Endzone_frame195.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58093_002653_Sideline_frame194.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58093_004050_Endzone_frame79.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58093_004050_Sideline_frame324.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58094_000126_Endzone_frame93.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58094_000126_Sideline_frame624.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58094_001323_Endzone_frame60.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58094_001323_Sideline_frame153.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58094_002022_Endzone_frame281.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58094_002022_Sideline_frame379.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58094_003097_Endzone_frame389.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58094_003097_Sideline_frame30.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58095_000245_Endzone_frame685.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58095_000245_Sideline_frame552.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58095_000916_Endzone_frame307.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58095_000916_Sideline_frame282.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58095_002347_Endzone_frame264.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58095_002347_Sideline_frame107.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58096_001528_Endzone_frame80.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58096_001528_Sideline_frame367.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58096_002159_Endzone_frame303.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58096_002159_Sideline_frame357.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58096_003034_Endzone_frame43.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58096_003034_Sideline_frame337.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58096_003798_Endzone_frame111.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58096_003798_Sideline_frame80.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58097_000231_Endzone_frame56.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58097_000231_Sideline_frame202.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58097_000520_Endzone_frame36.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58097_000520_Sideline_frame272.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58097_001286_Endzone_frame315.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58097_001286_Sideline_frame237.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58097_001409_Endzone_frame126.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58097_001409_Sideline_frame443.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58098_000189_Endzone_frame230.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58098_000189_Sideline_frame17.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58098_001119_Endzone_frame231.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58098_001119_Sideline_frame262.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58098_001780_Endzone_frame266.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58098_001780_Sideline_frame473.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58098_004236_Endzone_frame199.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58098_004236_Sideline_frame323.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58099_001596_Endzone_frame411.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58099_001596_Sideline_frame247.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58099_002952_Endzone_frame243.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58099_002952_Sideline_frame199.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58099_003061_Endzone_frame487.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58099_003061_Sideline_frame561.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58100_000333_Endzone_frame309.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58100_000333_Sideline_frame166.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58100_001539_Endzone_frame203.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58100_001539_Sideline_frame294.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58100_002544_Endzone_frame63.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58100_002544_Sideline_frame484.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58100_003060_Endzone_frame107.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58100_003060_Sideline_frame134.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58101_000659_Endzone_frame86.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58101_000659_Sideline_frame222.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58101_001471_Endzone_frame54.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58101_001471_Sideline_frame20.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58101_001959_Endzone_frame194.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58101_001959_Sideline_frame235.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58101_003299_Endzone_frame66.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58101_003299_Sideline_frame358.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58102_003168_Endzone_frame286.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58102_003168_Sideline_frame67.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58102_003718_Endzone_frame174.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58102_003718_Sideline_frame54.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58102_003918_Endzone_frame77.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58102_003918_Sideline_frame168.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58103_000495_Endzone_frame318.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58103_000495_Sideline_frame159.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58103_000784_Endzone_frame310.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58103_000784_Sideline_frame291.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58103_001175_Endzone_frame33.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58103_001175_Sideline_frame28.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58103_001548_Endzone_frame210.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58103_001548_Sideline_frame37.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58104_001666_Endzone_frame629.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58104_001666_Sideline_frame253.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58104_001786_Endzone_frame185.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58104_001786_Sideline_frame111.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58104_002066_Endzone_frame88.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58104_002066_Sideline_frame128.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58105_001189_Endzone_frame72.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58105_001189_Sideline_frame552.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58105_001697_Endzone_frame256.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58105_001697_Sideline_frame167.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58105_002568_Endzone_frame214.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58105_002568_Sideline_frame223.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58105_003368_Endzone_frame578.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58105_003368_Sideline_frame795.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58106_000386_Endzone_frame195.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58106_000386_Sideline_frame336.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58106_001111_Endzone_frame210.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58106_001111_Sideline_frame273.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58106_001583_Endzone_frame135.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58106_001583_Sideline_frame227.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58106_003091_Endzone_frame299.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58106_003091_Sideline_frame231.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58107_000256_Endzone_frame15.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58107_000256_Sideline_frame136.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58107_000318_Endzone_frame305.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58107_000318_Sideline_frame154.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58107_001413_Endzone_frame240.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58107_001413_Sideline_frame147.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58108_000429_Sideline_frame0534.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58108_000445_Endzone_frame0778.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58108_001202_Endzone_frame0991.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58108_001272_Endzone_frame303.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58108_001272_Sideline_frame58.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58108_001320_Endzone_frame104.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58108_001320_Sideline_frame16.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58108_001823_Sideline_frame0748.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58108_001973_Sideline_frame0783.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58108_002357_Endzone_frame0945.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58108_002641_Endzone_frame0978.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58108_002662_Endzone_frame0807.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58108_003950_Sideline_frame0456.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58108_003971_Endzone_frame26.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58108_003971_Sideline_frame178.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58109_000072_Sideline_frame0693.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58109_000505_Sideline_frame0566.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58109_000529_Endzone_frame0608.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58109_000552_Endzone_frame0707.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58109_000716_Endzone_frame231.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58109_000716_Sideline_frame244.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58109_001111_Endzone_frame287.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58109_001111_Sideline_frame80.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58109_001159_Endzone_frame0670.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58109_001191_Endzone_frame0785.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58109_002365_Sideline_frame0575.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58109_002791_Sideline_frame0805.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58109_003174_Endzone_frame342.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58109_003174_Sideline_frame33.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58109_003894_Endzone_frame0984.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58110_000339_Endzone_frame0763.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58110_000360_Sideline_frame0905.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58110_000625_Endzone_frame313.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58110_000625_Sideline_frame251.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58110_000847_Sideline_frame1072.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58110_000884_Endzone_frame0609.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58110_001590_Endzone_frame0608.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58110_002165_Endzone_frame39.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58110_002165_Sideline_frame296.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58110_002394_Sideline_frame1083.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58110_002570_Sideline_frame0872.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58110_002591_Endzone_frame284.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58110_002591_Sideline_frame334.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58110_002997_Sideline_frame1650.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58110_003268_Endzone_frame254.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58110_003268_Sideline_frame243.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58110_004008_Endzone_frame0806.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58111_000152_Endzone_frame0561.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58111_000463_Endzone_frame191.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58111_000463_Sideline_frame58.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58111_000484_Sideline_frame0983.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58111_000676_Sideline_frame0569.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58111_001047_Sideline_frame1151.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58111_002052_Endzone_frame0465.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58111_002228_Endzone_frame55.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58111_002228_Sideline_frame256.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58111_002837_Sideline_frame1184.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58111_003087_Endzone_frame0770.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58111_003167_Endzone_frame179.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58111_003167_Sideline_frame220.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58111_003829_Sideline_frame0784.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58111_004424_Endzone_frame104.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58111_004424_Sideline_frame52.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58111_004534_Endzone_frame0217.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58112_000036_Sideline_frame0862.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58112_000441_Sideline_frame0891.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58112_000964_Endzone_frame0494.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58112_001885_Endzone_frame1608.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58112_002201_Endzone_frame275.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58112_002201_Sideline_frame326.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58112_002915_Endzone_frame1040.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58112_002915_Endzone_frame138.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58112_002915_Sideline_frame44.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58112_003137_Endzone_frame0627.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58112_003632_Sideline_frame1571.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58112_003771_Endzone_frame0968.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58112_003771_Endzone_frame389.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58112_003771_Sideline_frame703.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58112_004306_Endzone_frame340.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58112_004306_Sideline_frame219.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58113_000424_Sideline_frame0968.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58113_000936_Endzone_frame321.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58113_000936_Sideline_frame279.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58113_001028_Endzone_frame0723.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58113_001959_Endzone_frame0881.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58113_002544_Sideline_frame1893.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58113_003030_Endzone_frame364.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58113_003030_Sideline_frame186.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58113_003054_Sideline_frame0664.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58113_003160_Endzone_frame0856.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58113_003181_Endzone_frame185.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58113_003181_Sideline_frame35.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58113_003393_Endzone_frame300.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58113_003393_Sideline_frame450.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58113_003482_Endzone_frame0604.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58113_003511_Endzone_frame0771.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58114_000124_Endzone_frame0571.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58114_000219_Endzone_frame119.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58114_000219_Sideline_frame1174.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58114_000219_Sideline_frame351.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58114_000519_Endzone_frame0985.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58114_000543_Endzone_frame0573.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58114_001062_Sideline_frame0721.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58114_001300_Endzone_frame315.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58114_001300_Sideline_frame172.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58114_001358_Sideline_frame0898.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58114_001598_Endzone_frame141.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58114_001598_Sideline_frame3.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58114_001662_Endzone_frame1434.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58114_001882_Endzone_frame485.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58114_001882_Sideline_frame470.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58114_002796_Endzone_frame0915.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58114_003195_Sideline_frame0674.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58115_000220_Endzone_frame40.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58115_000220_Sideline_frame571.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58115_000422_Sideline_frame1425.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58115_000976_Sideline_frame0951.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58115_001000_Endzone_frame0784.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58115_001122_Endzone_frame1038.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58115_001326_Endzone_frame193.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58115_001326_Sideline_frame101.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58115_001452_Endzone_frame1030.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58115_002676_Endzone_frame246.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58115_002676_Sideline_frame318.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58115_002741_Endzone_frame0633.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58115_003220_Sideline_frame1183.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58115_003685_Endzone_frame0487.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58115_003968_Sideline_frame1011.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58116_000269_Endzone_frame71.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58116_000269_Sideline_frame48.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58116_001457_Endzone_frame199.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58116_001457_Sideline_frame449.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58116_001761_Endzone_frame0888.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58116_002147_Endzone_frame489.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58116_002147_Sideline_frame376.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58116_002220_Endzone_frame0639.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58116_002430_Endzone_frame0548.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58116_002691_Sideline_frame1039.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58116_003461_Endzone_frame242.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58116_003461_Sideline_frame91.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58116_003537_Sideline_frame1427.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58116_003559_Sideline_frame0530.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58116_003618_Sideline_frame1049.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58116_003726_Sideline_frame0471.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58116_003915_Endzone_frame0646.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58117_000146_Endzone_frame0925.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58117_000146_Endzone_frame205.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58117_000146_Sideline_frame308.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58117_000357_Endzone_frame0652.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58117_000462_Sideline_frame1035.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58117_000666_Sideline_frame0864.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58117_000700_Endzone_frame1660.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58117_002168_Endzone_frame57.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58117_002168_Sideline_frame22.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58117_003029_Endzone_frame230.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58117_003029_Sideline_frame78.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58117_003303_Endzone_frame0586.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58117_003349_Sideline_frame0675.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58117_003365_Sideline_frame0986.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58117_003839_Endzone_frame0908.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58118_000310_Sideline_frame0849.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58118_000395_Endzone_frame1247.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58118_000440_Endzone_frame0761.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58118_001274_Sideline_frame0521.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58118_001770_Endzone_frame207.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58118_001770_Sideline_frame93.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58118_003185_Sideline_frame0863.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58118_003348_Endzone_frame0786.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58118_003617_Endzone_frame1127.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58118_003676_Sideline_frame0279.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58118_003906_Sideline_frame0205.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58118_003956_Endzone_frame230.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58118_003956_Sideline_frame21.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58118_003991_Endzone_frame338.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58118_003991_Sideline_frame359.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58118_004122_Endzone_frame93.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58118_004122_Sideline_frame243.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58119_000139_Sideline_frame1470.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58119_000964_Endzone_frame0910.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58119_001066_Endzone_frame0829.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58119_001185_Endzone_frame374.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58119_001185_Sideline_frame18.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58119_002268_Endzone_frame246.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58119_002268_Sideline_frame415.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58119_002583_Endzone_frame1086.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58119_002623_Endzone_frame0782.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58119_003263_Endzone_frame385.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58119_003263_Sideline_frame1256.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58119_003263_Sideline_frame466.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58119_003326_Endzone_frame342.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58119_003326_Sideline_frame429.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58119_003734_Sideline_frame0837.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58119_004039_Sideline_frame0139.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58120_000453_Endzone_frame291.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58120_000453_Sideline_frame106.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58120_000841_Endzone_frame1190.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58120_000957_Endzone_frame296.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58120_000957_Sideline_frame98.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58120_000979_Sideline_frame1029.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58120_001424_Endzone_frame0581.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58120_001607_Endzone_frame0654.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58120_001850_Endzone_frame0488.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58120_003226_Sideline_frame1470.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58120_003450_Sideline_frame1265.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58120_003658_Endzone_frame269.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58120_003658_Sideline_frame5.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58120_004023_Endzone_frame156.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58120_004023_Sideline_frame1133.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58120_004023_Sideline_frame403.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58120_004171_Sideline_frame0814.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58121_000239_Endzone_frame0748.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58121_000430_Sideline_frame0571.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58121_000523_Sideline_frame1410.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58121_001323_Endzone_frame187.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58121_001323_Sideline_frame418.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58121_001552_Endzone_frame95.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58121_001552_Sideline_frame715.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58121_002059_Endzone_frame0962.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58121_002594_Endzone_frame0691.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58121_002814_Sideline_frame1153.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58121_003207_Sideline_frame0761.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58121_003758_Endzone_frame0980.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58121_003999_Sideline_frame0578.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58122_000036_Sideline_frame0749.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58122_000469_Sideline_frame1238.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58122_000697_Endzone_frame0965.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58122_000838_Endzone_frame370.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58122_000838_Sideline_frame387.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58122_001219_Endzone_frame0469.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58122_001273_Endzone_frame1125.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58122_001273_Endzone_frame129.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58122_001273_Sideline_frame487.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58122_001771_Endzone_frame110.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58122_001771_Sideline_frame290.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58122_002191_Sideline_frame1277.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58122_002470_Endzone_frame0764.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58122_002944_Sideline_frame0711.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58122_003018_Endzone_frame245.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58122_003018_Sideline_frame185.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58122_003063_Endzone_frame0806.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58123_000646_Sideline_frame1440.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58123_001090_Sideline_frame0958.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58123_001457_Endzone_frame335.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58123_001457_Sideline_frame13.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58123_001785_Endzone_frame0881.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58123_001952_Endzone_frame0470.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58123_001976_Endzone_frame0377.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58123_002346_Sideline_frame0811.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58123_002639_Endzone_frame0575.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58123_002831_Sideline_frame0600.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58123_003092_Sideline_frame0707.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58123_003458_Endzone_frame87.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58123_003458_Sideline_frame476.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58124_000220_Endzone_frame282.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58124_000220_Sideline_frame229.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58124_000724_Endzone_frame1048.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58124_000790_Endzone_frame232.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58124_000790_Sideline_frame162.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58124_000881_Sideline_frame0667.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58124_001139_Sideline_frame1168.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58124_001861_Sideline_frame1155.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58124_001986_Endzone_frame0742.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58124_002007_Endzone_frame1145.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58124_002122_Endzone_frame0813.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58124_002166_Sideline_frame0650.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58124_002200_Endzone_frame203.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58124_002200_Sideline_frame169.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58124_002895_Endzone_frame0600.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58125_000213_Endzone_frame97.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58125_000213_Sideline_frame481.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58125_000893_Endzone_frame281.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58125_000893_Sideline_frame298.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58125_001003_Sideline_frame1626.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58125_001046_Sideline_frame0698.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58125_002309_Sideline_frame1192.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58125_002632_Endzone_frame0986.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58125_003037_Endzone_frame1059.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58125_003152_Endzone_frame0831.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58125_003302_Endzone_frame57.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58125_003302_Sideline_frame311.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58125_003485_Sideline_frame0854.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58125_003622_Sideline_frame0728.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58125_003767_Endzone_frame0834.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58126_000543_Sideline_frame0927.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58126_000663_Endzone_frame1159.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58126_000982_Sideline_frame1123.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58126_001824_Sideline_frame1246.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58126_002842_Endzone_frame182.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58126_002842_Sideline_frame227.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58126_002930_Endzone_frame1319.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58126_002930_Endzone_frame71.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58126_002930_Sideline_frame463.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58126_003142_Endzone_frame0861.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58126_003142_Endzone_frame87.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58126_003142_Sideline_frame408.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58126_003469_Endzone_frame0731.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58126_003917_Endzone_frame0921.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58127_001058_Sideline_frame0599.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58127_001324_Endzone_frame0729.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58127_001544_Endzone_frame0526.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58127_001780_Endzone_frame27.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58127_001780_Sideline_frame85.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58127_002187_Endzone_frame161.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58127_002187_Sideline_frame151.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58127_002402_Sideline_frame0828.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58127_002962_Sideline_frame0849.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58127_003204_Endzone_frame0921.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58127_003225_Endzone_frame0744.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58127_003289_Endzone_frame436.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58127_003289_Sideline_frame175.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58127_003589_Sideline_frame0564.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58127_004204_Endzone_frame0338.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58128_000568_Endzone_frame166.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58128_000568_Sideline_frame198.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58128_000618_Endzone_frame0436.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58128_001644_Endzone_frame0941.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58128_002103_Sideline_frame1011.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58128_002118_Sideline_frame0697.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58128_002484_Sideline_frame0555.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58128_002751_Endzone_frame0701.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58128_003074_Endzone_frame89.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58128_003074_Sideline_frame178.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58128_003125_Endzone_frame0521.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58128_003189_Sideline_frame1281.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58128_003367_Endzone_frame208.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58128_003367_Sideline_frame81.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58128_004192_Endzone_frame0830.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58129_000309_Sideline_frame1029.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58129_000359_Endzone_frame0876.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58129_000648_Endzone_frame449.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58129_000648_Sideline_frame301.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58129_000696_Endzone_frame176.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58129_000696_Sideline_frame45.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58129_001364_Endzone_frame0791.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58129_001835_Sideline_frame0563.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58129_002676_Endzone_frame202.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58129_002676_Sideline_frame459.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58129_002957_Sideline_frame1561.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58129_003486_Endzone_frame0960.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58129_003826_Endzone_frame1160.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58129_003845_Sideline_frame0572.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58129_003964_Endzone_frame356.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58129_003964_Sideline_frame452.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58130_000084_Endzone_frame0921.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58130_000169_Sideline_frame1132.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58130_000211_Endzone_frame177.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58130_000211_Sideline_frame415.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58130_000493_Sideline_frame0921.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58130_001234_Endzone_frame1110.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58130_001499_Sideline_frame1299.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58130_002220_Endzone_frame210.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58130_002220_Sideline_frame218.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58130_002391_Endzone_frame144.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58130_002391_Sideline_frame145.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58130_002494_Endzone_frame165.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58130_002494_Sideline_frame158.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58130_002515_Sideline_frame1020.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58130_002622_Endzone_frame1557.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58130_002770_Sideline_frame1051.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58130_003453_Endzone_frame0413.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58131_000240_Endzone_frame180.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58131_000240_Sideline_frame369.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58131_000674_Endzone_frame1041.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58131_001389_Sideline_frame0880.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58131_001479_Sideline_frame1018.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58131_001939_Endzone_frame218.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58131_001939_Sideline_frame267.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58131_002066_Endzone_frame0593.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58131_002114_Endzone_frame0526.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58131_002457_Endzone_frame316.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58131_002457_Sideline_frame415.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58131_002974_Endzone_frame0808.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58131_003885_Sideline_frame1371.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58132_000568_Sideline_frame1574.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58132_000678_Endzone_frame195.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58132_000678_Sideline_frame127.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58132_001471_Sideline_frame0963.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58132_001935_Sideline_frame1129.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58132_002900_Endzone_frame0627.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58132_002921_Endzone_frame0714.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58132_003017_Endzone_frame76.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58132_003017_Sideline_frame374.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58132_003240_Endzone_frame0722.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58132_003511_Sideline_frame1109.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58132_003532_Endzone_frame407.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58132_003532_Sideline_frame137.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58132_004587_Sideline_frame0722.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58132_005109_Endzone_frame1134.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58133_000051_Endzone_frame1077.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58133_000217_Endzone_frame1209.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58133_000554_Endzone_frame0597.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58133_001317_Sideline_frame1299.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58133_001341_Sideline_frame1506.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58133_001560_Endzone_frame126.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58133_001560_Sideline_frame79.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58133_002345_Sideline_frame1097.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58133_002426_Endzone_frame158.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58133_002426_Sideline_frame465.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58133_003103_Endzone_frame0796.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58133_003647_Endzone_frame236.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58133_003647_Sideline_frame272.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58133_003966_Endzone_frame333.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58133_003966_Sideline_frame0598.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58133_003966_Sideline_frame264.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58134_000202_Endzone_frame0499.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58134_000607_Endzone_frame52.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58134_000607_Sideline_frame413.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58134_000878_Endzone_frame0553.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58134_001670_Endzone_frame175.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58134_001670_Sideline_frame19.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58134_002598_Sideline_frame0630.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58134_003246_Endzone_frame0772.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58134_003305_Endzone_frame1124.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58134_003375_Endzone_frame361.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58134_003375_Sideline_frame116.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58134_003542_Sideline_frame0563.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58134_003646_Endzone_frame0860.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58134_003951_Endzone_frame237.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58134_003951_Sideline_frame153.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58134_004078_Sideline_frame0636.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58135_000449_Endzone_frame0938.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58135_001342_Sideline_frame0640.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58135_002243_Endzone_frame0789.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58135_002623_Endzone_frame0841.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58135_002695_Endzone_frame38.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58135_002695_Sideline_frame216.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58135_002849_Sideline_frame1202.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58135_003027_Endzone_frame0865.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58135_003428_Sideline_frame193.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58135_003472_Endzone_frame150.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58135_003472_Sideline_frame284.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58135_003578_Sideline_frame0768.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58135_004016_Sideline_frame0548.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58135_004218_Endzone_frame503.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58135_004218_Sideline_frame324.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58135_004582_Sideline_frame0661.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58136_000274_Sideline_frame0759.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58136_000501_Endzone_frame0752.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58136_000789_Sideline_frame0887.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58136_000852_Endzone_frame2.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58136_000852_Sideline_frame380.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58136_001247_Endzone_frame0865.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58136_001754_Sideline_frame0728.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58136_001860_Endzone_frame168.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58136_001860_Sideline_frame517.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58136_001984_Endzone_frame361.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58136_001984_Sideline_frame206.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58136_002114_Sideline_frame1265.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58136_002542_Sideline_frame0817.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58136_002740_Endzone_frame37.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58136_002740_Sideline_frame215.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58136_003456_Endzone_frame0595.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58136_003521_Endzone_frame1448.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58137_000282_Sideline_frame0647.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58137_000946_Endzone_frame203.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58137_000946_Sideline_frame243.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58137_000967_Sideline_frame1131.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58137_001056_Endzone_frame288.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58137_001056_Sideline_frame250.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58137_001123_Endzone_frame0738.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58137_001377_Endzone_frame0521.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58137_001553_Sideline_frame0953.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58137_002909_Endzone_frame119.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58137_002909_Sideline_frame297.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58137_003287_Endzone_frame284.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58137_003287_Sideline_frame669.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58137_003371_Endzone_frame0858.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58137_003822_Sideline_frame0750.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58137_003909_Sideline_frame1238.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58137_004358_Endzone_frame0841.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58138_000122_Endzone_frame302.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58138_000122_Sideline_frame449.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58138_000564_Endzone_frame1592.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58138_001371_Endzone_frame518.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58138_001371_Sideline_frame251.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58138_001416_Endzone_frame90.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58138_001416_Sideline_frame200.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58138_001962_Sideline_frame0679.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58138_002059_Endzone_frame0315.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58138_002233_Endzone_frame1089.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58138_002726_Endzone_frame0784.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58138_002840_Sideline_frame0872.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58138_003317_Sideline_frame1026.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58138_003526_Endzone_frame0746.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58138_003696_Endzone_frame113.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58138_003696_Sideline_frame78.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58138_004109_Sideline_frame0540.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58139_000194_Endzone_frame0550.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58139_000300_Sideline_frame0764.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58139_000952_Endzone_frame0799.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58139_001292_Endzone_frame298.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58139_001292_Sideline_frame615.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58139_001481_Sideline_frame0652.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58139_001500_Sideline_frame0863.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58139_002202_Sideline_frame1484.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58139_003401_Sideline_frame1088.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58139_003624_Endzone_frame50.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58139_003624_Sideline_frame179.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58139_003710_Endzone_frame0422.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58139_003877_Endzone_frame0485.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58139_003995_Endzone_frame93.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58139_003995_Sideline_frame71.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58140_000233_Endzone_frame0530.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58140_000721_Sideline_frame0721.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58140_001146_Sideline_frame0742.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58140_001189_Endzone_frame0899.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58140_001431_Sideline_frame1176.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58140_002278_Endzone_frame0631.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58140_002458_Endzone_frame1613.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58140_002506_Endzone_frame276.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58140_002506_Sideline_frame322.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58140_002551_Sideline_frame0693.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58140_002672_Endzone_frame269.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58140_002672_Sideline_frame199.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58140_003200_Endzone_frame0655.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58140_003263_Endzone_frame72.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58140_003263_Sideline_frame1.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58140_003411_Endzone_frame19.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58141_000045_Endzone_frame0780.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58141_000392_Endzone_frame0711.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58141_000420_Endzone_frame159.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58141_000420_Sideline_frame201.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58141_000446_Endzone_frame375.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58141_000446_Sideline_frame537.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58141_001381_Sideline_frame1595.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58141_001609_Endzone_frame0963.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58141_001868_Endzone_frame0737.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58141_002025_Endzone_frame216.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58141_002025_Sideline_frame489.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58141_002398_Endzone_frame82.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58141_002398_Sideline_frame260.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58141_002478_Endzone_frame0731.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58141_002520_Sideline_frame1244.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58141_002567_Sideline_frame0976.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58141_003798_Sideline_frame1310.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58142_000036_Endzone_frame0.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58142_000036_Sideline_frame53.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58142_000193_Endzone_frame0682.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58142_000362_Sideline_frame1169.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58142_000688_Endzone_frame9.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58142_000688_Sideline_frame217.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58142_001085_Endzone_frame1007.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58142_001286_Sideline_frame1012.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58142_001404_Endzone_frame14.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58142_001404_Sideline_frame289.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58142_002022_Endzone_frame0726.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58142_002591_Endzone_frame0662.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58142_003866_Endzone_frame141.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58142_003866_Sideline_frame221.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58142_004054_Endzone_frame0652.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58142_004217_Sideline_frame0592.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58143_000079_Endzone_frame14.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58143_000079_Sideline_frame136.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58143_000337_Endzone_frame281.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58143_000337_Sideline_frame304.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58143_000373_Sideline_frame0849.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58143_000748_Sideline_frame0526.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58143_000790_Sideline_frame1194.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58143_001036_Endzone_frame0529.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58143_001547_Sideline_frame0708.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58143_001637_Endzone_frame318.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58143_001637_Sideline_frame0584.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58143_001637_Sideline_frame116.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58143_001818_Endzone_frame325.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58143_001818_Sideline_frame375.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58143_002425_Endzone_frame0758.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58143_002827_Endzone_frame1020.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58143_002942_Endzone_frame0645.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58144_000389_Sideline_frame1104.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58144_000613_Endzone_frame1221.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58144_000774_Endzone_frame147.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58144_000774_Sideline_frame164.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58144_001150_Sideline_frame0784.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58144_001437_Endzone_frame359.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58144_001437_Sideline_frame29.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58144_002463_Endzone_frame0587.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58144_002668_Sideline_frame0879.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58144_003034_Endzone_frame1182.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58144_003198_Endzone_frame0809.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58144_003352_Sideline_frame0632.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58144_003584_Endzone_frame36.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58144_003584_Sideline_frame315.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58144_004084_Endzone_frame1319.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58145_000179_Endzone_frame0663.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58145_000591_Endzone_frame183.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58145_000591_Sideline_frame1028.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58145_000591_Sideline_frame26.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58145_000890_Endzone_frame0556.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58145_001969_Endzone_frame432.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58145_001969_Sideline_frame198.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58145_002136_Endzone_frame77.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58145_002136_Sideline_frame235.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58145_002331_Endzone_frame0480.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58145_002945_Endzone_frame0763.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58145_003339_Sideline_frame0591.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58145_003542_Sideline_frame0839.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58145_004222_Endzone_frame0669.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58146_000143_Sideline_frame0710.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58146_000236_Endzone_frame1366.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58146_000435_Endzone_frame0844.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58146_000674_Endzone_frame186.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58146_000674_Sideline_frame199.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58146_001564_Sideline_frame0435.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58146_001748_Sideline_frame0614.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58146_002143_Endzone_frame0754.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58146_002633_Sideline_frame0744.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58146_002671_Endzone_frame304.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58146_002671_Sideline_frame196.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58146_003135_Endzone_frame0685.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58146_003400_Endzone_frame329.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58146_003400_Sideline_frame45.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58146_003444_Endzone_frame0759.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58146_003809_Endzone_frame399.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58146_003809_Sideline_frame32.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58147_000465_Endzone_frame0717.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58147_000858_Sideline_frame0784.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58147_001154_Endzone_frame0694.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58147_001178_Sideline_frame0722.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58147_001419_Sideline_frame0653.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58147_001650_Endzone_frame177.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58147_001650_Sideline_frame216.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58147_002218_Endzone_frame0789.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58147_002363_Sideline_frame0815.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58147_002980_Sideline_frame1340.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58147_003327_Endzone_frame35.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58147_003327_Sideline_frame293.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58147_003598_Endzone_frame0829.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58147_003728_Endzone_frame262.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58147_003728_Sideline_frame93.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58148_000447_Sideline_frame1768.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58148_000535_Endzone_frame0816.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58148_000815_Endzone_frame233.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58148_000815_Sideline_frame121.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58148_000961_Sideline_frame0746.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58148_001686_Endzone_frame0475.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58148_002045_Sideline_frame0518.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58148_002081_Endzone_frame219.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58148_002081_Sideline_frame117.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58148_002232_Endzone_frame0821.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58148_002357_Endzone_frame0897.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58148_002741_Endzone_frame65.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58148_002741_Sideline_frame151.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58148_002885_Endzone_frame0512.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58148_002951_Sideline_frame1012.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58148_003068_Endzone_frame5.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58148_003068_Sideline_frame406.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58149_000240_Endzone_frame320.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58149_000240_Sideline_frame240.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58149_000368_Endzone_frame0871.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58149_000691_Sideline_frame0870.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58149_001014_Endzone_frame1173.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58149_001082_Endzone_frame209.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58149_001082_Sideline_frame73.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58149_001526_Sideline_frame0569.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58149_001547_Endzone_frame53.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58149_001547_Sideline_frame436.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58149_001606_Sideline_frame0603.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58149_001758_Sideline_frame0999.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58149_001800_Endzone_frame0727.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58149_002634_Sideline_frame1675.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58149_003528_Endzone_frame0532.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58150_000244_Sideline_frame0919.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58150_000935_Sideline_frame0819.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58150_001048_Endzone_frame464.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58150_001048_Sideline_frame185.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58150_001363_Endzone_frame1058.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58150_002167_Endzone_frame1389.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58150_002212_Sideline_frame1396.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58150_002661_Sideline_frame0862.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58150_002899_Sideline_frame0815.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58150_002922_Endzone_frame304.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58150_002922_Sideline_frame450.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58150_003725_Endzone_frame1180.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58150_003819_Endzone_frame0896.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58150_003856_Endzone_frame119.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58150_003856_Sideline_frame376.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58150_003897_Endzone_frame19.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58150_003897_Sideline_frame153.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58151_000336_Endzone_frame0989.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58151_000394_Endzone_frame260.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58151_000394_Sideline_frame297.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58151_000582_Sideline_frame1051.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58151_000677_Endzone_frame0900.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58151_000907_Endzone_frame1313.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58151_001451_Endzone_frame1082.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58151_001870_Sideline_frame1235.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58151_001920_Endzone_frame423.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58151_001920_Sideline_frame197.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58151_003491_Endzone_frame136.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58151_003491_Sideline_frame0854.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58151_003491_Sideline_frame32.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58151_003933_Endzone_frame0832.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58151_004564_Sideline_frame0604.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58152_000098_Endzone_frame326.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58152_000098_Sideline_frame488.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58152_000400_Sideline_frame0828.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58152_000731_Sideline_frame0597.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58152_000774_Sideline_frame0927.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58152_001457_Endzone_frame0599.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58152_001879_Endzone_frame0911.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58152_002460_Sideline_frame0966.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58152_002546_Endzone_frame0599.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58152_002572_Endzone_frame282.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58152_002572_Sideline_frame196.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58152_002751_Endzone_frame385.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58152_002751_Sideline_frame171.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58152_003155_Sideline_frame0751.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58152_003852_Endzone_frame267.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58152_003852_Sideline_frame95.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58152_004038_Endzone_frame0632.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58153_000172_Endzone_frame321.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58153_000172_Sideline_frame139.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58153_000196_Endzone_frame163.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58153_000196_Sideline_frame224.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58153_000543_Endzone_frame74.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58153_000543_Sideline_frame231.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58153_000764_Endzone_frame0504.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58153_001727_Endzone_frame0966.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58153_001944_Sideline_frame1324.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58153_002272_Sideline_frame1000.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58153_002758_Endzone_frame0782.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58153_002858_Endzone_frame0994.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58153_003278_Endzone_frame237.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58153_003278_Sideline_frame485.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58153_003319_Sideline_frame1146.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58153_003880_Endzone_frame0377.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58154_000294_Endzone_frame225.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58154_000294_Sideline_frame16.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58154_000328_Endzone_frame294.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58154_000328_Sideline_frame377.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58154_000697_Endzone_frame0914.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58154_000719_Sideline_frame0719.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58154_000911_Endzone_frame0952.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58154_001407_Endzone_frame0661.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58154_002806_Endzone_frame1088.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58154_003844_Sideline_frame0799.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58154_003951_Endzone_frame0599.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58154_003951_Endzone_frame171.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58154_003951_Sideline_frame89.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58154_004025_Endzone_frame119.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58154_004025_Sideline_frame256.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58154_004542_Sideline_frame0694.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58154_004629_Sideline_frame0880.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58155_000183_Sideline_frame0739.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58155_000283_Sideline_frame0956.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58155_000331_Sideline_frame0808.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58155_000392_Endzone_frame322.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58155_000392_Sideline_frame224.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58155_000580_Endzone_frame49.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58155_000580_Sideline_frame438.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58155_000745_Endzone_frame0756.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58155_000964_Endzone_frame1182.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58155_001751_Endzone_frame0946.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58155_002385_Sideline_frame1119.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58155_002557_Endzone_frame0600.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58155_002771_Endzone_frame1075.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58155_003279_Endzone_frame147.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58155_003279_Sideline_frame135.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58155_003700_Endzone_frame34.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58155_003700_Sideline_frame309.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58156_000337_Endzone_frame4.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58156_000337_Sideline_frame517.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58156_000546_Sideline_frame0560.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58156_000738_Sideline_frame0792.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58156_001091_Sideline_frame0514.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58156_001210_Sideline_frame305.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58156_002138_Endzone_frame0617.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58156_002201_Sideline_frame1052.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58156_002682_Endzone_frame290.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58156_002682_Sideline_frame320.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58156_002952_Sideline_frame0764.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58156_003458_Endzone_frame0767.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58156_003861_Endzone_frame0745.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58156_004046_Endzone_frame90.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58156_004046_Sideline_frame70.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58156_004275_Endzone_frame0762.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58157_000458_Sideline_frame0690.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58157_000752_Endzone_frame192.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58157_000752_Sideline_frame234.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58157_001040_Endzone_frame1319.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58157_001457_Sideline_frame0971.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58157_001501_Endzone_frame316.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58157_001501_Sideline_frame140.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58157_001604_Endzone_frame457.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58157_001604_Sideline_frame573.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58157_001802_Sideline_frame0970.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58157_002221_Endzone_frame0622.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58157_002442_Endzone_frame1224.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58157_002442_Endzone_frame638.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58157_002442_Sideline_frame127.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58157_002496_Endzone_frame1288.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58157_003440_Sideline_frame0880.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58157_003481_Sideline_frame0522.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58158_000978_Endzone_frame0529.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58158_001212_Sideline_frame0881.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58158_001361_Endzone_frame215.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58158_001361_Sideline_frame324.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58158_002215_Endzone_frame0666.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58158_002445_Sideline_frame0791.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58158_002648_Endzone_frame0473.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58158_002854_Endzone_frame184.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58158_002854_Sideline_frame96.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58158_003142_Sideline_frame1207.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58158_003175_Endzone_frame7.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58158_003175_Sideline_frame133.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58158_003397_Endzone_frame190.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58158_003397_Sideline_frame198.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58158_003419_Endzone_frame1417.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58158_003920_Sideline_frame0793.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58158_004074_Sideline_frame0857.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58159_000199_Endzone_frame1072.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58159_000463_Sideline_frame0774.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58159_000666_Endzone_frame4.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58159_000666_Sideline_frame269.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58159_001367_Sideline_frame0720.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58159_001504_Endzone_frame0633.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58159_001879_Sideline_frame0634.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58159_003062_Endzone_frame243.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58159_003062_Sideline_frame360.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58159_003122_Endzone_frame1028.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58159_003247_Endzone_frame1126.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58159_003659_Endzone_frame292.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58159_003659_Sideline_frame231.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58159_003955_Sideline_frame1113.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58160_000115_Sideline_frame0883.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58160_000689_Endzone_frame114.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58160_000689_Sideline_frame35.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58160_001044_Sideline_frame1275.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58160_001830_Endzone_frame0885.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58160_002040_Endzone_frame0662.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58160_002081_Sideline_frame0635.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58160_002167_Endzone_frame218.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58160_002167_Sideline_frame246.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58160_002230_Endzone_frame259.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58160_002230_Sideline_frame408.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58160_002431_Endzone_frame17.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58160_002431_Sideline_frame319.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58160_002476_Sideline_frame0947.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58160_002883_Endzone_frame0639.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58160_003266_Sideline_frame0668.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58160_003537_Endzone_frame0495.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58161_000363_Sideline_frame0889.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58161_000626_Endzone_frame103.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58161_000626_Sideline_frame197.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58161_000941_Endzone_frame62.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58161_000941_Sideline_frame82.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58161_001661_Endzone_frame0537.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58161_001962_Endzone_frame0648.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58161_002297_Endzone_frame22.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58161_002297_Sideline_frame233.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58161_002717_Endzone_frame0400.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58161_002894_Endzone_frame0980.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58161_003003_Endzone_frame1217.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58161_003063_Sideline_frame0893.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58161_003736_Sideline_frame0557.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58161_003879_Endzone_frame81.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58161_003879_Sideline_frame13.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58161_004246_Sideline_frame0445.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58162_000423_Endzone_frame0950.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58162_000514_Endzone_frame8.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58162_000514_Sideline_frame45.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58162_000542_Sideline_frame0695.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58162_000613_Sideline_frame1135.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58162_000751_Endzone_frame0849.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58162_001122_Endzone_frame256.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58162_001122_Sideline_frame74.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58162_001422_Endzone_frame0656.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58162_001995_Endzone_frame135.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58162_001995_Sideline_frame435.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58162_002400_Endzone_frame0726.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58162_003001_Sideline_frame0768.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58162_004490_Endzone_frame0629.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58162_004597_Sideline_frame0927.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58162_004637_Endzone_frame103.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58162_004637_Sideline_frame279.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58163_000101_Endzone_frame0948.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58163_000252_Sideline_frame318.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58163_000542_Sideline_frame0662.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58163_001067_Sideline_frame0767.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58163_001154_Sideline_frame0786.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58163_001864_Endzone_frame0974.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58163_002247_Endzone_frame298.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58163_002247_Sideline_frame246.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58163_002321_Sideline_frame0833.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58163_002718_Endzone_frame0633.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58163_002784_Endzone_frame94.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58163_002784_Sideline_frame319.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58163_002851_Sideline_frame0494.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58163_002948_Endzone_frame148.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58163_002948_Sideline_frame48.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58163_003368_Endzone_frame1139.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58164_000431_Endzone_frame0328.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58164_001297_Sideline_frame0780.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58164_001321_Endzone_frame70.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58164_001321_Sideline_frame190.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58164_001732_Endzone_frame61.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58164_001732_Sideline_frame200.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58164_001828_Endzone_frame1163.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58164_001901_Endzone_frame0909.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58164_002689_Sideline_frame0578.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58164_003118_Endzone_frame216.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58164_003118_Sideline_frame9.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58164_003155_Endzone_frame199.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58164_003155_Sideline_frame136.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58164_003214_Sideline_frame1093.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58164_003368_Endzone_frame0584.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58164_003996_Sideline_frame0684.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58165_000072_Endzone_frame0808.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58165_000595_Sideline_frame1255.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58165_000792_Endzone_frame259.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58165_000792_Sideline_frame38.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58165_000813_Sideline_frame1155.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58165_000870_Sideline_frame1123.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58165_000909_Sideline_frame0998.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58165_001300_Endzone_frame127.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58165_001300_Sideline_frame12.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58165_001505_Endzone_frame0548.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58165_001684_Sideline_frame0989.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58165_002020_Endzone_frame55.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58165_002020_Sideline_frame303.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58165_003112_Endzone_frame0795.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58165_003320_Endzone_frame0997.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58166_000102_Sideline_frame1102.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58166_000379_Endzone_frame0757.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58166_001371_Sideline_frame0729.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58166_001486_Endzone_frame0963.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58166_002467_Sideline_frame0454.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58166_002483_Endzone_frame0440.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58166_003232_Endzone_frame0758.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58166_003413_Sideline_frame0411.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58166_003528_Endzone_frame2771.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58167_000223_Endzone_frame0509.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58167_000310_Sideline_frame0875.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58167_001253_Sideline_frame1018.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58167_001908_Endzone_frame224.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58167_001908_Sideline_frame232.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58167_002365_Sideline_frame1205.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58167_002809_Sideline_frame1804.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58167_002835_Endzone_frame93.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58167_002835_Sideline_frame927.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58167_002854_Endzone_frame0799.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58167_003086_Endzone_frame0691.jpg \n extracting: train_local/nfl-health-and-safety-helmet-assignment/images/58167_003154_Endzone_frame0808.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58167_003881_Endzone_frame221.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/images/58167_003881_Sideline_frame533.jpg \n inflating: train_local/nfl-health-and-safety-helmet-assignment/sample_submission.csv \n inflating: train_local/nfl-health-and-safety-helmet-assignment/test/57906_000718_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/test/57906_000718_Sideline.mp4 \n extracting: train_local/nfl-health-and-safety-helmet-assignment/test/57995_000109_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/test/57995_000109_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/test/58102_002798_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/test/58102_002798_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/test_baseline_helmets.csv \n inflating: train_local/nfl-health-and-safety-helmet-assignment/test_player_tracking.csv \n extracting: train_local/nfl-health-and-safety-helmet-assignment/train/57583_000082_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57583_000082_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57584_000336_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57584_000336_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57584_002674_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57584_002674_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57586_000540_Endzone.mp4 \n extracting: train_local/nfl-health-and-safety-helmet-assignment/train/57586_000540_Sideline.mp4 \n extracting: train_local/nfl-health-and-safety-helmet-assignment/train/57586_001934_Endzone.mp4 \n extracting: train_local/nfl-health-and-safety-helmet-assignment/train/57586_001934_Sideline.mp4 \n extracting: train_local/nfl-health-and-safety-helmet-assignment/train/57586_004152_Endzone.mp4 \n extracting: train_local/nfl-health-and-safety-helmet-assignment/train/57586_004152_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57594_000923_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57594_000923_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57596_002686_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57596_002686_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57597_000658_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57597_000658_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57597_001242_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57597_001242_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57676_003572_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57676_003572_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57679_003316_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57679_003316_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57680_002206_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57680_002206_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57680_003470_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57680_003470_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57682_002630_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57682_002630_Sideline.mp4 \n extracting: train_local/nfl-health-and-safety-helmet-assignment/train/57684_001985_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57684_001985_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57686_002546_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57686_002546_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57700_001264_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57700_001264_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57775_000933_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57775_000933_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57778_004244_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57778_004244_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57781_000252_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57781_000252_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57782_000600_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57782_000600_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57783_003374_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57783_003374_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57784_001741_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57784_001741_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57785_002026_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57785_002026_Sideline.mp4 \n extracting: train_local/nfl-health-and-safety-helmet-assignment/train/57786_003085_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57786_003085_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57787_003413_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57787_003413_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57788_000781_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57788_000781_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57790_002792_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57790_002792_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57790_002839_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57790_002839_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57904_001367_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57904_001367_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57905_002404_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57905_002404_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57906_000718_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57906_000718_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57907_003615_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57907_003615_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57910_001164_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57910_001164_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57911_000147_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57911_000147_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57911_002492_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57911_002492_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57912_001325_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57912_001325_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57913_000218_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57913_000218_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57915_003093_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57915_003093_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57992_000301_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57992_000301_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57992_000350_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57992_000350_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57993_000475_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57993_000475_Sideline.mp4 \n extracting: train_local/nfl-health-and-safety-helmet-assignment/train/57995_000109_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57995_000109_Sideline.mp4 \n extracting: train_local/nfl-health-and-safety-helmet-assignment/train/57997_003691_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57997_003691_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57998_002181_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/57998_002181_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58000_001306_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58000_001306_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58005_001254_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58005_001254_Sideline.mp4 \n extracting: train_local/nfl-health-and-safety-helmet-assignment/train/58005_001612_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58005_001612_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58048_000086_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58048_000086_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58093_001923_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58093_001923_Sideline.mp4 \n extracting: train_local/nfl-health-and-safety-helmet-assignment/train/58094_000423_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58094_000423_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58094_002819_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58094_002819_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58095_004022_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58095_004022_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58098_001193_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58098_001193_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58102_002798_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58102_002798_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58103_003494_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58103_003494_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58104_000352_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58104_000352_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58106_002918_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58106_002918_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58107_004362_Endzone.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train/58107_004362_Sideline.mp4 \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train_baseline_helmets.csv \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train_labels.csv \n inflating: train_local/nfl-health-and-safety-helmet-assignment/train_player_tracking.csv \n"
],
[
"import random\nimport numpy as np\nfrom pathlib import Path\nimport datetime\nimport pandas as pd\nfrom tqdm.notebook import tqdm\nfrom sklearn.model_selection import train_test_split\nimport cv2\nimport json\nimport matplotlib.pyplot as plt\nfrom IPython.core.display import Video, display\nimport subprocess\nimport gc\nimport shutil",
"_____no_output_____"
],
[
"os.listdir(\"./train_local\")",
"_____no_output_____"
]
],
[
[
"# prepare data as COCO format (standard format of obeject detection)\n# reference https://www.kaggle.com/eneszvo/mmdet-cascadercnn-helmet-detection-for-beginners",
"_____no_output_____"
]
],
[
[
"import pandas as pd\n\n# Load image level csv file\nextra_df = pd.read_csv('./train_local/image_labels.csv')\nprint('Number of ground truth bounding boxes: ', len(extra_df))\n\n# Number of unique labels\nlabel_to_id = {label: i for i, label in enumerate(extra_df.label.unique())}\nprint('Unique labels: ', label_to_id)",
"_____no_output_____"
],
[
"extra_df.head()",
"_____no_output_____"
],
[
"def create_ann_file(df, category_id):\n \n now = datetime.datetime.now()\n\n data = dict(\n info=dict(\n description='NFL-Helmet-Assignment',\n url=None,\n version=None,\n year=now.year,\n contributor=None,\n date_created=now.strftime('%Y-%m-%d %H:%M:%S.%f'),\n ),\n licenses=[dict(\n url=None,\n id=0,\n name=None,\n )],\n images=[\n # license, url, file_name, height, width, date_captured, id\n ],\n type='instances',\n annotations=[\n # segmentation, area, iscrowd, image_id, bbox, category_id, id\n ],\n categories=[\n # supercategory, id, name\n ],\n )\n \n class_name_to_id = {}\n labels = [\"__ignore__\",\n 'Helmet',\n 'Helmet-Blurred', \n 'Helmet-Difficult', \n 'Helmet-Sideline',\n 'Helmet-Partial']\n\n for i, each_label in enumerate(labels):\n class_id = i - 1 # starts with -1\n class_name = each_label\n if class_id == -1:\n assert class_name == '__ignore__'\n continue\n class_name_to_id[class_name] = class_id\n data['categories'].append(dict(\n supercategory=None,\n id=class_id,\n name=class_name,\n ))\n \n box_id = 0\n for i, image in tqdm(enumerate(os.listdir(TRAIN_PATH))):\n\n img = cv2.imread(TRAIN_PATH+'/'+image)\n height, width, _ = img.shape\n\n data['images'].append({\n 'license':0, \n 'url': None,\n 'file_name': image,\n 'height': height,\n 'width': width,\n 'date_camputured': None,\n 'id': i\n })\n\n df_temp = df[df.image == image]\n for index, row in df_temp.iterrows():\n\n area = round(row.width*row.height, 1)\n bbox =[row.left, row.top, row.width, row.height]\n\n data['annotations'].append({\n 'id': box_id,\n 'image_id': i,\n 'category_id': category_id[row.label],\n 'area': area,\n 'bbox':bbox,\n 'iscrowd':0\n })\n box_id+=1\n \n return data",
"_____no_output_____"
],
[
"from sklearn.model_selection import train_test_split\n\nTRAIN_PATH = './train_local/images'\nextra_df = pd.read_csv('./train_local/image_labels.csv')\n\ncategory_id = {'Helmet':0, 'Helmet-Blurred':1,\n 'Helmet-Difficult':2, 'Helmet-Sideline':3,\n 'Helmet-Partial':4}\n\ndf_train, df_val = train_test_split(extra_df, test_size=0.2, random_state=36)\n\nann_file_train = create_ann_file(df_train, category_id)\nann_file_val = create_ann_file(df_val, category_id)",
"_____no_output_____"
],
[
"print('train:',df_train.shape,'val:',df_val.shape)",
"_____no_output_____"
],
[
"#save as json to gdrive\nwith open('/content/gdrive/MyDrive/NFL_Helmet/ann_file_train.json', 'w') as f:\n json.dump(ann_file_train, f, indent=4)\n \nwith open('/content/gdrive/MyDrive/NFL_Helmet/ann_file_val.json', 'w') as f:\n json.dump(ann_file_val, f, indent=4)",
"_____no_output_____"
]
]
] |
[
"code",
"markdown",
"code"
] |
[
[
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code"
]
] |
cbd9cd422b1c415493314aba1cbe8c256f3dcfbc
| 392,936 |
ipynb
|
Jupyter Notebook
|
assignment 1/files/knn.ipynb
|
kendricktan/cs231n
|
dea03cf3e31d2a4aa30f423ca3c9e0cbbfe27328
|
[
"MIT"
] | null | null | null |
assignment 1/files/knn.ipynb
|
kendricktan/cs231n
|
dea03cf3e31d2a4aa30f423ca3c9e0cbbfe27328
|
[
"MIT"
] | null | null | null |
assignment 1/files/knn.ipynb
|
kendricktan/cs231n
|
dea03cf3e31d2a4aa30f423ca3c9e0cbbfe27328
|
[
"MIT"
] | null | null | null | 654.893333 | 291,694 | 0.933753 |
[
[
[
"# k-Nearest Neighbor (kNN) exercise\n\n*Complete and hand in this completed worksheet (including its outputs and any supporting code outside of the worksheet) with your assignment submission. For more details see the [assignments page](http://vision.stanford.edu/teaching/cs231n/assignments.html) on the course website.*\n\nThe kNN classifier consists of two stages:\n\n- During training, the classifier takes the training data and simply remembers it\n- During testing, kNN classifies every test image by comparing to all training images and transfering the labels of the k most similar training examples\n- The value of k is cross-validated\n\nIn this exercise you will implement these steps and understand the basic Image Classification pipeline, cross-validation, and gain proficiency in writing efficient, vectorized code.",
"_____no_output_____"
]
],
[
[
"# Run some setup code for this notebook.\n\nimport random\nimport numpy as np\nfrom cs231n.data_utils import load_CIFAR10\nimport matplotlib.pyplot as plt\n\n# This is a bit of magic to make matplotlib figures appear inline in the notebook\n# rather than in a new window.\n%matplotlib inline\nplt.rcParams['figure.figsize'] = (10.0, 8.0) # set default size of plots\nplt.rcParams['image.interpolation'] = 'nearest'\nplt.rcParams['image.cmap'] = 'gray'\n\n# Some more magic so that the notebook will reload external python modules;\n# see http://stackoverflow.com/questions/1907993/autoreload-of-modules-in-ipython\n%load_ext autoreload\n%autoreload 2",
"The autoreload extension is already loaded. To reload it, use:\n %reload_ext autoreload\n"
],
[
"# Load the raw CIFAR-10 data.\ncifar10_dir = 'cs231n/datasets/cifar-10-batches-py'\nX_train, y_train, X_test, y_test = load_CIFAR10(cifar10_dir)\n\n# As a sanity check, we print out the size of the training and test data.\nprint 'Training data shape: ', X_train.shape\nprint 'Training labels shape: ', y_train.shape\nprint 'Test data shape: ', X_test.shape\nprint 'Test labels shape: ', y_test.shape",
"Training data shape: (50000, 32, 32, 3)\nTraining labels shape: (50000,)\nTest data shape: (10000, 32, 32, 3)\nTest labels shape: (10000,)\n"
],
[
"# Visualize some examples from the dataset.\n# We show a few examples of training images from each class.\nclasses = ['plane', 'car', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck']\nnum_classes = len(classes)\nsamples_per_class = 7\nfor y, cls in enumerate(classes):\n idxs = np.flatnonzero(y_train == y)\n idxs = np.random.choice(idxs, samples_per_class, replace=False)\n for i, idx in enumerate(idxs):\n plt_idx = i * num_classes + y + 1\n plt.subplot(samples_per_class, num_classes, plt_idx)\n plt.imshow(X_train[idx].astype('uint8'))\n plt.axis('off')\n if i == 0:\n plt.title(cls)\nplt.show()",
"_____no_output_____"
],
[
"# Subsample the data for more efficient code execution in this exercise\nnum_training = 5000\nmask = range(num_training)\nX_train = X_train[mask]\ny_train = y_train[mask]\n\nnum_test = 500\nmask = range(num_test)\nX_test = X_test[mask]\ny_test = y_test[mask]",
"_____no_output_____"
],
[
"# Reshape the image data into rows\nX_train = np.reshape(X_train, (X_train.shape[0], -1))\nX_test = np.reshape(X_test, (X_test.shape[0], -1))\nprint X_train.shape, X_test.shape",
"(5000, 3072) (500, 3072)\n"
],
[
"from cs231n.classifiers import KNearestNeighbor\n\n# Create a kNN classifier instance. \n# Remember that training a kNN classifier is a noop: \n# the Classifier simply remembers the data and does no further processing \nclassifier = KNearestNeighbor()\nclassifier.train(X_train, y_train)",
"_____no_output_____"
]
],
[
[
"We would now like to classify the test data with the kNN classifier. Recall that we can break down this process into two steps: \n\n1. First we must compute the distances between all test examples and all train examples. \n2. Given these distances, for each test example we find the k nearest examples and have them vote for the label\n\nLets begin with computing the distance matrix between all training and test examples. For example, if there are **Ntr** training examples and **Nte** test examples, this stage should result in a **Nte x Ntr** matrix where each element (i,j) is the distance between the i-th test and j-th train example.\n\nFirst, open `cs231n/classifiers/k_nearest_neighbor.py` and implement the function `compute_distances_two_loops` that uses a (very inefficient) double loop over all pairs of (test, train) examples and computes the distance matrix one element at a time.",
"_____no_output_____"
]
],
[
[
"# Open cs231n/classifiers/k_nearest_neighbor.py and implement\n# compute_distances_two_loops.\n\n# Test your implementation:\ndists = classifier.compute_distances_two_loops(X_test)\nprint dists.shape",
"(500, 5000)\n"
],
[
"# We can visualize the distance matrix: each row is a single test example and\n# its distances to training examples\nplt.imshow(dists, interpolation='none')\nplt.show()",
"_____no_output_____"
]
],
[
[
"**Inline Question #1:** Notice the structured patterns in the distance matrix, where some rows or columns are visible brighter. (Note that with the default color scheme black indicates low distances while white indicates high distances.)\n\n- What in the data is the cause behind the distinctly bright rows?\n- What causes the columns?",
"_____no_output_____"
],
[
"**Your Answer**: The larger the distance between the images, the brighter the value of the cell will be. In other words the larger the value, the less similar the training image is to the test image. The bright row states how the test image isn't similar to any of the training images, while the bright column states that how the training images isn't similar to any of the test images. \n\n",
"_____no_output_____"
]
],
[
[
"# Now implement the function predict_labels and run the code below:\n# We use k = 1 (which is Nearest Neighbor).\ny_test_pred = classifier.predict_labels(dists, k=1)\n\n# Compute and print the fraction of correctly predicted examples\nnum_correct = np.sum(y_test_pred == y_test)\naccuracy = float(num_correct) / num_test\nprint 'Got %d / %d correct => accuracy: %f' % (num_correct, num_test, accuracy)",
"Got 137 / 500 correct => accuracy: 0.274000\n"
]
],
[
[
"You should expect to see approximately `27%` accuracy. Now lets try out a larger `k`, say `k = 5`:",
"_____no_output_____"
]
],
[
[
"y_test_pred = classifier.predict_labels(dists, k=5)\nnum_correct = np.sum(y_test_pred == y_test)\naccuracy = float(num_correct) / num_test\nprint 'Got %d / %d correct => accuracy: %f' % (num_correct, num_test, accuracy)",
"Got 139 / 500 correct => accuracy: 0.278000\n"
]
],
[
[
"You should expect to see a slightly better performance than with `k = 1`.",
"_____no_output_____"
]
],
[
[
"# Now lets speed up distance matrix computation by using partial vectorization\n# with one loop. Implement the function compute_distances_one_loop and run the\n# code below:\ndists_one = classifier.compute_distances_one_loop(X_test)\n\n# To ensure that our vectorized implementation is correct, we make sure that it\n# agrees with the naive implementation. There are many ways to decide whether\n# two matrices are similar; one of the simplest is the Frobenius norm. In case\n# you haven't seen it before, the Frobenius norm of two matrices is the square\n# root of the squared sum of differences of all elements; in other words, reshape\n# the matrices into vectors and compute the Euclidean distance between them.\ndifference = np.linalg.norm(dists - dists_one, ord='fro')\nprint 'Difference was: %f' % (difference, )\nif difference < 0.001:\n print 'Good! The distance matrices are the same'\nelse:\n print 'Uh-oh! The distance matrices are different'",
"(500, 5000)\nDifference was: 0.000000\nGood! The distance matrices are the same\n"
],
[
"# Now implement the fully vectorized version inside compute_distances_no_loops\n# and run the code\ndists_two = classifier.compute_distances_no_loops(X_test)\n\n# check that the distance matrix agrees with the one we computed before:\ndifference = np.linalg.norm(dists - dists_two, ord='fro')\nprint 'Difference was: %f' % (difference, )\nif difference < 0.001:\n print 'Good! The distance matrices are the same'\nelse:\n print 'Uh-oh! The distance matrices are different'",
"Difference was: 0.000000\nGood! The distance matrices are the same\n"
],
[
"# Let's compare how fast the implementations are\ndef time_function(f, *args):\n \"\"\"\n Call a function f with args and return the time (in seconds) that it took to execute.\n \"\"\"\n import time\n tic = time.time()\n f(*args)\n toc = time.time()\n return toc - tic\n\ntwo_loop_time = time_function(classifier.compute_distances_two_loops, X_test)\nprint 'Two loop version took %f seconds' % two_loop_time\n\none_loop_time = time_function(classifier.compute_distances_one_loop, X_test)\nprint 'One loop version took %f seconds' % one_loop_time\n\nno_loop_time = time_function(classifier.compute_distances_no_loops, X_test)\nprint 'No loop version took %f seconds' % no_loop_time\n\n# you should see significantly faster performance with the fully vectorized implementation",
"_____no_output_____"
]
],
[
[
"### Cross-validation\n\nWe have implemented the k-Nearest Neighbor classifier but we set the value k = 5 arbitrarily. We will now determine the best value of this hyperparameter with cross-validation.",
"_____no_output_____"
]
],
[
[
"num_folds = 5\nk_choices = [1, 3, 5, 8, 10, 12, 15, 20, 50, 100]\n\nX_train_folds = []\ny_train_folds = []\n################################################################################\n# TODO: #\n# Split up the training data into folds. After splitting, X_train_folds and #\n# y_train_folds should each be lists of length num_folds, where #\n# y_train_folds[i] is the label vector for the points in X_train_folds[i]. #\n# Hint: Look up the numpy array_split function. #\n################################################################################\nindex_split = np.array_split(range(X_train.shape[0]), num_folds)\nX_train_folds = [X_train[i] for i in index_split]\nY_train_folds = [y_train[i] for i in index_split]\n\n################################################################################\n# END OF YOUR CODE #\n################################################################################\n\n# A dictionary holding the accuracies for different values of k that we find\n# when running cross-validation. After running cross-validation,\n# k_to_accuracies[k] should be a list of length num_folds giving the different\n# accuracy values that we found when using that value of k.\nk_to_accuracies = {}\n\n################################################################################\n# TODO: #\n# Perform k-fold cross validation to find the best value of k. For each #\n# possible value of k, run the k-nearest-neighbor algorithm num_folds times, #\n# where in each case you use all but one of the folds as training data and the #\n# last fold as a validation set. Store the accuracies for all fold and all #\n# values of k in the k_to_accuracies dictionary. #\n################################################################################\nfor k in k_choices:\n for fold in range(num_folds): \n val_X_text = X_train_folds[fold]\n val_Y_test = Y_train_folds[fold]\n \n X_train_temp = np.concatenate(X_train_folds[:fold] + X_train_folds[fold+1:])\n Y_train_temp = np.concatenate(Y_train_folds[:fold] + Y_train_folds[fold+1:])\n \n # Out KNN classifier\n knn = KNearestNeighbor()\n knn.train(X_train_temp, Y_train_temp)\n \n # Computing the distance\n dists_temp = knn.compute_distances_two_loops(val_X_text)\n y_test_pred_temp = knn.predict_labels(dists_temp, k=k)\n \n # Accuracy\n num_correct = np.sum(y_test_pred_temp==val_Y_test)\n num_test = val_X_text.shape[0]\n acc = float(num_correct)/num_test\n print(\"k=\",k,\"Fold=\",fold,\"Accuracy=\",accuracy)\n k_to_accuracies[k] = k_to_accuracies.get(k,[]) + [accuracy]\n \n \n################################################################################\n# END OF YOUR CODE #\n################################################################################\n\n# Print out the computed accuracies\nfor k in sorted(k_to_accuracies):\n for accuracy in k_to_accuracies[k]:\n print 'k = %d, accuracy = %f' % (k, accuracy)",
"('k=', 1, 'Fold=', 0, 'Accuracy=', 0.278)\n('k=', 1, 'Fold=', 1, 'Accuracy=', 0.278)"
],
[
"# Based on the cross-validation results above, choose the best value for k, \n# retrain the classifier using all the training data, and test it on the test\n# data. You should be able to get above 28% accuracy on the test data.\nbest_k = 1\n\nclassifier = KNearestNeighbor()\nclassifier.train(X_train, y_train)\ny_test_pred = classifier.predict(X_test, k=best_k)\n\n# Compute and display the accuracy\nnum_correct = np.sum(y_test_pred == y_test)\naccuracy = float(num_correct) / num_test\nprint 'Got %d / %d correct => accuracy: %f' % (num_correct, num_test, accuracy)",
"_____no_output_____"
],
[
"# plot the raw observations\nfor k in k_choices:\n accuracies = k_to_accuracies[k]\n plt.scatter([k] * len(accuracies), accuracies)\n\n# plot the trend line with error bars that correspond to standard deviation\naccuracies_mean = np.array([np.mean(v) for k,v in sorted(k_to_accuracies.items())])\naccuracies_std = np.array([np.std(v) for k,v in sorted(k_to_accuracies.items())])\nplt.errorbar(k_choices, accuracies_mean, yerr=accuracies_std)\nplt.title('Cross-validation on k')\nplt.xlabel('k')\nplt.ylabel('Cross-validation accuracy')\nplt.show()",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
]
] |
cbd9d8dc00c6a210adf66ac98283ec4fcf594322
| 14,238 |
ipynb
|
Jupyter Notebook
|
00-Pytorch/Original/cifar10_tutorial.ipynb
|
biroeniko/hello-jupyter
|
ac1f62e5f020f6e9d58de49dba12320514ede151
|
[
"MIT"
] | null | null | null |
00-Pytorch/Original/cifar10_tutorial.ipynb
|
biroeniko/hello-jupyter
|
ac1f62e5f020f6e9d58de49dba12320514ede151
|
[
"MIT"
] | null | null | null |
00-Pytorch/Original/cifar10_tutorial.ipynb
|
biroeniko/hello-jupyter
|
ac1f62e5f020f6e9d58de49dba12320514ede151
|
[
"MIT"
] | null | null | null | 54.972973 | 1,922 | 0.583298 |
[
[
[
"%matplotlib inline",
"_____no_output_____"
]
],
[
[
"\nTraining a Classifier\n=====================\n\nThis is it. You have seen how to define neural networks, compute loss and make\nupdates to the weights of the network.\n\nNow you might be thinking,\n\nWhat about data?\n----------------\n\nGenerally, when you have to deal with image, text, audio or video data,\nyou can use standard python packages that load data into a numpy array.\nThen you can convert this array into a ``torch.*Tensor``.\n\n- For images, packages such as Pillow, OpenCV are useful\n- For audio, packages such as scipy and librosa\n- For text, either raw Python or Cython based loading, or NLTK and\n SpaCy are useful\n\nSpecifically for vision, we have created a package called\n``torchvision``, that has data loaders for common datasets such as\nImagenet, CIFAR10, MNIST, etc. and data transformers for images, viz.,\n``torchvision.datasets`` and ``torch.utils.data.DataLoader``.\n\nThis provides a huge convenience and avoids writing boilerplate code.\n\nFor this tutorial, we will use the CIFAR10 dataset.\nIt has the classes: ‘airplane’, ‘automobile’, ‘bird’, ‘cat’, ‘deer’,\n‘dog’, ‘frog’, ‘horse’, ‘ship’, ‘truck’. The images in CIFAR-10 are of\nsize 3x32x32, i.e. 3-channel color images of 32x32 pixels in size.\n\n.. figure:: /_static/img/cifar10.png\n :alt: cifar10\n\n cifar10\n\n\nTraining an image classifier\n----------------------------\n\nWe will do the following steps in order:\n\n1. Load and normalizing the CIFAR10 training and test datasets using\n ``torchvision``\n2. Define a Convolution Neural Network\n3. Define a loss function\n4. Train the network on the training data\n5. Test the network on the test data\n\n1. Loading and normalizing CIFAR10\n^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n\nUsing ``torchvision``, it’s extremely easy to load CIFAR10.\n\n",
"_____no_output_____"
]
],
[
[
"import torch\nimport torchvision\nimport torchvision.transforms as transforms",
"_____no_output_____"
]
],
[
[
"The output of torchvision datasets are PILImage images of range [0, 1].\nWe transform them to Tensors of normalized range [-1, 1].\n\n",
"_____no_output_____"
]
],
[
[
"transform = transforms.Compose(\n [transforms.ToTensor(),\n transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])\n\ntrainset = torchvision.datasets.CIFAR10(root='./data', train=True,\n download=True, transform=transform)\ntrainloader = torch.utils.data.DataLoader(trainset, batch_size=4,\n shuffle=True, num_workers=2)\n\ntestset = torchvision.datasets.CIFAR10(root='./data', train=False,\n download=True, transform=transform)\ntestloader = torch.utils.data.DataLoader(testset, batch_size=4,\n shuffle=False, num_workers=2)\n\nclasses = ('plane', 'car', 'bird', 'cat',\n 'deer', 'dog', 'frog', 'horse', 'ship', 'truck')",
"_____no_output_____"
]
],
[
[
"Let us show some of the training images, for fun.\n\n",
"_____no_output_____"
]
],
[
[
"import matplotlib.pyplot as plt\nimport numpy as np\n\n# functions to show an image\n\n\ndef imshow(img):\n img = img / 2 + 0.5 # unnormalize\n npimg = img.numpy()\n plt.imshow(np.transpose(npimg, (1, 2, 0)))\n\n\n# get some random training images\ndataiter = iter(trainloader)\nimages, labels = dataiter.next()\n\n# show images\nimshow(torchvision.utils.make_grid(images))\n# print labels\nprint(' '.join('%5s' % classes[labels[j]] for j in range(4)))",
"_____no_output_____"
]
],
[
[
"2. Define a Convolution Neural Network\n^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\nCopy the neural network from the Neural Networks section before and modify it to\ntake 3-channel images (instead of 1-channel images as it was defined).\n\n",
"_____no_output_____"
]
],
[
[
"import torch.nn as nn\nimport torch.nn.functional as F\n\n\nclass Net(nn.Module):\n def __init__(self):\n super(Net, self).__init__()\n self.conv1 = nn.Conv2d(3, 6, 5)\n self.pool = nn.MaxPool2d(2, 2)\n self.conv2 = nn.Conv2d(6, 16, 5)\n self.fc1 = nn.Linear(16 * 5 * 5, 120)\n self.fc2 = nn.Linear(120, 84)\n self.fc3 = nn.Linear(84, 10)\n\n def forward(self, x):\n x = self.pool(F.relu(self.conv1(x)))\n x = self.pool(F.relu(self.conv2(x)))\n x = x.view(-1, 16 * 5 * 5)\n x = F.relu(self.fc1(x))\n x = F.relu(self.fc2(x))\n x = self.fc3(x)\n return x\n\n\nnet = Net()",
"_____no_output_____"
]
],
[
[
"3. Define a Loss function and optimizer\n^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\nLet's use a Classification Cross-Entropy loss and SGD with momentum.\n\n",
"_____no_output_____"
]
],
[
[
"import torch.optim as optim\n\ncriterion = nn.CrossEntropyLoss()\noptimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9)",
"_____no_output_____"
]
],
[
[
"4. Train the network\n^^^^^^^^^^^^^^^^^^^^\n\nThis is when things start to get interesting.\nWe simply have to loop over our data iterator, and feed the inputs to the\nnetwork and optimize.\n\n",
"_____no_output_____"
]
],
[
[
"for epoch in range(2): # loop over the dataset multiple times\n\n running_loss = 0.0\n for i, data in enumerate(trainloader, 0):\n # get the inputs\n inputs, labels = data\n\n # zero the parameter gradients\n optimizer.zero_grad()\n\n # forward + backward + optimize\n outputs = net(inputs)\n loss = criterion(outputs, labels)\n loss.backward()\n optimizer.step()\n\n # print statistics\n running_loss += loss.item()\n if i % 2000 == 1999: # print every 2000 mini-batches\n print('[%d, %5d] loss: %.3f' %\n (epoch + 1, i + 1, running_loss / 2000))\n running_loss = 0.0\n\nprint('Finished Training')",
"_____no_output_____"
]
],
[
[
"5. Test the network on the test data\n^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n\nWe have trained the network for 2 passes over the training dataset.\nBut we need to check if the network has learnt anything at all.\n\nWe will check this by predicting the class label that the neural network\noutputs, and checking it against the ground-truth. If the prediction is\ncorrect, we add the sample to the list of correct predictions.\n\nOkay, first step. Let us display an image from the test set to get familiar.\n\n",
"_____no_output_____"
]
],
[
[
"dataiter = iter(testloader)\nimages, labels = dataiter.next()\n\n# print images\nimshow(torchvision.utils.make_grid(images))\nprint('GroundTruth: ', ' '.join('%5s' % classes[labels[j]] for j in range(4)))",
"_____no_output_____"
]
],
[
[
"Okay, now let us see what the neural network thinks these examples above are:\n\n",
"_____no_output_____"
]
],
[
[
"outputs = net(images)",
"_____no_output_____"
]
],
[
[
"The outputs are energies for the 10 classes.\nHigher the energy for a class, the more the network\nthinks that the image is of the particular class.\nSo, let's get the index of the highest energy:\n\n",
"_____no_output_____"
]
],
[
[
"_, predicted = torch.max(outputs, 1)\n\nprint('Predicted: ', ' '.join('%5s' % classes[predicted[j]]\n for j in range(4)))",
"_____no_output_____"
]
],
[
[
"The results seem pretty good.\n\nLet us look at how the network performs on the whole dataset.\n\n",
"_____no_output_____"
]
],
[
[
"correct = 0\ntotal = 0\nwith torch.no_grad():\n for data in testloader:\n images, labels = data\n outputs = net(images)\n _, predicted = torch.max(outputs.data, 1)\n total += labels.size(0)\n correct += (predicted == labels).sum().item()\n\nprint('Accuracy of the network on the 10000 test images: %d %%' % (\n 100 * correct / total))",
"_____no_output_____"
]
],
[
[
"That looks waaay better than chance, which is 10% accuracy (randomly picking\na class out of 10 classes).\nSeems like the network learnt something.\n\nHmmm, what are the classes that performed well, and the classes that did\nnot perform well:\n\n",
"_____no_output_____"
]
],
[
[
"class_correct = list(0. for i in range(10))\nclass_total = list(0. for i in range(10))\nwith torch.no_grad():\n for data in testloader:\n images, labels = data\n outputs = net(images)\n _, predicted = torch.max(outputs, 1)\n c = (predicted == labels).squeeze()\n for i in range(4):\n label = labels[i]\n class_correct[label] += c[i].item()\n class_total[label] += 1\n\n\nfor i in range(10):\n print('Accuracy of %5s : %2d %%' % (\n classes[i], 100 * class_correct[i] / class_total[i]))",
"_____no_output_____"
]
],
[
[
"Okay, so what next?\n\nHow do we run these neural networks on the GPU?\n\nTraining on GPU\n----------------\nJust like how you transfer a Tensor on to the GPU, you transfer the neural\nnet onto the GPU.\n\nLet's first define our device as the first visible cuda device if we have\nCUDA available:\n\n",
"_____no_output_____"
]
],
[
[
"device = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n\n# Assume that we are on a CUDA machine, then this should print a CUDA device:\n\nprint(device)",
"_____no_output_____"
]
],
[
[
"The rest of this section assumes that `device` is a CUDA device.\n\nThen these methods will recursively go over all modules and convert their\nparameters and buffers to CUDA tensors:\n\n.. code:: python\n\n net.to(device)\n\n\nRemember that you will have to send the inputs and targets at every step\nto the GPU too:\n\n.. code:: python\n\n inputs, labels = inputs.to(device), labels.to(device)\n\nWhy dont I notice MASSIVE speedup compared to CPU? Because your network\nis realllly small.\n\n**Exercise:** Try increasing the width of your network (argument 2 of\nthe first ``nn.Conv2d``, and argument 1 of the second ``nn.Conv2d`` –\nthey need to be the same number), see what kind of speedup you get.\n\n**Goals achieved**:\n\n- Understanding PyTorch's Tensor library and neural networks at a high level.\n- Train a small neural network to classify images\n\nTraining on multiple GPUs\n-------------------------\nIf you want to see even more MASSIVE speedup using all of your GPUs,\nplease check out :doc:`data_parallel_tutorial`.\n\nWhere do I go next?\n-------------------\n\n- :doc:`Train neural nets to play video games </intermediate/reinforcement_q_learning>`\n- `Train a state-of-the-art ResNet network on imagenet`_\n- `Train a face generator using Generative Adversarial Networks`_\n- `Train a word-level language model using Recurrent LSTM networks`_\n- `More examples`_\n- `More tutorials`_\n- `Discuss PyTorch on the Forums`_\n- `Chat with other users on Slack`_\n\n\n",
"_____no_output_____"
]
]
] |
[
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown"
] |
[
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
]
] |
cbd9db5776382a70407440286be53a18a6ea89f7
| 8,147 |
ipynb
|
Jupyter Notebook
|
notebooks/exploratory_data_analysis/contact_map_clustering.ipynb
|
FordyceLab/tessellate
|
a3f0c38b4392027a7503828f48d65ec02eb24698
|
[
"MIT"
] | null | null | null |
notebooks/exploratory_data_analysis/contact_map_clustering.ipynb
|
FordyceLab/tessellate
|
a3f0c38b4392027a7503828f48d65ec02eb24698
|
[
"MIT"
] | 7 |
2021-03-31T19:41:46.000Z
|
2022-01-13T02:39:45.000Z
|
notebooks/exploratory_data_analysis/contact_map_clustering.ipynb
|
FordyceLab/tessellate
|
a3f0c38b4392027a7503828f48d65ec02eb24698
|
[
"MIT"
] | null | null | null | 52.902597 | 4,300 | 0.749233 |
[
[
[
"# Purpose\n\nTo reduce the size of the model while maintaining maximal information content, I will cluster all of the channels of each contact map and take a respresentative channel from each cluster.",
"_____no_output_____"
]
],
[
[
"# A little magic\n%load_ext autoreload\n%autoreload 2\n%matplotlib inline",
"_____no_output_____"
],
[
"# Imports\nimport h5py\nfrom matplotlib import pyplot as plt\nimport numpy as np\nfrom scipy.cluster.hierarchy import dendrogram, linkage\nfrom scipy.sparse import coo_matrix\nimport sklearn.cluster\nimport torch\nimport itertools\nfrom tqdm import *",
"_____no_output_____"
]
],
[
[
"## Handle reading the HDF5 file containing all of the data",
"_____no_output_____"
]
],
[
[
"# Make a counts matrix to store co-clustering of \ncounts = np.zeros((12, 12))\n\nwith h5py.File('../../data/contacts.hdf5', 'r') as h5file:\n keys = list(h5file.keys())\n for entry in tqdm(keys):\n atomtypes = h5file[entry]['memberships'][:]\n memberships = h5file[entry]['memberships'][:]\n target = torch.from_numpy(h5file[entry]['target'][:].astype(np.int64))\n target[:, 2] = target[:, 2] - 3\n \n n_res = np.max(memberships[:, 0]) + 1\n val = torch.ones(len(target))\n size = torch.Size([n_res, n_res, 12])\n \n contact_map = torch.sparse.FloatTensor(target.t(), val, size)\n contact_map = contact_map.to_dense().numpy().transpose((2, 0, 1)).reshape(12, -1)\n \n \n n_clust = 4\n clustering = sklearn.cluster.AgglomerativeClustering(n_clust).fit(contact_map)\n labels = clustering.labels_\n \n for i in range(n_clust):\n channels = np.where(labels == i)[0]\n for j in itertools.product(channels, repeat=2):\n counts[int(j[0]), int(j[1])] += 1",
"100%|██████████| 60895/60895 [3:31:54<00:00, 4.79it/s]\n"
],
[
"plt.imshow(counts / len(keys), cmap='gray')",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
]
] |
cbd9e68ceb2df618e6b65b3d7a7d2ab1dee7c183
| 13,192 |
ipynb
|
Jupyter Notebook
|
Deep Learning/Boltzzman Machines/Boltzmann_Machine.ipynb
|
d4rk-lucif3r/Machine-Learning-Models
|
403c7a2a37420f1ce99985422fb44e2c330742f5
|
[
"MIT"
] | 3 |
2021-01-04T08:42:00.000Z
|
2021-05-30T11:39:58.000Z
|
Deep Learning/Boltzzman Machines/Boltzmann_Machine.ipynb
|
d4rk-lucif3r/Machine-Learning-Models
|
403c7a2a37420f1ce99985422fb44e2c330742f5
|
[
"MIT"
] | 3 |
2021-05-29T17:09:41.000Z
|
2021-05-29T17:10:32.000Z
|
Deep Learning/Boltzzman Machines/Boltzmann_Machine.ipynb
|
d4rk-lucif3r/Machine-Learning-Models
|
403c7a2a37420f1ce99985422fb44e2c330742f5
|
[
"MIT"
] | 2 |
2021-05-26T03:54:13.000Z
|
2021-08-17T06:34:27.000Z
| 28.554113 | 123 | 0.451637 |
[
[
[
"# Boltzmann Machine\r\n",
"_____no_output_____"
],
[
"## Downloading the dataset\r\n",
"_____no_output_____"
],
[
"### ML-100K\r\n",
"_____no_output_____"
]
],
[
[
"# !wget \"http://files.grouplens.org/datasets/movielens/ml-100k.zip\"\r\n# !unzip ml-100k.zip\r\n# !ls",
"_____no_output_____"
]
],
[
[
"### ML-1M\r\n",
"_____no_output_____"
]
],
[
[
"# !wget \"http://files.grouplens.org/datasets/movielens/ml-1m.zip\"\r\n# !unzip ml-1m.zip\r\n# !ls",
"_____no_output_____"
]
],
[
[
"## Importing the libraries\r\n",
"_____no_output_____"
]
],
[
[
"import numpy as np\r\nimport pandas as pd\r\nimport torch\r\nimport torch.nn as nn\r\nimport torch.nn.parallel\r\nimport torch.optim as optim\r\nimport torch.utils.data\r\nfrom torch.autograd import Variable",
"_____no_output_____"
]
],
[
[
"## Importing the dataset\r\n\r\n",
"_____no_output_____"
]
],
[
[
"# We won't be using this dataset.\r\nmovies = pd.read_csv('ml-1m/movies.dat', sep = '::', header = None, engine = 'python', encoding = 'latin-1')\r\nusers = pd.read_csv('ml-1m/users.dat', sep = '::', header = None, engine = 'python', encoding = 'latin-1')\r\nratings = pd.read_csv('ml-1m/ratings.dat', sep = '::', header = None, engine = 'python', encoding = 'latin-1')",
"_____no_output_____"
]
],
[
[
"## Preparing the training set and the test set\n",
"_____no_output_____"
]
],
[
[
"training_set = pd.read_csv('ml-100k/u1.base', delimiter = '\\t')\r\ntraining_set = np.array(training_set, dtype = 'int')\r\ntest_set = pd.read_csv('ml-100k/u1.test', delimiter = '\\t')\r\ntest_set = np.array(test_set, dtype = 'int')",
"_____no_output_____"
]
],
[
[
"## Getting the number of users and movies\n",
"_____no_output_____"
]
],
[
[
"nb_users = int(max(max(training_set[:,0]), max(test_set[:,0])))\r\nnb_movies = int(max(max(training_set[:,1]), max(test_set[:,1])))",
"_____no_output_____"
],
[
"print(nb_users,'|',nb_movies)",
"943 | 1682\n"
]
],
[
[
"## Converting the data into an array with users in lines and movies in columns\n",
"_____no_output_____"
]
],
[
[
"def convert(data):\r\n new_data = []\r\n for id_users in range(1, nb_users + 1):\r\n id_movies = data[:,1][data[:,0] == id_users]\r\n id_ratings = data[:,2][data[:,0] == id_users]\r\n ratings = np.zeros(nb_movies)\r\n ratings[id_movies - 1] = id_ratings\r\n new_data.append(list(ratings))\r\n return new_data\r\ntraining_set = convert(training_set)\r\ntest_set = convert(test_set)",
"_____no_output_____"
]
],
[
[
"## Converting the data into Torch Tensors\r\n\r\n",
"_____no_output_____"
]
],
[
[
"training_set = torch.FloatTensor(training_set)\r\ntest_set = torch.FloatTensor(test_set)",
"_____no_output_____"
]
],
[
[
"## Converting the ratings into binary ratings 1 (Liked) or 0 (Not Liked)\n",
"_____no_output_____"
]
],
[
[
"training_set[training_set == 0] = -1\r\ntraining_set[training_set == 1] = 0\r\ntraining_set[training_set == 2] = 0\r\ntraining_set[training_set >= 3] = 1\r\ntest_set[test_set == 0] = -1\r\ntest_set[test_set == 1 ] = 0\r\ntest_set[test_set == 2] = 0\r\ntest_set[test_set >= 3] = 1 ",
"_____no_output_____"
]
],
[
[
"## Creating the architecture of the Neural Network\n",
"_____no_output_____"
]
],
[
[
"class RBM():\r\n def __init__(self, nv, nh):\r\n self.W = torch.randn(nh, nv)\r\n self.a = torch.randn(1, nh)\r\n self.b = torch.randn(1, nv)\r\n def sample_h(self, x):\r\n wx = torch.mm(x, self.W.t())\r\n activation = wx + self.a.expand_as(wx)\r\n p_h_given_v = torch.sigmoid(activation)\r\n return p_h_given_v, torch.bernoulli(p_h_given_v)\r\n def sample_v(self, y):\r\n wy = torch.mm(y, self.W)\r\n activation = wy + self.b.expand_as(wy)\r\n p_v_given_h = torch.sigmoid(activation)\r\n return p_v_given_h, torch.bernoulli(p_v_given_h)\r\n def train(self, v0, vk, ph0, phk):\r\n self.W += (torch.mm(v0.t(), ph0) - torch.mm(vk.t(), phk)).t()\r\n self.b += torch.sum((v0 - vk), 0)\r\n self.a += torch.sum((ph0 - phk), 0)\r\n",
"_____no_output_____"
],
[
"nv = len(training_set[0])\r\nnh = 100\r\nbatch_size = 100\r\n",
"_____no_output_____"
],
[
"rbm = RBM(nv, nh)",
"_____no_output_____"
]
],
[
[
"## Training the RBM\n",
"_____no_output_____"
]
],
[
[
"nb_epoch = 10\r\nfor epoch in range(1, nb_epoch + 1):\r\n train_loss = 0\r\n train_rmse = 0\r\n s = 0.\r\n for id_user in range(0, nb_users - batch_size, batch_size):\r\n vk = training_set[id_user:id_user+batch_size]\r\n v0 = training_set[id_user:id_user+batch_size]\r\n ph0,_ = rbm.sample_h(v0)\r\n for k in range(10):\r\n _,hk = rbm.sample_h(vk)\r\n _,vk = rbm.sample_v(hk)\r\n vk[v0<0] = v0[v0<0]\r\n phk,_ = rbm.sample_h(vk)\r\n rbm.train(v0, vk, ph0, phk)\r\n train_loss += torch.mean(torch.abs(v0[v0>=0] - vk[v0>=0]))#\r\n train_rmse += np.sqrt(torch.mean((v0[v0 >= 0] - vk[v0 >= 0])**2))\r\n s += 1.\r\n print('epoch : {} | Loss : {} | RMSE : {}'.format(epoch, train_loss/s, train_rmse/s))\r\n",
"epoch : 1 | Loss : 0.3479700982570648 | RMSE : 0.5840339660644531\nepoch : 2 | Loss : 0.21596258878707886 | RMSE : 0.46370697021484375\nepoch : 3 | Loss : 0.24536579847335815 | RMSE : 0.4949502944946289\nepoch : 4 | Loss : 0.24698251485824585 | RMSE : 0.49677684903144836\nepoch : 5 | Loss : 0.24985729157924652 | RMSE : 0.49960410594940186\nepoch : 6 | Loss : 0.24552202224731445 | RMSE : 0.49522528052330017\nepoch : 7 | Loss : 0.24806632101535797 | RMSE : 0.4979269206523895\nepoch : 8 | Loss : 0.24915306270122528 | RMSE : 0.4989830255508423\nepoch : 9 | Loss : 0.24926723539829254 | RMSE : 0.49917200207710266\nepoch : 10 | Loss : 0.2471865862607956 | RMSE : 0.4969930648803711\n"
]
],
[
[
"## Testing the RBM\n",
"_____no_output_____"
]
],
[
[
"\r\n\r\ntest_loss = 0\r\ntest_rmse = 0\r\ns = 0.\r\nfor id_user in range(nb_users):\r\n v = training_set[id_user:id_user+1]\r\n vt = test_set[id_user:id_user+1]\r\n if len(vt[vt>=0]) > 0:\r\n _,h = rbm.sample_h(v)\r\n _,v = rbm.sample_v(h)\r\n test_loss += torch.mean(torch.abs(vt[vt>=0] - v[vt>=0]))\r\n test_rmse += np.sqrt(torch.mean((vt[vt>=0] - v[vt>=0])**2)) \r\n s += 1.\r\nprint('loss : {} | RMSE : {}'.format(test_loss/s, test_rmse/s))\r\n",
"loss : 0.24995048344135284 | RMSE : 0.47680291533470154\n"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
]
] |
cbda085107bac9ca4ab85946d29e47497494018f
| 10,675 |
ipynb
|
Jupyter Notebook
|
src/data/make_dataset.ipynb
|
changsoooo/kisra-1
|
dce2132df97d2b20d9019fa3421540895a097183
|
[
"FTL"
] | null | null | null |
src/data/make_dataset.ipynb
|
changsoooo/kisra-1
|
dce2132df97d2b20d9019fa3421540895a097183
|
[
"FTL"
] | null | null | null |
src/data/make_dataset.ipynb
|
changsoooo/kisra-1
|
dce2132df97d2b20d9019fa3421540895a097183
|
[
"FTL"
] | null | null | null | 27.30179 | 172 | 0.377705 |
[
[
[
"import requests\nimport json\nimport pandas as pd\nimport numpy as np\nfrom pandas import json_normalize",
"_____no_output_____"
],
[
"url = 'https://finance.naver.com/api/sise/etfItemList.nhn'\njson_data = json.loads(requests.get(url).text)\ndf = json_normalize(json_data['result']['etfItemList'])",
"_____no_output_____"
]
],
[
[
"## 1. 전처리\n#### 문자열 처리 쉽도록 종목코드 앞에 'A' 붙임",
"_____no_output_____"
]
],
[
[
"df['itemcode'] = 'A' + df['itemcode']\ndf = df.sort_values(by='etfTabCode', axis=0)",
"_____no_output_____"
],
[
"df['exchange_nm'] = 'KRX'",
"_____no_output_____"
],
[
"df = df.rename(columns = {'itemcode':'code', 'itemname':'name', 'nowVal':'price', 'quant':'volumne_sh', 'amonut':'trading_amt_mln', 'marketSum':'market_cap_100m'})\ndf = df.drop(['etfTabCode', 'risefall', 'changeVal', 'changeRate', 'threeMonthEarnRate'], axis=1)",
"_____no_output_____"
],
[
"df",
"_____no_output_____"
]
],
[
[
"# 2. 종목 필터링",
"_____no_output_____"
],
[
"## 1) 투자목적 적합성\n### (1) 레버리지, 인버스 배제\n- 종목명에 금지어가 포함돼 있으면 유니버스에서 원천 삭제(`restricted_tokens`)\n- 레버리지, 인버스는 단기 투자목적에만 적합하므로 배제",
"_____no_output_____"
]
],
[
[
"restricted_tokens = ['레버리지', '인버스', '2X']\ndf = df[~df.itemname.str.contains('|'.join(restricted_tokens))]",
"_____no_output_____"
]
],
[
[
"## 2) 유동성\n### (1) 시총/거래량\n- 시총(`limit_market_cap`) >= 300억원\n- 거래금액(`limit_trading_amt`) >= 1억원",
"_____no_output_____"
]
],
[
[
"limit_market_cap = 300 # 300억원\nlimit_trading_amt = 100 # 1억원 (100백만원)",
"_____no_output_____"
],
[
"df = df[np.logical_and(df.market_cap_100m >= limit_market_cap, df.trading_amt_mln >= limit_trading_amt)]",
"_____no_output_____"
]
],
[
[
"# 3. 파일 저장",
"_____no_output_____"
]
],
[
[
"df.reset_index(inplace=True, drop=True)",
"_____no_output_____"
],
[
"df.to_csv('universe.csv', encoding='utf-8-sig', index=False)",
"_____no_output_____"
]
]
] |
[
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
]
] |
cbda224d4e4e2c0bc0e396e6eebd376877de7f28
| 30,999 |
ipynb
|
Jupyter Notebook
|
Session_01/template/Session_01_notebook_student.ipynb
|
rob3rtroessler/CS-14115
|
06c66288ca319eea9f73ff87a507cd0ec43faf82
|
[
"MIT"
] | 3 |
2021-06-26T18:36:33.000Z
|
2021-07-02T19:07:23.000Z
|
Session_01/template/Session_01_notebook_student.ipynb
|
rob3rtroessler/CS-14115
|
06c66288ca319eea9f73ff87a507cd0ec43faf82
|
[
"MIT"
] | null | null | null |
Session_01/template/Session_01_notebook_student.ipynb
|
rob3rtroessler/CS-14115
|
06c66288ca319eea9f73ff87a507cd0ec43faf82
|
[
"MIT"
] | 2 |
2021-06-29T16:59:06.000Z
|
2021-06-30T22:49:20.000Z
| 27.554667 | 730 | 0.537243 |
[
[
[
"<div style='background: #FF7B47; padding: 10px; border: thin solid darblue; border-radius: 5px; margin-bottom: 2vh'>\n \n# Session 01 - Notebook\n\nLike most session notebooks in this course, this notebook is divided into two parts. Part one is a 'manual' that will allow you to code along with the new code that we introduce at the beginning of each session. The second part is the actual lab/assignment part, where you will work through a few practical tasks and write small but useful programs.",
"_____no_output_____"
],
[
"<div style='background: #FF7B47; padding: 10px; border: thin solid darblue; border-radius: 5px'>\n \n## Part 1 - Manual",
"_____no_output_____"
],
[
"<div style='background: lightsalmon; padding: 10px; border: thin solid darblue; border-radius: 5px'>\nA.1 - \"hello, world!\"",
"_____no_output_____"
]
],
[
[
"print('Hello, world!')",
"Hello, world!\n"
]
],
[
[
"<div style='background: lightsalmon; padding: 10px; border: thin solid darblue; border-radius: 5px'>\nA.2 - basic datatypes - strings and numeric variables",
"_____no_output_____"
]
],
[
[
"# your code here\nprint(2)\nprint(2+4)\nprint(2**4)",
"2\n6\n16\n"
],
[
"my_var = 2",
"_____no_output_____"
],
[
"print(my_var*2)",
"4\n"
]
],
[
[
"<div style='background: lightsalmon; padding: 10px; border: thin solid darblue; border-radius: 5px'>\nA.3 - basic operations",
"_____no_output_____"
]
],
[
[
"# your code here\nprint(my_var *2)\nprint(my_var **4)",
"4\n16\n"
]
],
[
[
"<div style='background: lightsalmon; padding: 10px; border: thin solid darblue; border-radius: 5px'>\nA.4 - advanced data types to store collections of data",
"_____no_output_____"
]
],
[
[
"# your code here - lists",
"_____no_output_____"
],
[
"# your code here - dictionaries\n\nmy_dict = {\n \"brand\": \"Ford\",\n \"model\": \"Mustang\",\n \"year\": 1964,\n \"example\" : {'test':'here', 'test_2': 'here_2'} \n}\n\nprint(my_dict['brand'])\nprint(my_dict['example'])\nprint(my_dict['example']['test'])",
"Ford\n{'test': 'here', 'test_2': 'here_2'}\nhere\n"
]
],
[
[
"<div style='background: lightsalmon; padding: 10px; border: thin solid darblue; border-radius: 5px'>\nA.5 - for loops",
"_____no_output_____"
]
],
[
[
"# your code here\nfruits = ['apple', 'banana', 'strawberry', 'peach']\n\n# we are iterating over each element in the data structure\nfor fruit in fruits:\n print(fruit)\n",
"apple\nbanana\nstrawberry\npeach\n"
],
[
"fruits_in_fridge = {\n 'apple': 2,\n 'banana': 3, \n 'strawberry':10,\n 'peach': 4\n}\n\nfor key in fruits_in_fridge:\n print(key) # printing the keys\n print(fruits_in_fridge[key])",
"apple\n2\nbanana\n3\nstrawberry\n10\npeach\n4\n"
]
],
[
[
"<div style='background: lightsalmon; padding: 10px; border: thin solid darblue; border-radius: 5px'>\nA.6 - Python If ... Else",
"_____no_output_____"
]
],
[
[
"# your code here\n\na = 33\nb = 33\nif b > a:\n print(\"b is greater than a\")\nelif a == b:\n print(\"a and b are equal\")\nelse:\n print(\"a is greater than b\")",
"a and b are equal\n"
]
],
[
[
"<div style='background: lightsalmon; padding: 10px; border: thin solid darblue; border-radius: 5px'>\nA.6 - Functions",
"_____no_output_____"
]
],
[
[
"# Your code here\ndef say_hello_to(name):\n print(\"hello, \" + name)\n",
"_____no_output_____"
],
[
"say_hello_to('Nathan')",
"hello, Nathan\n"
],
[
"list_of_names = ['Megan', 'Robert', 'Jermain', 'Angela', 'Amr', 'Anthony', 'Rex', 'Nathan']\n\nfor classmate in list_of_names:\n say_hello_to(classmate)",
"hello, Megan\nhello, Robert\nhello, Jermain\nhello, Angela\nhello, Amr\nhello, Anthony\nhello, Rex\nhello, Nathan\n"
]
],
[
[
"<div style='background: #6A9EB4; padding: 10px; border: thin solid darblue; border-radius: 5px'>\n\n## Part 2 - Lab\n \nDuring today's lab you will write code that will help the College to perform the house lottery more efficiently and assist the house administrations in a variety of tasks.",
"_____no_output_____"
],
[
"<div style='background: #ADCAD6; padding: 10px; border: thin solid darblue; border-radius: 5px'>\n\n## Task #1 - automatize the house lottery\n\nIn the precirculated template folder, you will find the file students.csv with all rising sophomores that will enter the house lottery, i.e. they will get assigned to one of the twelve undergraduate houses. So far, the college has done this task manually but they hope that you can help them to automtize that process. Please load the csv and add another column 'house_id'. Pyhton's csv package will come in handy to load the csv file and treat each row as a list. Having loaded the file, add a random house id to each student and save that information in a new csv file. You might find the python package 'random' quite useful to automatize the lottery process. We've imported the package for you and provided an example.\n",
"_____no_output_____"
],
[
"<div style='background: #ADCAD6; padding: 10px; border: thin solid darblue; border-radius: 5px'>\nExamples and precirculated code:",
"_____no_output_____"
]
],
[
[
"# house ids lookup tables\n\nhouse_to_id = {\n 'Adams House': 0,\n 'Cabot House': 1,\n 'Currier House' : 2,\n 'Dunster House': 3,\n 'Eliot House': 4,\n 'Kirkland House': 5,\n 'Leverett House': 6,\n 'Lowell House': 7,\n 'Mather House': 8,\n 'Pforzheimer House':9,\n 'Quincy House': 10,\n 'Winthrop House': 11\n}\n\nid_to_house = {\n 0: 'Adams House',\n 1: 'Cabot House',\n 2: 'Currier House',\n 3: 'Dunster House',\n 4: 'Eliot House',\n 5: 'Kirkland House',\n 6: 'Leverett House',\n 7: 'Lowell House',\n 8: 'Mather House',\n 9: 'Pforzheimer House',\n 10: 'Quincy House',\n 11: 'Winthrop House'\n}",
"_____no_output_____"
],
[
"# importing useful python packages\nimport random\nimport csv",
"_____no_output_____"
],
[
"# some example code snippets how to load a csv file and how to write into one\n\n# read\nfile_read = open(\"data/students.csv\", \"r\")\nreader = csv.reader(file_read)\n\nfor row in reader:\n print(row)\n break # breaking after first element feel free to check out the entire data stucture\n \nfile_read.close()\n\n# write - notice that the file doesn't have to exist beforehand! csv write will create the file automatically, which is very useful!\nfile_write = open('students_with_house.csv', 'w', newline='')\n\nwriter = csv.writer(file_write)\n\n# we just write one row here. It might be useful to put this line into a loop when automatizing things\nwriter.writerow(['first_name', 'last_name', 'HUID','email', 'house_id'])\n\nfile_write.close()\n\n",
"['Oona', 'Swigart', '16076420', '[email protected]']\n"
],
[
"# example - generate a random integer between 1 and 10.\nexample_random = random.randint(1,10)\nprint(example_random)",
"6\n"
]
],
[
[
"<div style='background: #ADCAD6; padding: 10px; border: thin solid darblue; border-radius: 5px'>\nYour turn - load the csv file, create a random number for each student between 0-11 and store all students in a new csv file with their respective house assignments. A for loop might come in handy.",
"_____no_output_____"
]
],
[
[
"# your code here\nfile_read = open(\"data/students.csv\", \"r\")\nreader = csv.reader(file_read)\n\n# write - notice that the file doesn't have to exist beforehand! csv write will create the file automatically, which is very useful!\nfile_write = open('students_with_house.csv', 'w', newline='')\nwriter = csv.writer(file_write)\n\nfor row in reader:\n \n student = row\n house_tmp = random.randint(0,11)\n student.append(house_tmp)\n writer.writerow(student)\n\nfile_write.close()\n ",
"_____no_output_____"
]
],
[
[
"<div style='background: #ADCAD6; padding: 10px; border: thin solid darblue; border-radius: 5px'>\nWrite a small program that makes sure that you've successfully created and populated a csv with all students and their assigned houses.",
"_____no_output_____"
]
],
[
[
"# your code here\nfile_test = open('students_with_house.csv', 'r')\nreader = csv.reader(file_test)\n\nfor row in reader:\n print(row)\n break\n",
"['Oona', 'Swigart', '16076420', '[email protected]', '3']\n"
]
],
[
[
"<div style='background: #ADCAD6; padding: 10px; border: thin solid darblue; border-radius: 5px'>\n\n## Task #2 - generate a file for a house on demand\n\nOK, you've helped the college out with the lottery but now the house administrators are struggling a bit because they have all 2000 students in one file but only care about the students that were assigned to their particular house. Write a small programm that solves that task on demand and generates a csv for them with only their students. You can write a program that does this task on demand for a given house, or you can generate a csv for each house in advance.\n",
"_____no_output_____"
]
],
[
[
"# your code here\n\nhouse = 'Adams House'\nhouse_id = house_to_id['Adams House']\n\nfile_read = open('students_with_house.csv', 'r')\nreader = csv.reader(file_read)\n\nfile_write = open('adams_students.csv', 'w', newline='')\nwriter = csv.writer(file_write)\n\nfor row in reader:\n if int(row[4]) == house_id:\n writer.writerow(row)\n \nfile_write.close()",
"_____no_output_____"
],
[
"def make_house_file(house):\n \n house_id = house_to_id[house]\n\n file_read = open('students_with_house.csv', 'r')\n reader = csv.reader(file_read)\n\n file_write = open(house + '_students.csv', 'w', newline='')\n writer = csv.writer(file_write)\n\n for row in reader:\n if int(row[4]) == house_id:\n writer.writerow(row)\n\n file_write.close()",
"_____no_output_____"
],
[
"for house_name in house_to_id:\n make_house_file(house_name)",
"_____no_output_____"
]
],
[
[
"<div style='background: #CBE0A4; padding: 10px; border: thin solid darblue; border-radius: 5px'>\n\n## Bonus Tasks\n \n1. calculate vacant rooms per house\n2. write a program that computes the number of students assigned per house in a given csv\n3. write a function that checks whether there are problems with the numbers of students assigned to each house\n4. write code that assigns students randomly but in such a way that there are no capacity issues.",
"_____no_output_____"
],
[
"<div style='background: #CBE0A4; padding: 10px; border: thin solid darblue; border-radius: 5px'>\nSome house administrators have complaned that the list of students is too long to accomodate all new sophomores assigned to their houses. Since some houses are bigger and others are smaller, we cannot simply generate integers and get away with the randomly generated number of students in each house. Rather, we have to check more carefolly whether there is still capacity. Below, find two useful dictionaries hat should help you to solve this task.",
"_____no_output_____"
]
],
[
[
"# bonus is house with exact capacities\n\nhouse_capacity = {\n 'Adams House': 411,\n 'Cabot House': 362,\n 'Currier House' : 356,\n 'Dunster House': 428,\n 'Eliot House': 450,\n 'Kirkland House': 400,\n 'Leverett House': 480,\n 'Lowell House': 450,\n 'Mather House': 426,\n 'Pforzheimer House':360,\n 'Quincy House': 420,\n 'Winthrop House': 500\n}\n\n# number of occupied rooms after seniors have left\nhouse_occupied = {\n 'Adams House': 236,\n 'Cabot House': 213,\n 'Currier House' : 217,\n 'Dunster House': 296,\n 'Eliot House': 288,\n 'Kirkland House': 224,\n 'Leverett House': 233,\n 'Lowell House': 242,\n 'Mather House': 217,\n 'Pforzheimer House':195,\n 'Quincy House': 253,\n 'Winthrop House': 310\n}",
"_____no_output_____"
],
[
"house_names = []\n\nfor house in house_capacity:\n house_names.append(house)",
"_____no_output_____"
]
],
[
[
"<div style='background: #CBE0A4; padding: 10px; border: thin solid darblue; border-radius: 5px'>Let's start by writing a small program that helps us to calculate the vacant rooms for each house. Try to use a dictionary structure that contains all information for each house. Feel free to also write a few lines that check how many vacant rooms there are in total.",
"_____no_output_____"
]
],
[
[
"vacant_rooms = {}\n\n\n# your code here\n\nfor house in house_names:\n vacant_rooms[house] = house_capacity[house] - house_occupied[house]\n \n# your code ends here\n\nprint(vacant_rooms)",
"{'Adams House': 175, 'Cabot House': 149, 'Currier House': 139, 'Dunster House': 132, 'Eliot House': 162, 'Kirkland House': 176, 'Leverett House': 247, 'Lowell House': 208, 'Mather House': 209, 'Pforzheimer House': 165, 'Quincy House': 167, 'Winthrop House': 190}\n"
],
[
"# your code here\n\n# take each house name\n# add their values\n\ntotal = 0\n\nfor house in house_names:\n total = total + vacant_rooms[house]\n \n # total += vacant_rooms[house]\n \n \nprint(total)",
"2119\n"
]
],
[
[
"<div style='background: #CBE0A4; padding: 10px; border: thin solid darblue; border-radius: 5px'>Let's now write a small function that calculates the number of students assigned per house with our old method and returns a dictionary with that information",
"_____no_output_____"
]
],
[
[
"helper_dict = {\n \"A\": 222,\n \"B\": 123\n}\n",
"_____no_output_____"
],
[
"def calculate_students_per_house(filename):\n\n # helper dict\n helper_dict = {\n 'Adams House': 0,\n 'Cabot House': 0,\n 'Currier House' : 0,\n 'Dunster House': 0,\n 'Eliot House': 0,\n 'Kirkland House': 0,\n 'Leverett House': 0,\n 'Lowell House': 0,\n 'Mather House': 0,\n 'Pforzheimer House':0,\n 'Quincy House': 0,\n 'Winthrop House': 0\n }\n\n # your code here\n \n file_read = open(filename, 'r')\n \n reader = csv.reader(file_read)\n \n for row in reader:\n house_id = int(row[4])\n house_name = id_to_house[house_id]\n \n helper_dict[house_name] = helper_dict[house_name] + 1\n \n # your code ends here\n\n return helper_dict\n ",
"_____no_output_____"
],
[
"assigned_students_random = calculate_students_per_house('students_with_house.csv')\n\nprint(assigned_students_random)",
"{'Adams House': 160, 'Cabot House': 159, 'Currier House': 171, 'Dunster House': 151, 'Eliot House': 164, 'Kirkland House': 204, 'Leverett House': 155, 'Lowell House': 177, 'Mather House': 172, 'Pforzheimer House': 156, 'Quincy House': 151, 'Winthrop House': 180}\n"
]
],
[
[
"<div style='background: #CBE0A4; padding: 10px; border: thin solid darblue; border-radius: 5px'>Next, let's check by how much we were off for each house with our random approach. ",
"_____no_output_____"
]
],
[
[
"def house_assignment_check(assignements_per_house_dict):\n \n # your code here\n \n for house in house_names:\n difference = vacant_rooms[house] - assignements_per_house_dict[house]\n\n if difference < 0:\n print(f'there is a problem with {house}. We have assigned {abs(difference)} too many students')\n else:\n print(f'there is no problemwith {house}')",
"_____no_output_____"
],
[
"house_assignment_check(assigned_students_random)",
"there is no problemwith Adams House\nthere is a problem with Cabot House. We have assigned 10 too many students\nthere is a problem with Currier House. We have assigned 32 too many students\nthere is a problem with Dunster House. We have assigned 19 too many students\nthere is a problem with Eliot House. We have assigned 2 too many students\nthere is a problem with Kirkland House. We have assigned 28 too many students\nthere is no problemwith Leverett House\nthere is no problemwith Lowell House\nthere is no problemwith Mather House\nthere is no problemwith Pforzheimer House\nthere is no problemwith Quincy House\nthere is no problemwith Winthrop House\n"
]
],
[
[
"<div style='background: #CBE0A4; padding: 10px; border: thin solid darblue; border-radius: 5px'>Finally, let's write a function that assignes houses more carefully. We can still generate random integers to assign a house, but we need to check whether that house still has capacity. For that reason, please create a function called assign_house() that generates not only a random number, but also checks whether that number is valid, i.e. if that house still has capacity. If there's no capacity, that function should call itself again until it generates a house (id) that still has capacity.",
"_____no_output_____"
]
],
[
[
"vacant_rooms = {'Adams House': 175,\n 'Cabot House': 149,\n 'Currier House': 139, \n 'Dunster House': 132, \n 'Eliot House': 162, \n 'Kirkland House': 176, 'Leverett House': 247, 'Lowell House': 208, 'Mather House': 209, 'Pforzheimer House': 165, 'Quincy House': 167, 'Winthrop House': 190}",
"_____no_output_____"
],
[
"# solution\n\ndef assign_house():\n \n house_id = random.randint(0,11)\n house_name = id_to_house[house_id]\n \n if vacant_rooms[house_name] > 0:\n vacant_rooms[house_name] -= 1 \n return house_id\n \n else:\n return assign_house()\n \n \n \n ",
"_____no_output_____"
],
[
"# next, load the students.csv file, read it row by row and use the \"assign_house()\" function to generate a house for each student.\n\n# your code here\n\n\n# your code here\nfile_read = open(\"data/students.csv\", \"r\")\nreader = csv.reader(file_read)\n\n# write - notice that the file doesn't have to exist beforehand! csv write will create the file automatically, which is very useful!\nfile_write = open('students_with_house_correct_capaity.csv', 'w', newline='')\nwriter = csv.writer(file_write)\n\nfor row in reader:\n \n student = row\n correct_house = assign_house()\n student.append(correct_house)\n writer.writerow(student)\n\nfile_write.close()\n \n ",
"_____no_output_____"
],
[
"# finally, check whether your new solution is working flawlesly\n\n# your code here\nfile_read = open(\"students_with_house_correct_capaity.csv\", \"r\")\nreader = csv.reader(file_read)\n\nfor row in reader:\n print(row)\n break",
"['Oona', 'Swigart', '16076420', '[email protected]', '6']\n"
],
[
"assigned_students_per_house = calculate_students_per_house('students_with_house_correct_capaity.csv')\nassigned_students_per_house",
"_____no_output_____"
],
[
"vacant_rooms = {'Adams House': 175,\n 'Cabot House': 149,\n 'Currier House': 139, \n 'Dunster House': 132, \n 'Eliot House': 162, \n 'Kirkland House': 176, \n 'Leverett House': 247,\n 'Lowell House': 208,\n 'Mather House': 209, \n 'Pforzheimer House': 165,\n 'Quincy House': 167, \n 'Winthrop House': 190}",
"_____no_output_____"
],
[
"house_assignment_check(assigned_students_per_house)",
"there is no problemwith Adams House\nthere is no problemwith Cabot House\nthere is no problemwith Currier House\nthere is no problemwith Dunster House\nthere is no problemwith Eliot House\nthere is no problemwith Kirkland House\nthere is no problemwith Leverett House\nthere is no problemwith Lowell House\nthere is no problemwith Mather House\nthere is no problemwith Pforzheimer House\nthere is no problemwith Quincy House\nthere is no problemwith Winthrop House\n"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown",
"markdown",
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code"
]
] |
cbda4350244faf0540eb8a553dc1dbcebcdd389f
| 2,296 |
ipynb
|
Jupyter Notebook
|
spear/labeling/testing/lf_set_test.ipynb
|
ayushbits/spear
|
22d55b5f707c01321a1ebe39aa5d2d554ea8e017
|
[
"MIT"
] | 1 |
2021-11-08T07:52:03.000Z
|
2021-11-08T07:52:03.000Z
|
spear/labeling/testing/lf_set_test.ipynb
|
ayushbits/spear
|
22d55b5f707c01321a1ebe39aa5d2d554ea8e017
|
[
"MIT"
] | null | null | null |
spear/labeling/testing/lf_set_test.ipynb
|
ayushbits/spear
|
22d55b5f707c01321a1ebe39aa5d2d554ea8e017
|
[
"MIT"
] | null | null | null | 29.435897 | 105 | 0.651568 |
[
[
[
"empty"
]
]
] |
[
"empty"
] |
[
[
"empty"
]
] |
cbda60c6c1542fb2b58c2251ccbcffcf1562bc44
| 42,029 |
ipynb
|
Jupyter Notebook
|
docs/notebooks/vmapped_log_probs.ipynb
|
lucyfan7266/jax
|
36fea96185a3c222e1909be2704875625b422f27
|
[
"ECL-2.0",
"Apache-2.0"
] | 1 |
2021-03-06T16:44:45.000Z
|
2021-03-06T16:44:45.000Z
|
docs/notebooks/vmapped_log_probs.ipynb
|
lucyfan7266/jax
|
36fea96185a3c222e1909be2704875625b422f27
|
[
"ECL-2.0",
"Apache-2.0"
] | null | null | null |
docs/notebooks/vmapped_log_probs.ipynb
|
lucyfan7266/jax
|
36fea96185a3c222e1909be2704875625b422f27
|
[
"ECL-2.0",
"Apache-2.0"
] | null | null | null | 70.048333 | 26,042 | 0.800233 |
[
[
[
"# Autobatching log-densities example\n\n[](https://colab.sandbox.google.com/github/google/jax/blob/master/docs/notebooks/vmapped_log_probs.ipynb)\n\nThis notebook demonstrates a simple Bayesian inference example where autobatching makes user code easier to write, easier to read, and less likely to include bugs.\n\nInspired by a notebook by @davmre.",
"_____no_output_____"
]
],
[
[
"import functools\nimport itertools\nimport re\nimport sys\nimport time\n\nfrom matplotlib.pyplot import *\n\nimport jax\n\nfrom jax import lax\nimport jax.numpy as jnp\nimport jax.scipy as jsp\nfrom jax import random\n\nimport numpy as np\nimport scipy as sp",
"_____no_output_____"
]
],
[
[
"## Generate a fake binary classification dataset",
"_____no_output_____"
]
],
[
[
"np.random.seed(10009)\n\nnum_features = 10\nnum_points = 100\n\ntrue_beta = np.random.randn(num_features).astype(jnp.float32)\nall_x = np.random.randn(num_points, num_features).astype(jnp.float32)\ny = (np.random.rand(num_points) < sp.special.expit(all_x.dot(true_beta))).astype(jnp.int32)",
"_____no_output_____"
],
[
"y",
"_____no_output_____"
]
],
[
[
"## Write the log-joint function for the model\n\nWe'll write a non-batched version, a manually batched version, and an autobatched version.",
"_____no_output_____"
],
[
"### Non-batched",
"_____no_output_____"
]
],
[
[
"def log_joint(beta):\n result = 0.\n # Note that no `axis` parameter is provided to `jnp.sum`.\n result = result + jnp.sum(jsp.stats.norm.logpdf(beta, loc=0., scale=1.))\n result = result + jnp.sum(-jnp.log(1 + jnp.exp(-(2*y-1) * jnp.dot(all_x, beta))))\n return result",
"_____no_output_____"
],
[
"log_joint(np.random.randn(num_features))",
"_____no_output_____"
],
[
"# This doesn't work, because we didn't write `log_prob()` to handle batching.\ntry:\n batch_size = 10\n batched_test_beta = np.random.randn(batch_size, num_features)\n\n log_joint(np.random.randn(batch_size, num_features))\nexcept ValueError as e:\n print(\"Caught expected exception \" + str(e))",
"Caught expected exception Incompatible shapes for broadcasting: ((100, 10), (1, 100))\n"
]
],
[
[
"### Manually batched",
"_____no_output_____"
]
],
[
[
"def batched_log_joint(beta):\n result = 0.\n # Here (and below) `sum` needs an `axis` parameter. At best, forgetting to set axis\n # or setting it incorrectly yields an error; at worst, it silently changes the\n # semantics of the model.\n result = result + jnp.sum(jsp.stats.norm.logpdf(beta, loc=0., scale=1.),\n axis=-1)\n # Note the multiple transposes. Getting this right is not rocket science,\n # but it's also not totally mindless. (I didn't get it right on the first\n # try.)\n result = result + jnp.sum(-jnp.log(1 + jnp.exp(-(2*y-1) * jnp.dot(all_x, beta.T).T)),\n axis=-1)\n return result",
"_____no_output_____"
],
[
"batch_size = 10\nbatched_test_beta = np.random.randn(batch_size, num_features)\n\nbatched_log_joint(batched_test_beta)",
"_____no_output_____"
]
],
[
[
"### Autobatched with vmap\n\nIt just works.",
"_____no_output_____"
]
],
[
[
"vmap_batched_log_joint = jax.vmap(log_joint)\nvmap_batched_log_joint(batched_test_beta)",
"_____no_output_____"
]
],
[
[
"## Self-contained variational inference example\n\nA little code is copied from above.",
"_____no_output_____"
],
[
"### Set up the (batched) log-joint function",
"_____no_output_____"
]
],
[
[
"@jax.jit\ndef log_joint(beta):\n result = 0.\n # Note that no `axis` parameter is provided to `jnp.sum`.\n result = result + jnp.sum(jsp.stats.norm.logpdf(beta, loc=0., scale=10.))\n result = result + jnp.sum(-jnp.log(1 + jnp.exp(-(2*y-1) * jnp.dot(all_x, beta))))\n return result\n\nbatched_log_joint = jax.jit(jax.vmap(log_joint))",
"_____no_output_____"
]
],
[
[
"### Define the ELBO and its gradient",
"_____no_output_____"
]
],
[
[
"def elbo(beta_loc, beta_log_scale, epsilon):\n beta_sample = beta_loc + jnp.exp(beta_log_scale) * epsilon\n return jnp.mean(batched_log_joint(beta_sample), 0) + jnp.sum(beta_log_scale - 0.5 * np.log(2*np.pi))\n \nelbo = jax.jit(elbo)\nelbo_val_and_grad = jax.jit(jax.value_and_grad(elbo, argnums=(0, 1)))",
"_____no_output_____"
]
],
[
[
"### Optimize the ELBO using SGD",
"_____no_output_____"
]
],
[
[
"def normal_sample(key, shape):\n \"\"\"Convenience function for quasi-stateful RNG.\"\"\"\n new_key, sub_key = random.split(key)\n return new_key, random.normal(sub_key, shape)\n\nnormal_sample = jax.jit(normal_sample, static_argnums=(1,))\n\nkey = random.PRNGKey(10003)\n\nbeta_loc = jnp.zeros(num_features, jnp.float32)\nbeta_log_scale = jnp.zeros(num_features, jnp.float32)\n\nstep_size = 0.01\nbatch_size = 128\nepsilon_shape = (batch_size, num_features)\nfor i in range(1000):\n key, epsilon = normal_sample(key, epsilon_shape)\n elbo_val, (beta_loc_grad, beta_log_scale_grad) = elbo_val_and_grad(\n beta_loc, beta_log_scale, epsilon)\n beta_loc += step_size * beta_loc_grad\n beta_log_scale += step_size * beta_log_scale_grad\n if i % 10 == 0:\n print('{}\\t{}'.format(i, elbo_val))",
"0\t-180.85391235351562\n10\t-113.06047058105469\n20\t-102.73725891113281\n30\t-99.78732299804688\n40\t-98.90898895263672\n50\t-98.29743957519531\n60\t-98.18630981445312\n70\t-97.5797348022461\n80\t-97.28599548339844\n90\t-97.46998596191406\n100\t-97.47715759277344\n110\t-97.5806884765625\n120\t-97.49433898925781\n130\t-97.50270080566406\n140\t-96.86398315429688\n150\t-97.44197082519531\n160\t-97.06938934326172\n170\t-96.84031677246094\n180\t-97.21339416503906\n190\t-97.56500244140625\n200\t-97.26395416259766\n210\t-97.11984252929688\n220\t-97.39595794677734\n230\t-97.16830444335938\n240\t-97.11840057373047\n250\t-97.24346160888672\n260\t-97.29786682128906\n270\t-96.69286346435547\n280\t-96.96443176269531\n290\t-97.3005599975586\n300\t-96.63589477539062\n310\t-97.0351791381836\n320\t-97.52906799316406\n330\t-97.2880630493164\n340\t-97.07324981689453\n350\t-97.15620422363281\n360\t-97.25880432128906\n370\t-97.19515228271484\n380\t-97.13092803955078\n390\t-97.11730194091797\n400\t-96.93872833251953\n410\t-97.26676940917969\n420\t-97.35321044921875\n430\t-97.2100830078125\n440\t-97.28434753417969\n450\t-97.16310119628906\n460\t-97.26123809814453\n470\t-97.21342468261719\n480\t-97.23995971679688\n490\t-97.1491470336914\n500\t-97.23527526855469\n510\t-96.93415832519531\n520\t-97.21208190917969\n530\t-96.82577514648438\n540\t-97.01283264160156\n550\t-96.9417724609375\n560\t-97.16526794433594\n570\t-97.29165649414062\n580\t-97.42940521240234\n590\t-97.24371337890625\n600\t-97.15219116210938\n610\t-97.4984359741211\n620\t-96.99072265625\n630\t-96.88955688476562\n640\t-96.89968872070312\n650\t-97.13794708251953\n660\t-97.43705749511719\n670\t-96.99232482910156\n680\t-97.15624237060547\n690\t-97.1869125366211\n700\t-97.1115951538086\n710\t-97.78104400634766\n720\t-97.23224639892578\n730\t-97.16204071044922\n740\t-96.99580383300781\n750\t-96.66720581054688\n760\t-97.16795349121094\n770\t-97.51432037353516\n780\t-97.28899383544922\n790\t-96.91226959228516\n800\t-97.17100524902344\n810\t-97.29046630859375\n820\t-97.16242980957031\n830\t-97.19109344482422\n840\t-97.5638427734375\n850\t-97.00192260742188\n860\t-96.86555480957031\n870\t-96.76338195800781\n880\t-96.83660125732422\n890\t-97.121826171875\n900\t-97.09553527832031\n910\t-97.06825256347656\n920\t-97.1194839477539\n930\t-96.87931823730469\n940\t-97.45622253417969\n950\t-96.69277954101562\n960\t-97.29376983642578\n970\t-97.33528137207031\n980\t-97.349609375\n990\t-97.09675598144531\n"
]
],
[
[
"### Display the results\n\nCoverage isn't quite as good as we might like, but it's not bad, and nobody said variational inference was exact.",
"_____no_output_____"
]
],
[
[
"figure(figsize=(7, 7))\nplot(true_beta, beta_loc, '.', label='Approximated Posterior Means')\nplot(true_beta, beta_loc + 2*jnp.exp(beta_log_scale), 'r.', label='Approximated Posterior $2\\sigma$ Error Bars')\nplot(true_beta, beta_loc - 2*jnp.exp(beta_log_scale), 'r.')\nplot_scale = 3\nplot([-plot_scale, plot_scale], [-plot_scale, plot_scale], 'k')\nxlabel('True beta')\nylabel('Estimated beta')\nlegend(loc='best')",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
]
] |
cbda80397bf22d1585363f7391befb41f833e1ea
| 23,673 |
ipynb
|
Jupyter Notebook
|
playground.ipynb
|
Mokusama/sec_analyse
|
9736b843f81f355f27c6390d7d69491df8f94200
|
[
"MIT"
] | null | null | null |
playground.ipynb
|
Mokusama/sec_analyse
|
9736b843f81f355f27c6390d7d69491df8f94200
|
[
"MIT"
] | null | null | null |
playground.ipynb
|
Mokusama/sec_analyse
|
9736b843f81f355f27c6390d7d69491df8f94200
|
[
"MIT"
] | null | null | null | 81.071918 | 11,784 | 0.783171 |
[
[
[
"a = ['aa','bb','cc']",
"_____no_output_____"
],
[
"q =0 \nfor i in a:\n instr = f'def {i}():print({q})'\n q+= 1\n instr2 = f'{i}()'\n exec(instr)\n exec(instr2)",
"0\n1\n2\n"
],
[
"aa()",
"0\n"
],
[
"bb()",
"1\n"
],
[
"import matplotlib.pyplot as plt\nimport numpy as np\nimport nltk\nimport pandas as pd\nimport mytoolkit as mt",
"_____no_output_____"
],
[
"pp1 =mt.pandas_read_sec_csv('/Users/wangmu/Documents/Science/mG1/数据/sec/20190624/6D-11631.CSV')\n\n\n",
"_____no_output_____"
],
[
"pp1[pp1.time[2.5:5.0]]",
"_____no_output_____"
],
[
"x = 10\nexpr = \"\"\"\nz = 30\nsum = x + y + z\nprint(sum)\n\"\"\"\ndef func():\n y = 20\n exec(expr)\n exec(expr, {'x': 1, 'y': 2})\n exec(expr, {'x': 1, 'y': 2}, {'y': 3, 'z': 4})\n \nfunc()\n",
"60\n33\n34\n"
],
[
"import mytoolkit as mt\nimport matplotlib.pyplot as plt",
"_____no_output_____"
],
[
"pp1 =mt.pandas_read_sec_csv('/Users/wangmu/Documents/Science/mG1/数据/sec/20190624/6D-11631.CSV')\n",
"_____no_output_____"
],
[
"pos = pp1.where((pp1.time<5.0)&(pp1.time>0.0))",
"_____no_output_____"
],
[
"type((pp1.time<1.0)&(pp1.time>0.5))",
"_____no_output_____"
],
[
"mt.normalize_peak2zero(pos)\nplt.plot(pos.time,pos.peak)",
"_____no_output_____"
],
[
"pos = mt.pick_x_region",
"_____no_output_____"
],
[
"pos.peak.max()",
"_____no_output_____"
]
]
] |
[
"code"
] |
[
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
]
] |
cbda813be0e6e1dbf96e12d657d19b402fabb5e8
| 271,075 |
ipynb
|
Jupyter Notebook
|
PCA_Cary5000_deepUV.ipynb
|
juliaaaking/urea_machine
|
e5783c32bd74ea997d72695a635c5f1fb52e3a17
|
[
"MIT"
] | null | null | null |
PCA_Cary5000_deepUV.ipynb
|
juliaaaking/urea_machine
|
e5783c32bd74ea997d72695a635c5f1fb52e3a17
|
[
"MIT"
] | null | null | null |
PCA_Cary5000_deepUV.ipynb
|
juliaaaking/urea_machine
|
e5783c32bd74ea997d72695a635c5f1fb52e3a17
|
[
"MIT"
] | null | null | null | 653.192771 | 74,992 | 0.95205 |
[
[
[
"# PCA with Cary5000 data for deep-UV spectra (190-300 nm)",
"_____no_output_____"
]
],
[
[
"# Import packages\n\nimport matplotlib.pyplot as plt\nimport numpy as np\nimport pandas as pd\nimport seaborn as sns\n\nfrom mpl_toolkits import mplot3d\n\nfrom sklearn.decomposition import PCA\nfrom sklearn.preprocessing import StandardScaler\n\nplt.style.use('ggplot')\n\n# Set seed\nseed = 4",
"_____no_output_____"
],
[
"# Import data\ndata = pd.read_csv('Datasets/urea_saline_cary5000.csv')",
"_____no_output_____"
],
[
"# Define features and targets\nX = data.drop(data.columns[0:2204], axis=1)\ny = data['Urea Concentration (mM)']",
"_____no_output_____"
],
[
"# Normalize data\nsc = StandardScaler()\nX = sc.fit_transform(X)",
"_____no_output_____"
],
[
"# Do PCA\npca = PCA(n_components=10, random_state=seed)\nX_pca = pca.fit_transform(X)\n\nprint(\"Variance explained by all 10 PC's =\", sum(pca.explained_variance_ratio_ *100))",
"Variance explained by all 10 PC's = 99.99857806849782\n"
],
[
"# Elbow Plot\nplt.plot(np.cumsum(pca.explained_variance_ratio_), color='blueviolet')\nplt.xlabel('Number of components')\nplt.ylabel('Explained variance (%)')\nplt.savefig('elbow_plot.png', dpi=100)",
"_____no_output_____"
],
[
"np.cumsum(pca.explained_variance_ratio_)",
"_____no_output_____"
],
[
"# If we apply PCA with n_components=2\npca_2 = PCA(n_components=2, random_state=seed)\nX_pca_2 = pca_2.fit_transform(X)\n\n\n# Plot it\nplt.figure(figsize=(10, 7))\nsns.scatterplot(x=X_pca_2[:, 0], y=X_pca_2[:, 1], s=70,\n hue=y, palette='viridis')\n\nplt.title('2D Scatterplot: 95.5% of the variability captured', pad=15)\nplt.xlabel('First prinicipal component')\nplt.ylabel('Second principal component')",
"_____no_output_____"
],
[
"# Plot it in 3D\npca_3 = PCA(n_components=3, random_state=seed)\nX_pca_3 = pca_3.fit_transform(X)\n\nfig = plt.figure(figsize = (12, 8))\nax = plt.axes(projection='3d')\n\nsctt = ax.scatter3D(X_pca_3[:, 0], X_pca_3[:, 1], X_pca_3[:, 2],\n c = y, s=50, alpha=0.6)\n\nplt.title('3D Scatterplot: 98.0% of the variability captured', pad=15)\nax.set_xlabel('First principal component')\nax.set_ylabel('Second principal component')\nax.set_zlabel('Third principal component')\nplt.savefig('3d_scatterplot.png')",
"_____no_output_____"
]
],
[
[
"## Drop outliers - data from 02/11/2022",
"_____no_output_____"
]
],
[
[
"data = pd.read_csv('Datasets/urea_saline_cary5000.csv')\ndata = data.drop(data.index[17:29])\ndata.reset_index(inplace=True)",
"_____no_output_____"
],
[
"# Define features and targets\nX = data.drop(data.columns[0:2205], axis=1)\ny = data['Urea Concentration (mM)']\n\n# Normalize data\nsc = StandardScaler()\nX = sc.fit_transform(X)\n\n# Do PCA\npca = PCA(n_components=10, random_state=seed)\nX_pca = pca.fit_transform(X)\n\nprint(\"Variance explained by all 10 PC's =\", sum(pca.explained_variance_ratio_ *100))\n\n# Elbow Plot\nplt.plot(np.cumsum(pca.explained_variance_ratio_), color='blueviolet')\nplt.xlabel('Number of components')\nplt.ylabel('Explained variance (%)')\nplt.savefig('elbow_plot.png', dpi=100)",
"Variance explained by all 10 PC's = 99.99954059859198\n"
],
[
"np.cumsum(pca.explained_variance_ratio_)",
"_____no_output_____"
],
[
"# If we apply PCA with n_components=2\npca_2 = PCA(n_components=2, random_state=seed)\nX_pca_2 = pca_2.fit_transform(X)\n\n\n# Plot it\nplt.figure(figsize=(10, 7))\nsns.scatterplot(x=X_pca_2[:, 0], y=X_pca_2[:, 1], s=70,\n hue=y, palette='viridis')\n\nplt.title('2D Scatterplot: 96.4% of the variability captured', pad=15)\nplt.xlabel('First prinicipal component')\nplt.ylabel('Second principal component')",
"_____no_output_____"
],
[
"# Plot it in 3D\npca_3 = PCA(n_components=3, random_state=seed)\nX_pca_3 = pca_3.fit_transform(X)\n\nfig = plt.figure(figsize = (12, 8))\nax = plt.axes(projection='3d')\n\nsctt = ax.scatter3D(X_pca_3[:, 0], X_pca_3[:, 1], X_pca_3[:, 2],\n c = y, s=50, alpha=0.6)\n\nplt.title('3D Scatterplot: 98.4% of the variability captured', pad=15)\nax.set_xlabel('First principal component')\nax.set_ylabel('Second principal component')\nax.set_zlabel('Third principal component')\nplt.savefig('3d_scatterplot.png')",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code"
]
] |
cbda8cb8bd0d7314110dbfbba1169b817f392c70
| 22,719 |
ipynb
|
Jupyter Notebook
|
intro-to-pytorch/Part 1 - Tensors in PyTorch (Exercises).ipynb
|
agungsantoso/deep-learning-v2-pytorch
|
3b244dc5c77ec76ecd0b1ad849c46262d652ec08
|
[
"MIT"
] | 3 |
2019-04-16T14:31:13.000Z
|
2020-09-02T02:23:41.000Z
|
intro-to-pytorch/Part 1 - Tensors in PyTorch (Exercises).ipynb
|
agungsantoso/deep-learning-v2-pytorch
|
3b244dc5c77ec76ecd0b1ad849c46262d652ec08
|
[
"MIT"
] | null | null | null |
intro-to-pytorch/Part 1 - Tensors in PyTorch (Exercises).ipynb
|
agungsantoso/deep-learning-v2-pytorch
|
3b244dc5c77ec76ecd0b1ad849c46262d652ec08
|
[
"MIT"
] | 5 |
2019-09-26T15:05:08.000Z
|
2021-12-16T11:00:33.000Z
| 39.857895 | 678 | 0.537568 |
[
[
[
"<a href=\"https://colab.research.google.com/github/agungsantoso/deep-learning-v2-pytorch/blob/master/intro-to-pytorch/Part%201%20-%20Tensors%20in%20PyTorch%20(Exercises).ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>",
"_____no_output_____"
],
[
"# Introduction to Deep Learning with PyTorch\n\nIn this notebook, you'll get introduced to [PyTorch](http://pytorch.org/), a framework for building and training neural networks. PyTorch in a lot of ways behaves like the arrays you love from Numpy. These Numpy arrays, after all, are just tensors. PyTorch takes these tensors and makes it simple to move them to GPUs for the faster processing needed when training neural networks. It also provides a module that automatically calculates gradients (for backpropagation!) and another module specifically for building neural networks. All together, PyTorch ends up being more coherent with Python and the Numpy/Scipy stack compared to TensorFlow and other frameworks.\n\n",
"_____no_output_____"
],
[
"## Neural Networks\n\nDeep Learning is based on artificial neural networks which have been around in some form since the late 1950s. The networks are built from individual parts approximating neurons, typically called units or simply \"neurons.\" Each unit has some number of weighted inputs. These weighted inputs are summed together (a linear combination) then passed through an activation function to get the unit's output.\n\n<img src=\"https://github.com/agungsantoso/deep-learning-v2-pytorch/blob/master/intro-to-pytorch/assets/simple_neuron.png?raw=1\" width=400px>\n\nMathematically this looks like: \n\n$$\n\\begin{align}\ny &= f(w_1 x_1 + w_2 x_2 + b) \\\\\ny &= f\\left(\\sum_i w_i x_i +b \\right)\n\\end{align}\n$$\n\nWith vectors this is the dot/inner product of two vectors:\n\n$$\nh = \\begin{bmatrix}\nx_1 \\, x_2 \\cdots x_n\n\\end{bmatrix}\n\\cdot \n\\begin{bmatrix}\n w_1 \\\\\n w_2 \\\\\n \\vdots \\\\\n w_n\n\\end{bmatrix}\n$$",
"_____no_output_____"
],
[
"## Tensors\n\nIt turns out neural network computations are just a bunch of linear algebra operations on *tensors*, a generalization of matrices. A vector is a 1-dimensional tensor, a matrix is a 2-dimensional tensor, an array with three indices is a 3-dimensional tensor (RGB color images for example). The fundamental data structure for neural networks are tensors and PyTorch (as well as pretty much every other deep learning framework) is built around tensors.\n\n<img src=\"https://github.com/agungsantoso/deep-learning-v2-pytorch/blob/master/intro-to-pytorch/assets/tensor_examples.svg?raw=1\" width=600px>\n\nWith the basics covered, it's time to explore how we can use PyTorch to build a simple neural network.",
"_____no_output_____"
]
],
[
[
"# http://pytorch.org/\nfrom os.path import exists\nfrom wheel.pep425tags import get_abbr_impl, get_impl_ver, get_abi_tag\nplatform = '{}{}-{}'.format(get_abbr_impl(), get_impl_ver(), get_abi_tag())\ncuda_output = !ldconfig -p|grep cudart.so|sed -e 's/.*\\.\\([0-9]*\\)\\.\\([0-9]*\\)$/cu\\1\\2/'\naccelerator = cuda_output[0] if exists('/dev/nvidia0') else 'cpu'\n\n!pip install -q http://download.pytorch.org/whl/{accelerator}/torch-0.4.1-{platform}-linux_x86_64.whl torchvision\nimport torch",
"_____no_output_____"
],
[
"# First, import PyTorch\nimport torch",
"_____no_output_____"
],
[
"def activation(x):\n \"\"\" Sigmoid activation function \n \n Arguments\n ---------\n x: torch.Tensor\n \"\"\"\n return 1/(1+torch.exp(-x))",
"_____no_output_____"
],
[
"### Generate some data\ntorch.manual_seed(7) # Set the random seed so things are predictable\n\n# Features are 3 random normal variables\nfeatures = torch.randn((1, 5))\n# True weights for our data, random normal variables again\nweights = torch.randn_like(features)\n# and a true bias term\nbias = torch.randn((1, 1))",
"_____no_output_____"
]
],
[
[
"Above I generated data we can use to get the output of our simple network. This is all just random for now, going forward we'll start using normal data. Going through each relevant line:\n\n`features = torch.randn((1, 5))` creates a tensor with shape `(1, 5)`, one row and five columns, that contains values randomly distributed according to the normal distribution with a mean of zero and standard deviation of one. \n\n`weights = torch.randn_like(features)` creates another tensor with the same shape as `features`, again containing values from a normal distribution.\n\nFinally, `bias = torch.randn((1, 1))` creates a single value from a normal distribution.\n\nPyTorch tensors can be added, multiplied, subtracted, etc, just like Numpy arrays. In general, you'll use PyTorch tensors pretty much the same way you'd use Numpy arrays. They come with some nice benefits though such as GPU acceleration which we'll get to later. For now, use the generated data to calculate the output of this simple single layer network. \n> **Exercise**: Calculate the output of the network with input features `features`, weights `weights`, and bias `bias`. Similar to Numpy, PyTorch has a [`torch.sum()`](https://pytorch.org/docs/stable/torch.html#torch.sum) function, as well as a `.sum()` method on tensors, for taking sums. Use the function `activation` defined above as the activation function.",
"_____no_output_____"
]
],
[
[
"## Calculate the output of this network using the weights and bias tensors\ny = activation(torch.sum(features * weights) + bias)",
"_____no_output_____"
]
],
[
[
"You can do the multiplication and sum in the same operation using a matrix multiplication. In general, you'll want to use matrix multiplications since they are more efficient and accelerated using modern libraries and high-performance computing on GPUs.\n\nHere, we want to do a matrix multiplication of the features and the weights. For this we can use [`torch.mm()`](https://pytorch.org/docs/stable/torch.html#torch.mm) or [`torch.matmul()`](https://pytorch.org/docs/stable/torch.html#torch.matmul) which is somewhat more complicated and supports broadcasting. If we try to do it with `features` and `weights` as they are, we'll get an error\n\n```python\n>> torch.mm(features, weights)\n\n---------------------------------------------------------------------------\nRuntimeError Traceback (most recent call last)\n<ipython-input-13-15d592eb5279> in <module>()\n----> 1 torch.mm(features, weights)\n\nRuntimeError: size mismatch, m1: [1 x 5], m2: [1 x 5] at /Users/soumith/minicondabuild3/conda-bld/pytorch_1524590658547/work/aten/src/TH/generic/THTensorMath.c:2033\n```\n\nAs you're building neural networks in any framework, you'll see this often. Really often. What's happening here is our tensors aren't the correct shapes to perform a matrix multiplication. Remember that for matrix multiplications, the number of columns in the first tensor must equal to the number of rows in the second column. Both `features` and `weights` have the same shape, `(1, 5)`. This means we need to change the shape of `weights` to get the matrix multiplication to work.\n\n**Note:** To see the shape of a tensor called `tensor`, use `tensor.shape`. If you're building neural networks, you'll be using this method often.\n\nThere are a few options here: [`weights.reshape()`](https://pytorch.org/docs/stable/tensors.html#torch.Tensor.reshape), [`weights.resize_()`](https://pytorch.org/docs/stable/tensors.html#torch.Tensor.resize_), and [`weights.view()`](https://pytorch.org/docs/stable/tensors.html#torch.Tensor.view).\n\n* `weights.reshape(a, b)` will return a new tensor with the same data as `weights` with size `(a, b)` sometimes, and sometimes a clone, as in it copies the data to another part of memory.\n* `weights.resize_(a, b)` returns the same tensor with a different shape. However, if the new shape results in fewer elements than the original tensor, some elements will be removed from the tensor (but not from memory). If the new shape results in more elements than the original tensor, new elements will be uninitialized in memory. Here I should note that the underscore at the end of the method denotes that this method is performed **in-place**. Here is a great forum thread to [read more about in-place operations](https://discuss.pytorch.org/t/what-is-in-place-operation/16244) in PyTorch.\n* `weights.view(a, b)` will return a new tensor with the same data as `weights` with size `(a, b)`.\n\nI usually use `.view()`, but any of the three methods will work for this. So, now we can reshape `weights` to have five rows and one column with something like `weights.view(5, 1)`.\n\n> **Exercise**: Calculate the output of our little network using matrix multiplication.",
"_____no_output_____"
]
],
[
[
"## Calculate the output of this network using matrix multiplication\ny = activation(torch.mm(features, weights.view(5, 1)) + bias)",
"_____no_output_____"
]
],
[
[
"### Stack them up!\n\nThat's how you can calculate the output for a single neuron. The real power of this algorithm happens when you start stacking these individual units into layers and stacks of layers, into a network of neurons. The output of one layer of neurons becomes the input for the next layer. With multiple input units and output units, we now need to express the weights as a matrix.\n\n<img src='https://github.com/agungsantoso/deep-learning-v2-pytorch/blob/master/intro-to-pytorch/assets/multilayer_diagram_weights.png?raw=1' width=450px>\n\nThe first layer shown on the bottom here are the inputs, understandably called the **input layer**. The middle layer is called the **hidden layer**, and the final layer (on the right) is the **output layer**. We can express this network mathematically with matrices again and use matrix multiplication to get linear combinations for each unit in one operation. For example, the hidden layer ($h_1$ and $h_2$ here) can be calculated \n\n$$\n\\vec{h} = [h_1 \\, h_2] = \n\\begin{bmatrix}\nx_1 \\, x_2 \\cdots \\, x_n\n\\end{bmatrix}\n\\cdot \n\\begin{bmatrix}\n w_{11} & w_{12} \\\\\n w_{21} &w_{22} \\\\\n \\vdots &\\vdots \\\\\n w_{n1} &w_{n2}\n\\end{bmatrix}\n$$\n\nThe output for this small network is found by treating the hidden layer as inputs for the output unit. The network output is expressed simply\n\n$$\ny = f_2 \\! \\left(\\, f_1 \\! \\left(\\vec{x} \\, \\mathbf{W_1}\\right) \\mathbf{W_2} \\right)\n$$",
"_____no_output_____"
]
],
[
[
"### Generate some data\ntorch.manual_seed(7) # Set the random seed so things are predictable\n\n# Features are 3 random normal variables\nfeatures = torch.randn((1, 3))\n\n# Define the size of each layer in our network\nn_input = features.shape[1] # Number of input units, must match number of input features\nn_hidden = 2 # Number of hidden units \nn_output = 1 # Number of output units\n\n# Weights for inputs to hidden layer\nW1 = torch.randn(n_input, n_hidden)\n# Weights for hidden layer to output layer\nW2 = torch.randn(n_hidden, n_output)\n\n# and bias terms for hidden and output layers\nB1 = torch.randn((1, n_hidden))\nB2 = torch.randn((1, n_output))",
"_____no_output_____"
]
],
[
[
"> **Exercise:** Calculate the output for this multi-layer network using the weights `W1` & `W2`, and the biases, `B1` & `B2`. ",
"_____no_output_____"
]
],
[
[
"## Your solution here\nh = activation(torch.mm(features, W1) + B1)\noutput = activation(torch.mm(h, W2) + B2)\noutput",
"_____no_output_____"
]
],
[
[
"If you did this correctly, you should see the output `tensor([[ 0.3171]])`.\n\nThe number of hidden units a parameter of the network, often called a **hyperparameter** to differentiate it from the weights and biases parameters. As you'll see later when we discuss training a neural network, the more hidden units a network has, and the more layers, the better able it is to learn from data and make accurate predictions.",
"_____no_output_____"
],
[
"## Numpy to Torch and back\n\nSpecial bonus section! PyTorch has a great feature for converting between Numpy arrays and Torch tensors. To create a tensor from a Numpy array, use `torch.from_numpy()`. To convert a tensor to a Numpy array, use the `.numpy()` method.",
"_____no_output_____"
]
],
[
[
"import numpy as np\na = np.random.rand(4,3)\na",
"_____no_output_____"
],
[
"b = torch.from_numpy(a)\nb",
"_____no_output_____"
],
[
"b.numpy()",
"_____no_output_____"
]
],
[
[
"The memory is shared between the Numpy array and Torch tensor, so if you change the values in-place of one object, the other will change as well.",
"_____no_output_____"
]
],
[
[
"# Multiply PyTorch Tensor by 2, in place\nb.mul_(2)",
"_____no_output_____"
],
[
"# Numpy array matches new values from Tensor\na",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
]
] |
cbdaabcbf147b7b96200e99c190f343a72f8434e
| 20,310 |
ipynb
|
Jupyter Notebook
|
assets/assignments/a4-stylegan.ipynb
|
bowangvector/2021
|
f5ec179840b5ccb20238c860de620cff47f508c4
|
[
"MIT"
] | 1 |
2021-09-08T02:24:00.000Z
|
2021-09-08T02:24:00.000Z
|
assets/assignments/a4-stylegan.ipynb
|
bowangvector/2021
|
f5ec179840b5ccb20238c860de620cff47f508c4
|
[
"MIT"
] | 1 |
2021-03-06T21:24:49.000Z
|
2021-03-06T21:24:49.000Z
|
assets/assignments/a4-stylegan.ipynb
|
bowangvector/2021
|
f5ec179840b5ccb20238c860de620cff47f508c4
|
[
"MIT"
] | 6 |
2021-01-21T15:41:30.000Z
|
2021-10-17T19:46:19.000Z
| 45.334821 | 381 | 0.501083 |
[
[
[
"## **University of Toronto - CSC413 - Neural Networks and Deep Learning** \r\n## **Programming Assignment 4 - StyleGAN2-Ada**\r\n\r\nThis is a self-contained notebook that allows you to play around with a pre-trained StyleGAN2-Ada generator\r\n\r\nDisclaimer: Some codes were borrowed from StyleGAN official documentation on Github https://github.com/NVlabs/stylegan\r\n\r\nMake sure to set your runtime to GPU\r\n\r\nRemember to save your progress periodically!",
"_____no_output_____"
]
],
[
[
"# Run this for Google CoLab (use TensorFlow 1.x)\r\n%tensorflow_version 1.x\r\n\r\n# clone StyleGAN2 Ada\r\n!git clone https://github.com/NVlabs/stylegan2-ada.git",
"_____no_output_____"
],
[
"#setup some environments (Do not change any of the following)\r\nimport sys\r\nimport pickle\r\nimport os\r\nimport numpy as np\r\nfrom IPython.display import Image\r\nimport PIL.Image\r\nfrom PIL import Image\r\nimport matplotlib.pyplot as plt\r\n\r\nsys.path.insert(0, \"/content/stylegan2-ada\") #do not remove this line\r\n\r\nimport dnnlib\r\nimport dnnlib.tflib as tflib\r\nimport IPython.display\r\nfrom google.colab import files",
"_____no_output_____"
]
],
[
[
"Next, we will load a pre-trained StyleGan2-ada network.\r\n\r\nEach of the following pre-trained network is specialized to generate one type of image.",
"_____no_output_____"
]
],
[
[
"# The pre-trained networks are stored as standard pickle files\r\n# Uncomment one of the following URL to begin\r\n# If you wish, you can also find other pre-trained networks online\r\n\r\n#URL = \"https://nvlabs-fi-cdn.nvidia.com/stylegan2-ada/pretrained/ffhq.pkl\" # Human faces\r\n#URL = \"https://nvlabs-fi-cdn.nvidia.com/stylegan2-ada/pretrained/cifar10.pkl\" # CIFAR10, these images are a bit too tiny for our experiment \r\n#URL = \"https://nvlabs-fi-cdn.nvidia.com/stylegan2-ada/pretrained/afhqwild.pkl\" # wild animal pictures\r\n#URL = \"https://nvlabs-fi-cdn.nvidia.com/stylegan2-ada/pretrained/metfaces.pkl\" # European portrait paintings\r\n#URL = \"https://nvlabs-fi-cdn.nvidia.com/stylegan2-ada/pretrained/afhqcat.pkl\" # cats\r\n#URL = \"https://nvlabs-fi-cdn.nvidia.com/stylegan2-ada/pretrained/afhqdog.pkl\" # dogs\r\ntflib.init_tf() #this creates a default Tensorflow session\r\n\r\n# we are now going to load the StyleGAN2-Ada model \r\n# The following code downloads the file and unpickles it to yield 3 instances of dnnlib.tflib.Network. \r\nwith dnnlib.util.open_url(URL) as fp:\r\n _G, _D, Gs = pickle.load(fp) \r\n# Here is a brief description of _G, _D, Gs, for details see the official StyleGAN documentation \r\n# _G = Instantaneous snapshot of the generator. Mainly useful for resuming a previous training run.\r\n# _D = Instantaneous snapshot of the discriminator. Mainly useful for resuming a previous training run.\r\n# Gs = Long-term average of the generator. Yields higher-quality results than the instantaneous snapshot.\r\n# We will work with Gs ",
"_____no_output_____"
]
],
[
[
"## Part 1 Sampling and Identifying Fakes \r\n\r\nOpen: https://github.com/NVlabs/stylegan and follow the instructions starting from *There are three ways to use the pre-trained generator....*\r\n\r\nComplete generate_latent_code and generate_images function in the Colab notebook to generate a small row of $3 - 5$ images. \r\n\r\nYou do not need to include these images into your PDF submission. \r\n\r\nIf you wish, you can try to use https://www.whichfaceisreal.com/learn.html as a guideline to spot any imperfections that you detect in these images, e.g., ``blob artifact\" and make a short remark for your attached images. ",
"_____no_output_____"
]
],
[
[
"# Sample a batch of latent codes {z_1, ...., z_B}, B is your batch size. \r\ndef generate_latent_code(SEED, BATCH, LATENT_DIMENSION = 512):\r\n \"\"\"\r\n This function returns a sample a batch of 512 dimensional random latent code\r\n\r\n - SEED: int\r\n - BATCH: int that specifies the number of latent codes, Recommended batch_size is 3 - 6\r\n - LATENT_DIMENSION is by default 512 (see Karras et al.)\r\n \r\n You should use np.random.RandomState to construct a random number generator, say rnd\r\n Then use rnd.randn along with your BATCH and LATENT_DIMENSION to generate your latent codes. \r\n This samples a batch of latent codes from a normal distribution \r\n https://numpy.org/doc/stable/reference/random/generated/numpy.random.RandomState.randn.html\r\n \r\n Return latent_codes, which is a 2D array with dimensions BATCH times LATENT_DIMENSION \r\n \"\"\"\r\n ################################################################################\r\n ########################## COMPLETE THE FOLLOWING ##############################\r\n ################################################################################\r\n latent_codes = ...\r\n ################################################################################\r\n return latent_codes",
"_____no_output_____"
],
[
"# Sample images from your latent codes https://github.com/NVlabs/stylegan\r\n# You can use their default settings\r\n\r\n################################################################################\r\n########################## COMPLETE THE FOLLOWING ##############################\r\n################################################################################\r\ndef generate_images(SEED, BATCH, TRUNCATION = 0.7):\r\n \"\"\"\r\n This function generates a batch of images from latent codes. \r\n \r\n - SEED: int\r\n - BATCH: int that specifies the number of latent codes to be generated\r\n - TRUNCATION: float between [-1, 1] that decides the amount of clipping to apply to the latent code distribution\r\n recommended setting is 0.7\r\n\r\n You will use Gs.run() to sample images. See https://github.com/NVlabs/stylegan for details\r\n You may use their default setting. \r\n \"\"\" \r\n # Sample a batch of latent code z using generate_latent_code function\r\n latent_codes = ...\r\n\r\n # Convert latent code into images by following https://github.com/NVlabs/stylegan\r\n fmt = dict(...) \r\n images = Gs.run(...)\r\n return PIL.Image.fromarray(np.concatenate(images, axis=1) , 'RGB')\r\n################################################################################",
"_____no_output_____"
],
[
"# Generate your images\r\ngenerate_images(...)",
"_____no_output_____"
]
],
[
[
"## **Part 2 Interpolation**\r\n\r\nComplete the interpolate_images function using linear interpolation between two latent codes,\r\n\\begin{equation}\r\n z = r z_1 + (1-r) z_2, r \\in [0, 1]\r\n\\end{equation}\r\nand feeding this interpolation through the StyleGAN2-Ada generator Gs as done in generate_images. Include a small row of interpolation in your PDF submission as a screen shot if necessary to keep the file size small. ",
"_____no_output_____"
]
],
[
[
"################################################################################\r\n########################## COMPLETE THE FOLLOWING ##############################\r\n################################################################################\r\ndef interpolate_images(SEED1, SEED2, INTERPOLATION, BATCH = 1, TRUNCATION = 0.7):\r\n \"\"\"\r\n - SEED1, SEED2: int, seed to use to generate the two latent codes\r\n - INTERPOLATION: int, the number of interpolation between the two images, recommended setting 6 - 10\r\n - BATCH: int, the number of latent code to generate. In this experiment, it is 1. \r\n - TRUNCATION: float between [-1, 1] that decides the amount of clipping to apply to the latent code distribution\r\n recommended setting is 0.7\r\n\r\n You will interpolate between two latent code that you generate using the above formula\r\n You can generate an interpolation variable using np.linspace\r\n https://numpy.org/doc/stable/reference/generated/numpy.linspace.html\r\n \r\n This function should return an interpolated image. Include a screenshot in your submission.\r\n \"\"\"\r\n latent_code_1 = ...\r\n latent_code_2 = ...\r\n images = Gs.run(...)\r\n \r\n return PIL.Image.fromarray(np.concatenate(images, axis=1) , 'RGB')\r\n################################################################################",
"_____no_output_____"
],
[
"# Create an interpolation of your generated images\r\ninterpolate_images(...)",
"_____no_output_____"
]
],
[
[
"After you have generated interpolated images, an interesting task would be to see how you can create a GIF. Feel free to explore a little bit more. ",
"_____no_output_____"
],
[
"## **Part 3 Style Mixing and Fine Control**\r\nIn the final part, you will reproduce the famous style mixing example from the original StyleGAN paper.\r\n\r\n### Step 1. We will first learn how to generate from sub-networks of the StyleGAN generator. ",
"_____no_output_____"
]
],
[
[
"# You will generate images from sub-networks of the StyleGAN generator\r\n# Similar to Gs, the sub-networks are represented as independent instances of dnnlib.tflib.Network\r\n# Complete the function by following \\url{https://github.com/NVlabs/stylegan} \r\n# And Look up Gs.components.mapping, Gs.components.synthesism, Gs.get_var\r\n# Remember to use the truncation trick as described in the handout after you obtain src_dlatents from Gs.components.mapping.run\r\ndef generate_from_subnetwork(src_seeds, LATENT_DIMENSION = 512):\r\n \"\"\"\r\n - src_seeds: a list of int, where each int is used to generate a latent code, e.g., [1,2,3]\r\n - LATENT_DIMENSION: by default 512\r\n\r\n You will complete the code snippet in the Write Your Code Here block\r\n This generates several images from a sub-network of the genrator. \r\n\r\n To prevent mistakes, we have provided the variable names which corresponds to the ones in the StyleGAN documentation\r\n You should use their convention. \r\n \"\"\"\r\n\r\n # default arguments to Gs.components.synthesis.run, this is given to you. \r\n synthesis_kwargs = {\r\n 'output_transform': dict(func=tflib.convert_images_to_uint8, nchw_to_nhwc=True),\r\n 'randomize_noise': False,\r\n 'minibatch_size': 4\r\n }\r\n ############################################################################\r\n ########################## WRITE YOUR CODE HERE ############################\r\n ############################################################################\r\n truncation = ...\r\n src_latents = ...\r\n src_dlatents = ...\r\n w_avg = ...\r\n src_dlatents = ... \r\n all_images = Gs.components.synthesis.run(...)\r\n ############################################################################\r\n return PIL.Image.fromarray(np.concatenate(all_images, axis=1) , 'RGB')",
"_____no_output_____"
],
[
"# generate several iamges from the sub-network\r\ngenerate_from_subnetwork(...)",
"_____no_output_____"
]
],
[
[
"### Step 2. Initialize the col_seeds, row_seeds and col_styles and generate a grid of image. \r\n\r\nA recommended example for your experiment is as follows:\r\n\r\n* col_seeds = [1, 2, 3, 4, 5]\r\n* row_seeds = [6]\r\n* col_styles = [1, 2, 3, 4, 5]\r\n\r\nand\r\n\r\n* col_seeds = [1, 2, 3, 4, 5]\r\n* row_seeds = [6]\r\n* col_styles = [8, 9, 10, 11, 12]\r\n\r\nYou will then incorporate your code from generate from sub_network into the cell below. \r\n\r\nExperiment with the col_styles variable. Explain what col_styles does, for instance, roughly describe what these numbers corresponds to. Create a simple experiment to backup your argument. Include **at maximum two** sets of images that illustrates the effect of changing col_styles and your explanation. Include them as screen shots to minimize the size of the file.\r\n\r\nMake reference to the original StyleGAN or the StyleGAN2 paper by Karras et al. as needed https://arxiv.org/pdf/1812.04948.pdf https://arxiv.org/pdf/1912.04958.pdf",
"_____no_output_____"
]
],
[
[
"################################################################################\r\n####################COMPLETE THE NEXT THREE LINES###############################\r\n################################################################################\r\ncol_seeds = ...\r\nrow_seeds = ...\r\ncol_styles = ...\r\n################################################################################\r\nsrc_seeds = list(set(row_seeds + col_seeds))\r\n\r\n# default arguments to Gs.components.synthesis.run, do not change\r\nsynthesis_kwargs = {\r\n 'output_transform': dict(func=tflib.convert_images_to_uint8, nchw_to_nhwc=True),\r\n 'randomize_noise': False,\r\n 'minibatch_size': 4\r\n}\r\n################################################################################\r\n########################## COMPLETE THE FOLLOWING ##############################\r\n################################################################################\r\n# Copy the #### WRITE YOUR CODE HERE #### portion from generate_from_subnetwork()\r\n\r\nall_images = Gs.components.synthesis.run(...) \r\n################################################################################\r\n\r\n# (Do not change)\r\nimage_dict = {(seed, seed): image for seed, image in zip(src_seeds, list(all_images))}\r\nw_dict = {seed: w for seed, w in zip(src_seeds, list(src_dlatents))} \r\n\r\n# Generating Images (Do not Change)\r\nfor row_seed in row_seeds:\r\n for col_seed in col_seeds:\r\n w = w_dict[row_seed].copy()\r\n w[col_styles] = w_dict[col_seed][col_styles]\r\n image = Gs.components.synthesis.run(w[np.newaxis], **synthesis_kwargs)[0]\r\n image_dict[(row_seed, col_seed)] = image\r\n\r\n# Create an Image Grid (Do not Change)\r\ndef create_grid_images(): \r\n _N, _C, H, W = Gs.output_shape\r\n canvas = PIL.Image.new('RGB', (W * (len(col_seeds) + 1), H * (len(row_seeds) + 1)), 'black')\r\n for row_idx, row_seed in enumerate([None] + row_seeds):\r\n for col_idx, col_seed in enumerate([None] + col_seeds):\r\n if row_seed is None and col_seed is None:\r\n continue\r\n key = (row_seed, col_seed)\r\n if row_seed is None:\r\n key = (col_seed, col_seed)\r\n if col_seed is None:\r\n key = (row_seed, row_seed)\r\n canvas.paste(PIL.Image.fromarray(image_dict[key], 'RGB'), (W * col_idx, H * row_idx))\r\n return canvas\r\n\r\n# The following code will create your image, save it as a png, and display the image\r\n# Run the following code after you have set your row_seed, col_seed and col_style\r\nimage_grid = create_grid_images()\r\nimage_grid.save('image_grid.png')\r\nim = Image.open(\"image_grid.png\")\r\nim",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
]
] |
cbdab238c4dabf616f98e0e638e71d88a4e6cdc1
| 34,107 |
ipynb
|
Jupyter Notebook
|
nbs/examples/migrating_lightning.ipynb
|
rubensmau/fastai2
|
f9231256e2a8372949123bda36e44cb0e1493aa2
|
[
"Apache-2.0"
] | 708 |
2019-11-23T13:48:38.000Z
|
2022-03-21T22:28:04.000Z
|
nbs/examples/migrating_lightning.ipynb
|
rubensmau/fastai2
|
f9231256e2a8372949123bda36e44cb0e1493aa2
|
[
"Apache-2.0"
] | 503 |
2019-11-23T21:16:56.000Z
|
2020-08-17T23:35:52.000Z
|
nbs/examples/migrating_lightning.ipynb
|
Smirenost/fastai2
|
f9231256e2a8372949123bda36e44cb0e1493aa2
|
[
"Apache-2.0"
] | 386 |
2019-11-23T19:28:03.000Z
|
2022-03-09T15:48:01.000Z
| 157.175115 | 27,748 | 0.893981 |
[
[
[
"#all_slow",
"_____no_output_____"
]
],
[
[
"# Tutorial - Migrating from Lightning\n\n> Incrementally adding fastai goodness to your Lightning training",
"_____no_output_____"
],
[
"We're going to use the MNIST training code from Lightning's 'Quick Start' (as at August 2020), converted to a module. See `migrating_lightning.py` for the Lightning code we are importing here.",
"_____no_output_____"
]
],
[
[
"from migrating_lightning import *\n\nfrom fastai2.vision.all import *",
"_____no_output_____"
]
],
[
[
"## Using fastai's training loop",
"_____no_output_____"
],
[
"We can use the Lightning module directly:",
"_____no_output_____"
]
],
[
[
"model = LitModel()",
"_____no_output_____"
]
],
[
[
"To use it in fastai, we first pull the DataLoaders from the module into a `DataLoaders` object:",
"_____no_output_____"
]
],
[
[
"data = DataLoaders(model.train_dataloader(), model.val_dataloader()).cuda()",
"_____no_output_____"
]
],
[
[
"We can now create a `Learner` and fit:",
"_____no_output_____"
]
],
[
[
"learn = Learner(data, model, loss_func=F.cross_entropy, opt_func=Adam, metrics=accuracy)\nlearn.fit_one_cycle(1, 0.001)",
"_____no_output_____"
]
],
[
[
"As you can see, migrating from Ignite allowed us to reduce the amount of code, and doesn't require you to change any of your existing data pipelines, optimizers, loss functions, models, etc. Once you've made this change, you can then benefit from fastai's rich set of callbacks, transforms, visualizations, and so forth.\n\nFor instance, in the Lightning example, Tensorboard support was defined a special-case \"logger\". In fastai, Tensorboard is just another `Callback` that you can add, with the parameter `cbs=Tensorboard`, when you create your `Learner`. The callbacks all work together, so you can add an remove any schedulers, loggers, visualizers, and so forth. You don't have to learn about special types of functionality for each - they are all just plain callbacks.\n\nNote that fastai is very different from Lightning, in that it is much more than just a training loop (although we're only using the training loop in this example) - it is a complete framework including GPU-accelerated transformations, end-to-end inference, integrated applications for vision, text, tabular, and collaborative filtering, and so forth. You can use any part of the framework on its own, or combine them together, as described in the [fastai paper](https://arxiv.org/abs/2002.04688).",
"_____no_output_____"
],
[
"### Taking advantage of fastai Data Blocks",
"_____no_output_____"
],
[
"One problem in the Lightning example is that it doesn't actually use a validation set - it's just using the training set a second time as a validation set.\n\nYou might prefer to use fastai's Data Block API, which makes it really easy to create, visualize, and test your input data processing. Here's how you can create input data for MNIST, for instance:",
"_____no_output_____"
]
],
[
[
"mnist = DataBlock(blocks=(ImageBlock(cls=PILImageBW), CategoryBlock), \n get_items=get_image_files, \n splitter=GrandparentSplitter(),\n get_y=parent_label)",
"_____no_output_____"
]
],
[
[
"Here, we're telling `DataBlock` that we have a B&W image input, and a category output, our input items are file names of images, the images are labeled based on the name of the parent folder, and they are split by training vs validation based on the grandparent folder name. It's important to actually look at your data, so fastai also makes it easy to visualize your inputs and outputs, for instance:",
"_____no_output_____"
]
],
[
[
"dls = mnist.dataloaders(untar_data(URLs.MNIST_TINY))\ndls.show_batch(max_n=9, figsize=(4,4))",
"_____no_output_____"
]
]
] |
[
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
]
] |
cbdac5a400b1f7d7b214f77cc5e8c48b364ba62d
| 4,518 |
ipynb
|
Jupyter Notebook
|
testing.ipynb
|
ClarkUniversity-NiuLab/example_package_niulab
|
d7f341811329efc4b243b455446b64e860f74122
|
[
"MIT"
] | null | null | null |
testing.ipynb
|
ClarkUniversity-NiuLab/example_package_niulab
|
d7f341811329efc4b243b455446b64e860f74122
|
[
"MIT"
] | null | null | null |
testing.ipynb
|
ClarkUniversity-NiuLab/example_package_niulab
|
d7f341811329efc4b243b455446b64e860f74122
|
[
"MIT"
] | null | null | null | 86.884615 | 1,335 | 0.67552 |
[
[
[
"import clarku_youtube_crawler \nimport clarku_youtube_crawler.crawlerRaw as cr\n\ntest = cr.RawCrawler()",
"_____no_output_____"
]
]
] |
[
"code"
] |
[
[
"code"
]
] |
cbdad516d9a966201030842fe2565f2f24a6e856
| 985,748 |
ipynb
|
Jupyter Notebook
|
Image Segmentaion/main_new.ipynb
|
Ayush12062000/Orange-Fruit-Recognition-Using-Image-Segmentation
|
c4bc789ce1b3986933cb5d5440dfefce8f527acc
|
[
"MIT"
] | 6 |
2021-04-19T09:35:42.000Z
|
2022-03-21T03:40:16.000Z
|
Image Segmentaion/main_new.ipynb
|
Ayush12062000/Orange-Fruit-Recognition-Using-Image-Segmentation
|
c4bc789ce1b3986933cb5d5440dfefce8f527acc
|
[
"MIT"
] | 5 |
2021-03-28T08:45:00.000Z
|
2021-05-21T07:27:45.000Z
|
Image Segmentaion/main_new.ipynb
|
Ayush12062000/Orange-Fruit-Recognition-Using-Image-Segmentation
|
c4bc789ce1b3986933cb5d5440dfefce8f527acc
|
[
"MIT"
] | 6 |
2021-04-19T09:35:47.000Z
|
2021-05-12T11:10:19.000Z
| 985,748 | 985,748 | 0.861751 |
[
[
[
"# Import Required Libraries\ntry:\n import tensorflow as tf\n import os\n import random\n import numpy as np\n from tqdm import tqdm \n from skimage.io import imread, imshow\n from skimage.transform import resize\n import matplotlib.pyplot as plt\n from tensorflow.keras.models import load_model\n from keras.models import model_from_json\n print(\"----Libraries Imported----\")\nexcept:\n print(\"----Libraries Not Imported----\")\n\n",
"----Libraries Imported----\n"
],
[
"# checking the content of the current directory\nos.listdir()\n",
"_____no_output_____"
],
[
"# Setting up path\nseed = 42\nnp.random.seed = seed\n\nIMG_WIDTH = 128\nIMG_HEIGHT = 128\nIMG_CHANNELS = 3\n\n\nTRAIN_PATH = 'E:/Projects 6th SEM/Orange-Fruit-Recognition-Using-Image-Segmentation/Image Segmentaion/train_data/'\nTEST_PATH = 'E:/Projects 6th SEM/Orange-Fruit-Recognition-Using-Image-Segmentation/Image Segmentaion/test_data/'\n\ntrain_ids = next(os.walk(TRAIN_PATH))[1]\ntest_ids = next(os.walk(TEST_PATH))[1]\n\nprint(train_ids)\nprint(test_ids)\n",
"['abcd1', 'abcd10', 'abcd11', 'abcd12', 'abcd13', 'abcd14', 'abcd15', 'abcd16', 'abcd17', 'abcd18', 'abcd19', 'abcd2', 'abcd20', 'abcd21', 'abcd22', 'abcd23', 'abcd3', 'abcd4', 'abcd5', 'abcd6', 'abcd7', 'abcd8', 'abcd9']\n['xyz1', 'xyz10', 'xyz11', 'xyz12', 'xyz13', 'xyz14', 'xyz15', 'xyz16', 'xyz17', 'xyz18', 'xyz19', 'xyz2', 'xyz20', 'xyz21', 'xyz22', 'xyz23', 'xyz24', 'xyz25', 'xyz26', 'xyz27', 'xyz28', 'xyz29', 'xyz3', 'xyz30', 'xyz31', 'xyz32', 'xyz33', 'xyz34', 'xyz35', 'xyz36', 'xyz37', 'xyz38', 'xyz39', 'xyz4', 'xyz40', 'xyz41', 'xyz42', 'xyz43', 'xyz44', 'xyz45', 'xyz46', 'xyz47', 'xyz48', 'xyz49', 'xyz5', 'xyz50', 'xyz51', 'xyz52', 'xyz53', 'xyz54', 'xyz55', 'xyz56', 'xyz57', 'xyz58', 'xyz59', 'xyz6', 'xyz60', 'xyz61', 'xyz62', 'xyz63', 'xyz64', 'xyz65', 'xyz66', 'xyz67', 'xyz68', 'xyz69', 'xyz7', 'xyz70', 'xyz71', 'xyz72', 'xyz73', 'xyz74', 'xyz75', 'xyz76', 'xyz77', 'xyz8', 'xyz9']\n"
],
[
"# Loading data\n\n# independent variable\nX_train = np.zeros((len(train_ids), IMG_HEIGHT, IMG_WIDTH, IMG_CHANNELS), dtype=np.uint8)\n# dependent variable (what we are trying to predict)\nY_train = np.zeros((len(train_ids), IMG_HEIGHT, IMG_WIDTH, 1), dtype=np.bool) \n\nprint('Resizing training images and masks')\nfor n, id_ in tqdm(enumerate(train_ids), total=len(train_ids)): \n path = TRAIN_PATH + id_\n img = imread(path + '/images/' + id_ + '.jpg')[:,:,:IMG_CHANNELS] \n img = resize(img, (IMG_HEIGHT, IMG_WIDTH), mode='constant', preserve_range=True)\n X_train[n] = img #Fill empty X_train with values from img\n mask = np.zeros((IMG_HEIGHT, IMG_WIDTH, 1), dtype=np.bool)\n for mask_file in next(os.walk(path + '/masks/'))[2]:\n mask_ = imread(path + '/masks/' + mask_file)\n mask_ = np.expand_dims(resize(mask_, (IMG_HEIGHT, IMG_WIDTH), mode='constant', preserve_range=True), axis=-1)\n mask = np.maximum(mask, mask_) \n \n Y_train[n] = mask \n\n# test images\nX_test = np.zeros((len(test_ids), IMG_HEIGHT, IMG_WIDTH, IMG_CHANNELS), dtype=np.uint8)\nsizes_test = []\nprint('Resizing test images') \nfor n, id_ in tqdm(enumerate(test_ids), total=len(test_ids)):\n path = TEST_PATH + id_\n img = imread(path + '/images/' + id_ + '.jpg')[:,:,:IMG_CHANNELS]\n sizes_test.append([img.shape[0], img.shape[1]])\n img = resize(img, (IMG_HEIGHT, IMG_WIDTH), mode='constant', preserve_range=True)\n X_test[n] = img\n\nprint('Done!')\n",
"Resizing training images and masks\n100%|██████████| 23/23 [00:53<00:00, 2.32s/it]\n 1%|▏ | 1/77 [00:00<00:07, 9.52it/s]Resizing test images\n100%|██████████| 77/77 [00:10<00:00, 7.35it/s]Done!\n\n"
],
[
"# Showing Random images from the dataset\n\nimage_x = random.randint(0, len(train_ids))\nimshow(X_train[image_x])\nplt.show()\nimshow(np.squeeze(Y_train[image_x]))\nplt.show()\n\n",
"_____no_output_____"
],
[
"from UNet_Model import Segmentation_model\n\nmodel = Segmentation_model()\nmodel.summary()\n",
"Model: \"model\"\n__________________________________________________________________________________________________\nLayer (type) Output Shape Param # Connected to \n==================================================================================================\ninput_1 (InputLayer) [(None, 128, 128, 3) 0 \n__________________________________________________________________________________________________\nlambda (Lambda) (None, 128, 128, 3) 0 input_1[0][0] \n__________________________________________________________________________________________________\nconv2d (Conv2D) (None, 128, 128, 16) 448 lambda[0][0] \n__________________________________________________________________________________________________\ndropout (Dropout) (None, 128, 128, 16) 0 conv2d[0][0] \n__________________________________________________________________________________________________\nconv2d_1 (Conv2D) (None, 128, 128, 16) 2320 dropout[0][0] \n__________________________________________________________________________________________________\nmax_pooling2d (MaxPooling2D) (None, 64, 64, 16) 0 conv2d_1[0][0] \n__________________________________________________________________________________________________\nconv2d_2 (Conv2D) (None, 64, 64, 32) 4640 max_pooling2d[0][0] \n__________________________________________________________________________________________________\ndropout_1 (Dropout) (None, 64, 64, 32) 0 conv2d_2[0][0] \n__________________________________________________________________________________________________\nconv2d_3 (Conv2D) (None, 64, 64, 32) 9248 dropout_1[0][0] \n__________________________________________________________________________________________________\nmax_pooling2d_1 (MaxPooling2D) (None, 32, 32, 32) 0 conv2d_3[0][0] \n__________________________________________________________________________________________________\nconv2d_4 (Conv2D) (None, 32, 32, 64) 18496 max_pooling2d_1[0][0] \n__________________________________________________________________________________________________\ndropout_2 (Dropout) (None, 32, 32, 64) 0 conv2d_4[0][0] \n__________________________________________________________________________________________________\nconv2d_5 (Conv2D) (None, 32, 32, 64) 36928 dropout_2[0][0] \n__________________________________________________________________________________________________\nmax_pooling2d_2 (MaxPooling2D) (None, 16, 16, 64) 0 conv2d_5[0][0] \n__________________________________________________________________________________________________\nconv2d_6 (Conv2D) (None, 16, 16, 128) 73856 max_pooling2d_2[0][0] \n__________________________________________________________________________________________________\ndropout_3 (Dropout) (None, 16, 16, 128) 0 conv2d_6[0][0] \n__________________________________________________________________________________________________\nconv2d_7 (Conv2D) (None, 16, 16, 128) 147584 dropout_3[0][0] \n__________________________________________________________________________________________________\nmax_pooling2d_3 (MaxPooling2D) (None, 8, 8, 128) 0 conv2d_7[0][0] \n__________________________________________________________________________________________________\nconv2d_8 (Conv2D) (None, 8, 8, 256) 295168 max_pooling2d_3[0][0] \n__________________________________________________________________________________________________\ndropout_4 (Dropout) (None, 8, 8, 256) 0 conv2d_8[0][0] \n__________________________________________________________________________________________________\nconv2d_9 (Conv2D) (None, 8, 8, 256) 590080 dropout_4[0][0] \n__________________________________________________________________________________________________\nconv2d_transpose (Conv2DTranspo (None, 16, 16, 128) 131200 conv2d_9[0][0] \n__________________________________________________________________________________________________\nconcatenate (Concatenate) (None, 16, 16, 256) 0 conv2d_transpose[0][0] \n conv2d_7[0][0] \n__________________________________________________________________________________________________\nconv2d_10 (Conv2D) (None, 16, 16, 128) 295040 concatenate[0][0] \n__________________________________________________________________________________________________\ndropout_5 (Dropout) (None, 16, 16, 128) 0 conv2d_10[0][0] \n__________________________________________________________________________________________________\nconv2d_11 (Conv2D) (None, 16, 16, 128) 147584 dropout_5[0][0] \n__________________________________________________________________________________________________\nconv2d_transpose_1 (Conv2DTrans (None, 32, 32, 64) 32832 conv2d_11[0][0] \n__________________________________________________________________________________________________\nconcatenate_1 (Concatenate) (None, 32, 32, 128) 0 conv2d_transpose_1[0][0] \n conv2d_5[0][0] \n__________________________________________________________________________________________________\nconv2d_12 (Conv2D) (None, 32, 32, 64) 73792 concatenate_1[0][0] \n__________________________________________________________________________________________________\ndropout_6 (Dropout) (None, 32, 32, 64) 0 conv2d_12[0][0] \n__________________________________________________________________________________________________\nconv2d_13 (Conv2D) (None, 32, 32, 64) 36928 dropout_6[0][0] \n__________________________________________________________________________________________________\nconv2d_transpose_2 (Conv2DTrans (None, 64, 64, 32) 8224 conv2d_13[0][0] \n__________________________________________________________________________________________________\nconcatenate_2 (Concatenate) (None, 64, 64, 64) 0 conv2d_transpose_2[0][0] \n conv2d_3[0][0] \n__________________________________________________________________________________________________\nconv2d_14 (Conv2D) (None, 64, 64, 32) 18464 concatenate_2[0][0] \n__________________________________________________________________________________________________\ndropout_7 (Dropout) (None, 64, 64, 32) 0 conv2d_14[0][0] \n__________________________________________________________________________________________________\nconv2d_15 (Conv2D) (None, 64, 64, 32) 9248 dropout_7[0][0] \n__________________________________________________________________________________________________\nconv2d_transpose_3 (Conv2DTrans (None, 128, 128, 16) 2064 conv2d_15[0][0] \n__________________________________________________________________________________________________\nconcatenate_3 (Concatenate) (None, 128, 128, 32) 0 conv2d_transpose_3[0][0] \n conv2d_1[0][0] \n__________________________________________________________________________________________________\nconv2d_16 (Conv2D) (None, 128, 128, 16) 4624 concatenate_3[0][0] \n__________________________________________________________________________________________________\ndropout_8 (Dropout) (None, 128, 128, 16) 0 conv2d_16[0][0] \n__________________________________________________________________________________________________\nconv2d_17 (Conv2D) (None, 128, 128, 16) 2320 dropout_8[0][0] \n__________________________________________________________________________________________________\nconv2d_18 (Conv2D) (None, 128, 128, 1) 17 conv2d_17[0][0] \n==================================================================================================\nTotal params: 1,941,105\nTrainable params: 1,941,105\nNon-trainable params: 0\n__________________________________________________________________________________________________\n"
],
[
"\n################################\n#Modelcheckpoint\n\nwith tf.device('/GPU:0'):\n results = model.fit(X_train, Y_train, validation_split=0.1, batch_size=4, epochs=100)\n\n\nprint('Training DONE')\n",
"Epoch 1/100\n5/5 [==============================] - 14s 2s/step - loss: 0.4693 - accuracy: 0.7957 - val_loss: 0.5569 - val_accuracy: 0.6373\nEpoch 2/100\n5/5 [==============================] - 1s 269ms/step - loss: 0.4102 - accuracy: 0.7963 - val_loss: 0.4773 - val_accuracy: 0.6373\nEpoch 3/100\n5/5 [==============================] - 1s 260ms/step - loss: 0.4068 - accuracy: 0.7963 - val_loss: 0.4523 - val_accuracy: 0.6373\nEpoch 4/100\n5/5 [==============================] - 1s 261ms/step - loss: 0.3940 - accuracy: 0.7962 - val_loss: 0.4506 - val_accuracy: 0.6373\nEpoch 5/100\n5/5 [==============================] - 1s 272ms/step - loss: 0.3920 - accuracy: 0.7955 - val_loss: 0.4480 - val_accuracy: 0.6371\nEpoch 6/100\n5/5 [==============================] - 1s 280ms/step - loss: 0.3884 - accuracy: 0.7951 - val_loss: 0.4344 - val_accuracy: 0.6372\nEpoch 7/100\n5/5 [==============================] - 1s 245ms/step - loss: 0.3796 - accuracy: 0.7945 - val_loss: 0.4131 - val_accuracy: 0.6372\nEpoch 8/100\n5/5 [==============================] - 1s 247ms/step - loss: 0.3789 - accuracy: 0.7943 - val_loss: 0.4113 - val_accuracy: 0.6383\nEpoch 9/100\n5/5 [==============================] - 1s 286ms/step - loss: 0.3667 - accuracy: 0.7966 - val_loss: 0.4147 - val_accuracy: 0.6372\nEpoch 10/100\n5/5 [==============================] - 1s 242ms/step - loss: 0.3735 - accuracy: 0.7939 - val_loss: 0.4098 - val_accuracy: 0.6371\nEpoch 11/100\n5/5 [==============================] - 1s 208ms/step - loss: 0.3737 - accuracy: 0.7941 - val_loss: 0.4050 - val_accuracy: 0.6373\nEpoch 12/100\n5/5 [==============================] - 1s 229ms/step - loss: 0.3676 - accuracy: 0.7944 - val_loss: 0.4116 - val_accuracy: 0.6372\nEpoch 13/100\n5/5 [==============================] - 1s 232ms/step - loss: 0.3772 - accuracy: 0.7945 - val_loss: 0.4186 - val_accuracy: 0.6375\nEpoch 14/100\n5/5 [==============================] - 1s 265ms/step - loss: 0.3644 - accuracy: 0.7952 - val_loss: 0.4117 - val_accuracy: 0.6374\nEpoch 15/100\n5/5 [==============================] - 2s 389ms/step - loss: 0.3650 - accuracy: 0.7956 - val_loss: 0.4129 - val_accuracy: 0.6375\nEpoch 16/100\n5/5 [==============================] - 1s 290ms/step - loss: 0.3664 - accuracy: 0.7953 - val_loss: 0.4369 - val_accuracy: 0.6373\nEpoch 17/100\n5/5 [==============================] - 1s 248ms/step - loss: 0.3612 - accuracy: 0.7961 - val_loss: 0.4311 - val_accuracy: 0.6373\nEpoch 18/100\n5/5 [==============================] - 1s 242ms/step - loss: 0.3620 - accuracy: 0.7963 - val_loss: 0.4237 - val_accuracy: 0.6373\nEpoch 19/100\n5/5 [==============================] - 2s 336ms/step - loss: 0.3585 - accuracy: 0.7960 - val_loss: 0.4277 - val_accuracy: 0.6374\nEpoch 20/100\n5/5 [==============================] - 1s 243ms/step - loss: 0.3562 - accuracy: 0.7960 - val_loss: 0.4203 - val_accuracy: 0.6374\nEpoch 21/100\n5/5 [==============================] - 1s 215ms/step - loss: 0.3534 - accuracy: 0.7962 - val_loss: 0.4181 - val_accuracy: 0.6375\nEpoch 22/100\n5/5 [==============================] - 1s 206ms/step - loss: 0.3524 - accuracy: 0.7963 - val_loss: 0.4178 - val_accuracy: 0.6375\nEpoch 23/100\n5/5 [==============================] - 1s 198ms/step - loss: 0.3507 - accuracy: 0.7960 - val_loss: 0.4198 - val_accuracy: 0.6374\nEpoch 24/100\n5/5 [==============================] - 1s 212ms/step - loss: 0.3528 - accuracy: 0.7961 - val_loss: 0.4144 - val_accuracy: 0.6374\nEpoch 25/100\n5/5 [==============================] - 1s 210ms/step - loss: 0.3486 - accuracy: 0.7961 - val_loss: 0.4323 - val_accuracy: 0.6373\nEpoch 26/100\n5/5 [==============================] - 1s 217ms/step - loss: 0.3486 - accuracy: 0.7962 - val_loss: 0.4245 - val_accuracy: 0.6372\nEpoch 27/100\n5/5 [==============================] - 1s 214ms/step - loss: 0.3466 - accuracy: 0.7963 - val_loss: 0.4179 - val_accuracy: 0.6373\nEpoch 28/100\n5/5 [==============================] - 1s 205ms/step - loss: 0.3404 - accuracy: 0.7962 - val_loss: 0.4440 - val_accuracy: 0.6373\nEpoch 29/100\n5/5 [==============================] - 1s 210ms/step - loss: 0.3434 - accuracy: 0.7984 - val_loss: 0.4246 - val_accuracy: 0.6371\nEpoch 30/100\n5/5 [==============================] - 1s 209ms/step - loss: 0.3444 - accuracy: 0.7963 - val_loss: 0.4236 - val_accuracy: 0.6372\nEpoch 31/100\n5/5 [==============================] - 1s 214ms/step - loss: 0.3428 - accuracy: 0.7965 - val_loss: 0.4230 - val_accuracy: 0.6373\nEpoch 32/100\n5/5 [==============================] - 1s 225ms/step - loss: 0.3390 - accuracy: 0.7953 - val_loss: 0.4399 - val_accuracy: 0.6375\nEpoch 33/100\n5/5 [==============================] - 1s 242ms/step - loss: 0.3391 - accuracy: 0.7968 - val_loss: 0.4351 - val_accuracy: 0.6375\nEpoch 34/100\n5/5 [==============================] - 1s 258ms/step - loss: 0.4114 - accuracy: 0.8158 - val_loss: 0.5267 - val_accuracy: 0.6374\nEpoch 35/100\n5/5 [==============================] - 1s 254ms/step - loss: 0.3603 - accuracy: 0.7963 - val_loss: 0.4345 - val_accuracy: 0.6373\nEpoch 36/100\n5/5 [==============================] - 1s 220ms/step - loss: 0.3578 - accuracy: 0.7963 - val_loss: 0.4309 - val_accuracy: 0.6373\nEpoch 37/100\n5/5 [==============================] - 1s 203ms/step - loss: 0.3482 - accuracy: 0.7962 - val_loss: 0.4368 - val_accuracy: 0.6373\nEpoch 38/100\n5/5 [==============================] - 1s 214ms/step - loss: 0.3526 - accuracy: 0.7964 - val_loss: 0.4280 - val_accuracy: 0.6374\nEpoch 39/100\n5/5 [==============================] - 1s 216ms/step - loss: 0.3459 - accuracy: 0.7963 - val_loss: 0.4263 - val_accuracy: 0.6373\nEpoch 40/100\n5/5 [==============================] - 1s 205ms/step - loss: 0.3448 - accuracy: 0.7964 - val_loss: 0.4261 - val_accuracy: 0.6373\nEpoch 41/100\n5/5 [==============================] - 1s 209ms/step - loss: 0.3421 - accuracy: 0.7963 - val_loss: 0.4310 - val_accuracy: 0.6373\nEpoch 42/100\n5/5 [==============================] - 1s 212ms/step - loss: 0.3411 - accuracy: 0.7963 - val_loss: 0.4379 - val_accuracy: 0.6373\nEpoch 43/100\n5/5 [==============================] - 1s 202ms/step - loss: 0.3398 - accuracy: 0.7963 - val_loss: 0.4342 - val_accuracy: 0.6373\nEpoch 44/100\n5/5 [==============================] - 1s 215ms/step - loss: 0.3404 - accuracy: 0.7964 - val_loss: 0.4836 - val_accuracy: 0.6373\nEpoch 45/100\n5/5 [==============================] - 1s 215ms/step - loss: 0.3441 - accuracy: 0.7963 - val_loss: 0.4310 - val_accuracy: 0.6373\nEpoch 46/100\n5/5 [==============================] - 1s 228ms/step - loss: 0.3371 - accuracy: 0.7963 - val_loss: 0.4634 - val_accuracy: 0.6373\nEpoch 47/100\n5/5 [==============================] - 1s 211ms/step - loss: 0.3340 - accuracy: 0.7963 - val_loss: 0.4714 - val_accuracy: 0.6373\nEpoch 48/100\n5/5 [==============================] - 1s 208ms/step - loss: 0.3285 - accuracy: 0.7963 - val_loss: 0.4825 - val_accuracy: 0.6373\nEpoch 49/100\n5/5 [==============================] - 1s 214ms/step - loss: 0.3260 - accuracy: 0.7964 - val_loss: 0.4996 - val_accuracy: 0.6373\nEpoch 50/100\n5/5 [==============================] - 1s 219ms/step - loss: 0.3360 - accuracy: 0.7963 - val_loss: 0.5108 - val_accuracy: 0.6373\nEpoch 51/100\n5/5 [==============================] - 1s 221ms/step - loss: 0.3278 - accuracy: 0.7963 - val_loss: 0.4978 - val_accuracy: 0.6373\nEpoch 52/100\n5/5 [==============================] - 1s 208ms/step - loss: 0.3264 - accuracy: 0.7962 - val_loss: 0.5620 - val_accuracy: 0.6372\nEpoch 53/100\n5/5 [==============================] - 1s 191ms/step - loss: 0.3294 - accuracy: 0.7963 - val_loss: 0.4837 - val_accuracy: 0.6373\nEpoch 54/100\n5/5 [==============================] - 1s 209ms/step - loss: 0.3281 - accuracy: 0.7962 - val_loss: 0.5186 - val_accuracy: 0.6373\nEpoch 55/100\n5/5 [==============================] - 1s 223ms/step - loss: 0.3295 - accuracy: 0.7961 - val_loss: 0.5031 - val_accuracy: 0.6374\nEpoch 56/100\n5/5 [==============================] - 1s 209ms/step - loss: 0.3270 - accuracy: 0.7962 - val_loss: 0.4958 - val_accuracy: 0.6373\nEpoch 57/100\n5/5 [==============================] - 1s 206ms/step - loss: 0.3212 - accuracy: 0.7963 - val_loss: 0.5656 - val_accuracy: 0.6374\nEpoch 58/100\n5/5 [==============================] - 1s 223ms/step - loss: 0.3202 - accuracy: 0.7963 - val_loss: 0.5291 - val_accuracy: 0.6374\nEpoch 59/100\n5/5 [==============================] - 1s 204ms/step - loss: 0.3210 - accuracy: 0.7964 - val_loss: 0.6244 - val_accuracy: 0.6373\nEpoch 60/100\n5/5 [==============================] - 1s 220ms/step - loss: 0.3218 - accuracy: 0.7961 - val_loss: 0.5594 - val_accuracy: 0.6374\nEpoch 61/100\n5/5 [==============================] - 1s 209ms/step - loss: 0.3167 - accuracy: 0.7967 - val_loss: 0.6029 - val_accuracy: 0.6374\nEpoch 62/100\n5/5 [==============================] - 1s 200ms/step - loss: 0.3230 - accuracy: 0.7963 - val_loss: 0.5387 - val_accuracy: 0.6374\nEpoch 63/100\n5/5 [==============================] - 1s 220ms/step - loss: 0.3218 - accuracy: 0.7965 - val_loss: 0.4799 - val_accuracy: 0.6374\nEpoch 64/100\n5/5 [==============================] - 1s 207ms/step - loss: 0.3220 - accuracy: 0.7968 - val_loss: 0.5874 - val_accuracy: 0.6373\nEpoch 65/100\n5/5 [==============================] - 1s 228ms/step - loss: 0.3178 - accuracy: 0.7966 - val_loss: 0.5526 - val_accuracy: 0.6374\nEpoch 66/100\n5/5 [==============================] - 1s 197ms/step - loss: 0.3170 - accuracy: 0.7969 - val_loss: 0.5377 - val_accuracy: 0.6374\nEpoch 67/100\n5/5 [==============================] - 1s 225ms/step - loss: 0.3131 - accuracy: 0.7968 - val_loss: 0.6341 - val_accuracy: 0.6373\nEpoch 68/100\n5/5 [==============================] - 1s 205ms/step - loss: 0.3137 - accuracy: 0.7970 - val_loss: 0.5577 - val_accuracy: 0.6373\nEpoch 69/100\n5/5 [==============================] - 1s 210ms/step - loss: 0.3129 - accuracy: 0.7977 - val_loss: 0.5492 - val_accuracy: 0.6375\nEpoch 70/100\n5/5 [==============================] - 1s 214ms/step - loss: 0.3129 - accuracy: 0.7980 - val_loss: 0.6363 - val_accuracy: 0.6375\nEpoch 71/100\n5/5 [==============================] - 1s 215ms/step - loss: 0.3103 - accuracy: 0.7979 - val_loss: 0.5313 - val_accuracy: 0.6373\nEpoch 72/100\n5/5 [==============================] - 1s 247ms/step - loss: 0.3103 - accuracy: 0.8071 - val_loss: 0.6312 - val_accuracy: 0.6375\nEpoch 73/100\n5/5 [==============================] - 1s 257ms/step - loss: 0.3138 - accuracy: 0.8091 - val_loss: 0.4656 - val_accuracy: 0.6396\nEpoch 74/100\n5/5 [==============================] - 1s 249ms/step - loss: 0.3151 - accuracy: 0.8096 - val_loss: 0.4699 - val_accuracy: 0.6428\nEpoch 75/100\n5/5 [==============================] - 1s 256ms/step - loss: 0.3129 - accuracy: 0.8095 - val_loss: 0.4976 - val_accuracy: 0.6491\nEpoch 76/100\n5/5 [==============================] - 1s 236ms/step - loss: 0.3129 - accuracy: 0.8076 - val_loss: 0.5045 - val_accuracy: 0.6442\nEpoch 77/100\n5/5 [==============================] - 1s 234ms/step - loss: 0.3088 - accuracy: 0.8071 - val_loss: 0.5577 - val_accuracy: 0.6394\nEpoch 78/100\n5/5 [==============================] - 1s 240ms/step - loss: 0.3074 - accuracy: 0.8091 - val_loss: 0.5578 - val_accuracy: 0.6446\nEpoch 79/100\n5/5 [==============================] - 1s 234ms/step - loss: 0.3063 - accuracy: 0.8098 - val_loss: 0.5718 - val_accuracy: 0.6494\nEpoch 80/100\n5/5 [==============================] - 1s 220ms/step - loss: 0.3028 - accuracy: 0.8114 - val_loss: 0.5784 - val_accuracy: 0.6555\nEpoch 81/100\n5/5 [==============================] - 1s 224ms/step - loss: 0.7179 - accuracy: 0.8140 - val_loss: 0.5065 - val_accuracy: 0.6488\nEpoch 82/100\n5/5 [==============================] - 1s 244ms/step - loss: 0.4162 - accuracy: 0.8065 - val_loss: 0.4153 - val_accuracy: 0.6549\nEpoch 83/100\n5/5 [==============================] - 1s 247ms/step - loss: 0.3736 - accuracy: 0.7963 - val_loss: 0.5170 - val_accuracy: 0.6373\nEpoch 84/100\n5/5 [==============================] - 1s 229ms/step - loss: 0.3869 - accuracy: 0.8010 - val_loss: 0.4159 - val_accuracy: 0.6472\nEpoch 85/100\n5/5 [==============================] - 1s 213ms/step - loss: 0.3613 - accuracy: 0.8013 - val_loss: 0.4054 - val_accuracy: 0.6612\nEpoch 86/100\n5/5 [==============================] - 1s 208ms/step - loss: 0.3592 - accuracy: 0.7999 - val_loss: 0.3976 - val_accuracy: 0.6464\nEpoch 87/100\n5/5 [==============================] - 1s 235ms/step - loss: 0.3578 - accuracy: 0.8040 - val_loss: 0.3973 - val_accuracy: 0.6443\nEpoch 88/100\n5/5 [==============================] - 1s 242ms/step - loss: 0.3560 - accuracy: 0.8046 - val_loss: 0.4323 - val_accuracy: 0.6374\nEpoch 89/100\n5/5 [==============================] - 1s 213ms/step - loss: 0.3531 - accuracy: 0.8016 - val_loss: 0.4229 - val_accuracy: 0.6377\nEpoch 90/100\n5/5 [==============================] - 1s 245ms/step - loss: 0.3552 - accuracy: 0.8009 - val_loss: 0.3989 - val_accuracy: 0.6419\nEpoch 91/100\n5/5 [==============================] - 1s 252ms/step - loss: 0.3498 - accuracy: 0.8029 - val_loss: 0.3970 - val_accuracy: 0.6490\nEpoch 92/100\n5/5 [==============================] - 1s 259ms/step - loss: 0.3469 - accuracy: 0.8035 - val_loss: 0.4070 - val_accuracy: 0.6388\nEpoch 93/100\n5/5 [==============================] - 1s 212ms/step - loss: 0.3458 - accuracy: 0.8030 - val_loss: 0.4062 - val_accuracy: 0.6416\nEpoch 94/100\n5/5 [==============================] - 1s 220ms/step - loss: 0.3489 - accuracy: 0.8031 - val_loss: 0.4209 - val_accuracy: 0.6374\nEpoch 95/100\n5/5 [==============================] - 1s 231ms/step - loss: 0.3486 - accuracy: 0.8007 - val_loss: 0.4324 - val_accuracy: 0.6373\nEpoch 96/100\n5/5 [==============================] - 1s 220ms/step - loss: 0.3454 - accuracy: 0.8033 - val_loss: 0.4167 - val_accuracy: 0.6452\nEpoch 97/100\n5/5 [==============================] - 1s 207ms/step - loss: 0.3429 - accuracy: 0.8077 - val_loss: 0.4079 - val_accuracy: 0.6481\nEpoch 98/100\n5/5 [==============================] - 1s 218ms/step - loss: 0.3409 - accuracy: 0.8096 - val_loss: 0.4100 - val_accuracy: 0.6503\nEpoch 99/100\n5/5 [==============================] - 1s 236ms/step - loss: 0.3406 - accuracy: 0.8091 - val_loss: 0.4133 - val_accuracy: 0.6509\nEpoch 100/100\n5/5 [==============================] - 1s 258ms/step - loss: 0.3388 - accuracy: 0.8073 - val_loss: 0.4366 - val_accuracy: 0.6379\nTraining DONE\n"
],
[
"# Plotting Training Results\n\nplt.plot(results.history['accuracy'][0:150])\nplt.plot(results.history['val_accuracy'][0:150])\nplt.title('model accuracy')\nplt.ylabel('accuracy')\nplt.xlabel('epoch')\nplt.legend(['training_accuracy', 'validation_accuracy'])\nplt.show()\n\n",
"_____no_output_____"
],
[
"\nplt.plot(results.history['loss'][0:150])\nplt.plot(results.history['val_loss'][0:150])\nplt.title('model loss')\nplt.ylabel('loss')\nplt.xlabel('epoch')\nplt.legend(['training_loss', 'validation_loss'])\nplt.show()\n\n",
"_____no_output_____"
],
[
"# Saving model\n\norange_model_json = model.to_json() \nwith open(\"Segmentation_model.json\", \"w\") as json_file: \n json_file.write(orange_model_json) \nmodel.save_weights(\"Orange_Fruit_Weights_segmentation.h5\")\n\n\n",
"_____no_output_____"
],
[
"# Loading Unet\nsegmentation_model = model_from_json(open(\"Segmentation_model.json\", \"r\").read()) \nsegmentation_model.load_weights('Orange_Fruit_Weights_segmentation.h5')\n",
"_____no_output_____"
],
[
"####################################\n\nidx = random.randint(0, len(X_train))\nprint(idx)\n\npreds_train = segmentation_model.predict(X_train[:int(X_train.shape[0]*0.9)], verbose=1)\npreds_val = segmentation_model.predict(X_train[int(X_train.shape[0]*0.9):], verbose=1)\npreds_test = segmentation_model.predict(X_test, verbose=1)\n\n \npreds_train_t = (preds_train > 0.5).astype(np.uint8)\npreds_val_t = (preds_val > 0.5).astype(np.uint8)\npreds_test_t = (preds_test > 0.5).astype(np.uint8)\n\n",
"6\n1/1 [==============================] - 7s 7s/step\n1/1 [==============================] - 0s 85ms/step\n3/3 [==============================] - 12s 2s/step\n"
],
[
"# Perform a sanity check on some random training samples\nix = random.randint(0, len(preds_train_t))\nimshow(X_train[ix])\nplt.show()\nimshow(np.squeeze(Y_train[ix]))\nplt.show()\nimshow(np.squeeze(preds_train_t[ix]))\nplt.show()\n\n# Perform a sanity check on some random validation samples\nix = random.randint(0, len(preds_val_t))\nimshow(X_train[int(X_train.shape[0]*0.9):][ix])\nplt.show()\nimshow(np.squeeze(Y_train[int(Y_train.shape[0]*0.9):][ix]))\nplt.show()\nimshow(np.squeeze(preds_val_t[ix]))\nplt.show()\n",
"_____no_output_____"
],
[
"# Loading Classification Model\n\nimport Prediction_file as pf\nclassification_model = pf.Loading_Model()\n",
"_____no_output_____"
],
[
"# Prediction\n\npath1 = 'Images/kiwi.jpg'\npath2 = 'Images/Orange.jpg'\n\npred1 = pf.predicting(path1,classification_model)\npred2 = pf.predicting(path2,classification_model)\n\n\n",
"_____no_output_____"
],
[
"from tensorflow.keras.preprocessing.image import load_img, img_to_array\ndef process_image(path):\n img = load_img(path, target_size = (IMG_WIDTH,IMG_HEIGHT))\n img_tensor = img_to_array(img)\n img_tensor = np.expand_dims(img_tensor, axis = 0)\n img_tensor/=255.0\n return img_tensor\n\nif pred2 > 0.5:\n p = segmentation_model.predict(process_image(path2), verbose=1)\n p_t = (p > 0.5).astype(np.uint8)\n imshow(np.squeeze(p_t))\n plt.show()\n \n",
"1/1 [==============================] - 3s 3s/step\n"
],
[
"p = segmentation_model.predict(process_image(path1), verbose=1)\np_t = (p > 0.5).astype(np.uint8)\nimshow(np.squeeze(p_t))\nplt.show()\n",
"1/1 [==============================] - 0s 73ms/step\n"
]
]
] |
[
"code"
] |
[
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
]
] |
cbdaf97b1f8d652dead1e83ed5c3813da5bb541b
| 903,973 |
ipynb
|
Jupyter Notebook
|
nbs/dl1/lesson6-pets-more-ex.ipynb
|
rforgione/course-v3
|
f496c2eea82153e23c20b3fcb2bd2c3409677cf7
|
[
"Apache-2.0"
] | null | null | null |
nbs/dl1/lesson6-pets-more-ex.ipynb
|
rforgione/course-v3
|
f496c2eea82153e23c20b3fcb2bd2c3409677cf7
|
[
"Apache-2.0"
] | null | null | null |
nbs/dl1/lesson6-pets-more-ex.ipynb
|
rforgione/course-v3
|
f496c2eea82153e23c20b3fcb2bd2c3409677cf7
|
[
"Apache-2.0"
] | null | null | null | 1,382.221713 | 519,384 | 0.952681 |
[
[
[
"# Lesson 6: pets revisited",
"_____no_output_____"
]
],
[
[
"%reload_ext autoreload\n%autoreload 2\n%matplotlib inline\n\nfrom fastai.vision import *",
"_____no_output_____"
]
],
[
[
"Set a batch size of 64.",
"_____no_output_____"
],
[
"Untar the data at `URLs.PETS`, and set the variable `path` to the returned path with `images` appended to the end.",
"_____no_output_____"
],
[
"## Data augmentation",
"_____no_output_____"
],
[
"Create a variable `tfms` that captures the output of the `get_transforms` function, with the following arguments:\n- max_rotate=20\n- max_zoom=1.3\n- max_lighting=0.4\n- max_warp=0.4\n- p_affine=1.\n- p_lighting=1.",
"_____no_output_____"
],
[
"Explain what each of these are.",
"_____no_output_____"
],
[
"Create an `ImageList` from the folder `path` split by a random 20% (using a random seed of 2). Assign it to the variable `src`. What kind of object does this return? Can you find (in code) why this is the case?",
"_____no_output_____"
],
[
"Write a function `get_data` that takes `src` and labels it using the regex `([^/]+)_\\d+.jpg$`, transforms it with `tfms`, takes `size` as an argument, takes `padding_mode` as an argument that defaults to `reflection`, creates a databunch with batch size `bs`, and normalizes using the imagenet stats. What type would you expect this to return?",
"_____no_output_____"
],
[
"Create a variable `data` that calls `get_data` with size 224, `bs=bs`, and padding type `zeros`.",
"_____no_output_____"
],
[
"Write a function `_plot` that plots the fourth image in the training dataset. Pass it to `plot_multi` to create a 3x3 grid of augmented images.",
"_____no_output_____"
],
[
"Create a new variable `data` with size 224 and the same bs.",
"_____no_output_____"
],
[
"data = get_data(224,bs)",
"_____no_output_____"
],
[
"Use the same process to plot a 3x3 grid of 8x8 images of augmented data. This time allow the default padding mode.",
"_____no_output_____"
],
[
"## Train a model",
"_____no_output_____"
],
[
"Call `gc.collect`. Can you explain what this does?",
"_____no_output_____"
],
[
"Create a `cnn_learner` named `learn` with data `data`, architecture resnet34, using the `error_rate` metric, and `bn_final` set to true.",
"_____no_output_____"
],
[
"Fit a cycle with 3 epochs, a slice up to 1e-2, with `pct_start=0.8`. Can you explain what `pct_start` does?",
"_____no_output_____"
],
[
"Unfreeze the neural net. Fit another cycle with two epochs under the slice (1e-6, 1e-3). Same pct_start.",
"_____no_output_____"
],
[
"Create a new `data` object with size 352.",
"_____no_output_____"
],
[
"Train for another cycle with 2 epochs, this time with a `max_lr` of `slice(1e-6, 1e-4)`.",
"_____no_output_____"
],
[
"Save the model under the name `352`.",
"_____no_output_____"
],
[
"## Convolution kernel",
"_____no_output_____"
],
[
"Create another new `data` with size 352 and batch size 16.",
"_____no_output_____"
],
[
"Create a new learner `learn` with the same specs as earlier, and load the weights from `352` to it. ",
"_____no_output_____"
],
[
"Set the variable `idx=0`. Set the values returned at position `idx` within the valid_ds and to `x` and `y`.",
"_____no_output_____"
],
[
"Call the `show` method on x.",
"_____no_output_____"
],
[
"Return the item at position `idx` in the `y` part of the `valid_ds`.",
"_____no_output_____"
],
[
"This is created for you, because it doesn't teach much. Maybe dig into the `expand` method.",
"_____no_output_____"
],
[
"Return the shape of `k`.",
"_____no_output_____"
]
],
[
[
"k.shape",
"_____no_output_____"
]
],
[
[
"Get the `x` value of the first item in `valid_ds`, get the `data` property and set it to `t`. What does the data property represent? ",
"_____no_output_____"
],
[
"Add a new dimension to `t` using the `None` index syntax.",
"_____no_output_____"
],
[
"Create an image called `edge` by convolving `t` with our filter `k`.",
"_____no_output_____"
],
[
"Run `show_image` over `edge`. Hint: you'll have to get the zeroth index of `edge` -- why?",
"_____no_output_____"
],
[
"Show the number of classes in `data`.",
"_____no_output_____"
],
[
"Print the model.",
"_____no_output_____"
],
[
"Print a model summary.",
"_____no_output_____"
],
[
"## Heatmap",
"_____no_output_____"
],
[
"Get the model out of our learner and set it to `eval` mode.",
"_____no_output_____"
],
[
"Get one item from the `x` data you created above. Call this `xb`. Hint: `one_item` returns a tuple, but we only need the first thing.",
"_____no_output_____"
],
[
"Create an image from a denormed version of xb. Again, you'll have to index into this. Be sure you can explain why. Call the output `xb_im`.",
"_____no_output_____"
],
[
"Put the `xb` variable on the GPU by calling `cuda()`.",
"_____no_output_____"
],
[
"Import fastai.callbacks.hooks.",
"_____no_output_____"
],
[
"Create a function `hooked_backward` that returns two objects `grad_a` and `grad_g` representing the activations and the gradients. Make sure to use `with` statements here so that the hooks are removed after we get our results.",
"_____no_output_____"
],
[
"Create two objects, `hook_a` and `hook_g` with the outputs of `hooked_backward`.",
"_____no_output_____"
],
[
"Assign the stored activation outputs to a variable called `acts`. Make sure to call `.cpu` to put this back on the CPU.",
"_____no_output_____"
],
[
"Take an average over the channel dimension to get a 2d shape. Print out the shape.",
"_____no_output_____"
],
[
"Write a function `show_heatmap` that does the following:\n- takes an argument hm\n- Creates a new matplotlib axis using `plt.subplots`\n- shows `xb_im` on the new axis\n- calls `ax.imshow` with arguments `alpha=0.6`, `extent=(0,352,352,0)`, `interpolation=bilinear`, `cmap=magma`. Look up what these mean.",
"_____no_output_____"
],
[
"Call `show_heatmap` on `avg_acts`.",
"_____no_output_____"
],
[
"## Grad-CAM",
"_____no_output_____"
],
[
"Paper: [Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization](https://arxiv.org/abs/1610.02391)",
"_____no_output_____"
]
],
[
[
"grad = hook_g.stored[0][0].cpu()\ngrad_chan = grad.mean(1).mean(1)\ngrad.shape,grad_chan.shape",
"_____no_output_____"
],
[
"mult = (acts*grad_chan[...,None,None]).mean(0)",
"_____no_output_____"
],
[
"show_heatmap(mult)",
"_____no_output_____"
],
[
"fn = path/'../other/bulldog_maine.jpg' #Replace with your own image",
"_____no_output_____"
],
[
"x = open_image(fn); x",
"_____no_output_____"
],
[
"xb,_ = data.one_item(x)\nxb_im = Image(data.denorm(xb)[0])\nxb = xb.cuda()",
"_____no_output_____"
],
[
"hook_a,hook_g = hooked_backward()",
"_____no_output_____"
],
[
"acts = hook_a.stored[0].cpu()\ngrad = hook_g.stored[0][0].cpu()\n\ngrad_chan = grad.mean(1).mean(1)\nmult = (acts*grad_chan[...,None,None]).mean(0)",
"_____no_output_____"
],
[
"show_heatmap(mult)",
"_____no_output_____"
],
[
"data.classes[0]",
"_____no_output_____"
],
[
"hook_a,hook_g = hooked_backward(0)",
"_____no_output_____"
],
[
"acts = hook_a.stored[0].cpu()\ngrad = hook_g.stored[0][0].cpu()\n\ngrad_chan = grad.mean(1).mean(1)\nmult = (acts*grad_chan[...,None,None]).mean(0)",
"_____no_output_____"
],
[
"show_heatmap(mult)",
"_____no_output_____"
]
],
[
[
"## fin",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown"
] |
[
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
]
] |
cbdb138fb7814775d7f1649d07519a711182b9f0
| 5,107 |
ipynb
|
Jupyter Notebook
|
code_snippets/Deploy_Previously_Created_Model.ipynb
|
sureindia-in/sagemaker-examples-good
|
be72218d4fd75bdb7f0df136026366fcd726fbe0
|
[
"Apache-2.0"
] | null | null | null |
code_snippets/Deploy_Previously_Created_Model.ipynb
|
sureindia-in/sagemaker-examples-good
|
be72218d4fd75bdb7f0df136026366fcd726fbe0
|
[
"Apache-2.0"
] | 1 |
2021-05-26T00:01:10.000Z
|
2021-05-26T00:01:10.000Z
|
code_snippets/Deploy_Previously_Created_Model.ipynb
|
sureindia-in/sagemaker-examples-good
|
be72218d4fd75bdb7f0df136026366fcd726fbe0
|
[
"Apache-2.0"
] | null | null | null | 30.041176 | 430 | 0.561386 |
[
[
[
"# Deploy a previously created model in SageMaker",
"_____no_output_____"
],
[
"Sagemaker decouples model creation/fitting and model deployment. **This short notebook shows how you can deploy a model that you have already created**. It is assumed that you have already created the model and it appears in the `Models` section of the SageMaker console. Obviously, before you deploy a model the model must exist, so please go back and make sure you have already fit/created the model before proceeding. \nFor more information about deploying models, see https://docs.aws.amazon.com/sagemaker/latest/dg/ex1-deploy-model.html ",
"_____no_output_____"
]
],
[
[
"import boto3\nfrom time import gmtime, strftime",
"_____no_output_____"
],
[
"# configs for model, endpoint and batch transform\nmodel_name = (\n \"ENTER MODEL NAME\" # enter name of a model from the 'Model panel' in the AWS SageMaker console.\n)\nsm = boto3.client(\"sagemaker\")",
"_____no_output_____"
]
],
[
[
"## Deploy using an inference endpoint",
"_____no_output_____"
]
],
[
[
"# set endpoint name/config.\nendpoint_config_name = \"DEMO-model-config-\" + strftime(\"%Y-%m-%d-%H-%M-%S\", gmtime())\nendpoint_name = \"DEMO-model-config-\" + strftime(\"%Y-%m-%d-%H-%M-%S\", gmtime())",
"_____no_output_____"
],
[
"create_endpoint_config_response = sm.create_endpoint_config(\n EndpointConfigName=endpoint_config_name,\n ProductionVariants=[\n {\n \"InstanceType\": \"ml.m4.xlarge\",\n \"InitialVariantWeight\": 1,\n \"InitialInstanceCount\": 1,\n \"ModelName\": model_name,\n \"VariantName\": \"AllTraffic\",\n }\n ],\n)\n\nprint(\"Endpoint Config Arn: \" + create_endpoint_config_response[\"EndpointConfigArn\"])\n\n\ncreate_endpoint_response = sm.create_endpoint(\n EndpointName=endpoint_name, EndpointConfigName=endpoint_config_name\n)\nprint(create_endpoint_response[\"EndpointArn\"])\n\nresp = sm.describe_endpoint(EndpointName=endpoint_name)\nstatus = resp[\"EndpointStatus\"]\nprint(\"Status: \" + status)",
"_____no_output_____"
]
],
[
[
"If you go to the AWS SageMaker service console now, you should see that the endpoint creation is in progress.",
"_____no_output_____"
],
[
"## Deploy using a batch transform job\n\nA batch transform job should be used for when you want to create inferences on a dateset and then shut down the resources when inference is finished. ",
"_____no_output_____"
]
],
[
[
"# config for batch transform\nbatch_job_name = \"ENTER_JOB_NAME\"\noutput_location = \"ENDER_OUTPUT_LOCATION\" # S3 bucket/location\ninput_location = \"ENTER_INPUT_LOCATION\" # S3 bucket/location",
"_____no_output_____"
],
[
"request = {\n \"TransformJobName\": batch_job_name,\n \"ModelName\": model_name,\n \"TransformOutput\": {\n \"S3OutputPath\": output_location,\n \"Accept\": \"text/csv\",\n \"AssembleWith\": \"Line\",\n },\n \"TransformInput\": {\n \"DataSource\": {\"S3DataSource\": {\"S3DataType\": \"S3Prefix\", \"S3Uri\": input_location}},\n \"ContentType\": \"text/csv\",\n \"SplitType\": \"Line\",\n \"CompressionType\": \"None\",\n },\n \"TransformResources\": {\n \"InstanceType\": \"ml.m4.xlarge\", # change this based on what resources you want to request\n \"InstanceCount\": 1,\n },\n}\nsm.create_transform_job(**request)",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code"
]
] |
cbdb13947a96b4c47c35b021a59bcfe21aabc614
| 256,372 |
ipynb
|
Jupyter Notebook
|
Day12_Detection_Algorithms_YOLOv2/Autonomous_driving_application_Car_detection_v3a.ipynb
|
activatedbonkers/100-Days-of-Code-Challenge
|
a1a376e5373d8fc5fda5df4004115760aa92bfda
|
[
"MIT"
] | 8 |
2020-07-12T22:45:30.000Z
|
2021-06-01T10:44:28.000Z
|
Day12_Detection_Algorithms_YOLOv2/Autonomous_driving_application_Car_detection_v3a.ipynb
|
activatedbonkers/100-Days-of-Code-Challenge
|
a1a376e5373d8fc5fda5df4004115760aa92bfda
|
[
"MIT"
] | null | null | null |
Day12_Detection_Algorithms_YOLOv2/Autonomous_driving_application_Car_detection_v3a.ipynb
|
activatedbonkers/100-Days-of-Code-Challenge
|
a1a376e5373d8fc5fda5df4004115760aa92bfda
|
[
"MIT"
] | 6 |
2020-06-29T18:36:27.000Z
|
2022-01-16T21:18:03.000Z
| 168.444152 | 179,682 | 0.848002 |
[
[
[
"# Autonomous driving - Car detection\n\nWelcome to your week 3 programming assignment. You will learn about object detection using the very powerful YOLO model. Many of the ideas in this notebook are described in the two YOLO papers: [Redmon et al., 2016](https://arxiv.org/abs/1506.02640) and [Redmon and Farhadi, 2016](https://arxiv.org/abs/1612.08242). \n\n**You will learn to**:\n- Use object detection on a car detection dataset\n- Deal with bounding boxes\n\n",
"_____no_output_____"
],
[
"## <font color='darkblue'>Updates</font>\n\n#### If you were working on the notebook before this update...\n* The current notebook is version \"3a\".\n* You can find your original work saved in the notebook with the previous version name (\"v3\") \n* To view the file directory, go to the menu \"File->Open\", and this will open a new tab that shows the file directory.\n\n#### List of updates\n* Clarified \"YOLO\" instructions preceding the code. \n* Added details about anchor boxes.\n* Added explanation of how score is calculated.\n* `yolo_filter_boxes`: added additional hints. Clarify syntax for argmax and max.\n* `iou`: clarify instructions for finding the intersection.\n* `iou`: give variable names for all 8 box vertices, for clarity. Adds `width` and `height` variables for clarity.\n* `iou`: add test cases to check handling of non-intersecting boxes, intersection at vertices, or intersection at edges.\n* `yolo_non_max_suppression`: clarify syntax for tf.image.non_max_suppression and keras.gather.\n* \"convert output of the model to usable bounding box tensors\": Provides a link to the definition of `yolo_head`.\n* `predict`: hint on calling sess.run.\n* Spelling, grammar, wording and formatting updates to improve clarity.",
"_____no_output_____"
],
[
"## Import libraries\nRun the following cell to load the packages and dependencies that you will find useful as you build the object detector!",
"_____no_output_____"
]
],
[
[
"import argparse\nimport os\nimport matplotlib.pyplot as plt\nfrom matplotlib.pyplot import imshow\nimport scipy.io\nimport scipy.misc\nimport numpy as np\nimport pandas as pd\nimport PIL\nimport tensorflow as tf\nfrom keras import backend as K\nfrom keras.layers import Input, Lambda, Conv2D\nfrom keras.models import load_model, Model\nfrom yolo_utils import read_classes, read_anchors, generate_colors, preprocess_image, draw_boxes, scale_boxes\nfrom yad2k.models.keras_yolo import yolo_head, yolo_boxes_to_corners, preprocess_true_boxes, yolo_loss, yolo_body\n\n%matplotlib inline",
"_____no_output_____"
]
],
[
[
"**Important Note**: As you can see, we import Keras's backend as K. This means that to use a Keras function in this notebook, you will need to write: `K.function(...)`.",
"_____no_output_____"
],
[
"## 1 - Problem Statement\n\nYou are working on a self-driving car. As a critical component of this project, you'd like to first build a car detection system. To collect data, you've mounted a camera to the hood (meaning the front) of the car, which takes pictures of the road ahead every few seconds while you drive around. \n\n<center>\n<video width=\"400\" height=\"200\" src=\"nb_images/road_video_compressed2.mp4\" type=\"video/mp4\" controls>\n</video>\n</center>\n\n<caption><center> Pictures taken from a car-mounted camera while driving around Silicon Valley. <br> We thank [drive.ai](htps://www.drive.ai/) for providing this dataset.\n</center></caption>\n\nYou've gathered all these images into a folder and have labelled them by drawing bounding boxes around every car you found. Here's an example of what your bounding boxes look like.\n\n<img src=\"nb_images/box_label.png\" style=\"width:500px;height:250;\">\n<caption><center> <u> **Figure 1** </u>: **Definition of a box**<br> </center></caption>\n\nIf you have 80 classes that you want the object detector to recognize, you can represent the class label $c$ either as an integer from 1 to 80, or as an 80-dimensional vector (with 80 numbers) one component of which is 1 and the rest of which are 0. The video lectures had used the latter representation; in this notebook, we will use both representations, depending on which is more convenient for a particular step. \n\nIn this exercise, you will learn how \"You Only Look Once\" (YOLO) performs object detection, and then apply it to car detection. Because the YOLO model is very computationally expensive to train, we will load pre-trained weights for you to use. ",
"_____no_output_____"
],
[
"## 2 - YOLO",
"_____no_output_____"
],
[
"\"You Only Look Once\" (YOLO) is a popular algorithm because it achieves high accuracy while also being able to run in real-time. This algorithm \"only looks once\" at the image in the sense that it requires only one forward propagation pass through the network to make predictions. After non-max suppression, it then outputs recognized objects together with the bounding boxes.\n\n### 2.1 - Model details\n\n#### Inputs and outputs\n- The **input** is a batch of images, and each image has the shape (m, 608, 608, 3)\n- The **output** is a list of bounding boxes along with the recognized classes. Each bounding box is represented by 6 numbers $(p_c, b_x, b_y, b_h, b_w, c)$ as explained above. If you expand $c$ into an 80-dimensional vector, each bounding box is then represented by 85 numbers. \n\n#### Anchor Boxes\n* Anchor boxes are chosen by exploring the training data to choose reasonable height/width ratios that represent the different classes. For this assignment, 5 anchor boxes were chosen for you (to cover the 80 classes), and stored in the file './model_data/yolo_anchors.txt'\n* The dimension for anchor boxes is the second to last dimension in the encoding: $(m, n_H,n_W,anchors,classes)$.\n* The YOLO architecture is: IMAGE (m, 608, 608, 3) -> DEEP CNN -> ENCODING (m, 19, 19, 5, 85). \n\n\n#### Encoding\nLet's look in greater detail at what this encoding represents. \n\n<img src=\"nb_images/architecture.png\" style=\"width:700px;height:400;\">\n<caption><center> <u> **Figure 2** </u>: **Encoding architecture for YOLO**<br> </center></caption>\n\nIf the center/midpoint of an object falls into a grid cell, that grid cell is responsible for detecting that object.",
"_____no_output_____"
],
[
"Since we are using 5 anchor boxes, each of the 19 x19 cells thus encodes information about 5 boxes. Anchor boxes are defined only by their width and height.\n\nFor simplicity, we will flatten the last two last dimensions of the shape (19, 19, 5, 85) encoding. So the output of the Deep CNN is (19, 19, 425).\n\n<img src=\"nb_images/flatten.png\" style=\"width:700px;height:400;\">\n<caption><center> <u> **Figure 3** </u>: **Flattening the last two last dimensions**<br> </center></caption>",
"_____no_output_____"
],
[
"#### Class score\n\nNow, for each box (of each cell) we will compute the following element-wise product and extract a probability that the box contains a certain class. \nThe class score is $score_{c,i} = p_{c} \\times c_{i}$: the probability that there is an object $p_{c}$ times the probability that the object is a certain class $c_{i}$.\n\n<img src=\"nb_images/probability_extraction.png\" style=\"width:700px;height:400;\">\n<caption><center> <u> **Figure 4** </u>: **Find the class detected by each box**<br> </center></caption>\n\n##### Example of figure 4\n* In figure 4, let's say for box 1 (cell 1), the probability that an object exists is $p_{1}=0.60$. So there's a 60% chance that an object exists in box 1 (cell 1). \n* The probability that the object is the class \"category 3 (a car)\" is $c_{3}=0.73$. \n* The score for box 1 and for category \"3\" is $score_{1,3}=0.60 \\times 0.73 = 0.44$. \n* Let's say we calculate the score for all 80 classes in box 1, and find that the score for the car class (class 3) is the maximum. So we'll assign the score 0.44 and class \"3\" to this box \"1\".\n\n#### Visualizing classes\nHere's one way to visualize what YOLO is predicting on an image:\n- For each of the 19x19 grid cells, find the maximum of the probability scores (taking a max across the 80 classes, one maximum for each of the 5 anchor boxes).\n- Color that grid cell according to what object that grid cell considers the most likely.\n\nDoing this results in this picture: \n\n<img src=\"nb_images/proba_map.png\" style=\"width:300px;height:300;\">\n<caption><center> <u> **Figure 5** </u>: Each one of the 19x19 grid cells is colored according to which class has the largest predicted probability in that cell.<br> </center></caption>\n\nNote that this visualization isn't a core part of the YOLO algorithm itself for making predictions; it's just a nice way of visualizing an intermediate result of the algorithm. \n",
"_____no_output_____"
],
[
"#### Visualizing bounding boxes\nAnother way to visualize YOLO's output is to plot the bounding boxes that it outputs. Doing that results in a visualization like this: \n\n<img src=\"nb_images/anchor_map.png\" style=\"width:200px;height:200;\">\n<caption><center> <u> **Figure 6** </u>: Each cell gives you 5 boxes. In total, the model predicts: 19x19x5 = 1805 boxes just by looking once at the image (one forward pass through the network)! Different colors denote different classes. <br> </center></caption>\n\n#### Non-Max suppression\nIn the figure above, we plotted only boxes for which the model had assigned a high probability, but this is still too many boxes. You'd like to reduce the algorithm's output to a much smaller number of detected objects. \n\nTo do so, you'll use **non-max suppression**. Specifically, you'll carry out these steps: \n- Get rid of boxes with a low score (meaning, the box is not very confident about detecting a class; either due to the low probability of any object, or low probability of this particular class).\n- Select only one box when several boxes overlap with each other and detect the same object.\n\n",
"_____no_output_____"
],
[
"### 2.2 - Filtering with a threshold on class scores\n\nYou are going to first apply a filter by thresholding. You would like to get rid of any box for which the class \"score\" is less than a chosen threshold. \n\nThe model gives you a total of 19x19x5x85 numbers, with each box described by 85 numbers. It is convenient to rearrange the (19,19,5,85) (or (19,19,425)) dimensional tensor into the following variables: \n- `box_confidence`: tensor of shape $(19 \\times 19, 5, 1)$ containing $p_c$ (confidence probability that there's some object) for each of the 5 boxes predicted in each of the 19x19 cells.\n- `boxes`: tensor of shape $(19 \\times 19, 5, 4)$ containing the midpoint and dimensions $(b_x, b_y, b_h, b_w)$ for each of the 5 boxes in each cell.\n- `box_class_probs`: tensor of shape $(19 \\times 19, 5, 80)$ containing the \"class probabilities\" $(c_1, c_2, ... c_{80})$ for each of the 80 classes for each of the 5 boxes per cell.\n\n#### **Exercise**: Implement `yolo_filter_boxes()`.\n1. Compute box scores by doing the elementwise product as described in Figure 4 ($p \\times c$). \nThe following code may help you choose the right operator: \n```python\na = np.random.randn(19*19, 5, 1)\nb = np.random.randn(19*19, 5, 80)\nc = a * b # shape of c will be (19*19, 5, 80)\n```\nThis is an example of **broadcasting** (multiplying vectors of different sizes).\n\n2. For each box, find:\n - the index of the class with the maximum box score\n - the corresponding box score\n \n **Useful references**\n * [Keras argmax](https://keras.io/backend/#argmax)\n * [Keras max](https://keras.io/backend/#max)\n\n **Additional Hints**\n * For the `axis` parameter of `argmax` and `max`, if you want to select the **last** axis, one way to do so is to set `axis=-1`. This is similar to Python array indexing, where you can select the last position of an array using `arrayname[-1]`.\n * Applying `max` normally collapses the axis for which the maximum is applied. `keepdims=False` is the default option, and allows that dimension to be removed. We don't need to keep the last dimension after applying the maximum here.\n * Even though the documentation shows `keras.backend.argmax`, use `keras.argmax`. Similarly, use `keras.max`.\n\n\n3. Create a mask by using a threshold. As a reminder: `([0.9, 0.3, 0.4, 0.5, 0.1] < 0.4)` returns: `[False, True, False, False, True]`. The mask should be True for the boxes you want to keep. \n\n4. Use TensorFlow to apply the mask to `box_class_scores`, `boxes` and `box_classes` to filter out the boxes we don't want. You should be left with just the subset of boxes you want to keep. \n\n **Useful reference**:\n * [boolean mask](https://www.tensorflow.org/api_docs/python/tf/boolean_mask) \n\n **Additional Hints**: \n * For the `tf.boolean_mask`, we can keep the default `axis=None`.\n\n**Reminder**: to call a Keras function, you should use `K.function(...)`.",
"_____no_output_____"
]
],
[
[
"# GRADED FUNCTION: yolo_filter_boxes\n\ndef yolo_filter_boxes(box_confidence, boxes, box_class_probs, threshold = .6):\n \"\"\"Filters YOLO boxes by thresholding on object and class confidence.\n \n Arguments:\n box_confidence -- tensor of shape (19, 19, 5, 1)\n boxes -- tensor of shape (19, 19, 5, 4)\n box_class_probs -- tensor of shape (19, 19, 5, 80)\n threshold -- real value, if [ highest class probability score < threshold], then get rid of the corresponding box\n \n Returns:\n scores -- tensor of shape (None,), containing the class probability score for selected boxes\n boxes -- tensor of shape (None, 4), containing (b_x, b_y, b_h, b_w) coordinates of selected boxes\n classes -- tensor of shape (None,), containing the index of the class detected by the selected boxes\n \n Note: \"None\" is here because you don't know the exact number of selected boxes, as it depends on the threshold. \n For example, the actual output size of scores would be (10,) if there are 10 boxes.\n \"\"\"\n \n # Step 1: Compute box scores\n ### START CODE HERE ### (≈ 1 line)\n box_scores = np.multiply(box_confidence, box_class_probs)\n ### END CODE HERE ###\n \n # Step 2: Find the box_classes using the max box_scores, keep track of the corresponding score\n ### START CODE HERE ### (≈ 2 lines)\n box_classes = K.argmax(box_scores, axis = -1)\n box_class_scores = K.max(box_scores, axis = -1) \n ### END CODE HERE ###\n \n # Step 3: Create a filtering mask based on \"box_class_scores\" by using \"threshold\". The mask should have the\n # same dimension as box_class_scores, and be True for the boxes you want to keep (with probability >= threshold)\n ### START CODE HERE ### (≈ 1 line)\n filtering_mask = box_class_scores >= threshold\n ### END CODE HERE ###\n \n # Step 4: Apply the mask to box_class_scores, boxes and box_classes\n ### START CODE HERE ### (≈ 3 lines)\n scores = tf.boolean_mask(box_class_scores, filtering_mask)\n boxes = tf.boolean_mask(boxes, filtering_mask)\n classes = tf.boolean_mask(box_classes, filtering_mask)\n ### END CODE HERE ###\n \n return scores, boxes, classes",
"_____no_output_____"
],
[
"with tf.Session() as test_a:\n box_confidence = tf.random_normal([19, 19, 5, 1], mean=1, stddev=4, seed = 1)\n boxes = tf.random_normal([19, 19, 5, 4], mean=1, stddev=4, seed = 1)\n box_class_probs = tf.random_normal([19, 19, 5, 80], mean=1, stddev=4, seed = 1)\n scores, boxes, classes = yolo_filter_boxes(box_confidence, boxes, box_class_probs, threshold = 0.5)\n print(\"scores[2] = \" + str(scores[2].eval()))\n print(\"boxes[2] = \" + str(boxes[2].eval()))\n print(\"classes[2] = \" + str(classes[2].eval()))\n print(\"scores.shape = \" + str(scores.shape))\n print(\"boxes.shape = \" + str(boxes.shape))\n print(\"classes.shape = \" + str(classes.shape))",
"scores[2] = 10.7506\nboxes[2] = [ 8.42653275 3.27136683 -0.5313437 -4.94137383]\nclasses[2] = 7\nscores.shape = (?,)\nboxes.shape = (?, 4)\nclasses.shape = (?,)\n"
]
],
[
[
"**Expected Output**:\n\n<table>\n <tr>\n <td>\n **scores[2]**\n </td>\n <td>\n 10.7506\n </td>\n </tr>\n <tr>\n <td>\n **boxes[2]**\n </td>\n <td>\n [ 8.42653275 3.27136683 -0.5313437 -4.94137383]\n </td>\n </tr>\n\n <tr>\n <td>\n **classes[2]**\n </td>\n <td>\n 7\n </td>\n </tr>\n <tr>\n <td>\n **scores.shape**\n </td>\n <td>\n (?,)\n </td>\n </tr>\n <tr>\n <td>\n **boxes.shape**\n </td>\n <td>\n (?, 4)\n </td>\n </tr>\n\n <tr>\n <td>\n **classes.shape**\n </td>\n <td>\n (?,)\n </td>\n </tr>\n\n</table>",
"_____no_output_____"
],
[
"**Note** In the test for `yolo_filter_boxes`, we're using random numbers to test the function. In real data, the `box_class_probs` would contain non-zero values between 0 and 1 for the probabilities. The box coordinates in `boxes` would also be chosen so that lengths and heights are non-negative.",
"_____no_output_____"
],
[
"### 2.3 - Non-max suppression ###\n\nEven after filtering by thresholding over the class scores, you still end up with a lot of overlapping boxes. A second filter for selecting the right boxes is called non-maximum suppression (NMS). ",
"_____no_output_____"
],
[
"<img src=\"nb_images/non-max-suppression.png\" style=\"width:500px;height:400;\">\n<caption><center> <u> **Figure 7** </u>: In this example, the model has predicted 3 cars, but it's actually 3 predictions of the same car. Running non-max suppression (NMS) will select only the most accurate (highest probability) of the 3 boxes. <br> </center></caption>\n",
"_____no_output_____"
],
[
"Non-max suppression uses the very important function called **\"Intersection over Union\"**, or IoU.\n<img src=\"nb_images/iou.png\" style=\"width:500px;height:400;\">\n<caption><center> <u> **Figure 8** </u>: Definition of \"Intersection over Union\". <br> </center></caption>\n\n#### **Exercise**: Implement iou(). Some hints:\n- In this code, we use the convention that (0,0) is the top-left corner of an image, (1,0) is the upper-right corner, and (1,1) is the lower-right corner. In other words, the (0,0) origin starts at the top left corner of the image. As x increases, we move to the right. As y increases, we move down.\n- For this exercise, we define a box using its two corners: upper left $(x_1, y_1)$ and lower right $(x_2,y_2)$, instead of using the midpoint, height and width. (This makes it a bit easier to calculate the intersection).\n- To calculate the area of a rectangle, multiply its height $(y_2 - y_1)$ by its width $(x_2 - x_1)$. (Since $(x_1,y_1)$ is the top left and $x_2,y_2$ are the bottom right, these differences should be non-negative.\n- To find the **intersection** of the two boxes $(xi_{1}, yi_{1}, xi_{2}, yi_{2})$: \n - Feel free to draw some examples on paper to clarify this conceptually.\n - The top left corner of the intersection $(xi_{1}, yi_{1})$ is found by comparing the top left corners $(x_1, y_1)$ of the two boxes and finding a vertex that has an x-coordinate that is closer to the right, and y-coordinate that is closer to the bottom.\n - The bottom right corner of the intersection $(xi_{2}, yi_{2})$ is found by comparing the bottom right corners $(x_2,y_2)$ of the two boxes and finding a vertex whose x-coordinate is closer to the left, and the y-coordinate that is closer to the top.\n - The two boxes **may have no intersection**. You can detect this if the intersection coordinates you calculate end up being the top right and/or bottom left corners of an intersection box. Another way to think of this is if you calculate the height $(y_2 - y_1)$ or width $(x_2 - x_1)$ and find that at least one of these lengths is negative, then there is no intersection (intersection area is zero). \n - The two boxes may intersect at the **edges or vertices**, in which case the intersection area is still zero. This happens when either the height or width (or both) of the calculated intersection is zero.\n\n\n**Additional Hints**\n\n- `xi1` = **max**imum of the x1 coordinates of the two boxes\n- `yi1` = **max**imum of the y1 coordinates of the two boxes\n- `xi2` = **min**imum of the x2 coordinates of the two boxes\n- `yi2` = **min**imum of the y2 coordinates of the two boxes\n- `inter_area` = You can use `max(height, 0)` and `max(width, 0)`\n",
"_____no_output_____"
]
],
[
[
"# GRADED FUNCTION: iou\n\ndef iou(box1, box2):\n \"\"\"Implement the intersection over union (IoU) between box1 and box2\n \n Arguments:\n box1 -- first box, list object with coordinates (box1_x1, box1_y1, box1_x2, box_1_y2)\n box2 -- second box, list object with coordinates (box2_x1, box2_y1, box2_x2, box2_y2)\n \"\"\"\n\n # Assign variable names to coordinates for clarity\n (box1_x1, box1_y1, box1_x2, box1_y2) = box1\n (box2_x1, box2_y1, box2_x2, box2_y2) = box2\n \n # Calculate the (yi1, xi1, yi2, xi2) coordinates of the intersection of box1 and box2. Calculate its Area.\n ### START CODE HERE ### (≈ 7 lines)\n xi1 = box1_x1 if box1_x1 > box2_x1 else box2_x1\n yi1 = box1_y1 if box1_y1 > box2_y1 else box2_y1\n xi2 = box1_x2 if box1_x2 < box2_x2 else box2_x2\n yi2 = box1_y2 if box1_y2 < box2_y2 else box2_y2\n inter_width = xi2 - xi1\n inter_height = yi2 - yi1\n inter_area = 0 if (inter_width < 0 or inter_height < 0) else inter_width * inter_height\n ### END CODE HERE ### \n\n # Calculate the Union area by using Formula: Union(A,B) = A + B - Inter(A,B)\n ### START CODE HERE ### (≈ 3 lines)\n box1_area = (box1_x2 - box1_x1) * (box1_y2 - box1_y1)\n box2_area = (box2_x2 - box2_x1) * (box2_y2 - box2_y1)\n union_area = (box1_area + box2_area) - inter_area\n ### END CODE HERE ###\n \n # compute the IoU\n ### START CODE HERE ### (≈ 1 line)\n iou = inter_area / union_area\n ### END CODE HERE ###\n \n return iou",
"_____no_output_____"
],
[
"## Test case 1: boxes intersect\nbox1 = (2, 1, 4, 3)\nbox2 = (1, 2, 3, 4) \nprint(\"iou for intersecting boxes = \" + str(iou(box1, box2)))\n\n## Test case 2: boxes do not intersect\nbox1 = (1,2,3,4)\nbox2 = (5,6,7,8)\nprint(\"iou for non-intersecting boxes = \" + str(iou(box1,box2)))\n\n## Test case 3: boxes intersect at vertices only\nbox1 = (1,1,2,2)\nbox2 = (2,2,3,3)\nprint(\"iou for boxes that only touch at vertices = \" + str(iou(box1,box2)))\n\n## Test case 4: boxes intersect at edge only\nbox1 = (1,1,3,3)\nbox2 = (2,3,3,4)\nprint(\"iou for boxes that only touch at edges = \" + str(iou(box1,box2)))",
"iou for intersecting boxes = 0.14285714285714285\niou for non-intersecting boxes = 0.0\niou for boxes that only touch at vertices = 0.0\niou for boxes that only touch at edges = 0.0\n"
]
],
[
[
"**Expected Output**:\n\n```\niou for intersecting boxes = 0.14285714285714285\niou for non-intersecting boxes = 0.0\niou for boxes that only touch at vertices = 0.0\niou for boxes that only touch at edges = 0.0\n```",
"_____no_output_____"
],
[
"#### YOLO non-max suppression\n\nYou are now ready to implement non-max suppression. The key steps are: \n1. Select the box that has the highest score.\n2. Compute the overlap of this box with all other boxes, and remove boxes that overlap significantly (iou >= `iou_threshold`).\n3. Go back to step 1 and iterate until there are no more boxes with a lower score than the currently selected box.\n\nThis will remove all boxes that have a large overlap with the selected boxes. Only the \"best\" boxes remain.\n\n**Exercise**: Implement yolo_non_max_suppression() using TensorFlow. TensorFlow has two built-in functions that are used to implement non-max suppression (so you don't actually need to use your `iou()` implementation):\n\n** Reference documentation ** \n\n- [tf.image.non_max_suppression()](https://www.tensorflow.org/api_docs/python/tf/image/non_max_suppression)\n```\ntf.image.non_max_suppression(\n boxes,\n scores,\n max_output_size,\n iou_threshold=0.5,\n name=None\n)\n```\nNote that in the version of tensorflow used here, there is no parameter `score_threshold` (it's shown in the documentation for the latest version) so trying to set this value will result in an error message: *got an unexpected keyword argument 'score_threshold.*\n\n- [K.gather()](https://www.tensorflow.org/api_docs/python/tf/keras/backend/gather) \nEven though the documentation shows `tf.keras.backend.gather()`, you can use `keras.gather()`. \n```\nkeras.gather(\n reference,\n indices\n)\n```",
"_____no_output_____"
]
],
[
[
"# GRADED FUNCTION: yolo_non_max_suppression\n\ndef yolo_non_max_suppression(scores, boxes, classes, max_boxes = 10, iou_threshold = 0.5):\n \"\"\"\n Applies Non-max suppression (NMS) to set of boxes\n \n Arguments:\n scores -- tensor of shape (None,), output of yolo_filter_boxes()\n boxes -- tensor of shape (None, 4), output of yolo_filter_boxes() that have been scaled to the image size (see later)\n classes -- tensor of shape (None,), output of yolo_filter_boxes()\n max_boxes -- integer, maximum number of predicted boxes you'd like\n iou_threshold -- real value, \"intersection over union\" threshold used for NMS filtering\n \n Returns:\n scores -- tensor of shape (, None), predicted score for each box\n boxes -- tensor of shape (4, None), predicted box coordinates\n classes -- tensor of shape (, None), predicted class for each box\n \n Note: The \"None\" dimension of the output tensors has obviously to be less than max_boxes. Note also that this\n function will transpose the shapes of scores, boxes, classes. This is made for convenience.\n \"\"\"\n \n max_boxes_tensor = K.variable(max_boxes, dtype='int32') # tensor to be used in tf.image.non_max_suppression()\n K.get_session().run(tf.variables_initializer([max_boxes_tensor])) # initialize variable max_boxes_tensor\n \n # Use tf.image.non_max_suppression() to get the list of indices corresponding to boxes you keep\n ### START CODE HERE ### (≈ 1 line)\n nms_indices = tf.image.non_max_suppression(boxes, scores, max_boxes, iou_threshold)\n ### END CODE HERE ###\n \n # Use K.gather() to select only nms_indices from scores, boxes and classes\n ### START CODE HERE ### (≈ 3 lines)\n scores = K.gather(scores, nms_indices)\n boxes = K.gather(boxes, nms_indices)\n classes = K.gather(classes, nms_indices)\n ### END CODE HERE ###\n \n return scores, boxes, classes",
"_____no_output_____"
],
[
"with tf.Session() as test_b:\n scores = tf.random_normal([54,], mean=1, stddev=4, seed = 1)\n boxes = tf.random_normal([54, 4], mean=1, stddev=4, seed = 1)\n classes = tf.random_normal([54,], mean=1, stddev=4, seed = 1)\n scores, boxes, classes = yolo_non_max_suppression(scores, boxes, classes)\n print(\"scores[2] = \" + str(scores[2].eval()))\n print(\"boxes[2] = \" + str(boxes[2].eval()))\n print(\"classes[2] = \" + str(classes[2].eval()))\n print(\"scores.shape = \" + str(scores.eval().shape))\n print(\"boxes.shape = \" + str(boxes.eval().shape))\n print(\"classes.shape = \" + str(classes.eval().shape))",
"scores[2] = 6.9384\nboxes[2] = [-5.299932 3.13798141 4.45036697 0.95942086]\nclasses[2] = -2.24527\nscores.shape = (10,)\nboxes.shape = (10, 4)\nclasses.shape = (10,)\n"
]
],
[
[
"**Expected Output**:\n\n<table>\n <tr>\n <td>\n **scores[2]**\n </td>\n <td>\n 6.9384\n </td>\n </tr>\n <tr>\n <td>\n **boxes[2]**\n </td>\n <td>\n [-5.299932 3.13798141 4.45036697 0.95942086]\n </td>\n </tr>\n\n <tr>\n <td>\n **classes[2]**\n </td>\n <td>\n -2.24527\n </td>\n </tr>\n <tr>\n <td>\n **scores.shape**\n </td>\n <td>\n (10,)\n </td>\n </tr>\n <tr>\n <td>\n **boxes.shape**\n </td>\n <td>\n (10, 4)\n </td>\n </tr>\n\n <tr>\n <td>\n **classes.shape**\n </td>\n <td>\n (10,)\n </td>\n </tr>\n\n</table>",
"_____no_output_____"
],
[
"### 2.4 Wrapping up the filtering\n\nIt's time to implement a function taking the output of the deep CNN (the 19x19x5x85 dimensional encoding) and filtering through all the boxes using the functions you've just implemented. \n\n**Exercise**: Implement `yolo_eval()` which takes the output of the YOLO encoding and filters the boxes using score threshold and NMS. There's just one last implementational detail you have to know. There're a few ways of representing boxes, such as via their corners or via their midpoint and height/width. YOLO converts between a few such formats at different times, using the following functions (which we have provided): \n\n```python\nboxes = yolo_boxes_to_corners(box_xy, box_wh) \n```\nwhich converts the yolo box coordinates (x,y,w,h) to box corners' coordinates (x1, y1, x2, y2) to fit the input of `yolo_filter_boxes`\n```python\nboxes = scale_boxes(boxes, image_shape)\n```\nYOLO's network was trained to run on 608x608 images. If you are testing this data on a different size image--for example, the car detection dataset had 720x1280 images--this step rescales the boxes so that they can be plotted on top of the original 720x1280 image. \n\nDon't worry about these two functions; we'll show you where they need to be called. ",
"_____no_output_____"
]
],
[
[
"# GRADED FUNCTION: yolo_eval\n\ndef yolo_eval(yolo_outputs, image_shape = (720., 1280.), max_boxes=10, score_threshold=.6, iou_threshold=.5):\n \"\"\"\n Converts the output of YOLO encoding (a lot of boxes) to your predicted boxes along with their scores, box coordinates and classes.\n \n Arguments:\n yolo_outputs -- output of the encoding model (for image_shape of (608, 608, 3)), contains 4 tensors:\n box_confidence: tensor of shape (None, 19, 19, 5, 1)\n box_xy: tensor of shape (None, 19, 19, 5, 2)\n box_wh: tensor of shape (None, 19, 19, 5, 2)\n box_class_probs: tensor of shape (None, 19, 19, 5, 80)\n image_shape -- tensor of shape (2,) containing the input shape, in this notebook we use (608., 608.) (has to be float32 dtype)\n max_boxes -- integer, maximum number of predicted boxes you'd like\n score_threshold -- real value, if [ highest class probability score < threshold], then get rid of the corresponding box\n iou_threshold -- real value, \"intersection over union\" threshold used for NMS filtering\n \n Returns:\n scores -- tensor of shape (None, ), predicted score for each box\n boxes -- tensor of shape (None, 4), predicted box coordinates\n classes -- tensor of shape (None,), predicted class for each box\n \"\"\"\n \n ### START CODE HERE ### \n \n # Retrieve outputs of the YOLO model (≈1 line)\n box_confidence, box_xy, box_wh, box_class_probs = yolo_outputs\n\n # Convert boxes to be ready for filtering functions (convert boxes box_xy and box_wh to corner coordinates)\n boxes = yolo_boxes_to_corners(box_xy, box_wh)\n\n # Use one of the functions you've implemented to perform Score-filtering with a threshold of score_threshold (≈1 line)\n scores, boxes, classes = yolo_filter_boxes(box_confidence, boxes, box_class_probs, threshold = score_threshold)\n \n # Scale boxes back to original image shape.\n boxes = scale_boxes(boxes, image_shape)\n\n # Use one of the functions you've implemented to perform Non-max suppression with \n # maximum number of boxes set to max_boxes and a threshold of iou_threshold (≈1 line)\n scores, boxes, classes = yolo_non_max_suppression(scores, boxes, classes)\n \n ### END CODE HERE ###\n \n return scores, boxes, classes",
"_____no_output_____"
],
[
"with tf.Session() as test_b:\n yolo_outputs = (tf.random_normal([19, 19, 5, 1], mean=1, stddev=4, seed = 1),\n tf.random_normal([19, 19, 5, 2], mean=1, stddev=4, seed = 1),\n tf.random_normal([19, 19, 5, 2], mean=1, stddev=4, seed = 1),\n tf.random_normal([19, 19, 5, 80], mean=1, stddev=4, seed = 1))\n scores, boxes, classes = yolo_eval(yolo_outputs)\n print(\"scores[2] = \" + str(scores[2].eval()))\n print(\"boxes[2] = \" + str(boxes[2].eval()))\n print(\"classes[2] = \" + str(classes[2].eval()))\n print(\"scores.shape = \" + str(scores.eval().shape))\n print(\"boxes.shape = \" + str(boxes.eval().shape))\n print(\"classes.shape = \" + str(classes.eval().shape))",
"scores[2] = 138.791\nboxes[2] = [ 1292.32971191 -278.52166748 3876.98925781 -835.56494141]\nclasses[2] = 54\nscores.shape = (10,)\nboxes.shape = (10, 4)\nclasses.shape = (10,)\n"
]
],
[
[
"**Expected Output**:\n\n<table>\n <tr>\n <td>\n **scores[2]**\n </td>\n <td>\n 138.791\n </td>\n </tr>\n <tr>\n <td>\n **boxes[2]**\n </td>\n <td>\n [ 1292.32971191 -278.52166748 3876.98925781 -835.56494141]\n </td>\n </tr>\n\n <tr>\n <td>\n **classes[2]**\n </td>\n <td>\n 54\n </td>\n </tr>\n <tr>\n <td>\n **scores.shape**\n </td>\n <td>\n (10,)\n </td>\n </tr>\n <tr>\n <td>\n **boxes.shape**\n </td>\n <td>\n (10, 4)\n </td>\n </tr>\n\n <tr>\n <td>\n **classes.shape**\n </td>\n <td>\n (10,)\n </td>\n </tr>\n\n</table>",
"_____no_output_____"
],
[
"## Summary for YOLO:\n- Input image (608, 608, 3)\n- The input image goes through a CNN, resulting in a (19,19,5,85) dimensional output. \n- After flattening the last two dimensions, the output is a volume of shape (19, 19, 425):\n - Each cell in a 19x19 grid over the input image gives 425 numbers. \n - 425 = 5 x 85 because each cell contains predictions for 5 boxes, corresponding to 5 anchor boxes, as seen in lecture. \n - 85 = 5 + 80 where 5 is because $(p_c, b_x, b_y, b_h, b_w)$ has 5 numbers, and 80 is the number of classes we'd like to detect\n- You then select only few boxes based on:\n - Score-thresholding: throw away boxes that have detected a class with a score less than the threshold\n - Non-max suppression: Compute the Intersection over Union and avoid selecting overlapping boxes\n- This gives you YOLO's final output. ",
"_____no_output_____"
],
[
"## 3 - Test YOLO pre-trained model on images",
"_____no_output_____"
],
[
"In this part, you are going to use a pre-trained model and test it on the car detection dataset. We'll need a session to execute the computation graph and evaluate the tensors.",
"_____no_output_____"
]
],
[
[
"sess = K.get_session()",
"_____no_output_____"
]
],
[
[
"### 3.1 - Defining classes, anchors and image shape.\n\n* Recall that we are trying to detect 80 classes, and are using 5 anchor boxes. \n* We have gathered the information on the 80 classes and 5 boxes in two files \"coco_classes.txt\" and \"yolo_anchors.txt\". \n* We'll read class names and anchors from text files.\n* The car detection dataset has 720x1280 images, which we've pre-processed into 608x608 images. ",
"_____no_output_____"
]
],
[
[
"class_names = read_classes(\"model_data/coco_classes.txt\")\nanchors = read_anchors(\"model_data/yolo_anchors.txt\")\nimage_shape = (720., 1280.) ",
"_____no_output_____"
]
],
[
[
"### 3.2 - Loading a pre-trained model\n\n* Training a YOLO model takes a very long time and requires a fairly large dataset of labelled bounding boxes for a large range of target classes. \n* You are going to load an existing pre-trained Keras YOLO model stored in \"yolo.h5\". \n* These weights come from the official YOLO website, and were converted using a function written by Allan Zelener. References are at the end of this notebook. Technically, these are the parameters from the \"YOLOv2\" model, but we will simply refer to it as \"YOLO\" in this notebook.\n\nRun the cell below to load the model from this file.",
"_____no_output_____"
]
],
[
[
"yolo_model = load_model(\"model_data/yolo.h5\")",
"/opt/conda/lib/python3.6/site-packages/keras/models.py:251: UserWarning: No training configuration found in save file: the model was *not* compiled. Compile it manually.\n warnings.warn('No training configuration found in save file: '\n"
]
],
[
[
"This loads the weights of a trained YOLO model. Here's a summary of the layers your model contains.",
"_____no_output_____"
]
],
[
[
"yolo_model.summary()",
"____________________________________________________________________________________________________\nLayer (type) Output Shape Param # Connected to \n====================================================================================================\ninput_1 (InputLayer) (None, 608, 608, 3) 0 \n____________________________________________________________________________________________________\nconv2d_1 (Conv2D) (None, 608, 608, 32) 864 input_1[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_1 (BatchNorm (None, 608, 608, 32) 128 conv2d_1[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_1 (LeakyReLU) (None, 608, 608, 32) 0 batch_normalization_1[0][0] \n____________________________________________________________________________________________________\nmax_pooling2d_1 (MaxPooling2D) (None, 304, 304, 32) 0 leaky_re_lu_1[0][0] \n____________________________________________________________________________________________________\nconv2d_2 (Conv2D) (None, 304, 304, 64) 18432 max_pooling2d_1[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_2 (BatchNorm (None, 304, 304, 64) 256 conv2d_2[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_2 (LeakyReLU) (None, 304, 304, 64) 0 batch_normalization_2[0][0] \n____________________________________________________________________________________________________\nmax_pooling2d_2 (MaxPooling2D) (None, 152, 152, 64) 0 leaky_re_lu_2[0][0] \n____________________________________________________________________________________________________\nconv2d_3 (Conv2D) (None, 152, 152, 128) 73728 max_pooling2d_2[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_3 (BatchNorm (None, 152, 152, 128) 512 conv2d_3[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_3 (LeakyReLU) (None, 152, 152, 128) 0 batch_normalization_3[0][0] \n____________________________________________________________________________________________________\nconv2d_4 (Conv2D) (None, 152, 152, 64) 8192 leaky_re_lu_3[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_4 (BatchNorm (None, 152, 152, 64) 256 conv2d_4[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_4 (LeakyReLU) (None, 152, 152, 64) 0 batch_normalization_4[0][0] \n____________________________________________________________________________________________________\nconv2d_5 (Conv2D) (None, 152, 152, 128) 73728 leaky_re_lu_4[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_5 (BatchNorm (None, 152, 152, 128) 512 conv2d_5[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_5 (LeakyReLU) (None, 152, 152, 128) 0 batch_normalization_5[0][0] \n____________________________________________________________________________________________________\nmax_pooling2d_3 (MaxPooling2D) (None, 76, 76, 128) 0 leaky_re_lu_5[0][0] \n____________________________________________________________________________________________________\nconv2d_6 (Conv2D) (None, 76, 76, 256) 294912 max_pooling2d_3[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_6 (BatchNorm (None, 76, 76, 256) 1024 conv2d_6[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_6 (LeakyReLU) (None, 76, 76, 256) 0 batch_normalization_6[0][0] \n____________________________________________________________________________________________________\nconv2d_7 (Conv2D) (None, 76, 76, 128) 32768 leaky_re_lu_6[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_7 (BatchNorm (None, 76, 76, 128) 512 conv2d_7[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_7 (LeakyReLU) (None, 76, 76, 128) 0 batch_normalization_7[0][0] \n____________________________________________________________________________________________________\nconv2d_8 (Conv2D) (None, 76, 76, 256) 294912 leaky_re_lu_7[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_8 (BatchNorm (None, 76, 76, 256) 1024 conv2d_8[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_8 (LeakyReLU) (None, 76, 76, 256) 0 batch_normalization_8[0][0] \n____________________________________________________________________________________________________\nmax_pooling2d_4 (MaxPooling2D) (None, 38, 38, 256) 0 leaky_re_lu_8[0][0] \n____________________________________________________________________________________________________\nconv2d_9 (Conv2D) (None, 38, 38, 512) 1179648 max_pooling2d_4[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_9 (BatchNorm (None, 38, 38, 512) 2048 conv2d_9[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_9 (LeakyReLU) (None, 38, 38, 512) 0 batch_normalization_9[0][0] \n____________________________________________________________________________________________________\nconv2d_10 (Conv2D) (None, 38, 38, 256) 131072 leaky_re_lu_9[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_10 (BatchNor (None, 38, 38, 256) 1024 conv2d_10[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_10 (LeakyReLU) (None, 38, 38, 256) 0 batch_normalization_10[0][0] \n____________________________________________________________________________________________________\nconv2d_11 (Conv2D) (None, 38, 38, 512) 1179648 leaky_re_lu_10[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_11 (BatchNor (None, 38, 38, 512) 2048 conv2d_11[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_11 (LeakyReLU) (None, 38, 38, 512) 0 batch_normalization_11[0][0] \n____________________________________________________________________________________________________\nconv2d_12 (Conv2D) (None, 38, 38, 256) 131072 leaky_re_lu_11[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_12 (BatchNor (None, 38, 38, 256) 1024 conv2d_12[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_12 (LeakyReLU) (None, 38, 38, 256) 0 batch_normalization_12[0][0] \n____________________________________________________________________________________________________\nconv2d_13 (Conv2D) (None, 38, 38, 512) 1179648 leaky_re_lu_12[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_13 (BatchNor (None, 38, 38, 512) 2048 conv2d_13[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_13 (LeakyReLU) (None, 38, 38, 512) 0 batch_normalization_13[0][0] \n____________________________________________________________________________________________________\nmax_pooling2d_5 (MaxPooling2D) (None, 19, 19, 512) 0 leaky_re_lu_13[0][0] \n____________________________________________________________________________________________________\nconv2d_14 (Conv2D) (None, 19, 19, 1024) 4718592 max_pooling2d_5[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_14 (BatchNor (None, 19, 19, 1024) 4096 conv2d_14[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_14 (LeakyReLU) (None, 19, 19, 1024) 0 batch_normalization_14[0][0] \n____________________________________________________________________________________________________\nconv2d_15 (Conv2D) (None, 19, 19, 512) 524288 leaky_re_lu_14[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_15 (BatchNor (None, 19, 19, 512) 2048 conv2d_15[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_15 (LeakyReLU) (None, 19, 19, 512) 0 batch_normalization_15[0][0] \n____________________________________________________________________________________________________\nconv2d_16 (Conv2D) (None, 19, 19, 1024) 4718592 leaky_re_lu_15[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_16 (BatchNor (None, 19, 19, 1024) 4096 conv2d_16[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_16 (LeakyReLU) (None, 19, 19, 1024) 0 batch_normalization_16[0][0] \n____________________________________________________________________________________________________\nconv2d_17 (Conv2D) (None, 19, 19, 512) 524288 leaky_re_lu_16[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_17 (BatchNor (None, 19, 19, 512) 2048 conv2d_17[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_17 (LeakyReLU) (None, 19, 19, 512) 0 batch_normalization_17[0][0] \n____________________________________________________________________________________________________\nconv2d_18 (Conv2D) (None, 19, 19, 1024) 4718592 leaky_re_lu_17[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_18 (BatchNor (None, 19, 19, 1024) 4096 conv2d_18[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_18 (LeakyReLU) (None, 19, 19, 1024) 0 batch_normalization_18[0][0] \n____________________________________________________________________________________________________\nconv2d_19 (Conv2D) (None, 19, 19, 1024) 9437184 leaky_re_lu_18[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_19 (BatchNor (None, 19, 19, 1024) 4096 conv2d_19[0][0] \n____________________________________________________________________________________________________\nconv2d_21 (Conv2D) (None, 38, 38, 64) 32768 leaky_re_lu_13[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_19 (LeakyReLU) (None, 19, 19, 1024) 0 batch_normalization_19[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_21 (BatchNor (None, 38, 38, 64) 256 conv2d_21[0][0] \n____________________________________________________________________________________________________\nconv2d_20 (Conv2D) (None, 19, 19, 1024) 9437184 leaky_re_lu_19[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_21 (LeakyReLU) (None, 38, 38, 64) 0 batch_normalization_21[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_20 (BatchNor (None, 19, 19, 1024) 4096 conv2d_20[0][0] \n____________________________________________________________________________________________________\nspace_to_depth_x2 (Lambda) (None, 19, 19, 256) 0 leaky_re_lu_21[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_20 (LeakyReLU) (None, 19, 19, 1024) 0 batch_normalization_20[0][0] \n____________________________________________________________________________________________________\nconcatenate_1 (Concatenate) (None, 19, 19, 1280) 0 space_to_depth_x2[0][0] \n leaky_re_lu_20[0][0] \n____________________________________________________________________________________________________\nconv2d_22 (Conv2D) (None, 19, 19, 1024) 11796480 concatenate_1[0][0] \n____________________________________________________________________________________________________\nbatch_normalization_22 (BatchNor (None, 19, 19, 1024) 4096 conv2d_22[0][0] \n____________________________________________________________________________________________________\nleaky_re_lu_22 (LeakyReLU) (None, 19, 19, 1024) 0 batch_normalization_22[0][0] \n____________________________________________________________________________________________________\nconv2d_23 (Conv2D) (None, 19, 19, 425) 435625 leaky_re_lu_22[0][0] \n====================================================================================================\nTotal params: 50,983,561\nTrainable params: 50,962,889\nNon-trainable params: 20,672\n____________________________________________________________________________________________________\n"
]
],
[
[
"**Note**: On some computers, you may see a warning message from Keras. Don't worry about it if you do--it is fine.\n\n**Reminder**: this model converts a preprocessed batch of input images (shape: (m, 608, 608, 3)) into a tensor of shape (m, 19, 19, 5, 85) as explained in Figure (2).",
"_____no_output_____"
],
[
"### 3.3 - Convert output of the model to usable bounding box tensors\n\nThe output of `yolo_model` is a (m, 19, 19, 5, 85) tensor that needs to pass through non-trivial processing and conversion. The following cell does that for you.\n\nIf you are curious about how `yolo_head` is implemented, you can find the function definition in the file ['keras_yolo.py'](https://github.com/allanzelener/YAD2K/blob/master/yad2k/models/keras_yolo.py). The file is located in your workspace in this path 'yad2k/models/keras_yolo.py'.",
"_____no_output_____"
]
],
[
[
"yolo_outputs = yolo_head(yolo_model.output, anchors, len(class_names))",
"_____no_output_____"
]
],
[
[
"You added `yolo_outputs` to your graph. This set of 4 tensors is ready to be used as input by your `yolo_eval` function.",
"_____no_output_____"
],
[
"### 3.4 - Filtering boxes\n\n`yolo_outputs` gave you all the predicted boxes of `yolo_model` in the correct format. You're now ready to perform filtering and select only the best boxes. Let's now call `yolo_eval`, which you had previously implemented, to do this. ",
"_____no_output_____"
]
],
[
[
"scores, boxes, classes = yolo_eval(yolo_outputs, image_shape)",
"_____no_output_____"
]
],
[
[
"### 3.5 - Run the graph on an image\n\nLet the fun begin. You have created a graph that can be summarized as follows:\n\n1. <font color='purple'> yolo_model.input </font> is given to `yolo_model`. The model is used to compute the output <font color='purple'> yolo_model.output </font>\n2. <font color='purple'> yolo_model.output </font> is processed by `yolo_head`. It gives you <font color='purple'> yolo_outputs </font>\n3. <font color='purple'> yolo_outputs </font> goes through a filtering function, `yolo_eval`. It outputs your predictions: <font color='purple'> scores, boxes, classes </font>\n\n**Exercise**: Implement predict() which runs the graph to test YOLO on an image.\nYou will need to run a TensorFlow session, to have it compute `scores, boxes, classes`.\n\nThe code below also uses the following function:\n```python\nimage, image_data = preprocess_image(\"images/\" + image_file, model_image_size = (608, 608))\n```\nwhich outputs:\n- image: a python (PIL) representation of your image used for drawing boxes. You won't need to use it.\n- image_data: a numpy-array representing the image. This will be the input to the CNN.\n\n**Important note**: when a model uses BatchNorm (as is the case in YOLO), you will need to pass an additional placeholder in the feed_dict {K.learning_phase(): 0}.\n\n#### Hint: Using the TensorFlow Session object\n* Recall that above, we called `K.get_Session()` and saved the Session object in `sess`.\n* To evaluate a list of tensors, we call `sess.run()` like this:\n```\nsess.run(fetches=[tensor1,tensor2,tensor3],\n feed_dict={yolo_model.input: the_input_variable,\n K.learning_phase():0\n }\n```\n* Notice that the variables `scores, boxes, classes` are not passed into the `predict` function, but these are global variables that you will use within the `predict` function.",
"_____no_output_____"
]
],
[
[
"def predict(sess, image_file):\n \"\"\"\n Runs the graph stored in \"sess\" to predict boxes for \"image_file\". Prints and plots the predictions.\n \n Arguments:\n sess -- your tensorflow/Keras session containing the YOLO graph\n image_file -- name of an image stored in the \"images\" folder.\n \n Returns:\n out_scores -- tensor of shape (None, ), scores of the predicted boxes\n out_boxes -- tensor of shape (None, 4), coordinates of the predicted boxes\n out_classes -- tensor of shape (None, ), class index of the predicted boxes\n \n Note: \"None\" actually represents the number of predicted boxes, it varies between 0 and max_boxes. \n \"\"\"\n\n # Preprocess your image\n image, image_data = preprocess_image(\"images/\" + image_file, model_image_size = (608, 608))\n\n # Run the session with the correct tensors and choose the correct placeholders in the feed_dict.\n # You'll need to use feed_dict={yolo_model.input: ... , K.learning_phase(): 0})\n ### START CODE HERE ### (≈ 1 line)\n out_scores, out_boxes, out_classes = sess.run(fetches = [scores, boxes, classes], feed_dict = {yolo_model.input: image_data, K.learning_phase(): 0})\n ### END CODE HERE ###\n\n # Print predictions info\n print('Found {} boxes for {}'.format(len(out_boxes), image_file))\n # Generate colors for drawing bounding boxes.\n colors = generate_colors(class_names)\n # Draw bounding boxes on the image file\n draw_boxes(image, out_scores, out_boxes, out_classes, class_names, colors)\n # Save the predicted bounding box on the image\n image.save(os.path.join(\"out\", image_file), quality=90)\n # Display the results in the notebook\n output_image = scipy.misc.imread(os.path.join(\"out\", image_file))\n imshow(output_image)\n \n return out_scores, out_boxes, out_classes",
"_____no_output_____"
]
],
[
[
"Run the following cell on the \"test.jpg\" image to verify that your function is correct.",
"_____no_output_____"
]
],
[
[
"out_scores, out_boxes, out_classes = predict(sess, \"test.jpg\")",
"Found 7 boxes for test.jpg\ncar 0.60 (925, 285) (1045, 374)\ncar 0.66 (706, 279) (786, 350)\nbus 0.67 (5, 266) (220, 407)\ncar 0.70 (947, 324) (1280, 705)\ncar 0.74 (159, 303) (346, 440)\ncar 0.80 (761, 282) (942, 412)\ncar 0.89 (367, 300) (745, 648)\n"
]
],
[
[
"**Expected Output**:\n\n<table>\n <tr>\n <td>\n **Found 7 boxes for test.jpg**\n </td>\n </tr>\n <tr>\n <td>\n **car**\n </td>\n <td>\n 0.60 (925, 285) (1045, 374)\n </td>\n </tr>\n <tr>\n <td>\n **car**\n </td>\n <td>\n 0.66 (706, 279) (786, 350)\n </td>\n </tr>\n <tr>\n <td>\n **bus**\n </td>\n <td>\n 0.67 (5, 266) (220, 407)\n </td>\n </tr>\n <tr>\n <td>\n **car**\n </td>\n <td>\n 0.70 (947, 324) (1280, 705)\n </td>\n </tr>\n <tr>\n <td>\n **car**\n </td>\n <td>\n 0.74 (159, 303) (346, 440)\n </td>\n </tr>\n <tr>\n <td>\n **car**\n </td>\n <td>\n 0.80 (761, 282) (942, 412)\n </td>\n </tr>\n <tr>\n <td>\n **car**\n </td>\n <td>\n 0.89 (367, 300) (745, 648)\n </td>\n </tr>\n</table>",
"_____no_output_____"
],
[
"The model you've just run is actually able to detect 80 different classes listed in \"coco_classes.txt\". To test the model on your own images:\n 1. Click on \"File\" in the upper bar of this notebook, then click \"Open\" to go on your Coursera Hub.\n 2. Add your image to this Jupyter Notebook's directory, in the \"images\" folder\n 3. Write your image's name in the cell above code\n 4. Run the code and see the output of the algorithm!\n\nIf you were to run your session in a for loop over all your images. Here's what you would get:\n\n<center>\n<video width=\"400\" height=\"200\" src=\"nb_images/pred_video_compressed2.mp4\" type=\"video/mp4\" controls>\n</video>\n</center>\n\n<caption><center> Predictions of the YOLO model on pictures taken from a camera while driving around the Silicon Valley <br> Thanks [drive.ai](https://www.drive.ai/) for providing this dataset! </center></caption>",
"_____no_output_____"
],
[
"\n## <font color='darkblue'>What you should remember:\n \n- YOLO is a state-of-the-art object detection model that is fast and accurate\n- It runs an input image through a CNN which outputs a 19x19x5x85 dimensional volume. \n- The encoding can be seen as a grid where each of the 19x19 cells contains information about 5 boxes.\n- You filter through all the boxes using non-max suppression. Specifically: \n - Score thresholding on the probability of detecting a class to keep only accurate (high probability) boxes\n - Intersection over Union (IoU) thresholding to eliminate overlapping boxes\n- Because training a YOLO model from randomly initialized weights is non-trivial and requires a large dataset as well as lot of computation, we used previously trained model parameters in this exercise. If you wish, you can also try fine-tuning the YOLO model with your own dataset, though this would be a fairly non-trivial exercise. ",
"_____no_output_____"
],
[
"**References**: The ideas presented in this notebook came primarily from the two YOLO papers. The implementation here also took significant inspiration and used many components from Allan Zelener's GitHub repository. The pre-trained weights used in this exercise came from the official YOLO website. \n- Joseph Redmon, Santosh Divvala, Ross Girshick, Ali Farhadi - [You Only Look Once: Unified, Real-Time Object Detection](https://arxiv.org/abs/1506.02640) (2015)\n- Joseph Redmon, Ali Farhadi - [YOLO9000: Better, Faster, Stronger](https://arxiv.org/abs/1612.08242) (2016)\n- Allan Zelener - [YAD2K: Yet Another Darknet 2 Keras](https://github.com/allanzelener/YAD2K)\n- The official YOLO website (https://pjreddie.com/darknet/yolo/) ",
"_____no_output_____"
],
[
"**Car detection dataset**:\n<a rel=\"license\" href=\"http://creativecommons.org/licenses/by/4.0/\"><img alt=\"Creative Commons License\" style=\"border-width:0\" src=\"https://i.creativecommons.org/l/by/4.0/88x31.png\" /></a><br /><span xmlns:dct=\"http://purl.org/dc/terms/\" property=\"dct:title\">The Drive.ai Sample Dataset</span> (provided by drive.ai) is licensed under a <a rel=\"license\" href=\"http://creativecommons.org/licenses/by/4.0/\">Creative Commons Attribution 4.0 International License</a>. We are grateful to Brody Huval, Chih Hu and Rahul Patel for providing this data. ",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown"
] |
[
[
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown",
"markdown"
]
] |
cbdb23743a5012d09eb90f65388d5a2bc4271c35
| 424,208 |
ipynb
|
Jupyter Notebook
|
bert-pipeline/.ipynb_checkpoints/Predicting_Stressor_with_BERT_on_TF_Hub-checkpoint.ipynb
|
PervasiveWellbeingTech/Popbots-mTurk-HITS
|
aa7162d69824a44c97c48a321cf4f3eb1e9a0ec1
|
[
"MIT"
] | null | null | null |
bert-pipeline/.ipynb_checkpoints/Predicting_Stressor_with_BERT_on_TF_Hub-checkpoint.ipynb
|
PervasiveWellbeingTech/Popbots-mTurk-HITS
|
aa7162d69824a44c97c48a321cf4f3eb1e9a0ec1
|
[
"MIT"
] | 23 |
2020-04-23T17:47:46.000Z
|
2020-09-03T17:09:24.000Z
|
bert-pipeline/.ipynb_checkpoints/Predicting_Stressor_with_BERT_on_TF_Hub-checkpoint.ipynb
|
PervasiveWellbeingTech/Popbots-mTurk-HITS
|
aa7162d69824a44c97c48a321cf4f3eb1e9a0ec1
|
[
"MIT"
] | 1 |
2021-06-28T17:06:42.000Z
|
2021-06-28T17:06:42.000Z
| 49.551221 | 48,044 | 0.509425 |
[
[
[
"# Licences / Notes",
"_____no_output_____"
]
],
[
[
"# Copyright 2019 Google Inc.\n\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n\n# http://www.apache.org/licenses/LICENSE-2.0\n\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.",
"_____no_output_____"
],
[
"#Adapted by Thierry Lincoln in November,2019 from this Colab notebook:\n#https://colab.research.google.com/github/google-research/bert/blob/master/predicting_movie_reviews_with_bert_on_tf_hub.ipynb.\n#Changes includes \n\n# - Reading our stressor data and parsing it properly\n# - reconfiguring the last layer to include N neurons corresponding to N categories\n# - correcting the probability output so that it follows [0,1] proper pattern \n# - better analysis with confusion matrix\n# - exporting to pb format for tensorflow serving api",
"_____no_output_____"
]
],
[
[
"Intro:\n\nIf you’ve been following Natural Language Processing over the past year, you’ve probably heard of BERT: Bidirectional Encoder Representations from Transformers. It’s a neural network architecture designed by Google researchers that’s totally transformed what’s state-of-the-art for NLP tasks, like text classification, translation, summarization, and question answering.\n\nNow that BERT's been added to [TF Hub](https://www.tensorflow.org/hub) as a loadable module, it's easy(ish) to add into existing Tensorflow text pipelines. In an existing pipeline, BERT can replace text embedding layers like ELMO and GloVE. Alternatively, [finetuning](http://wiki.fast.ai/index.php/Fine_tuning) BERT can provide both an accuracy boost and faster training time in many cases.\n\nSome code was adapted from [this colab notebook](https://colab.sandbox.google.com/github/tensorflow/tpu/blob/master/tools/colab/bert_finetuning_with_cloud_tpus.ipynb). Let's get started!",
"_____no_output_____"
],
[
"# Loading Libraries",
"_____no_output_____"
]
],
[
[
"os.environ['LD_LIBRARY_PATH'] = '/usr/local/cuda-10.0/lib64:/usr/local/cuda-10.0/lib'\n#os.environ[\"CUDA_DEVICE_ORDER\"] = \"PCI_BUS_ID\"\nos.environ[\"CUDA_VISIBLE_DEVICES\"] = \"0\" #(or \"1\" or \"2\")",
"_____no_output_____"
],
[
"import sys\nprint(sys.executable)",
"/commuter/thierrylincoln/Tf1.1_py36/bin/python3.6\n"
],
[
"#export LD_LIBRARY_PATH=/usr/local/cuda-10.0/lib64:/usr/local/cuda-10.0/lib\n#export CUDA_VISIBLE_DEVICES=0",
"_____no_output_____"
],
[
"from sklearn.model_selection import train_test_split\nfrom sklearn.model_selection import StratifiedShuffleSplit\n\nimport pandas as pd\nimport tensorflow as tf\nimport tensorflow_hub as hub\nfrom datetime import datetime\n\nimport matplotlib.pyplot as plt\nfrom sklearn.utils.multiclass import unique_labels\nfrom sklearn.metrics import f1_score,confusion_matrix,classification_report,accuracy_score\n\n\npd.set_option('display.max_rows', 500)\npd.set_option('display.max_columns', 500)\npd.set_option('display.max_colwidth', 1000)",
"_____no_output_____"
],
[
"print(tf.__version__) #needs to be version 1.15.0, version 2.0 doesn't work with this notebook",
"1.15.0\n"
],
[
"\nconfig = tf.ConfigProto()\n#config.graph_options.optimizer_options.global_jit_level = tf.OptimizerOptions.ON_1\n#config.gpu_options.visible_device_list=\"0\"\n\n\n\nfrom tensorflow.python.client import device_lib\n\ndevice_lib.list_local_devices()",
"_____no_output_____"
]
],
[
[
"In addition to the standard libraries we imported above, we'll need to install BERT's python package.",
"_____no_output_____"
]
],
[
[
"#!pip install bert-tensorflow",
"_____no_output_____"
],
[
"import bert\nfrom bert import run_classifier_with_tfhub\nfrom bert import optimization\nfrom bert import tokenization\nimport numpy as np\n",
"WARNING:tensorflow:From /commuter/thierrylincoln/Tf1.1_py36/lib/python3.6/site-packages/bert/optimization.py:87: The name tf.train.Optimizer is deprecated. Please use tf.compat.v1.train.Optimizer instead.\n\n"
]
],
[
[
"Below, we'll set an output directory location to store our model output and checkpoints. This can be a local directory, in which case you'd set OUTPUT_DIR to the name of the directory you'd like to create. \n\n\nSet DO_DELETE to rewrite the OUTPUT_DIR if it exists. Otherwise, Tensorflow will load existing model checkpoints from that directory (if they exist).",
"_____no_output_____"
],
[
"## Utils functions ",
"_____no_output_____"
]
],
[
[
"def get_test_experiment_df(test):\n test_predictions = [x[0]['probabilities'] for x in zip(getListPrediction(in_sentences=list(test['Answer.Stressor'])))]\n test_live_labels = np.array(test_predictions).argmax(axis=1)\n test['Predicted'] = [label_list_text[x] for x in test_live_labels] # appending the labels to the dataframe\n\n probabilities_df_live = pd.DataFrame(test_predictions) # creating a proabilities dataset\n probabilities_df_live.columns = label_list_text # naming the columns\n\n test.reset_index(inplace=True,drop=True) # resetting index \n\n experiment_df = pd.concat([test,probabilities_df_live],axis=1, ignore_index=False)\n return test,experiment_df",
"_____no_output_____"
],
[
"def getListPrediction(in_sentences):\n #1\n input_examples = [bert.run_classifier.InputExample(guid=\"\", text_a = x, text_b = None, label = 0) for x in in_sentences] # here, \"\" is just a dummy label\n \n #2\n input_features = bert.run_classifier.convert_examples_to_features(input_examples, label_list, MAX_SEQ_LENGTH, tokenizer)\n \n #3\n predict_input_fn = bert.run_classifier.input_fn_builder(features=input_features, seq_length=MAX_SEQ_LENGTH, is_training=False, drop_remainder=False)\n \n print(input_features[0].input_ids)\n #4\n predictions = estimator.predict(input_fn=predict_input_fn,yield_single_examples=True)\n \n return predictions",
"_____no_output_____"
],
[
"is_normalize_active=False\n\ndef get_confusion_matrix(y_test,predicted,labels):\n class_names=labels\n # plotting confusion matrix\n np.set_printoptions(precision=2)\n\n # Plot non-normalized confusion matrix\n plot_confusion_matrix(y_test, predicted, classes=class_names,\n title='Confusion matrix, without normalization')\n\n # Plot normalized confusion matrix\n plot_confusion_matrix(y_test, predicted, classes=class_names, normalize=True,\n title='Normalized confusion matrix')\n plt.show()\ndef plot_confusion_matrix(y_true, y_pred, classes,\n normalize=False,\n title=None,\n cmap=plt.cm.Blues):\n \"\"\"\n This function prints and plots the confusion matrix.\n Normalization can be applied by setting `normalize=True`.\n \"\"\"\n if not title:\n if normalize:\n title = 'Normalized confusion matrix'\n else:\n title = 'Confusion matrix, without normalization'\n\n # Compute confusion matrix\n cm = confusion_matrix(y_true, y_pred)\n # Only use the labels that appear in the data\n classes =classes\n if normalize:\n cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]\n #print(\"Normalized confusion matrix\")\n else:\n test =1\n #print('Confusion matrix, without normalization')\n\n fig, ax = plt.subplots()\n im = ax.imshow(cm, interpolation='nearest', cmap=cmap)\n #ax.figure.colorbar(im, ax=ax)\n # We want to show all ticks...\n ax.set(xticks=np.arange(cm.shape[1]),\n yticks=np.arange(cm.shape[0]),\n # ... and label them with the respective list entries\n xticklabels=classes, yticklabels=classes,\n title=title,\n ylabel='True label',\n xlabel='Predicted label')\n\n # Rotate the tick labels and set their alignment.\n plt.setp(ax.get_xticklabels(), rotation=45, ha=\"right\",\n rotation_mode=\"anchor\")\n\n # Loop over data dimensions and create text annotations.\n fmt = '.2f' if normalize else 'd'\n thresh = cm.max() / 2.\n for i in range(cm.shape[0]):\n for j in range(cm.shape[1]):\n ax.text(j, i, format(cm[i, j], fmt),\n ha=\"center\", va=\"center\",\n color=\"white\" if cm[i, j] > thresh else \"black\")\n #fig.tight_layout()\n return ax\n",
"_____no_output_____"
]
],
[
[
"# Loading the data",
"_____no_output_____"
]
],
[
[
"def data_prep_bert(df,test_size):\n \n print(\"Filling missing values\")\n df[DATA_COLUMN] = df[DATA_COLUMN].fillna('_NA_')\n \n print(\"Splitting dataframe with shape {} into training and test datasets\".format(df.shape))\n X_train, X_test = train_test_split(df, test_size=test_size, random_state=2018,stratify = df[LABEL_COLUMN_RAW])\n\n return X_train, X_test\n",
"_____no_output_____"
],
[
"def open_dataset(NAME,mapping_index,excluded_categories):\n df = pd.read_csv(PATH+NAME+'.csv',sep =',')\n df.head(10)\n df = df[df[LABEL_COLUMN_RAW].notna()]\n \n \n \n #df.columns = [LABEL_COLUMN_RAW,'Severity',DATA_COLUMN,'Source']\n \n if excluded_categories is not None:\n for category in excluded_categories:\n\n df = df[df[LABEL_COLUMN_RAW] !=category]\n\n label_list=[]\n label_list_final =[]\n if(mapping_index is None):\n df[LABEL_COLUMN_RAW] = df[LABEL_COLUMN_RAW].astype('category')\n df[LABEL_COLUMN], mapping_index = pd.Series(df[LABEL_COLUMN_RAW]).factorize() #uses pandas factorize() to convert to numerical index\n \n \n \n label_list_final = [None] * len(mapping_index.categories)\n label_list_number = [None] * len(mapping_index.categories)\n \n for index,ele in enumerate(list(mapping_index.categories)):\n lindex = mapping_index.get_loc(ele)\n label_list_number[lindex] = lindex\n label_list_final[lindex] = ele\n else:\n df[LABEL_COLUMN] = df[LABEL_COLUMN_RAW].apply(lambda x: mapping_index.get_loc(x))\n \n frequency_dict = df[LABEL_COLUMN_RAW].value_counts().to_dict()\n df[\"class_freq\"] = df[LABEL_COLUMN_RAW].apply(lambda x: frequency_dict[x])\n \n \n return df,mapping_index,label_list_number,label_list_final\n ",
"_____no_output_____"
]
],
[
[
"# Require user changes > Start Here ",
"_____no_output_____"
],
[
"### Experiment Name",
"_____no_output_____"
]
],
[
[
"PATH = './datasets/'\nTODAY_DATE = \"21_04_2020/\"\nEXPERIMENT_NAME = 'test'\nEXPERIMENTS_PATH = PATH + 'experiments/'+TODAY_DATE+EXPERIMENT_NAME\nif not os.path.exists(PATH + 'experiments/'+TODAY_DATE):\n os.mkdir(PATH + 'experiments/'+TODAY_DATE)\nif not os.path.exists(EXPERIMENTS_PATH):\n os.mkdir(EXPERIMENTS_PATH)",
"_____no_output_____"
]
],
[
[
"### Model Hyperparameters",
"_____no_output_____"
]
],
[
[
"# Compute train and warmup steps from batch size\n# These hyperparameters are copied from this colab notebook (https://colab.sandbox.google.com/github/tensorflow/tpu/blob/master/tools/colab/bert_finetuning_with_cloud_tpus.ipynb)\nBATCH_SIZE = 32\nLEARNING_RATE = 2e-5\nNUM_TRAIN_EPOCHS = 3.0\n# Warmup is a period of time where hte learning rate \n# is small and gradually increases--usually helps training.\nWARMUP_PROPORTION = 0.1\n# Model configs\nSAVE_CHECKPOINTS_STEPS = 1000\nSAVE_SUMMARY_STEPS = 100\n\n# We'll set sequences to be at most 64 tokens long.\nMAX_SEQ_LENGTH = 64\n\n\nOUTPUT_DIR = './models/'+EXPERIMENT_NAME+ '/' #_01_04_2020/",
"_____no_output_____"
],
[
"\nDATASET_NAME = '2200_master_csv_mturk_and_live_for_training'\nDATA_COLUMN = 'Answer.Stressor'\nLABEL_COLUMN_RAW = 'Consolidated Updated Categories'#'Answer.Label'\nLABEL_COLUMN = 'label_numeric'\n\ntest_on_mturk_and_popbots_live = True # include live data in training + include mturk in testing\n\n\n#dataset,mapping_index,label_list, label_list_text = open_dataset('mturk900balanced',None)\n\nEXCLUDED_CATEGORIES = ['Other'] #None # # if nothing to exclude put None, THIS ALWAYS MUST BE A LIST \n\ndataset,mapping_index,label_list, label_list_text = open_dataset(DATASET_NAME,None,EXCLUDED_CATEGORIES)\n\n\n \n\nif test_on_mturk_and_popbots_live:\n \n mturk = dataset[dataset['Source']== 'mTurk']\n live = dataset[dataset['Source']== 'Popbots']\n live = live.sample(frac=1).reset_index(drop=True) # shuffle live\n \n TEST_PERCENTAGE = len(live)/(2*len(mturk)) # given to set the percentage of mturk used as test set to have 50/50\n \n print(f\"Test percentage is {TEST_PERCENTAGE}\")\n\n train,test = data_prep_bert(mturk,TEST_PERCENTAGE) # test size from mturk \n train = train.append(live.loc[0:int(len(live)/2)]) # taking 1/2 of that dataset for training\n test = test.append(live.loc[int(len(live)/2):int(len(live))]) # taking 1/2 of live dataset for testing\nelse:\n # or taking live only for testing\n train,test = dataset[dataset['Source']== 'mTurk'],dataset[dataset['Source']== 'Popbots'] \n\n\n#print(f\"Dataset has {len(dataset)} training examples\")\nprint(f\"Normal label list is {label_list}\")\nprint(f\"The labels text is {label_list_text}\")\n\n#Export train test to csv\n#train.to_csv(PATH+'900_CSV_SPLITTED/train.csv')\n#test.to_csv(PATH+'900_CSV_SPLITTED/test.csv')",
"Test percentage is 0.038159958180867745\nFilling missing values\nSplitting dataframe with shape (1913, 13) into training and test datasets\nNormal label list is [0, 1, 2, 3, 4, 5, 6, 7]\nThe labels text is ['Work', 'Family Issues', 'Emotional Turmoil', 'Financial Problem', 'Health or Physical Pain', 'School', 'Social Relationships', 'Everyday Decision Making']\n"
]
],
[
[
"### Train set and test set analysis",
"_____no_output_____"
]
],
[
[
"def print_dataset_info(train,test):\n print(f\"Train size {len(train)} with {len(train[train['Source']== 'Popbots'])} from Popbots and {len(train[train['Source']== 'mTurk'])} from mturk\")\n print(f\"Test size {len(test)} with {len(test[test['Source']== 'Popbots'])} from Popbots and {len(test[test['Source']== 'mTurk'])} from mturk\")\n \n print('\\nTraining distribution:')\n print(pd.pivot_table(train[[LABEL_COLUMN_RAW, 'Source']],index=[LABEL_COLUMN_RAW, 'Source'],columns=None, aggfunc=len)) #.to_clipboard(excel=True)\n \n print('\\nTesting distribution:')\n print(pd.pivot_table(test[[LABEL_COLUMN_RAW, 'Source']],index=[LABEL_COLUMN_RAW, 'Source'],columns=None, aggfunc=len)) #.to_clipboard(excel=True)",
"_____no_output_____"
],
[
"train = train.sample(frac=1).reset_index(drop=True) #reshuffle everything\ntest = test.sample(frac=1).reset_index(drop=True)",
"_____no_output_____"
],
[
"print_dataset_info(train,test)",
"Train size 1914 with 74 from Popbots and 1840 from mturk\nTest size 146 with 73 from Popbots and 73 from mturk\n\nTraining distribution:\nConsolidated Updated Categories Source \nEmotional Turmoil Popbots 1\n mTurk 208\nEveryday Decision Making mTurk 60\nFamily Issues Popbots 6\n mTurk 234\nFinancial Problem Popbots 5\n mTurk 396\nHealth or Physical Pain Popbots 15\n mTurk 142\nSchool Popbots 8\n mTurk 132\nSocial Relationships Popbots 9\n mTurk 34\nWork Popbots 30\n mTurk 634\ndtype: int64\n\nTesting distribution:\nConsolidated Updated Categories Source \nEmotional Turmoil Popbots 4\n mTurk 8\nEveryday Decision Making mTurk 2\nFamily Issues Popbots 3\n mTurk 9\nFinancial Problem mTurk 16\nHealth or Physical Pain Popbots 17\n mTurk 6\nSchool Popbots 4\n mTurk 5\nSocial Relationships Popbots 16\n mTurk 2\nWork Popbots 29\n mTurk 25\ndtype: int64\n"
]
],
[
[
"### Step to reduce the most dominant categories and balance the dataset",
"_____no_output_____"
]
],
[
[
"samplig_cutoff = 100 # all the categories which had less than 100 example won't be sampled down\n\n\nREVERSE_FREQ = 'Max_reverse_sampling_chance'\ntrain[REVERSE_FREQ] = train['class_freq'].apply(lambda x: (max(train['class_freq'])/x)*(max(train['class_freq'])/x)) \n\nsampling_boolean = (train['Source'] != 'Popbots') & (train['class_freq'].astype(float) > samplig_cutoff) \n\n\ntrain_to_be_balanced = train[sampling_boolean]\ntrain_not_resampled = train[~sampling_boolean]\n\ntrain_temp = train_to_be_balanced.sample(n=(1500-len(train_not_resampled)), weights=REVERSE_FREQ, random_state=2020)\ntrain = pd.concat([train_temp,train_not_resampled])",
"_____no_output_____"
],
[
"print_dataset_info(train,test)",
"Train size 1500 with 74 from Popbots and 1426 from mturk\nTest size 146 with 73 from Popbots and 73 from mturk\n\nTraining distribution:\nConsolidated Updated Categories Source \nEmotional Turmoil Popbots 1\n mTurk 208\nEveryday Decision Making mTurk 60\nFamily Issues Popbots 6\n mTurk 231\nFinancial Problem Popbots 5\n mTurk 327\nHealth or Physical Pain Popbots 15\n mTurk 142\nSchool Popbots 8\n mTurk 132\nSocial Relationships Popbots 9\n mTurk 34\nWork Popbots 30\n mTurk 292\ndtype: int64\n\nTesting distribution:\nConsolidated Updated Categories Source \nEmotional Turmoil Popbots 4\n mTurk 8\nEveryday Decision Making mTurk 2\nFamily Issues Popbots 3\n mTurk 9\nFinancial Problem mTurk 16\nHealth or Physical Pain Popbots 17\n mTurk 6\nSchool Popbots 4\n mTurk 5\nSocial Relationships Popbots 16\n mTurk 2\nWork Popbots 29\n mTurk 25\ndtype: int64\n"
],
[
"mapping_index.categories",
"_____no_output_____"
],
[
"train.to_csv(EXPERIMENTS_PATH+'/TRAIN_'+DATASET_NAME+'.csv')\ntest.to_csv(EXPERIMENTS_PATH+'/TEST_'+DATASET_NAME+'.csv')",
"_____no_output_____"
]
],
[
[
"# Require user changes > STOP Here ",
"_____no_output_____"
],
[
"# Data Preprocessing",
"_____no_output_____"
],
[
"For us, our input data is the 'sentence' column and our label is the 'polarity' column ",
"_____no_output_____"
],
[
"#Data Preprocessing\nWe'll need to transform our data into a format BERT understands. This involves two steps. First, we create `InputExample`'s using the constructor provided in the BERT library.\n\n- `text_a` is the text we want to classify, which in this case, is the `Request` field in our Dataframe. \n- `text_b` is used if we're training a model to understand the relationship between sentences (i.e. is `text_b` a translation of `text_a`? Is `text_b` an answer to the question asked by `text_a`?). This doesn't apply to our task, so we can leave `text_b` blank.\n- `label` is the label for our example, i.e. True, False",
"_____no_output_____"
]
],
[
[
"# Use the InputExample class from BERT's run_classifier code to create examples from the data\ntrain_InputExamples = train.apply(lambda x: bert.run_classifier.InputExample(guid=None, # Globally unique ID for bookkeeping, unused in this example\n text_a = x[DATA_COLUMN], \n text_b = None, \n label = x[LABEL_COLUMN]), axis = 1)\n\ntest_InputExamples = test.apply(lambda x: bert.run_classifier.InputExample(guid=None, \n text_a = x[DATA_COLUMN], \n text_b = None, \n label = x[LABEL_COLUMN]), axis = 1)",
"_____no_output_____"
]
],
[
[
"Next, we need to preprocess our data so that it matches the data BERT was \n\n1. List item\n2. List item\n\ntrained on. For this, we'll need to do a couple of things (but don't worry--this is also included in the Python library):\n\n\n1. Lowercase our text (if we're using a BERT lowercase model)\n2. Tokenize it (i.e. \"sally says hi\" -> [\"sally\", \"says\", \"hi\"])\n3. Break words into WordPieces (i.e. \"calling\" -> [\"call\", \"##ing\"])\n4. Map our words to indexes using a vocab file that BERT provides\n5. Add special \"CLS\" and \"SEP\" tokens (see the [readme](https://github.com/google-research/bert))\n6. Append \"index\" and \"segment\" tokens to each input (see the [BERT paper](https://arxiv.org/pdf/1810.04805.pdf))\n\nHappily, we don't have to worry about most of these details.\n\n\n",
"_____no_output_____"
],
[
"To start, we'll need to load a vocabulary file and lowercasing information directly from the BERT tf hub module:",
"_____no_output_____"
]
],
[
[
"# This is a path to an uncased (all lowercase) version of BERT\nBERT_MODEL_HUB = \"https://tfhub.dev/google/bert_uncased_L-12_H-768_A-12/1\"\n\ndef create_tokenizer_from_hub_module():\n \"\"\"Get the vocab file and casing info from the Hub module.\"\"\"\n with tf.Graph().as_default():\n bert_module = hub.Module(BERT_MODEL_HUB)\n tokenization_info = bert_module(signature=\"tokenization_info\", as_dict=True)\n with tf.Session() as sess:\n vocab_file, do_lower_case = sess.run([tokenization_info[\"vocab_file\"],\n tokenization_info[\"do_lower_case\"]])\n \n return bert.tokenization.FullTokenizer(\n vocab_file=vocab_file, do_lower_case=do_lower_case)\n\ntokenizer = create_tokenizer_from_hub_module()",
"INFO:tensorflow:Saver not created because there are no variables in the graph to restore\n"
]
],
[
[
"Great--we just learned that the BERT model we're using expects lowercase data (that's what stored in tokenization_info[\"do_lower_case\"]) and we also loaded BERT's vocab file. We also created a tokenizer, which breaks words into word pieces:",
"_____no_output_____"
]
],
[
[
"tokenizer.tokenize(\"This here's an example of using the BERT tokenizer\")",
"_____no_output_____"
]
],
[
[
"Using our tokenizer, we'll call `run_classifier.convert_examples_to_features` on our InputExamples to convert them into features BERT understands.",
"_____no_output_____"
]
],
[
[
"\n# Convert our train and test features to InputFeatures that BERT understands.\ntrain_features = bert.run_classifier.convert_examples_to_features(train_InputExamples, label_list, MAX_SEQ_LENGTH, tokenizer)\ntest_features = bert.run_classifier.convert_examples_to_features(test_InputExamples, label_list, MAX_SEQ_LENGTH, tokenizer)",
"INFO:tensorflow:Writing example 0 of 1500\n"
]
],
[
[
"# Creating a model",
"_____no_output_____"
],
[
"Now that we've prepared our data, let's focus on building a model. `create_model` does just this below. First, it loads the BERT tf hub module again (this time to extract the computation graph). Next, it creates a single new layer that will be trained to adapt BERT to our classification task. This strategy of using a mostly trained model is called [fine-tuning](http://wiki.fast.ai/index.php/Fine_tuning).",
"_____no_output_____"
],
[
"To understand the `pooled ouput` vs `sequence output` refer to https://www.kaggle.com/questions-and-answers/86510",
"_____no_output_____"
]
],
[
[
"def create_model(is_predicting, input_ids, input_mask, segment_ids, labels,\n num_labels):\n \"\"\"Creates a classification model.\"\"\"\n\n bert_module = hub.Module(\n BERT_MODEL_HUB,\n trainable=True)\n bert_inputs = dict(\n input_ids=input_ids,\n input_mask=input_mask,\n segment_ids=segment_ids)\n bert_outputs = bert_module(\n inputs=bert_inputs,\n signature=\"tokens\",\n as_dict=True)\n\n # Use \"pooled_output\" for classification tasks on an entire sentence.\n # Use \"sequence_outputs\" for token-level output.\n output_layer = bert_outputs[\"pooled_output\"]\n\n hidden_size = output_layer.shape[-1].value\n\n # Create our own layer to tune for politeness data.\n output_weights = tf.get_variable(\n \"output_weights\", [num_labels, hidden_size],\n initializer=tf.truncated_normal_initializer(stddev=0.02))\n\n output_bias = tf.get_variable(\n \"output_bias\", [num_labels], initializer=tf.zeros_initializer())\n\n with tf.variable_scope(\"loss\"):\n\n # Dropout helps prevent overfitting\n output_layer = tf.nn.dropout(output_layer, keep_prob=0.9)\n \n # does the Ax multiplication\n logits = tf.matmul(output_layer, output_weights, transpose_b=True)\n # add the bias eg: Ax+B\n logits = tf.nn.bias_add(logits, output_bias)\n \n \n ########################### HERE ADDITIONAL LAYERS CAN BE ADDED ######################\n \n # compute the log softmax for each neurons/logit\n log_probs = tf.nn.log_softmax(logits, axis=-1)\n #compute the normal softmax to get the probabilities\n probs = tf.nn.softmax(logits, axis=-1)\n \n # Convert labels into one-hot encoding\n one_hot_labels = tf.one_hot(labels, depth=num_labels, dtype=tf.float32)\n \n #classes_weights = tf.constant([1.0,1.0,1.0,1.0,1.0,1.0,0.7], dtype=tf.float32)\n #sample_weights = tf.multiply(one_hot_labels, classes_weights)\n \n \n predicted_labels = tf.squeeze(tf.argmax(log_probs, axis=-1, output_type=tf.int32))\n # If we're predicting, we want predicted labels and the probabiltiies.\n if is_predicting:\n return (predicted_labels, log_probs,probs)\n\n # If we're train/eval, compute loss between predicted and actual label\n #per_example_loss = -tf.reduce_sum(one_hot_labels * log_probs, axis=-1)\n per_example_loss = -tf.reduce_sum(one_hot_labels * log_probs, axis=-1)\n \n loss = tf.reduce_mean(per_example_loss)\n return (loss, predicted_labels, log_probs)\n",
"_____no_output_____"
]
],
[
[
"Next we'll wrap our model function in a `model_fn_builder` function that adapts our model to work for training, evaluation, and prediction.",
"_____no_output_____"
]
],
[
[
"# model_fn_builder actually creates our model function\n# using the passed parameters for num_labels, learning_rate, etc.\ndef model_fn_builder(num_labels, learning_rate, num_train_steps,\n num_warmup_steps):\n \"\"\"Returns `model_fn` closure for TPUEstimator.\"\"\"\n def model_fn(features, labels, mode, params): # pylint: disable=unused-argument\n \"\"\"The `model_fn` for TPUEstimator.\"\"\"\n\n input_ids = features[\"input_ids\"]\n input_mask = features[\"input_mask\"]\n segment_ids = features[\"segment_ids\"]\n label_ids = features[\"label_ids\"]\n\n is_predicting = (mode == tf.estimator.ModeKeys.PREDICT)\n \n # TRAIN and EVAL\n if not is_predicting:\n\n (loss, predicted_labels, log_probs) = create_model(\n is_predicting, input_ids, input_mask, segment_ids, label_ids, num_labels)\n\n train_op = bert.optimization.create_optimizer(\n loss, learning_rate, num_train_steps, num_warmup_steps, use_tpu=False)\n\n # Calculate evaluation metrics. \n def metric_fn(label_ids, predicted_labels):\n accuracy = tf.metrics.accuracy(label_ids, predicted_labels)\n \"\"\"\n f1_score = tf.contrib.metrics.f1_score(\n label_ids,\n predicted_labels)\n \n auc = tf.metrics.auc(\n label_ids,\n predicted_labels)\"\"\"\n recall = tf.metrics.recall(\n label_ids,\n predicted_labels)\n precision = tf.metrics.precision(\n label_ids,\n predicted_labels) \n true_pos = tf.metrics.true_positives(\n label_ids,\n predicted_labels)\n true_neg = tf.metrics.true_negatives(\n label_ids,\n predicted_labels) \n false_pos = tf.metrics.false_positives(\n label_ids,\n predicted_labels) \n false_neg = tf.metrics.false_negatives(\n label_ids,\n predicted_labels)\n return {\n \"eval_accuracy\": accuracy,\n #\"f1_score\": f1_score,\n #\"auc\": auc,\n \"precision\": precision,\n \"recall\": recall,\n \"true_positives\": true_pos,\n \"true_negatives\": true_neg,\n \"false_positives\": false_pos,\n \"false_negatives\": false_neg\n }\n\n eval_metrics = metric_fn(label_ids, predicted_labels)\n\n if mode == tf.estimator.ModeKeys.TRAIN:\n return tf.estimator.EstimatorSpec(mode=mode,\n loss=loss,\n train_op=train_op)\n else:\n return tf.estimator.EstimatorSpec(mode=mode,\n loss=loss,\n eval_metric_ops=eval_metrics)\n else:\n (predicted_labels, log_probs,probs) = create_model(\n is_predicting, input_ids, input_mask, segment_ids, label_ids, num_labels)\n\n predictions = {\n 'probabilities': probs#,\n #'labels': predicted_labels\n }\n return tf.estimator.EstimatorSpec(mode, predictions=predictions)\n\n # Return the actual model function in the closure\n return model_fn\n",
"_____no_output_____"
],
[
"# Compute # train and warmup steps from batch size\nnum_train_steps = int(len(train_features) / BATCH_SIZE * NUM_TRAIN_EPOCHS)\nnum_warmup_steps = int(num_train_steps * WARMUP_PROPORTION)",
"_____no_output_____"
],
[
"# Specify outpit directory and number of checkpoint steps to save\n\nrun_config = tf.estimator.RunConfig(\n model_dir=OUTPUT_DIR,\n save_summary_steps=SAVE_SUMMARY_STEPS,\n save_checkpoints_steps=SAVE_CHECKPOINTS_STEPS)",
"_____no_output_____"
],
[
"model_fn = model_fn_builder(\n num_labels=len(label_list),\n learning_rate=LEARNING_RATE,\n num_train_steps=num_train_steps,\n num_warmup_steps=num_warmup_steps)\n\nestimator = tf.estimator.Estimator(\n model_fn=model_fn,\n config=run_config,\n params={\"batch_size\": BATCH_SIZE})\n",
"INFO:tensorflow:Using config: {'_model_dir': './models/test/', '_tf_random_seed': None, '_save_summary_steps': 100, '_save_checkpoints_steps': 1000, '_save_checkpoints_secs': None, '_session_config': allow_soft_placement: true\ngraph_options {\n rewrite_options {\n meta_optimizer_iterations: ONE\n }\n}\n, '_keep_checkpoint_max': 5, '_keep_checkpoint_every_n_hours': 10000, '_log_step_count_steps': 100, '_train_distribute': None, '_device_fn': None, '_protocol': None, '_eval_distribute': None, '_experimental_distribute': None, '_experimental_max_worker_delay_secs': None, '_session_creation_timeout_secs': 7200, '_service': None, '_cluster_spec': <tensorflow.python.training.server_lib.ClusterSpec object at 0x7faa081fcd30>, '_task_type': 'worker', '_task_id': 0, '_global_id_in_cluster': 0, '_master': '', '_evaluation_master': '', '_is_chief': True, '_num_ps_replicas': 0, '_num_worker_replicas': 1}\n"
]
],
[
[
"Next we create an input builder function that takes our training feature set (`train_features`) and produces a generator. This is a pretty standard design pattern for working with Tensorflow [Estimators](https://www.tensorflow.org/guide/estimators).",
"_____no_output_____"
]
],
[
[
"# Create an input function for training. drop_remainder = True for using TPUs.\ntrain_input_fn = bert.run_classifier.input_fn_builder(\n features=train_features,\n seq_length=MAX_SEQ_LENGTH,\n is_training=True,\n drop_remainder=False)",
"_____no_output_____"
]
],
[
[
"# Training the model",
"_____no_output_____"
]
],
[
[
"print(f'Beginning Training!')\ncurrent_time = datetime.now()\nestimator.train(input_fn=train_input_fn, max_steps=num_train_steps)\nprint(\"Training took time \", datetime.now() - current_time)",
"Beginning Training!\nINFO:tensorflow:Calling model_fn.\n"
]
],
[
[
"# Evaluating the model on Test Set",
"_____no_output_____"
]
],
[
[
"test_input_fn = bert.run_classifier.input_fn_builder(\n features=test_features,\n seq_length=MAX_SEQ_LENGTH,\n is_training=False,\n drop_remainder=False)",
"_____no_output_____"
],
[
"#estimator.evaluate(input_fn=test_input_fn, steps=None)",
"_____no_output_____"
],
[
"#fetching all the probabilities for each line of the test set\ntest_probabilities = [x[0]['probabilities'] for x in zip(estimator.predict(test_input_fn,yield_single_examples=True))]\n\n#taking the argmex for the highest category\ntest_final_labels = np.array(test_probabilities).argmax(axis=1)",
"INFO:tensorflow:Calling model_fn.\n"
]
],
[
[
"### Classification Report",
"_____no_output_____"
]
],
[
[
"report = pd.DataFrame(classification_report(list(test[LABEL_COLUMN]),list(test_final_labels),zero_division=0, output_dict=True)).T\n\nprint(report)",
" precision recall f1-score support\n0 0.921569 0.783333 0.846847 60.000000\n1 0.692308 0.750000 0.720000 12.000000\n2 0.222222 0.666667 0.333333 9.000000\n3 0.866667 0.684211 0.764706 19.000000\n4 0.625000 0.789474 0.697674 19.000000\n5 0.812500 1.000000 0.896552 13.000000\n6 0.000000 0.000000 0.000000 12.000000\n7 0.000000 0.000000 0.000000 2.000000\naccuracy 0.705479 0.705479 0.705479 0.705479\nmacro avg 0.517533 0.584211 0.532389 146.000000\nweighted avg 0.715794 0.705479 0.697885 146.000000\n"
]
],
[
[
"### Confusion Matrix",
"_____no_output_____"
]
],
[
[
"get_confusion_matrix(y_test=test[LABEL_COLUMN],predicted=test_final_labels,labels=label_list_text)",
"_____no_output_____"
]
],
[
[
"### Exporting test set with probabilities",
"_____no_output_____"
]
],
[
[
"test, experiment_df = get_test_experiment_df(test)",
"INFO:tensorflow:Writing example 0 of 146\n"
],
[
"experiment_df.to_csv(EXPERIMENTS_PATH+'/test_with_probabilities.csv') ",
"_____no_output_____"
]
],
[
[
"### RUN ALL CELLS ABOVE ON HERE",
"_____no_output_____"
]
],
[
[
"experiment_df[experiment_df['Predicted'] != experiment_df['Answer.Label']].head(10) # change head(n) to see more ",
"_____no_output_____"
]
],
[
[
"# Exporting the model as Pb format",
"_____no_output_____"
]
],
[
[
"def export_model(dir_path):\n MAX_SEQ_LEN = 128\n def serving_input_receiver_fn():\n \"\"\"An input receiver that expects a serialized tf.Example.\"\"\"\n reciever_tensors = {\n \"input_ids\": tf.placeholder(dtype=tf.int32,\n shape=[1, MAX_SEQ_LEN])\n }\n features = {\n \"label_ids\":tf.placeholder(tf.int32, [None], name='label_ids'),\n \"input_ids\": reciever_tensors['input_ids'],\n \"input_mask\": 1 - tf.cast(tf.equal(reciever_tensors['input_ids'], 0), dtype=tf.int32),\n \"segment_ids\": tf.zeros(dtype=tf.int32,\n shape=[1, MAX_SEQ_LEN])\n \n }\n return tf.estimator.export.ServingInputReceiver(features, reciever_tensors)\n\n estimator._export_to_tpu = False\n estimator.export_saved_model(dir_path, serving_input_receiver_fn)",
"_____no_output_____"
],
[
"export_model('./tfmode/pbformat/')",
"_____no_output_____"
]
],
[
[
"## Getting analysis for a another dataset",
"_____no_output_____"
]
],
[
[
"test_all_live = pd.read_csv(PATH+'PopbotsLive_TestSet_213.csv')\n\ntest_all_live, experiment_df_live = get_test_experiment_df(test_all_live)",
"INFO:tensorflow:Writing example 0 of 213\n"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
]
] |
cbdb245b40a49235ccc76a673884a60af8510929
| 2,679 |
ipynb
|
Jupyter Notebook
|
Chapter6_ObjectOriented/inheritance.ipynb
|
Hasideluxe/UdemyPythonPro
|
67f3b55ebdacc9024ffbcf4fd4d3268a9d3bd760
|
[
"MIT"
] | null | null | null |
Chapter6_ObjectOriented/inheritance.ipynb
|
Hasideluxe/UdemyPythonPro
|
67f3b55ebdacc9024ffbcf4fd4d3268a9d3bd760
|
[
"MIT"
] | null | null | null |
Chapter6_ObjectOriented/inheritance.ipynb
|
Hasideluxe/UdemyPythonPro
|
67f3b55ebdacc9024ffbcf4fd4d3268a9d3bd760
|
[
"MIT"
] | null | null | null | 19.698529 | 79 | 0.463606 |
[
[
[
"class A:\n def __init__(self):\n print(\"Init A called!\")\n self.a = 1\n\nclass B(A):\n def __init__(self):\n super().__init__()\n print(\"Init B called!\")\n self.b = self.a * 2\n\nclass C(A):\n def __init__(self):\n super().__init__()\n print(\"Init C called!\")\n self.c = self.a / 2\n\nclass D(B, C):\n def __init__(self):\n super().__init__()\n print(\"Init D called!\")\n self.d = self.b + self.c",
"_____no_output_____"
],
[
"a = A()",
"Init A called!\n"
],
[
"b = B()",
"Init A called!\nInit B called!\n"
],
[
"c = C()",
"Init A called!\nInit C called!\n"
],
[
"d = D()",
"Init A called!\nInit C called!\nInit B called!\nInit D called!\n"
],
[
"D.mro()",
"_____no_output_____"
]
]
] |
[
"code"
] |
[
[
"code",
"code",
"code",
"code",
"code",
"code"
]
] |
cbdb2999e2669ab37dfdee9bfbc874bff33136ad
| 81,227 |
ipynb
|
Jupyter Notebook
|
AI/2 - Numpy, Pandas, Matplotlib /5. Matplotlib and Seaborn/5. Violin_and_box_plot_Practice/Violin_and_Box_Plot_Practice.ipynb
|
2series/DataScience-Courses
|
5ee71305721a61dfc207d8d7de67a9355530535d
|
[
"MIT"
] | 1 |
2020-02-23T07:40:39.000Z
|
2020-02-23T07:40:39.000Z
|
AI/2 - Numpy, Pandas, Matplotlib /5. Matplotlib and Seaborn/5. Violin_and_box_plot_Practice/Violin_and_Box_Plot_Practice.ipynb
|
2series/DataScience-Courses
|
5ee71305721a61dfc207d8d7de67a9355530535d
|
[
"MIT"
] | null | null | null |
AI/2 - Numpy, Pandas, Matplotlib /5. Matplotlib and Seaborn/5. Violin_and_box_plot_Practice/Violin_and_Box_Plot_Practice.ipynb
|
2series/DataScience-Courses
|
5ee71305721a61dfc207d8d7de67a9355530535d
|
[
"MIT"
] | 3 |
2019-12-05T11:04:58.000Z
|
2020-02-26T10:42:08.000Z
| 248.400612 | 35,248 | 0.895823 |
[
[
[
"## RIHAD VARIAWA, Data Scientist - Who has fun LEARNING, EXPLORING & GROWING\n## prerequisite package imports\nimport numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\nimport seaborn as sb\n\n%matplotlib inline\n\nfrom solutions_biv import violinbox_solution_1",
"_____no_output_____"
]
],
[
[
"We'll continue to make use of the fuel economy dataset in this workspace.",
"_____no_output_____"
]
],
[
[
"fuel_econ = pd.read_csv('./data/fuel_econ.csv')\nfuel_econ.head()",
"_____no_output_____"
]
],
[
[
"**Task**: What is the relationship between the size of a car and the size of its engine? The cars in this dataset are categorized into one of five different vehicle classes based on size. Starting from the smallest, they are: {Minicompact Cars, Subcompact Cars, Compact Cars, Midsize Cars, and Large Cars}. The vehicle classes can be found in the 'VClass' variable, while the engine sizes are in the 'displ' column (in liters). **Hint**: Make sure that the order of vehicle classes makes sense in your plot!",
"_____no_output_____"
]
],
[
[
"sedan_classes = ['Minicompact Cars','Subcompact Cars','Compact Cars','Midsize Cars','Large Cars']\nfuel_econ['VClass'] = pd.Categorical(fuel_econ['VClass'], categories=sedan_classes, ordered=True)\nbase_color = sb.color_palette()[0]\nsb.violinplot(data = fuel_econ, x = 'VClass', y = 'displ', color = base_color);\nplt.xlabel('VClass');\nplt.ylabel('displ');\nplt.xticks(rotation = 15);",
"_____no_output_____"
],
[
"# run this cell to check your work against ours\nviolinbox_solution_1()",
"I used a violin plot to depict the data in this case; you might have chosen a box plot instead. One of the interesting things about the relationship between variables is that it isn't consistent. Compact cars tend to have smaller engine sizes than the minicompact and subcompact cars, even though those two vehicle sizes are smaller. The box plot would make it easier to see that the median displacement for the two smallest vehicle classes is greater than the third quartile of the compact car class.\n"
]
]
] |
[
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
]
] |
cbdb351118aece63331642182345878be92d4862
| 1,276 |
ipynb
|
Jupyter Notebook
|
_unittests/ut_helpgen/notebooks_comment/example_with_comments.ipynb
|
janjagusch/pyquickhelper
|
d42e1579ea20f5add9a9cd2b6d2d0a3533aee40b
|
[
"MIT"
] | 18 |
2015-11-10T08:09:23.000Z
|
2022-02-16T11:46:45.000Z
|
_unittests/ut_helpgen/notebooks_comment/example_with_comments.ipynb
|
janjagusch/pyquickhelper
|
d42e1579ea20f5add9a9cd2b6d2d0a3533aee40b
|
[
"MIT"
] | 321 |
2015-06-14T21:34:28.000Z
|
2021-11-28T17:10:03.000Z
|
_unittests/ut_helpgen/notebooks_comment/example_with_comments.ipynb
|
janjagusch/pyquickhelper
|
d42e1579ea20f5add9a9cd2b6d2d0a3533aee40b
|
[
"MIT"
] | 10 |
2015-06-20T01:35:00.000Z
|
2022-01-19T15:54:32.000Z
| 17.013333 | 57 | 0.49373 |
[
[
[
"# example with comments",
"_____no_output_____"
],
[
"This markdown contains a comment.\n\n[//]: # (RST: .. index:: comment, notebook, rst)\n\nIt looks like this:",
"_____no_output_____"
]
],
[
[
"# [//]: # (RST: .. index:: comment, notebook, rst)",
"_____no_output_____"
],
[
"import pyquickhelper",
"_____no_output_____"
]
]
] |
[
"markdown",
"code"
] |
[
[
"markdown",
"markdown"
],
[
"code",
"code"
]
] |
cbdb3f2c2c7b39c718bc4713fecca4f4bbfa4a61
| 822,289 |
ipynb
|
Jupyter Notebook
|
Preprocessing_SDG_AUS_NZ.ipynb
|
JP-Thoma/Kangaroos-Kiwis-love-UN-SDGs
|
107a62a663566389afe09cd5fae522fcbcddc7d5
|
[
"MIT"
] | null | null | null |
Preprocessing_SDG_AUS_NZ.ipynb
|
JP-Thoma/Kangaroos-Kiwis-love-UN-SDGs
|
107a62a663566389afe09cd5fae522fcbcddc7d5
|
[
"MIT"
] | null | null | null |
Preprocessing_SDG_AUS_NZ.ipynb
|
JP-Thoma/Kangaroos-Kiwis-love-UN-SDGs
|
107a62a663566389afe09cd5fae522fcbcddc7d5
|
[
"MIT"
] | null | null | null | 57.174871 | 4,816 | 0.321892 |
[
[
[
"# Country Comparison Australia - New Zealand\n\nIn the following the data of Australia and New Zealand will be cleaned, preprocessed and compared. We chose these two countries for its geographical proximity, historical, political and economic ties as well as for our domain knowledge of the area. Moreover, the data we see in our Master's program often focuses on the US an Europe, therefore we decided to chang our focus on another important part of the world. \n\nHence, we obtained the complete SDG (Sustainable Development Goals) dataset for both countries for the period of 2006 to 2020 from the World Bank Database. ",
"_____no_output_____"
],
[
"## Data Preprocessing\n\nThis step is necessary because of the high prevalence of missing data. By the end of this script a new dataset with cleaned data will be create whose variables will be used for visual analysis in Tableau.",
"_____no_output_____"
],
[
"#### Importing Libraries and Dataset",
"_____no_output_____"
]
],
[
[
"import pandas as pd\nimport numpy as np",
"_____no_output_____"
],
[
"raw_csv_data = pd.read_csv(\"AUS_NZ_Complete_2006-2020.csv\")\n\nraw_csv_data",
"_____no_output_____"
]
],
[
[
"#### Restructuring Dataset",
"_____no_output_____"
]
],
[
[
"#Create subsets of data for each country\n\nraw_csv_AUS = raw_csv_data[raw_csv_data[\"Country Code\"]==\"AUS\"]\n\nraw_csv_NZL = raw_csv_data[raw_csv_data[\"Country Code\"]==\"NZL\"]",
"_____no_output_____"
],
[
"raw_csv_AUS",
"_____no_output_____"
],
[
"raw_csv_NZL",
"_____no_output_____"
],
[
"#drop columns: Country Name, Country Code, Series Code \n\nraw_csv_AUS.drop([\"Country Name\", \"Country Code\", \"Series Code\"], axis=1, inplace=True)\n\nraw_csv_NZL.drop([\"Country Name\", \"Country Code\", \"Series Code\"], axis=1, inplace=True)",
"C:\\Users\\thaer\\Anaconda3\\lib\\site-packages\\pandas\\core\\frame.py:4908: SettingWithCopyWarning: \nA value is trying to be set on a copy of a slice from a DataFrame\n\nSee the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n errors=errors,\n"
],
[
"raw_csv_AUS",
"_____no_output_____"
],
[
"raw_csv_NZL",
"_____no_output_____"
],
[
"#transpose data \n\ntransposed_csv_AUS = raw_csv_AUS.transpose().reset_index()\ntransposed_csv_AUS.columns = np.arange(len(transposed_csv_AUS.columns))\n\ntransposed_csv_NZL = raw_csv_NZL.transpose().reset_index()\ntransposed_csv_NZL.columns = np.arange(len(transposed_csv_NZL.columns))",
"_____no_output_____"
],
[
"#rename column headers with first row & change series name to year\n\ntransposed_csv_AUS.rename(columns=transposed_csv_AUS.iloc[0], inplace = True)\ntransposed_csv_AUS.drop(index=transposed_csv_AUS.index[0], axis=0, inplace=True)\n\ntransposed_csv_NZL.rename(columns=transposed_csv_NZL.iloc[0], inplace = True)\ntransposed_csv_NZL.drop(index=transposed_csv_NZL.index[0], axis=0, inplace=True)\n\ntransposed_csv_AUS.rename(columns={\"Series Name\": \"Year\"}, inplace=True)\ntransposed_csv_NZL.rename(columns={\"Series Name\": \"Year\"}, inplace=True)",
"_____no_output_____"
],
[
"transposed_csv_AUS",
"_____no_output_____"
],
[
"transposed_csv_NZL",
"_____no_output_____"
],
[
"#Clean year column data\n\ntype(transposed_csv_AUS[\"Year\"][1])\n\ntransposed_csv_AUS[\"Year\"][1][:4]",
"_____no_output_____"
],
[
"transposed_csv_AUS.Year=transposed_csv_AUS.Year.str[:4]\n\ntransposed_csv_NZL.Year=transposed_csv_NZL.Year.str[:4]",
"_____no_output_____"
],
[
"transposed_csv_AUS",
"_____no_output_____"
],
[
"transposed_csv_NZL",
"_____no_output_____"
],
[
"#Replace \"..\" with Nan\n\ntransposed_csv_AUS=transposed_csv_AUS.replace('..', np.nan)\n\ntransposed_csv_NZL=transposed_csv_NZL.replace('..', np.nan)",
"_____no_output_____"
]
],
[
[
"#### Data Exploration",
"_____no_output_____"
]
],
[
[
"#dataset checkpoint\n\nAUS_structured = transposed_csv_AUS.copy()\n\nNZL_structured = transposed_csv_NZL.copy()",
"_____no_output_____"
],
[
"AUS_structured.info()",
"<class 'pandas.core.frame.DataFrame'>\nInt64Index: 15 entries, 1 to 15\nColumns: 1444 entries, Year to Young people (ages 15-24) newly infected with HIV\ndtypes: float64(389), object(1055)\nmemory usage: 169.9+ KB\n"
],
[
"NZL_structured.info()",
"<class 'pandas.core.frame.DataFrame'>\nInt64Index: 15 entries, 1 to 15\nColumns: 1444 entries, Year to Young people (ages 15-24) newly infected with HIV\ndtypes: float64(402), object(1042)\nmemory usage: 169.3+ KB\n"
],
[
"AUS_structured.describe()",
"_____no_output_____"
],
[
"NZL_structured.describe()",
"_____no_output_____"
],
[
"# select non numeric columns AUS\nal_non_numeric = AUS_structured.select_dtypes(exclude=[np.number])\nnon_numeric_cols = al_non_numeric.columns.values\nprint(non_numeric_cols)",
"['Year'\n 'Access to clean fuels and technologies for cooking (% of population)'\n 'Access to electricity (% of population)' ...\n 'Women Business and the Law Index Score (scale 1-100)'\n \"Women's share of population ages 15+ living with HIV (%)\"\n 'Young people (ages 15-24) newly infected with HIV']\n"
],
[
"# select non numeric columns NZL\nal_non_numeric = NZL_structured.select_dtypes(exclude=[np.number])\nnon_numeric_cols = al_non_numeric.columns.values\nprint(non_numeric_cols)",
"['Year'\n 'Access to clean fuels and technologies for cooking (% of population)'\n 'Access to electricity (% of population)' ...\n 'Women Business and the Law Index Score (scale 1-100)'\n \"Women's share of population ages 15+ living with HIV (%)\"\n 'Young people (ages 15-24) newly infected with HIV']\n"
],
[
"# % of missing AUS\nfor col in AUS_structured.columns:\n pct_missing = np.mean(AUS_structured[col].isnull())\n print('{} - {}%'.format(col, round(pct_missing*100)))",
"Year - 0%\nAccess to clean fuels and technologies for cooking (% of population) - 27%\nAccess to electricity (% of population) - 7%\nAccess to electricity, rural (% of rural population) - 7%\nAccess to electricity, urban (% of urban population) - 7%\nAccount ownership at a financial institution or with a mobile-money-service provider (% of population ages 15+) - 80%\nAccount ownership at a financial institution or with a mobile-money-service provider, female (% of population ages 15+) - 80%\nAccount ownership at a financial institution or with a mobile-money-service provider, male (% of population ages 15+) - 80%\nAccount ownership at a financial institution or with a mobile-money-service provider, older adults (% of population ages 25+) - 80%\nAccount ownership at a financial institution or with a mobile-money-service provider, poorest 40% (% of population ages 15+) - 80%\nAccount ownership at a financial institution or with a mobile-money-service provider, primary education or less (% of population ages 15+) - 80%\nAccount ownership at a financial institution or with a mobile-money-service provider, richest 60% (% of population ages 15+) - 80%\nAccount ownership at a financial institution or with a mobile-money-service provider, secondary education or more (% of population ages 15+) - 80%\nAccount ownership at a financial institution or with a mobile-money-service provider, young adults (% of population ages 15-24) - 80%\nAdequacy of social insurance programs (% of total welfare of beneficiary households) - 100%\nAdequacy of social protection and labor programs (% of total welfare of beneficiary households) - 100%\nAdequacy of social safety net programs (% of total welfare of beneficiary households) - 100%\nAdequacy of unemployment benefits and ALMP (% of total welfare of beneficiary households) - 100%\nAdjusted net enrollment rate, primary (% of primary school age children) - 20%\nAdjusted net enrollment rate, primary, female (% of primary school age children) - 80%\nAdjusted net enrollment rate, primary, male (% of primary school age children) - 80%\nAdjusted net national income (annual % growth) - 100%\nAdjusted net national income (constant 2015 US$) - 93%\nAdjusted net national income (current US$) - 7%\nAdjusted net national income per capita (annual % growth) - 100%\nAdjusted net national income per capita (constant 2015 US$) - 93%\nAdjusted net national income per capita (current US$) - 7%\nAdjusted net savings, excluding particulate emission damage (% of GNI) - 7%\nAdjusted net savings, excluding particulate emission damage (current US$) - 7%\nAdjusted net savings, including particulate emission damage (% of GNI) - 7%\nAdjusted net savings, including particulate emission damage (current US$) - 7%\nAdjusted savings: carbon dioxide damage (% of GNI) - 7%\nAdjusted savings: carbon dioxide damage (current US$) - 7%\nAdjusted savings: consumption of fixed capital (% of GNI) - 7%\nAdjusted savings: consumption of fixed capital (current US$) - 7%\nAdjusted savings: education expenditure (% of GNI) - 7%\nAdjusted savings: education expenditure (current US$) - 7%\nAdjusted savings: energy depletion (% of GNI) - 7%\nAdjusted savings: energy depletion (current US$) - 7%\nAdjusted savings: gross savings (% of GNI) - 7%\nAdjusted savings: mineral depletion (% of GNI) - 7%\nAdjusted savings: mineral depletion (current US$) - 7%\nAdjusted savings: natural resources depletion (% of GNI) - 7%\nAdjusted savings: net forest depletion (% of GNI) - 7%\nAdjusted savings: net forest depletion (current US$) - 7%\nAdjusted savings: net national savings (% of GNI) - 7%\nAdjusted savings: net national savings (current US$) - 7%\nAdjusted savings: particulate emission damage (% of GNI) - 7%\nAdjusted savings: particulate emission damage (current US$) - 7%\nAdolescent fertility rate (births per 1,000 women ages 15-19) - 7%\nAdolescents out of school (% of lower secondary school age) - 53%\nAdolescents out of school, female (% of female lower secondary school age) - 60%\nAdolescents out of school, male (% of male lower secondary school age) - 60%\nAdults (ages 15+) and children (ages 0-14) newly infected with HIV - 0%\nAdults (ages 15-49) newly infected with HIV - 0%\nAge dependency ratio (% of working-age population) - 0%\nAge dependency ratio, old (% of working-age population) - 0%\nAge dependency ratio, young (% of working-age population) - 0%\nAgricultural irrigated land (% of total agricultural land) - 13%\nAgricultural land (% of land area) - 13%\nAgricultural land (sq. km) - 13%\nAgricultural machinery, tractors - 100%\nAgricultural machinery, tractors per 100 sq. km of arable land - 100%\nAgricultural methane emissions (% of total) - 80%\nAgricultural methane emissions (thousand metric tons of CO2 equivalent) - 13%\nAgricultural nitrous oxide emissions (% of total) - 80%\nAgricultural nitrous oxide emissions (thousand metric tons of CO2 equivalent) - 13%\nAgricultural raw materials exports (% of merchandise exports) - 7%\nAgricultural raw materials imports (% of merchandise imports) - 7%\nAgriculture, forestry, and fishing, value added (% of GDP) - 0%\nAgriculture, forestry, and fishing, value added (annual % growth) - 0%\nAgriculture, forestry, and fishing, value added (constant 2015 US$) - 0%\nAgriculture, forestry, and fishing, value added (constant LCU) - 0%\nAgriculture, forestry, and fishing, value added (current LCU) - 0%\nAgriculture, forestry, and fishing, value added (current US$) - 0%\nAgriculture, forestry, and fishing, value added per worker (constant 2015 US$) - 7%\nAir transport, freight (million ton-km) - 7%\nAir transport, passengers carried - 7%\nAir transport, registered carrier departures worldwide - 7%\nAlternative and nuclear energy (% of total energy use) - 33%\nAnnual freshwater withdrawals, agriculture (% of total freshwater withdrawal) - 80%\nAnnual freshwater withdrawals, domestic (% of total freshwater withdrawal) - 80%\nAnnual freshwater withdrawals, industry (% of total freshwater withdrawal) - 80%\nAnnual freshwater withdrawals, total (% of internal resources) - 80%\nAnnual freshwater withdrawals, total (billion cubic meters) - 80%\nAnnualized average growth rate in per capita real survey mean consumption or income, bottom 40% of population (%) - 100%\nAnnualized average growth rate in per capita real survey mean consumption or income, total population (%) - 100%\nAntiretroviral therapy coverage (% of people living with HIV) - 0%\nAntiretroviral therapy coverage for PMTCT (% of pregnant women living with HIV) - 100%\nAquaculture production (metric tons) - 13%\nArable land (% of land area) - 13%\nArable land (hectares per person) - 13%\nArable land (hectares) - 13%\nARI treatment (% of children under 5 taken to a health provider) - 100%\nArmed forces personnel (% of total labor force) - 13%\nArmed forces personnel, total - 13%\nArms exports (SIPRI trend indicator values) - 0%\nArms imports (SIPRI trend indicator values) - 0%\nAutomated teller machines (ATMs) (per 100,000 adults) - 0%\nAverage precipitation in depth (mm per year) - 80%\nAverage time to clear exports through customs (days) - 100%\nAverage transaction cost of sending remittances from a specific country (%) - 33%\nAverage transaction cost of sending remittances to a specific country (%) - 100%\nAverage working hours of children, study and work, ages 7-14 (hours per week) - 100%\nAverage working hours of children, study and work, female, ages 7-14 (hours per week) - 100%\nAverage working hours of children, study and work, male, ages 7-14 (hours per week) - 100%\nAverage working hours of children, working only, ages 7-14 (hours per week) - 100%\nAverage working hours of children, working only, female, ages 7-14 (hours per week) - 100%\nAverage working hours of children, working only, male, ages 7-14 (hours per week) - 100%\nBank capital to assets ratio (%) - 0%\nBank liquid reserves to bank assets ratio (%) - 0%\nBank nonperforming loans to total gross loans (%) - 0%\nBattle-related deaths (number of people) - 93%\nBenefit incidence of social insurance programs to poorest quintile (% of total social insurance benefits) - 100%\nBenefit incidence of social protection and labor programs to poorest quintile (% of total SPL benefits) - 100%\nBenefit incidence of social safety net programs to poorest quintile (% of total safety net benefits) - 100%\nBenefit incidence of unemployment benefits and ALMP to poorest quintile (% of total U/ALMP benefits) - 100%\nBinding coverage, all products (%) - 7%\nBinding coverage, manufactured products (%) - 7%\nBinding coverage, primary products (%) - 7%\nBird species, threatened - 93%\nBirth rate, crude (per 1,000 people) - 7%\nBirths attended by skilled health staff (% of total) - 20%\nBorrowers from commercial banks (per 1,000 adults) - 100%\nBound rate, simple mean, all products (%) - 7%\nBound rate, simple mean, manufactured products (%) - 7%\nBound rate, simple mean, primary products (%) - 7%\nBribery incidence (% of firms experiencing at least one bribe payment request) - 100%\nBroad money (% of GDP) - 0%\nBroad money (current LCU) - 0%\nBroad money growth (annual %) - 0%\nBroad money to total reserves ratio - 0%\nBusiness extent of disclosure index (0=less disclosure to 10=more disclosure) - 7%\nCapture fisheries production (metric tons) - 13%\nCause of death, by communicable diseases and maternal, prenatal and nutrition conditions (% of total) - 80%\nCause of death, by injury (% of total) - 80%\nCause of death, by non-communicable diseases (% of total) - 80%\nCentral government debt, total (% of GDP) - 27%\nCentral government debt, total (current LCU) - 27%\nCereal production (metric tons) - 13%\nCereal yield (kg per hectare) - 13%\nChanges in inventories (constant LCU) - 0%\nChanges in inventories (current LCU) - 0%\nChanges in inventories (current US$) - 0%\nCharges for the use of intellectual property, payments (BoP, current US$) - 0%\nCharges for the use of intellectual property, receipts (BoP, current US$) - 0%\nChemicals (% of value added in manufacturing) - 13%\nChild employment in agriculture (% of economically active children ages 7-14) - 100%\nChild employment in agriculture, female (% of female economically active children ages 7-14) - 100%\nChild employment in agriculture, male (% of male economically active children ages 7-14) - 100%\nChild employment in manufacturing (% of economically active children ages 7-14) - 100%\nChild employment in manufacturing, female (% of female economically active children ages 7-14) - 100%\nChild employment in manufacturing, male (% of male economically active children ages 7-14) - 100%\nChild employment in services (% of economically active children ages 7-14) - 100%\nChild employment in services, female (% of female economically active children ages 7-14) - 100%\nChild employment in services, male (% of male economically active children ages 7-14) - 100%\nChildren (0-14) living with HIV - 100%\nChildren (ages 0-14) newly infected with HIV - 100%\nChildren in employment, female (% of female children ages 7-14) - 100%\nChildren in employment, male (% of male children ages 7-14) - 100%\nChildren in employment, self-employed (% of children in employment, ages 7-14) - 100%\nChildren in employment, self-employed, female (% of female children in employment, ages 7-14) - 100%\nChildren in employment, self-employed, male (% of male children in employment, ages 7-14) - 100%\nChildren in employment, study and work (% of children in employment, ages 7-14) - 100%\nChildren in employment, study and work, female (% of female children in employment, ages 7-14) - 100%\nChildren in employment, study and work, male (% of male children in employment, ages 7-14) - 100%\nChildren in employment, total (% of children ages 7-14) - 100%\nChildren in employment, unpaid family workers (% of children in employment, ages 7-14) - 100%\nChildren in employment, unpaid family workers, female (% of female children in employment, ages 7-14) - 100%\nChildren in employment, unpaid family workers, male (% of male children in employment, ages 7-14) - 100%\nChildren in employment, wage workers (% of children in employment, ages 7-14) - 100%\nChildren in employment, wage workers, female (% of female children in employment, ages 7-14) - 100%\nChildren in employment, wage workers, male (% of male children in employment, ages 7-14) - 100%\nChildren in employment, work only (% of children in employment, ages 7-14) - 100%\nChildren in employment, work only, female (% of female children in employment, ages 7-14) - 100%\nChildren in employment, work only, male (% of male children in employment, ages 7-14) - 100%\nChildren out of school (% of primary school age) - 7%\nChildren out of school, female (% of female primary school age) - 67%\nChildren out of school, male (% of male primary school age) - 67%\nChildren out of school, primary - 7%\nChildren out of school, primary, female - 67%\nChildren out of school, primary, male - 67%\nChildren with fever receiving antimalarial drugs (% of children under age 5 with fever) - 100%\nClaims on central government (annual growth as % of broad money) - 0%\nClaims on central government, etc. (% GDP) - 0%\nClaims on other sectors of the domestic economy (% of GDP) - 100%\nClaims on other sectors of the domestic economy (annual growth as % of broad money) - 0%\nClaims on private sector (annual growth as % of broad money) - 0%\nCO2 emissions (kg per 2015 US$ of GDP) - 13%\nCO2 emissions (kg per 2017 PPP $ of GDP) - 13%\nCO2 emissions (kg per PPP $ of GDP) - 13%\nCO2 emissions (kt) - 13%\nCO2 emissions (metric tons per capita) - 13%\nCO2 emissions from electricity and heat production, total (% of total fuel combustion) - 40%\nCO2 emissions from gaseous fuel consumption (% of total) - 27%\nCO2 emissions from gaseous fuel consumption (kt) - 27%\nCO2 emissions from liquid fuel consumption (% of total) - 27%\nCO2 emissions from liquid fuel consumption (kt) - 27%\nCO2 emissions from manufacturing industries and construction (% of total fuel combustion) - 40%\nCO2 emissions from other sectors, excluding residential buildings and commercial and public services (% of total fuel combustion) - 40%\nCO2 emissions from residential buildings and commercial and public services (% of total fuel combustion) - 40%\nCO2 emissions from solid fuel consumption (% of total) - 27%\nCO2 emissions from solid fuel consumption (kt) - 27%\nCO2 emissions from transport (% of total fuel combustion) - 40%\nCO2 intensity (kg per kg of oil equivalent energy use) - 33%\nCoal rents (% of GDP) - 7%\nCombustible renewables and waste (% of total energy) - 33%\nCommercial bank branches (per 100,000 adults) - 0%\nCommercial banks and other lending (PPG + PNG) (NFL, current US$) - 100%\nCommercial service exports (current US$) - 0%\nCommercial service imports (current US$) - 0%\nCommunications, computer, etc. (% of service exports, BoP) - 0%\nCommunications, computer, etc. (% of service imports, BoP) - 0%\nCommunity health workers (per 1,000 people) - 93%\nCompensation of employees (% of expense) - 7%\nCompensation of employees (current LCU) - 7%\nCompleteness of birth registration (%) - 80%\nCompleteness of birth registration, female (%) - 93%\nCompleteness of birth registration, male (%) - 93%\nCompleteness of birth registration, rural (%) - 100%\nCompleteness of birth registration, urban (%) - 100%\nCompleteness of death registration with cause-of-death information (%) - 80%\nCompulsory education, duration (years) - 0%\nComputer, communications and other services (% of commercial service exports) - 0%\nComputer, communications and other services (% of commercial service imports) - 0%\nCondom use, population ages 15-24, female (% of females ages 15-24) - 100%\nCondom use, population ages 15-24, male (% of males ages 15-24) - 100%\nConsumer price index (2010 = 100) - 0%\nConsumption of iodized salt (% of households) - 100%\nContainer port traffic (TEU: 20 foot equivalent units) - 7%\nContraceptive prevalence, any methods (% of women ages 15-49) - 73%\nContraceptive prevalence, modern methods (% of women ages 15-49) - 73%\nContributing family workers, female (% of female employment) (modeled ILO estimate) - 7%\nContributing family workers, male (% of male employment) (modeled ILO estimate) - 7%\nContributing family workers, total (% of total employment) (modeled ILO estimate) - 7%\nCost of business start-up procedures (% of GNI per capita) - 7%\nCost of business start-up procedures, female (% of GNI per capita) - 7%\nCost of business start-up procedures, male (% of GNI per capita) - 7%\nCost to export, border compliance (US$) - 60%\nCost to export, documentary compliance (US$) - 60%\nCost to import, border compliance (US$) - 60%\nCost to import, documentary compliance (US$) - 60%\nCoverage of social insurance programs (% of population) - 100%\nCoverage of social insurance programs in 2nd quintile (% of population) - 100%\nCoverage of social insurance programs in 3rd quintile (% of population) - 100%\nCoverage of social insurance programs in 4th quintile (% of population) - 100%\nCoverage of social insurance programs in poorest quintile (% of population) - 100%\nCoverage of social insurance programs in richest quintile (% of population) - 100%\nCoverage of social protection and labor programs (% of population) - 100%\nCoverage of social safety net programs (% of population) - 100%\nCoverage of social safety net programs in 2nd quintile (% of population) - 100%\nCoverage of social safety net programs in 3rd quintile (% of population) - 100%\nCoverage of social safety net programs in 4th quintile (% of population) - 100%\nCoverage of social safety net programs in poorest quintile (% of population) - 100%\nCoverage of social safety net programs in richest quintile (% of population) - 100%\nCoverage of unemployment benefits and ALMP (% of population) - 100%\nCoverage of unemployment benefits and ALMP in 2nd quintile (% of population) - 100%\nCoverage of unemployment benefits and ALMP in 3rd quintile (% of population) - 100%\nCoverage of unemployment benefits and ALMP in 4th quintile (% of population) - 100%\nCoverage of unemployment benefits and ALMP in poorest quintile (% of population) - 100%\nCoverage of unemployment benefits and ALMP in richest quintile (% of population) - 100%\nCPIA building human resources rating (1=low to 6=high) - 100%\nCPIA business regulatory environment rating (1=low to 6=high) - 100%\nCPIA debt policy rating (1=low to 6=high) - 100%\nCPIA economic management cluster average (1=low to 6=high) - 100%\nCPIA efficiency of revenue mobilization rating (1=low to 6=high) - 100%\nCPIA equity of public resource use rating (1=low to 6=high) - 100%\nCPIA financial sector rating (1=low to 6=high) - 100%\nCPIA fiscal policy rating (1=low to 6=high) - 100%\nCPIA gender equality rating (1=low to 6=high) - 100%\nCPIA macroeconomic management rating (1=low to 6=high) - 100%\nCPIA policies for social inclusion/equity cluster average (1=low to 6=high) - 100%\nCPIA policy and institutions for environmental sustainability rating (1=low to 6=high) - 100%\nCPIA property rights and rule-based governance rating (1=low to 6=high) - 100%\nCPIA public sector management and institutions cluster average (1=low to 6=high) - 100%\nCPIA quality of budgetary and financial management rating (1=low to 6=high) - 100%\nCPIA quality of public administration rating (1=low to 6=high) - 100%\nCPIA social protection rating (1=low to 6=high) - 100%\nCPIA structural policies cluster average (1=low to 6=high) - 100%\nCPIA trade rating (1=low to 6=high) - 100%\nCPIA transparency, accountability, and corruption in the public sector rating (1=low to 6=high) - 100%\nCrop production index (2014-2016 = 100) - 13%\nCurrent account balance (% of GDP) - 0%\nCurrent account balance (BoP, current US$) - 0%\nCurrent education expenditure, primary (% of total expenditure in primary public institutions) - 13%\nCurrent education expenditure, secondary (% of total expenditure in secondary public institutions) - 13%\nCurrent education expenditure, tertiary (% of total expenditure in tertiary public institutions) - 20%\nCurrent education expenditure, total (% of total expenditure in public institutions) - 87%\nCurrent health expenditure (% of GDP) - 13%\nCurrent health expenditure per capita (current US$) - 13%\nCurrent health expenditure per capita, PPP (current international $) - 13%\nCustoms and other import duties (% of tax revenue) - 7%\nCustoms and other import duties (current LCU) - 7%\nDeath rate, crude (per 1,000 people) - 7%\nDebt service (PPG and IMF only, % of exports of goods, services and primary income) - 100%\nDebt service on external debt, public and publicly guaranteed (PPG) (TDS, current US$) - 100%\nDebt service on external debt, total (TDS, current US$) - 100%\nDEC alternative conversion factor (LCU per US$) - 0%\nDemand for family planning satisfied by modern methods (% of married women with demand for family planning) - 100%\nDeposit interest rate (%) - 7%\nDepositors with commercial banks (per 1,000 adults) - 100%\nDepth of credit information index (0=low to 8=high) - 53%\nDiabetes prevalence (% of population ages 20 to 79) - 87%\nDiarrhea treatment (% of children under 5 receiving oral rehydration and continued feeding) - 100%\nDiarrhea treatment (% of children under 5 who received ORS packet) - 100%\nDisaster risk reduction progress score (1-5 scale; 5=best) - 93%\nDiscrepancy in expenditure estimate of GDP (constant LCU) - 100%\nDiscrepancy in expenditure estimate of GDP (current LCU) - 0%\nDomestic credit provided by financial sector (% of GDP) - 100%\nDomestic credit to private sector (% of GDP) - 0%\nDomestic credit to private sector by banks (% of GDP) - 0%\nDomestic general government health expenditure (% of current health expenditure) - 13%\nDomestic general government health expenditure (% of GDP) - 13%\nDomestic general government health expenditure (% of general government expenditure) - 13%\nDomestic general government health expenditure per capita (current US$) - 13%\nDomestic general government health expenditure per capita, PPP (current international $) - 13%\nDomestic private health expenditure (% of current health expenditure) - 13%\nDomestic private health expenditure per capita (current US$) - 13%\nDomestic private health expenditure per capita, PPP (current international $) - 13%\nDroughts, floods, extreme temperatures (% of population, average 1990-2009) - 93%\nEase of doing business rank (1=most business-friendly regulations) - 93%\nEase of doing business score (0 = lowest performance to 100 = best performance) - 67%\nEducational attainment, at least Bachelor's or equivalent, population 25+, female (%) (cumulative) - 47%\nEducational attainment, at least Bachelor's or equivalent, population 25+, male (%) (cumulative) - 47%\nEducational attainment, at least Bachelor's or equivalent, population 25+, total (%) (cumulative) - 47%\nEducational attainment, at least completed lower secondary, population 25+, female (%) (cumulative) - 0%\nEducational attainment, at least completed lower secondary, population 25+, male (%) (cumulative) - 0%\nEducational attainment, at least completed lower secondary, population 25+, total (%) (cumulative) - 0%\nEducational attainment, at least completed post-secondary, population 25+, female (%) (cumulative) - 0%\nEducational attainment, at least completed post-secondary, population 25+, male (%) (cumulative) - 0%\nEducational attainment, at least completed post-secondary, population 25+, total (%) (cumulative) - 0%\nEducational attainment, at least completed primary, population 25+ years, female (%) (cumulative) - 87%\nEducational attainment, at least completed primary, population 25+ years, male (%) (cumulative) - 87%\nEducational attainment, at least completed primary, population 25+ years, total (%) (cumulative) - 87%\nEducational attainment, at least completed short-cycle tertiary, population 25+, female (%) (cumulative) - 0%\nEducational attainment, at least completed short-cycle tertiary, population 25+, male (%) (cumulative) - 0%\nEducational attainment, at least completed short-cycle tertiary, population 25+, total (%) (cumulative) - 0%\nEducational attainment, at least completed upper secondary, population 25+, female (%) (cumulative) - 0%\nEducational attainment, at least completed upper secondary, population 25+, male (%) (cumulative) - 0%\nEducational attainment, at least completed upper secondary, population 25+, total (%) (cumulative) - 0%\nEducational attainment, at least Master's or equivalent, population 25+, female (%) (cumulative) - 47%\nEducational attainment, at least Master's or equivalent, population 25+, male (%) (cumulative) - 47%\nEducational attainment, at least Master's or equivalent, population 25+, total (%) (cumulative) - 47%\nEducational attainment, Doctoral or equivalent, population 25+, female (%) (cumulative) - 47%\nEducational attainment, Doctoral or equivalent, population 25+, male (%) (cumulative) - 47%\nEducational attainment, Doctoral or equivalent, population 25+, total (%) (cumulative) - 47%\nElectric power consumption (kWh per capita) - 40%\nElectric power transmission and distribution losses (% of output) - 40%\nElectricity production from coal sources (% of total) - 33%\nElectricity production from hydroelectric sources (% of total) - 33%\nElectricity production from natural gas sources (% of total) - 33%\nElectricity production from nuclear sources (% of total) - 33%\nElectricity production from oil sources (% of total) - 33%\nElectricity production from oil, gas and coal sources (% of total) - 33%\nElectricity production from renewable sources, excluding hydroelectric (% of total) - 33%\nElectricity production from renewable sources, excluding hydroelectric (kWh) - 33%\nEmployers, female (% of female employment) (modeled ILO estimate) - 7%\nEmployers, male (% of male employment) (modeled ILO estimate) - 7%\nEmployers, total (% of total employment) (modeled ILO estimate) - 7%\nEmployment in agriculture (% of total employment) (modeled ILO estimate) - 7%\nEmployment in agriculture, female (% of female employment) (modeled ILO estimate) - 7%\nEmployment in agriculture, male (% of male employment) (modeled ILO estimate) - 7%\nEmployment in industry (% of total employment) (modeled ILO estimate) - 7%\nEmployment in industry, female (% of female employment) (modeled ILO estimate) - 7%\nEmployment in industry, male (% of male employment) (modeled ILO estimate) - 7%\nEmployment in services (% of total employment) (modeled ILO estimate) - 7%\nEmployment in services, female (% of female employment) (modeled ILO estimate) - 7%\nEmployment in services, male (% of male employment) (modeled ILO estimate) - 7%\nEmployment to population ratio, 15+, female (%) (modeled ILO estimate) - 7%\nEmployment to population ratio, 15+, female (%) (national estimate) - 0%\nEmployment to population ratio, 15+, male (%) (modeled ILO estimate) - 7%\nEmployment to population ratio, 15+, male (%) (national estimate) - 0%\nEmployment to population ratio, 15+, total (%) (modeled ILO estimate) - 0%\nEmployment to population ratio, 15+, total (%) (national estimate) - 0%\nEmployment to population ratio, ages 15-24, female (%) (modeled ILO estimate) - 7%\nEmployment to population ratio, ages 15-24, female (%) (national estimate) - 0%\nEmployment to population ratio, ages 15-24, male (%) (modeled ILO estimate) - 7%\nEmployment to population ratio, ages 15-24, male (%) (national estimate) - 0%\nEmployment to population ratio, ages 15-24, total (%) (modeled ILO estimate) - 7%\nEmployment to population ratio, ages 15-24, total (%) (national estimate) - 0%\nEnergy imports, net (% of energy use) - 33%\nEnergy intensity level of primary energy (MJ/$2011 PPP GDP) - 33%\nEnergy related methane emissions (% of total) - 80%\nEnergy use (kg of oil equivalent per capita) - 33%\nEnergy use (kg of oil equivalent) per $1,000 GDP (constant 2017 PPP) - 33%\nExclusive breastfeeding (% of children under 6 months) - 100%\nExpenditure on primary education (% of government expenditure on education) - 27%\nExpenditure on secondary education (% of government expenditure on education) - 27%\nExpenditure on tertiary education (% of government expenditure on education) - 27%\nExpense (% of GDP) - 7%\nExpense (current LCU) - 7%\nExport unit value index (2000 = 100) - 7%\nExport value index (2000 = 100) - 7%\nExport volume index (2000 = 100) - 7%\nExports as a capacity to import (constant LCU) - 0%\nExports of goods and services (% of GDP) - 0%\nExports of goods and services (annual % growth) - 0%\nExports of goods and services (BoP, current US$) - 0%\nExports of goods and services (constant 2015 US$) - 0%\nExports of goods and services (constant LCU) - 0%\nExports of goods and services (current LCU) - 0%\nExports of goods and services (current US$) - 0%\nExports of goods, services and primary income (BoP, current US$) - 0%\nExternal balance on goods and services (% of GDP) - 0%\nExternal balance on goods and services (constant LCU) - 87%\nExternal balance on goods and services (current LCU) - 0%\nExternal balance on goods and services (current US$) - 0%\nExternal debt stocks (% of GNI) - 100%\nExternal debt stocks, long-term (DOD, current US$) - 100%\nExternal debt stocks, private nonguaranteed (PNG) (DOD, current US$) - 100%\nExternal debt stocks, public and publicly guaranteed (PPG) (DOD, current US$) - 100%\nExternal debt stocks, short-term (DOD, current US$) - 100%\nExternal debt stocks, total (DOD, current US$) - 100%\nExternal health expenditure (% of current health expenditure) - 13%\nExternal health expenditure per capita (current US$) - 13%\nExternal health expenditure per capita, PPP (current international $) - 13%\nFemale genital mutilation prevalence (%) - 100%\nFemale headed households (% of households with a female head) - 100%\nFemale share of employment in senior and middle management (%) - 100%\nFertility rate, total (births per woman) - 7%\nFertilizer consumption (% of fertilizer production) - 13%\nFertilizer consumption (kilograms per hectare of arable land) - 13%\nFinal consumption expenditure (% of GDP) - 0%\nFinal consumption expenditure (annual % growth) - 0%\nFinal consumption expenditure (constant 2015 US$) - 0%\nFinal consumption expenditure (constant LCU) - 0%\nFinal consumption expenditure (current LCU) - 0%\nFinal consumption expenditure (current US$) - 0%\nFinancial intermediary services indirectly Measured (FISIM) (constant LCU) - 100%\nFinancial intermediary services indirectly Measured (FISIM) (current LCU) - 100%\nFirms competing against unregistered firms (% of firms) - 100%\nFirms expected to give gifts in meetings with tax officials (% of firms) - 100%\nFirms experiencing electrical outages (% of firms) - 100%\nFirms experiencing losses due to theft and vandalism (% of firms) - 100%\nFirms formally registered when operations started (% of firms) - 100%\nFirms offering formal training (% of firms) - 100%\nFirms that do not report all sales for tax purposes (% of firms) - 100%\nFirms that spend on R&D (% of firms) - 100%\nFirms using banks to finance investment (% of firms) - 100%\nFirms using banks to finance working capital (% of firms) - 100%\nFirms visited or required meetings with tax officials (% of firms) - 100%\nFirms with female participation in ownership (% of firms) - 100%\nFirms with female top manager (% of firms) - 100%\nFish species, threatened - 93%\nFixed broadband subscriptions - 0%\nFixed broadband subscriptions (per 100 people) - 7%\nFixed telephone subscriptions - 0%\nFixed telephone subscriptions (per 100 people) - 0%\nFood exports (% of merchandise exports) - 7%\nFood imports (% of merchandise imports) - 7%\n"
],
[
"# % of missing NZL\nfor col in NZL_structured.columns:\n pct_missing = np.mean(NZL_structured[col].isnull())\n print('{} - {}%'.format(col, round(pct_missing*100)))",
"Year - 0%\nAccess to clean fuels and technologies for cooking (% of population) - 27%\nAccess to electricity (% of population) - 7%\nAccess to electricity, rural (% of rural population) - 7%\nAccess to electricity, urban (% of urban population) - 7%\nAccount ownership at a financial institution or with a mobile-money-service provider (% of population ages 15+) - 80%\nAccount ownership at a financial institution or with a mobile-money-service provider, female (% of population ages 15+) - 80%\nAccount ownership at a financial institution or with a mobile-money-service provider, male (% of population ages 15+) - 80%\nAccount ownership at a financial institution or with a mobile-money-service provider, older adults (% of population ages 25+) - 80%\nAccount ownership at a financial institution or with a mobile-money-service provider, poorest 40% (% of population ages 15+) - 80%\nAccount ownership at a financial institution or with a mobile-money-service provider, primary education or less (% of population ages 15+) - 80%\nAccount ownership at a financial institution or with a mobile-money-service provider, richest 60% (% of population ages 15+) - 80%\nAccount ownership at a financial institution or with a mobile-money-service provider, secondary education or more (% of population ages 15+) - 80%\nAccount ownership at a financial institution or with a mobile-money-service provider, young adults (% of population ages 15-24) - 80%\nAdequacy of social insurance programs (% of total welfare of beneficiary households) - 100%\nAdequacy of social protection and labor programs (% of total welfare of beneficiary households) - 100%\nAdequacy of social safety net programs (% of total welfare of beneficiary households) - 100%\nAdequacy of unemployment benefits and ALMP (% of total welfare of beneficiary households) - 100%\nAdjusted net enrollment rate, primary (% of primary school age children) - 73%\nAdjusted net enrollment rate, primary, female (% of primary school age children) - 80%\nAdjusted net enrollment rate, primary, male (% of primary school age children) - 80%\nAdjusted net national income (annual % growth) - 7%\nAdjusted net national income (constant 2015 US$) - 7%\nAdjusted net national income (current US$) - 7%\nAdjusted net national income per capita (annual % growth) - 7%\nAdjusted net national income per capita (constant 2015 US$) - 7%\nAdjusted net national income per capita (current US$) - 7%\nAdjusted net savings, excluding particulate emission damage (% of GNI) - 13%\nAdjusted net savings, excluding particulate emission damage (current US$) - 13%\nAdjusted net savings, including particulate emission damage (% of GNI) - 13%\nAdjusted net savings, including particulate emission damage (current US$) - 13%\nAdjusted savings: carbon dioxide damage (% of GNI) - 7%\nAdjusted savings: carbon dioxide damage (current US$) - 7%\nAdjusted savings: consumption of fixed capital (% of GNI) - 7%\nAdjusted savings: consumption of fixed capital (current US$) - 7%\nAdjusted savings: education expenditure (% of GNI) - 7%\nAdjusted savings: education expenditure (current US$) - 7%\nAdjusted savings: energy depletion (% of GNI) - 7%\nAdjusted savings: energy depletion (current US$) - 7%\nAdjusted savings: gross savings (% of GNI) - 13%\nAdjusted savings: mineral depletion (% of GNI) - 7%\nAdjusted savings: mineral depletion (current US$) - 7%\nAdjusted savings: natural resources depletion (% of GNI) - 7%\nAdjusted savings: net forest depletion (% of GNI) - 7%\nAdjusted savings: net forest depletion (current US$) - 7%\nAdjusted savings: net national savings (% of GNI) - 13%\nAdjusted savings: net national savings (current US$) - 13%\nAdjusted savings: particulate emission damage (% of GNI) - 7%\nAdjusted savings: particulate emission damage (current US$) - 7%\nAdolescent fertility rate (births per 1,000 women ages 15-19) - 7%\nAdolescents out of school (% of lower secondary school age) - 53%\nAdolescents out of school, female (% of female lower secondary school age) - 53%\nAdolescents out of school, male (% of male lower secondary school age) - 53%\nAdults (ages 15+) and children (ages 0-14) newly infected with HIV - 0%\nAdults (ages 15-49) newly infected with HIV - 0%\nAge dependency ratio (% of working-age population) - 0%\nAge dependency ratio, old (% of working-age population) - 0%\nAge dependency ratio, young (% of working-age population) - 0%\nAgricultural irrigated land (% of total agricultural land) - 80%\nAgricultural land (% of land area) - 13%\nAgricultural land (sq. km) - 13%\nAgricultural machinery, tractors - 100%\nAgricultural machinery, tractors per 100 sq. km of arable land - 100%\nAgricultural methane emissions (% of total) - 80%\nAgricultural methane emissions (thousand metric tons of CO2 equivalent) - 13%\nAgricultural nitrous oxide emissions (% of total) - 80%\nAgricultural nitrous oxide emissions (thousand metric tons of CO2 equivalent) - 13%\nAgricultural raw materials exports (% of merchandise exports) - 0%\nAgricultural raw materials imports (% of merchandise imports) - 0%\nAgriculture, forestry, and fishing, value added (% of GDP) - 13%\nAgriculture, forestry, and fishing, value added (annual % growth) - 7%\nAgriculture, forestry, and fishing, value added (constant 2015 US$) - 7%\nAgriculture, forestry, and fishing, value added (constant LCU) - 7%\nAgriculture, forestry, and fishing, value added (current LCU) - 13%\nAgriculture, forestry, and fishing, value added (current US$) - 13%\nAgriculture, forestry, and fishing, value added per worker (constant 2015 US$) - 7%\nAir transport, freight (million ton-km) - 7%\nAir transport, passengers carried - 7%\nAir transport, registered carrier departures worldwide - 7%\nAlternative and nuclear energy (% of total energy use) - 33%\nAnnual freshwater withdrawals, agriculture (% of total freshwater withdrawal) - 80%\nAnnual freshwater withdrawals, domestic (% of total freshwater withdrawal) - 80%\nAnnual freshwater withdrawals, industry (% of total freshwater withdrawal) - 80%\nAnnual freshwater withdrawals, total (% of internal resources) - 80%\nAnnual freshwater withdrawals, total (billion cubic meters) - 80%\nAnnualized average growth rate in per capita real survey mean consumption or income, bottom 40% of population (%) - 100%\nAnnualized average growth rate in per capita real survey mean consumption or income, total population (%) - 100%\nAntiretroviral therapy coverage (% of people living with HIV) - 0%\nAntiretroviral therapy coverage for PMTCT (% of pregnant women living with HIV) - 100%\nAquaculture production (metric tons) - 13%\nArable land (% of land area) - 13%\nArable land (hectares per person) - 13%\nArable land (hectares) - 13%\nARI treatment (% of children under 5 taken to a health provider) - 100%\nArmed forces personnel (% of total labor force) - 13%\nArmed forces personnel, total - 13%\nArms exports (SIPRI trend indicator values) - 67%\nArms imports (SIPRI trend indicator values) - 13%\nAutomated teller machines (ATMs) (per 100,000 adults) - 0%\nAverage precipitation in depth (mm per year) - 80%\nAverage time to clear exports through customs (days) - 100%\nAverage transaction cost of sending remittances from a specific country (%) - 33%\nAverage transaction cost of sending remittances to a specific country (%) - 100%\nAverage working hours of children, study and work, ages 7-14 (hours per week) - 100%\nAverage working hours of children, study and work, female, ages 7-14 (hours per week) - 100%\nAverage working hours of children, study and work, male, ages 7-14 (hours per week) - 100%\nAverage working hours of children, working only, ages 7-14 (hours per week) - 100%\nAverage working hours of children, working only, female, ages 7-14 (hours per week) - 100%\nAverage working hours of children, working only, male, ages 7-14 (hours per week) - 100%\nBank capital to assets ratio (%) - 100%\nBank liquid reserves to bank assets ratio (%) - 47%\nBank nonperforming loans to total gross loans (%) - 100%\nBattle-related deaths (number of people) - 100%\nBenefit incidence of social insurance programs to poorest quintile (% of total social insurance benefits) - 100%\nBenefit incidence of social protection and labor programs to poorest quintile (% of total SPL benefits) - 100%\nBenefit incidence of social safety net programs to poorest quintile (% of total safety net benefits) - 100%\nBenefit incidence of unemployment benefits and ALMP to poorest quintile (% of total U/ALMP benefits) - 100%\nBinding coverage, all products (%) - 13%\nBinding coverage, manufactured products (%) - 13%\nBinding coverage, primary products (%) - 13%\nBird species, threatened - 93%\nBirth rate, crude (per 1,000 people) - 7%\nBirths attended by skilled health staff (% of total) - 27%\nBorrowers from commercial banks (per 1,000 adults) - 100%\nBound rate, simple mean, all products (%) - 13%\nBound rate, simple mean, manufactured products (%) - 13%\nBound rate, simple mean, primary products (%) - 13%\nBribery incidence (% of firms experiencing at least one bribe payment request) - 100%\nBroad money (% of GDP) - 13%\nBroad money (current LCU) - 13%\nBroad money growth (annual %) - 20%\nBroad money to total reserves ratio - 13%\nBusiness extent of disclosure index (0=less disclosure to 10=more disclosure) - 7%\nCapture fisheries production (metric tons) - 13%\nCause of death, by communicable diseases and maternal, prenatal and nutrition conditions (% of total) - 80%\nCause of death, by injury (% of total) - 80%\nCause of death, by non-communicable diseases (% of total) - 80%\nCentral government debt, total (% of GDP) - 27%\nCentral government debt, total (current LCU) - 27%\nCereal production (metric tons) - 13%\nCereal yield (kg per hectare) - 13%\nChanges in inventories (constant LCU) - 100%\nChanges in inventories (current LCU) - 7%\nChanges in inventories (current US$) - 7%\nCharges for the use of intellectual property, payments (BoP, current US$) - 0%\nCharges for the use of intellectual property, receipts (BoP, current US$) - 0%\nChemicals (% of value added in manufacturing) - 13%\nChild employment in agriculture (% of economically active children ages 7-14) - 100%\nChild employment in agriculture, female (% of female economically active children ages 7-14) - 100%\n"
]
],
[
[
"#### Data Cleaning",
"_____no_output_____"
]
],
[
[
"for col in AUS_structured.columns:\n if round((np.mean(AUS_structured[col].isnull()))*100) > 0:\n AUS_structured.drop(col, axis=1, inplace=True)\n else:\n pass",
"_____no_output_____"
],
[
"AUS_structured",
"_____no_output_____"
],
[
"for col in NZL_structured.columns:\n if round((np.mean(NZL_structured[col].isnull()))*100) > 0:\n NZL_structured.drop(col, axis=1, inplace=True)\n else:\n pass",
"_____no_output_____"
],
[
"NZL_structured",
"_____no_output_____"
]
],
[
[
"#### Feature Selection",
"_____no_output_____"
]
],
[
[
"# out of the remainin column 85 have been identified based on domain knowledge which will be used for furter analysis in Tableau\n\n\nAUS_preprocessed = AUS_structured[[\"Year\",\"Adults (ages 15+) and children (ages 0-14) newly infected with HIV\",\"Adults (ages 15-49) newly infected with HIV\",\"Incidence of HIV, ages 15-24 (per 1,000 uninfected population ages 15-24)\",\"Incidence of HIV, ages 15-49 (per 1,000 uninfected population ages 15-49)\",\"Incidence of HIV, all (per 1,000 uninfected population)\",\"Prevalence of HIV, female (% ages 15-24)\",\"Prevalence of HIV, male (% ages 15-24)\",\"Prevalence of HIV, total (% of population ages 15-49)\",\"Women's share of population ages 15+ living with HIV (%)\",\"Young people (ages 15-24) newly infected with HIV\",\"Population ages 00-04, female (% of female population)\",\"Population ages 00-04, male (% of male population)\",\"Population ages 0-14 (% of total population)\",\"Population ages 0-14, female\",\"Population ages 0-14, female (% of female population)\",\"Population ages 0-14, male\",\"Population ages 0-14, male (% of male population)\",\"Population ages 0-14, total\",\"Population ages 05-09, female (% of female population)\",\"Population ages 05-09, male (% of male population)\",\"Population ages 10-14, female (% of female population)\",\"Population ages 10-14, male (% of male population)\",\"Population ages 15-19, female (% of female population)\",\"Population ages 15-19, male (% of male population)\",\"Population ages 15-64 (% of total population)\",\"Population ages 15-64, female\",\"Population ages 15-64, female (% of female population)\",\"Population ages 15-64, male\",\"Population ages 15-64, male (% of male population)\",\"Population ages 15-64, total\",\"Population ages 20-24, female (% of female population)\",\"Population ages 20-24, male (% of male population)\",\"Population ages 25-29, female (% of female population)\",\"Population ages 25-29, male (% of male population)\",\"Population ages 30-34, female (% of female population)\",\"Population ages 30-34, male (% of male population)\",\"Population ages 35-39, female (% of female population)\",\"Population ages 35-39, male (% of male population)\",\"Population ages 40-44, female (% of female population)\",\"Population ages 40-44, male (% of male population)\",\"Population ages 45-49, female (% of female population)\",\"Population ages 45-49, male (% of male population)\",\"Population ages 50-54, female (% of female population)\",\"Population ages 50-54, male (% of male population)\",\"Population ages 55-59, female (% of female population)\",\"Population ages 55-59, male (% of male population)\",\"Population ages 60-64, female (% of female population)\",\"Population ages 60-64, male (% of male population)\",\"Population ages 65 and above (% of total population)\",\"Population ages 65 and above, female\",\"Population ages 65 and above, female (% of female population)\",\"Population ages 65 and above, male\",\"Population ages 65 and above, male (% of male population)\",\"Population ages 65 and above, total\",\"Population ages 65-69, female (% of female population)\",\"Population ages 65-69, male (% of male population)\",\"Population ages 70-74, female (% of female population)\",\"Population ages 70-74, male (% of male population)\",\"Population ages 75-79, female (% of female population)\",\"Population ages 75-79, male (% of male population)\",\"Population ages 80 and above, female (% of female population)\",\"Population ages 80 and above, male (% of male population)\",\"Population, female\",\"Population, female (% of total population)\",\"Population, male\",\"Population, male (% of total population)\",\"Population, total\",\"Proportion of seats held by women in national parliaments (%)\",\"Women Business and the Law Index Score (scale 1-100)\",\"Employment to population ratio, 15+, female (%) (national estimate)\",\"Employment to population ratio, 15+, male (%) (national estimate)\",\"Employment to population ratio, 15+, total (%) (modeled ILO estimate)\",\"Employment to population ratio, 15+, total (%) (national estimate)\",\"Employment to population ratio, ages 15-24, female (%) (national estimate)\",\"Employment to population ratio, ages 15-24, male (%) (national estimate)\",\"Employment to population ratio, ages 15-24, total (%) (national estimate)\",\"GDP per person employed (constant 2017 PPP $)\",\"Unemployment, female (% of female labor force) (national estimate)\",\"Unemployment, male (% of male labor force) (national estimate)\",\"Unemployment, total (% of total labor force) (modeled ILO estimate)\",\"Unemployment, total (% of total labor force) (national estimate)\",\"Unemployment, youth female (% of female labor force ages 15-24) (national estimate)\",\"Unemployment, youth male (% of male labor force ages 15-24) (national estimate)\",\"Unemployment, youth total (% of total labor force ages 15-24) (national estimate)\"]]\n\nNZL_preprocessed = NZL_structured[[\"Year\",\"Adults (ages 15+) and children (ages 0-14) newly infected with HIV\",\"Adults (ages 15-49) newly infected with HIV\",\"Incidence of HIV, ages 15-24 (per 1,000 uninfected population ages 15-24)\",\"Incidence of HIV, ages 15-49 (per 1,000 uninfected population ages 15-49)\",\"Incidence of HIV, all (per 1,000 uninfected population)\",\"Prevalence of HIV, female (% ages 15-24)\",\"Prevalence of HIV, male (% ages 15-24)\",\"Prevalence of HIV, total (% of population ages 15-49)\",\"Women's share of population ages 15+ living with HIV (%)\",\"Young people (ages 15-24) newly infected with HIV\",\"Population ages 00-04, female (% of female population)\",\"Population ages 00-04, male (% of male population)\",\"Population ages 0-14 (% of total population)\",\"Population ages 0-14, female\",\"Population ages 0-14, female (% of female population)\",\"Population ages 0-14, male\",\"Population ages 0-14, male (% of male population)\",\"Population ages 0-14, total\",\"Population ages 05-09, female (% of female population)\",\"Population ages 05-09, male (% of male population)\",\"Population ages 10-14, female (% of female population)\",\"Population ages 10-14, male (% of male population)\",\"Population ages 15-19, female (% of female population)\",\"Population ages 15-19, male (% of male population)\",\"Population ages 15-64 (% of total population)\",\"Population ages 15-64, female\",\"Population ages 15-64, female (% of female population)\",\"Population ages 15-64, male\",\"Population ages 15-64, male (% of male population)\",\"Population ages 15-64, total\",\"Population ages 20-24, female (% of female population)\",\"Population ages 20-24, male (% of male population)\",\"Population ages 25-29, female (% of female population)\",\"Population ages 25-29, male (% of male population)\",\"Population ages 30-34, female (% of female population)\",\"Population ages 30-34, male (% of male population)\",\"Population ages 35-39, female (% of female population)\",\"Population ages 35-39, male (% of male population)\",\"Population ages 40-44, female (% of female population)\",\"Population ages 40-44, male (% of male population)\",\"Population ages 45-49, female (% of female population)\",\"Population ages 45-49, male (% of male population)\",\"Population ages 50-54, female (% of female population)\",\"Population ages 50-54, male (% of male population)\",\"Population ages 55-59, female (% of female population)\",\"Population ages 55-59, male (% of male population)\",\"Population ages 60-64, female (% of female population)\",\"Population ages 60-64, male (% of male population)\",\"Population ages 65 and above (% of total population)\",\"Population ages 65 and above, female\",\"Population ages 65 and above, female (% of female population)\",\"Population ages 65 and above, male\",\"Population ages 65 and above, male (% of male population)\",\"Population ages 65 and above, total\",\"Population ages 65-69, female (% of female population)\",\"Population ages 65-69, male (% of male population)\",\"Population ages 70-74, female (% of female population)\",\"Population ages 70-74, male (% of male population)\",\"Population ages 75-79, female (% of female population)\",\"Population ages 75-79, male (% of male population)\",\"Population ages 80 and above, female (% of female population)\",\"Population ages 80 and above, male (% of male population)\",\"Population, female\",\"Population, female (% of total population)\",\"Population, male\",\"Population, male (% of total population)\",\"Population, total\",\"Proportion of seats held by women in national parliaments (%)\",\"Women Business and the Law Index Score (scale 1-100)\",\"Employment to population ratio, 15+, female (%) (national estimate)\",\"Employment to population ratio, 15+, male (%) (national estimate)\",\"Employment to population ratio, 15+, total (%) (modeled ILO estimate)\",\"Employment to population ratio, 15+, total (%) (national estimate)\",\"Employment to population ratio, ages 15-24, female (%) (national estimate)\",\"Employment to population ratio, ages 15-24, male (%) (national estimate)\",\"Employment to population ratio, ages 15-24, total (%) (national estimate)\",\"GDP per person employed (constant 2017 PPP $)\",\"Unemployment, female (% of female labor force) (national estimate)\",\"Unemployment, male (% of male labor force) (national estimate)\",\"Unemployment, total (% of total labor force) (modeled ILO estimate)\",\"Unemployment, total (% of total labor force) (national estimate)\",\"Unemployment, youth female (% of female labor force ages 15-24) (national estimate)\",\"Unemployment, youth male (% of male labor force ages 15-24) (national estimate)\",\"Unemployment, youth total (% of total labor force ages 15-24) (national estimate)\"]]\n",
"_____no_output_____"
],
[
"AUS_preprocessed",
"_____no_output_____"
],
[
"NZL_preprocessed",
"_____no_output_____"
],
[
"#joined dataset\n\nAUS_preprocessed[\"Country\"]=\"Australia\"\n\nNZL_preprocessed[\"Country\"]=\"New Zealand\"",
"C:\\Users\\thaer\\Anaconda3\\lib\\site-packages\\pandas\\core\\frame.py:3607: SettingWithCopyWarning: \nA value is trying to be set on a copy of a slice from a DataFrame.\nTry using .loc[row_indexer,col_indexer] = value instead\n\nSee the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n self._set_item(key, value)\n"
],
[
"NZL_preprocessed",
"_____no_output_____"
],
[
"joint_preprocessed = AUS_preprocessed\n\njoint_preprocessed = joint_preprocessed.append(NZL_preprocessed)\n\njoint_preprocessed",
"_____no_output_____"
]
],
[
[
"#### Save Preprocessed Datasets as CSV",
"_____no_output_____"
]
],
[
[
"AUS_preprocessed.to_csv(\"AUS_preprocessed.csv\", index=False)",
"_____no_output_____"
],
[
"NZL_preprocessed.to_csv(\"NZL_preprocessed.csv\", index=False)",
"_____no_output_____"
],
[
"joint_preprocessed.to_csv(\"joint_preprocessed.csv\", index=False)",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown",
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
]
] |
cbdb4801eec272b402a482ddc21dde65f84b147c
| 518,380 |
ipynb
|
Jupyter Notebook
|
app.ipynb
|
philippjfr/clifford-app
|
34c04b3f550663a36947c5006dfce84bcf1e10d9
|
[
"BSD-3-Clause"
] | 1 |
2018-11-29T16:34:12.000Z
|
2018-11-29T16:34:12.000Z
|
app.ipynb
|
philippjfr/clifford-app
|
34c04b3f550663a36947c5006dfce84bcf1e10d9
|
[
"BSD-3-Clause"
] | null | null | null |
app.ipynb
|
philippjfr/clifford-app
|
34c04b3f550663a36947c5006dfce84bcf1e10d9
|
[
"BSD-3-Clause"
] | null | null | null | 3,410.394737 | 513,173 | 0.965151 |
[
[
[
"# Clifford attractors with Datashader, Panel, and interact\n\n[Datashader](http://datashader.org) renders data into images. \n[Clifford attractors](https://anaconda.org/jbednar/clifford_attractor/notebook) are a type of iterative equation that traces the path of a particle through a 2D space using functions of sine and cosine terms that make interesting \"attractor\" patterns (covering only some portions of the possible space, in certain shapes). Here we use Numpy and Pandas to calculate a dataframe consisting of ten million such locations, using Numba to make it 50X faster than bare Python. You can run this code yourself if you make a [conda](http://conda.pydata.org/miniconda.html) environment and run `conda install -c pyviz pyviz` first, then update to the GitHub master of [Panel](https://github.com/pyviz/panel) and [Param](https://github.com/ioam/param).",
"_____no_output_____"
]
],
[
[
"import numpy as np, pandas as pd\nfrom numba import jit\n\n@jit\ndef clifford_trajectory(a, b, c, d, x0, y0, n):\n xs, ys = np.zeros(n), np.zeros(n)\n xs[0], ys[0] = x0, y0\n for i in np.arange(n-1):\n xs[i+1] = np.sin(a * ys[i]) + c * np.cos(a * xs[i])\n ys[i+1] = np.sin(b * xs[i]) + d * np.cos(b * ys[i])\n return pd.DataFrame(dict(x=xs,y=ys))",
"_____no_output_____"
]
],
[
[
"We can visualize the resulting dataframe using [Datashader](http://datashader.org), with colormaps from [Colorcet](http://colorcet.pyviz.org) and interactive widgets from [Panel](http://panel.pyviz.org):",
"_____no_output_____"
]
],
[
[
"import datashader as ds\nfrom datashader import transfer_functions as tf\n\nfrom colorcet import palette_n\nps ={k:p[::-1] for k,p in palette_n.items()}\n\nimport panel as pp\npp.extension()\n\ndef clifford_plot(a=1.7, b=1.7, c=0.6, d=1.2, n=2000000, colormap=ps['kbc']):\n cvs = ds.Canvas(plot_width=600, plot_height=600)\n agg = cvs.points(clifford_trajectory(a, b, c, d, 0, 0, n), 'x', 'y')\n return tf.shade(agg, cmap=colormap)",
"_____no_output_____"
],
[
"clifford_plot(a=1.7, b=1.7, c=0.6, d=1.2, n=2000000, colormap=ps['kbc'])",
"_____no_output_____"
],
[
"pp.interact(clifford_plot, n=(1,20000000), colormap=ps)",
"_____no_output_____"
]
],
[
[
"Here `interact()` returns an object that lets us vary the arguments to the given function interactively and see the results. This object is actually a full `Panel` object, and so we can add other viewable objects to this panel if we like. For instance, let's define an object `i` that's the same as above but with a PNG logo and a description:",
"_____no_output_____"
]
],
[
[
"i=pp.interact(clifford_plot, a=(-2,2), b=(-2,2), c=(-2,2), d=(-2,2), \n n=(1,20000000), colormap=ps, panel_layout=pp.Row)\n\nlogo = \"https://tinyurl.com/y9c2zn65/logo_stacked_s.png\"\n\ntext = \"\"\"\n<H4>Use the widgets to vary the parameters of this \n<a href=\"https://anaconda.org/jbednar/clifford_attractor\">Clifford attractor</a>.\n<br><br>Note that many values result in nearly blank plots that contain only \na few scattered points.</H4>\"\"\"\n\nprint(pp.__version__)\npp.Row(pp.Column(logo, text, i[0]), i[1]).servable()",
"_____no_output_____"
]
],
[
[
"Here we could display the new object `i` here in the notebook by removing the semicolon. Instead, we've called `i.servable()` to declare that if someone later runs Bokeh Server on this notebook (using `bokeh serve --show DatashaderCliffordInteract.ipynb`), your browser will open a separate window like the following, ready to explore or share:",
"_____no_output_____"
]
],
[
[
"#! bokeh serve --show --port 5009 DatashaderCliffordInteract.ipynb",
"_____no_output_____"
]
],
[
[
"<img src='data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAnUAAAHRCAYAAAAMkk8xAAAgAElEQVR4Aey9B5hV13nv/dvl9Om9MsPADEPv6l2od6Fmx3Zc4hLbiUsSO+XGcXJ9702+m+pry1V2LFtWtWTZ6h0LSYBAgICBobeB6X3mtF2+511nDgwYECAkYWntec7sc/bea613/dcu//22Zfi+76MXjYBGQCOgEdAIaAQ0AhqBP2gEzD9o6bXwGgGNgEZAI6AR0AhoBDQCCgFb4/D+R0ArY9//Y6x7qBHQCGgENAIaAU3q3ofngOO6OI6L53nq4yMWduN92FPdJY2ARkAjoBHQCGgEsghoUpdF4n2wTqXTpNIOrufBmKukYWTJnHadfB8Mse6CRkAjoBHQCGgEjoqAJnVHheYPZ4do5OKJJKKhk0URuQNk7g+nH1pSjYBGQCOgEdAIaAROHgFN6k4eu9OipOM4itB5vp8hc6eFVFoIjYBGQCOgEdAIaATebQQ0qXu3ET+F7QmhG00kVY0HzaynsAFdlUZAI6AR0AhoBDQCfzAIvCOkTqIt32uS8V7IIG1mXNkO+q8JDuOxOFVyZU2uJ3qmZWQ8unwnWp8+XiOgEdAIaAQ0AhqB0wOBd4TUjScx71U3320ZsmQt48qWDU7I9D67L7s+FZiID92JmFylbdM0CQQCap2VQbR98pHlRDHzJSDjMNKarffUrn08z8cwTRXD++61e/ReCJ5HWg7HUI7Ljvvh+6R8th6pTs6dox7jgy/7jzJO2XqkfPb7ePmOVO/4/e/094xM8oJziloSXKUqOf/eZpVZvKS+I+Gb3X8kDLP7MiJk+nfItlMgX7Z72Xqzcsjv7PfsMXqtEdAIfLARME7VjBLZG6LAub+9g9zcHHJisfcE3ZGREYaGhqmoKH/H2x9/Y43H4+zYuYv+/n7Vbm5uLvV1E5B19rj+/gFisagiVycrnES5Cqk73hu6tB0MBhH5du/eQ29vr0p1Eg6HqampoaysVP12Xfe46xTZg6EQnuOQdpwTKndi/RYyZxMM2DipFI7qSwg8l3Q6nWFCJ1bhWx4tWlDB9lj4mpZ1BDLh47pepn7F0oRE25iGged7Sl6fgyRExsWybGw7U5fjCsE+dAzUMbaNbVmqXk+lq3EO6bc0ZVkWlmWo8kLes4uQKN/zVUR09hzM7jvWWo6V5VgYHKv8ofsMAraN60lktiDwdhc5JyyFies46uXmZGscPwZSh+976iVHvUSMMVA7YOO7nsLQNA3kXcY0UL9l/AR3OV7U9BKsZNk2lhpzIfSSViiD5cnKmC0nbUm7cr0J/bQsE3nBkZe7t16yVoRTSKzfutHfO+LUnle/V73eoBH4wCNwSkjd+IfF1m3bWb12HeecuZDq6qoDZObdQDorR9u+/by6bAVzZ89k8qQG1fR40nmqZMm2J/UtW/46P/7p3by5cQuVpcXqYbh7XwdNDRP4/Gc+wQXnn0dnVxc/+NFP+KuvfplwOHTSYgyPxjkRAiYP/PXrN3DvfQ+yau16CvPz1EN2cHhEEYrbbr6eK6+8nPz8fEU83vJBLoTHTbP0uecpnTKHaZOrSKfeAWInmgjLJjXUwQuvrGL22RdTVxBkxYtP41c2M3/qJJx0+hQRj+xwGIRCIVw3TTp9lD75PgP9/aRcTxE2KSkPb9MOUViQizEmN26CHTt2MTiSICe/mIa6akw8XDkZfV8R+8G+bnbt2YfjG1TW1FJeXIA7RlblWR0IWPT3dLJ7bweeYVJcVkl1eRG+6ypNlZyDcsz+PdvZsW+UubMbGRoYRHiEjKMQ1GAkSn5uDrZpHBcBP0gSTYXv8VCGLHqHrkWTZOGmhli5eiO1Dc3UlueRPoy4Hlrm2L+kv0Kahrr382brTppmzqUsN4jjSlvHLnv4XumXkM2hvm52t7WTdj0iOfnU1lQRDVrqt+E7rFm5hsLaiUyuLWXFi8+yddDimisuJC9oMTzYp66ttu4R5i44g8m1JfR07GPD+o10jricec45VBeESZ+EfFl5ZTxsE3Zt2cC+eIiFM5swvCTdPYNEc/OIRYJH1M5myyNnimERsC08eXF4G7IcrPPkvomlQIiosg6c6ICdXJO6lEbgA4XA2za/jic2W7ZsZX3LJmbNmKYInSD5lgThFMKdbau6qpKZ06eybkOLuoE0Nk5W2oHxsp6KZrPtPfDQw9z+Z9/grv/zNf7ma1+lvKwUwzDZv38/L7y0hM98+e/4H3/xp2xq3cz+9k71hn2y7YsmQPLQZds+Vj3SX9u2Wbr0Fb761//Ix//oFj76kQ9RXl5OOBSit6+XLVu38b0f/ISWTZv40p99kcLCAnXDPWr98oQxTAw/zV2LF3Ptr19hdvMEUsm0MBWlQVBlfR/RWYmWKruIPLJk6haTKojmQxbZJ5+slklIklRnmjbOYDu3LV7MS5vaaSwO8uK3F+Pd8QgLZkyBVErZzJRY49tSpmFTPegPbVe1prQnB9rKmpFNEy81zBNPPUv99Dk01Zcd0JwdqMO08JwRnvr1w2zrGcYWcyfgOimiRXV8+uOLyQsHSMf7ee6Jx3lp7U6K8mP09g1zzqWXc9UFC7AUAbTp2LuZ++9/lK6EQch0SRoxbr1tMTMbKhRJDgZMdm1azX0PP8WIHyRs+QylfC67/GouOGO6NCpgYvge6159js2BadSV7OG+Bx4kGcjB91wE5JTrM7l5FpdecCZVpULcM2RV+iRjIR/Be2x4FGnq2beb1t1dzJw7ixxbqGiGiMoqM36ZMRPws7ArjAxDjXlWQ2XZJr47zFPPPMu1t1RTV1kAqraM5ujQuqRu0TJmzgXVN2lw7PzI/jZNi6Gudh569Ck+Uz+dyvyQImCKu0h/pAXVmYOyZes40B6osdu2biWPPvU8XaMe+bEw/QODVNY1ccO1lzGxspRUPM7SZ59j9tW30FRbQtvO7bQOhLjU9cj3fJY9/wRPrmln/twZmIaJlx7lud/8mjc6U8yfPTMj+5j80rZgJZpTdaVkgcv2bxy2h8grmOKys+V1nmsrYOHsaSR7dvCj79/DosUf5pw5jSTHae7HYyhYiFYzPdLL8vXbqG9spqo4B8fN3EMOngNyqoy/PpXoGbI4hmlWJlkfwPEwnA9eJ6Z6ccn8zpxjeGlWL1tJqHICTbWVSsufrSfTmv6vEdAIvF0E3hapy94QRIjWzVto2djK1OYmpjQ1KrnG73+7gh5v+WybzVOaVBEhmXKrahJipx5e6jl4vNUd9bhsO6Khu/2L3+C5X36HSy+56JDjJ0+ehHwm1Nbyjf/1r0yqr6GmqvIt3qoPqeL3foh5Tj19xz0Qfu+gsYeEmFw3btzEX/zt/+Tv//pLXHP11cocKGZL0fSJ6bWuro6mpka+8c1v8Yt77uVzn/sT9XA6Up1q2xhHEyNa9XnnEQ0GxgiZmGPDimCIJsYK2AQN4VwZE6ngZQeCmPgZE6FlEwwapFIp1R1Tfgds0vIbQ5mLnVRS1S3augVnLyBoCfmAvLpFmDkxwgELL2nimxYBy1Rl5akpxwTDYaXNkoeXaAeEeonZWj1EDItwJKA0YvIcC4ZDOOkEvhkkYKW5+dYbueeJVcydWsfg4JAisWJGFS2HMh56cXqGE5xx1oXMmFxOMu2olwc7GCFsmZiGz/rXl/Lkyt189KMfZX5zFetWvMxPH3qc0rIyzps1meRgJ4898hvisVq+8MnryLcSPPbwgzzw8OOUf/qPqCyM0bNvG/fe/2sKms/lU1edT17A5ZXnn+Z7P72HsvKvMmtiMUkHUqO9bNw5xDk3NRIODNE37HDRNRcxZUIRTtqlr3Mvz7+whB/u2sun/vhWaotySDmuwlj6JCZjMScHTINkKq2293bu477fvEjDnDkUByziKQcZo6y2TzAWXMUc7KoHu/wO4nuiCfIyZmVLzL4ZbV1Bfh7hcFBpixLJVKYuQzSHrhob0b4Jv3fSGaIqdbuCt7JqystJAHwxuYtG1CYUDlNcWIAtJkhFTk2CtkXaSeP5xkHZ5AXINBFvTNO2lHxpxyMYCrBv65v89y8fpnLG2fzponMoy4uwf+9Onn78CX56/xN84RN3UBo1ycnPJ2iZYAW58raPcalvkBMN4zqj7NrZzrlXXMYtF5+J6TrEh9vZ2dnPVVffwqIzZ5AcGcbxfEKhgOqbECfLDiiNbVZjKS8Xcl7KJZ3tv1xrgoFoon1D+hMkEs2hICecuTTHMJe+C4u1TAvDMpXZN+OWkDH9y/VkiXYs0c/DDz/Khz9dx8SqIpzRhKonKG2ICduT6zOg3BoU4TuAWcbEmyGB/lHPAUWi5aVMzNEm6sVBOhQIBnHTKXxD3ANg6TNPUn7JjUybVK/GVzJrjt1SMv3S/zUCGoG3hcDbInXZt6xNm7fQ2rqZ6dObaZo8+YBA2f0HNrwLX8a3KcRObugtm1rVjV/I5ltwoeOSMEvokskkP737Hu76568dIHRZfyw5Rm7WvX19rF77Jg0TqlXdiUTmZnpcDR3hIKn/eBZpO5FIcv8DD3HLDVdw1ZVXKG2MlH355aX863/eyYP3/rfS5NVNqOUrX/oCn/zsl7ng/HOZN28uyWRiTGty9NbSIyNKIyKkORiw2Ld1HU889iRvbtpF9ZRpXH3t9UyfXIObShIIBelu284zjz3GirVbKKtr5PKrr2LutMlYeCRH+3jhhRf43SsrSPhBzr7ochZddCZ5ORnH/6HhQfWAF2k8zyQdH2TJU4/y0K8eJ7dqMldedy3zp09Sb/8By2Dbmyt4/PGn2LS7k8aZZ3DttVcwuaY0QyKcOK8teZKnn/sdvSmLcy+4hCuuuJTQ6G5+8fO7mT5lJi8+/gA9beu5/qYbKQ6m2Lmzk8LSMvJiAVLxUUZSHhMnN9E0uYSReIpgMKC0dSnHJxkf4I03Wlh44UXMn96A6TnMXng2F7Rs4PXVGzlz5mQ69+6idX+CP/rUJVQX5+D6eSy67BLWff+XbNq5j9rSRrZsfJPBYA2fueFyyqIGKdfgwsuuoqhmEoVRS5EF8ffq2LmPPseksqIEevpIY1FRUcXkhkoSyTSNkydRU13Gj370C5asWM/tV5xDwPTYs62VN9asZ3/PAAUlFcyfP4+m+nI2r3+dpW9spKGqiJce+w3VE6dy7vzppId7WLl2HRu37iSNTWPzNObNnkYsaCkiu2/XFlauWktb9yAllTUsWDCX+qpyHEWsbIZ6u1jy3FpWt7ZRUlHDWWcuoKY0Dw+L0f5uVq9dy+btezECEZpnzGLejEYCho8ZsOnrbGPF8pXs2NdDbdN0amNCHDPmZbnGcRKsW7WWVetbSbg2U6ZOU9qznHCA0e42XlvTSkVNLbtbVtORzuW2Gy5k7YoVmOXN3HbdFZQXBEmmHBqnzSZieXznrkfYubedimnlilgprTMeuzevoy2Zx8LJRbz86msMGGHSW9bzyPAA9ZVl7Nu+Di8UY9vaN+jfv4ezLziX0rDF5g1reGPtRvqGU1RMmMhZC2ZTXpyHl06x5rVXSOcUYsUHePWNrVx21dXMaChj07rVrFzbwmDKYN6c2Yykx17o5CI4cCMzMH2P1a/+Dr+oHCvez6o1mwjnFzF37nymNdXTu3srS15eRmVtDetWvUzXnirOP/9MIjZsXp9poz/uMLFxCmfMm01RToRkvIeXXltLaXk1fXtb2diZYPGtN1HoDfLa6jVs2rrrsHPABNMmPdzPijWradm6GyMYY8q0zDlip/p4+sWlOLml9G3dyEN93Zxx1tnUFkdIOyduOj/6XUnv0Qh8sBE46FF9kjhs3rKVzZu3KE2YEDohHaIFkvWRPifZzFGLHamN8TKIhk4+okkUWU/lsmPnTtZt3MJFF16gqpV2syY9WYuW58d3/TebN29VfjolRYUUFRYeMF2csCy+EJpMxOmxygqhFLPrnj17eGXZKq664vIxPzFXjY1o5v7x77+unOuFBAv5m9LUxA1XL+LV15aP+dUd+9TIvF1nolKDdoDdG5fTOGM+D65t58rrr2Rw45OcNXMSy9bvIJKTQ/fuVj4zaSqf/uXrXHTVVZg967hw4QyefL0V23L53l/ewI23f4TGeWexoLmCjy6+nO//6kV8Q947Mhof6bNoFHJKq/nax27hOw+8xPmXX8HAuvu4eOFc1u/oIjcaYsNrTzJz4Xm82Wtx3dWX0/rYnzO78Q427OgkJxrgtcd+ziXXLKZq+plcdvY0Pn77tXzuX+7BIUhObpS9m9sIRUPEIhFCQZtdm1v43//wf3h5/WYMCTgZGiUQMNm9eyu/evB+/vueB3n+5ZX0jaQU7qnBHvYPOTRPqMI2XaUhtOwgE6fU07O/C8dN0TfUTaCgjMrKAkUa0ukUOSXl1BRG6BkZIpny6Ni5m4nNTfTv3cx9997HT+/+JS+90UrjtDlMqChENE4WLnv37iBc2kB5fiSjJVMan7TSkoqmVHIZVtRM4oIzp7O1ZTMp32f/9nV863//B7uHfWbPnEqifRvfv/tB9nQNEg2HMppP1yUUDilTPe4oj91/Nz/4zctU1k9iUmUBzzz8AE8uXY0RCNG3fyc/u/sXtCcDLJw/i5E9G/inf/8xOzsHld9mJODyzLNP09qRYta0BnZsWM7PHnqCXnnHSfbxi7u+x0MvraehqZnKwgAP3HMPv1vVihWKMNzdxn0//zlLW9ponjaVxP7N/PrZ3+EHo+oyEOPwshee4vv3PUlOmZj2inj60Qf41dNLFcF1k8OsWrGUX9z/CG0jHrUVZbipQbbu7WTOvFkU54dIJERj7BMfjVM6oYmvfvXzTJ9Yrqbds8YCT0TL3LZ1LS+v3a40T9FIUGmmMS3ku2hzw5GwIlkEbEKRkNIur1+xhG//WEziBcyeMYkdq5fyg58/zL7eEaX53L15Iw8+9GuWrNpCZXUlsXCQrW+u4Lt33cegEWVaYw3LX3qWl9fvJSdkH3i5UV+E3Hkee7du5IEHf8Xv1u5g6vRpOL17+cHd97FtX4/Satry0pFOYwVCBCMhbMtgw4ol/NdPHsKNlqiXr+XPP8E9v36GISFZXoL1a1dy7/2/YuO+YSZUV2G5cR5/4Od8/9Gl486BB3lq6Wql5fZG+3js4ft46LlV1DQ0UV1o86tf/IzfvvQ6LjbRsPjTuaJWJBaNYJkZrfqx7mV6n0ZAI3BiCJyUpi6rqdrf3k7r5q3EYjH27W9n1+49qvXx2rKsOOLpIuavBfPmqKjUbB3Z/SeyzpZt7+hg5RtrlLPzkcrLcbKIGUNkFFnzcnNPWfu9vX2UlxRRXl6m2sn2O7uORCL8+Rf/VD3os/OxCtkTeU5mEU8c+Xsrc4X0W9rp7ummuKhA+TeKKUtpNMb8YWR/Vk45Xky1c2bP4unnnlfmGCGFWfyOKqvSwFj4boLffu/vKfvw33Hv//tbKgujXHnROcS8C/j6nb/l5Tv/jOWP/YinuY1ND32PydVFJK+8mJqiPFasbmXR3DoW3vIN1n1zPjPqK1VzFfkeV//9s3zhw9cgD1Ux48kiMnuJLopu+zv+69/+lqrCKIvOmc6KR6aybvMO5k7K44d/fQt3/N13+fbffpqCSICLzpmPc/5E7vzty9xZfzMbX/g/3PwPP+FP/vhDRG3YtmAea1r7CORX8KGPfozvfv7rLLpyMTdeMouR0TjJgkIuuPQCqouKlD/c0EhCaZ2WvvQSM2fNoDg2wGOP3MfmfZfw4RuvwBCToZiT7ZAKmlAGJsNQD9SUM8KIIwEWPiE7QtTO+OTJs9nHJhwMMDiQREhekhDdu9Zzd8sqZs6bQ23eMM8++ivWtu7ij2+9hsKImKvjtLZsZ9qsizPm6KwmV5nzDAzls6iiJsgvKSCR6mEklSYQK+CzX/oyM2ZNpSAUYMH0Cfzbf/0323e2c8W5czh3YIifPbGc8y+7kupogOHRfhrnX8g/3VhPQ005tu+QH0nz6+Wt3HTZWXR2dDHg5fKhyy9jWnUB86Y1MqtlCyHDV+bD5MgIzfMv4JYrxGRvMqEkhzvveZp9nQMUlpssuPg6bpnURE1FEX46SSQ9yGstu7jszOls37iabUMh/vRzdzClWgjZXIqefJQn3tivzHuD7bt4+nerue2PPs6iM6diGT6Tqor47n0vsuuMhUwIhfA9k/MuWsSV58/FtmxGO7cyFPeoKMhRZt+Mb2D2VcWkoLBQGQa9QyKsDYLhHGIBg3BBORddeinb122kpHkmV5w7F99xaZ5UyBurW2ieMYeLFk5joH07zzz/CudeeQ2LLzuHSMCiuaGWH37vLl5bv43F588iFLQoqWnmY3dcR3lBjFSij589uIQp58j5dBmFkSCzmhq45+5f4MhlcOAGMPbFEHeFAMVVTXz09hupKMph9pQqur53Nzv37KH5ggVcfn6ctS0PMHvuWZwxo5a+9h08+8KrXHXTbVx/yRkELZjeUMF//PBhtu48kzn1YSzfZ8bZ56kxC5vi3jDExDnn8a3r62moPngOPLq8lRsWncHeHRtZurGbj37mT1jYVANumoqifF5Yu4WR8+Zz8cWXsHXtBkqbpnPNeQtIj2ZM0weUjuoK1/80AhqBt4PASZG6bIN5eXnk5eUiKURqqqsoLCzEccRv6chaHol6kvQesmQJRbauE1lny+bm5DK9ecoBsnJ4HeKsLb44YgJta9unZBV5T+WiXJ4zfOP3qpU3d/nI8raA/r2aj2+D53oEg+K7JtFxyntcEcqWlo38+df/kVVLn1a/RfsnmEajEZKJ1PFVLkd5rtL2pZLDvHDXK3zj4W9Rkheiv6+f3PwSFt3yNf7XjU/R8y+fYOtrv+HPv/sf1JQVMDQ4oPyK7vjK/1JRgcFglHMvPJe1b6zknpefJ+15rHl1OQsnT1SkSD3HDj6/SAz18Nkrz6YoJ6S0oeG8Ci6YAykcEoPd/HglPPT/nUduECSFTG5xJbd+/RvcdO86/uMLN9F88df4ykc/ScHALi47dx5TZ87humtnq2CP0dERJHlIKpEkmU6RchyKK+u448MT1NNUtB12JMbExhmcfemlTKktw/BcJleW8cNHlrBjwQIao+J/d9hJITnmTEUdMiRi7IVDYDz0SDFFZa4fD4/hEYfFt3+YeU3VmL6rHMzv/PnjbNo2k/PnT6Nn13a2D7jcVFOtzNjHpPyS+sQ0FEGeUDMJjF2sXbGM0ZEko4kBUn4A2/Rx8ZSZWjRT4p+VTPnYoXymT21i6/ZtvLhpg/K927Oni4JYVGkHS0qLyDOG+OUv7mHezCnqRWLarLmI+XN4YIi0H6BxQi0hy2M04VFcWEhO2CIdHyWcP5FpTbB9x1Y2rhtWfpY7O4fJzalU2qWu/e3UNk6lrqyQRHyEQDjG1KapPPf6DhVo09szCHaE/vbdPP/sXmUW9Ef7SAwPIC9edVUmccdiQnU1IQsSkhJk7KXmSM4MMh5He6FRASAS1OO5pMTnT8i4pPURv0rHGYtcNnHTmfQtgz199KcDXDZ5MmHTZ2Q0QWFpFbOba9nU1o3jeHhOktKaCRTlCZYuyaE+OgYcLr64ifyQxcjIKIWllcxsqmZZt3OQ0x24Ug3SyQSlNTOVz51oGyN5JZTlRhhOS5ocX/kaisnWSaeUtn6op4+EbxPv72DJS88jcU5BbxQnOUxHbx/exAoSCZdJtTXEgrbSYNrBvIPnwMaD50BeLFcRuN6udoprJtNcW4aTTOBiMn3h+TTNcQnYBomsf63jqjEWrXvmpeNAR/QXjYBG4G0icFJcQwiA3PRi0Sjz585mxcpVDAwOqgAJSQfxbi2S723ixPpjNid+bzt37VIpRETWaDSqZM8Sw2MWPtrOsVfLoqJCOrv76OjsoGHixCPWmzXJ9vT0KjnmzZ1z0oRWtJ3y5x2Htk5YnGgK+/qH6OvrJycnJ6Pn832FwSXnLjggh2AhJvM9e9soL5fI3QOqgKMhoIiIIrRyhO/QCeSIQ7ZErY6ZSe1ADtBH0nUZHYqTFwupfG2yXx5e4XCMmG3S376Tb31pMf/50Gq++LVvMn1iFYFgiNEx7ZyqcNw/zwsoB3ahQ0KAfDeFGwdLfc8kUg5ZdsbUo9zwfYI5+dC9E3FLuuiWz/LqhFn8bskSvnnrDbQC//jdB/jzT9yoNJzjSZYg4bqi+ZRzXlIx+JTVTOSOOyYq02wqkcAKhmmYVEdZ2Gd4aBDybQwnyUg6rtKQyLUiEY+p0VGEwEYlIhSbeHKYoZRHLGKqlCb4KeLJNOV5IQJ2gJjlkVdSTXN9OW4yQcqwqW2aTH1pmM6hYTBhf9seCJdQXZ5/IFJXZFZtSkSruARYtoSD0rG/g3Akn4KgxZvLXuSnv3qG0tp66qqryI+MRWYqnDPf1Vfl/G4x0t/BQ/fdS0t7nMbGBiqLC1TwgJBIceYvr5vCZz/9Sdau28DGjWt56fnnKayezEduv4mKmIlnWlhmRg4ZM9cTVwAPOxhioGMnP/rhz+hIhpnSOIHSwnxllpRxkFPASTkqJ6L8EC6sXkJUfaqnOH4mt99ofBRSmVxxgVCYa66+gvKiGK47oJz3XSeTBkYqsSJRcsIGbd0DuMLMVMRt5iSTHHOJeBwrGMKSXYfR7sxR49za1EuqbJWDM3WpX3JdOeIuYaqXH5E7cy4YREJBnKGUhMKq/V4WDwkGcjIBOZKCRM45RTI9SV0zzvSaFSK7VlZYiYz3FXH3hCxK9LiSbUwuIaBjJ7eY7uU6l9yVfb1p0iq/XohLL7+c6ooCBCvR7EseSnUPswMk+jt48AjngJJQxsVxVDCUaEqlbd931e9QKKxeDn4vKEJhm+2AXmsENAKnAoEjq9SOo2a5IcgNSsjCGQvmK7+sJS+/kokWVA/1sRvYWKqKzENm/OPyOBo5jkOy9Y5fy01IFvFpE5mSyZSSURGbsTQOx1H1UQ/J3ovq6+uZObWRF176nTpWTJrZG7dsUDfDMX+cR3/7GP/8b99W245a8VvtMETBkEkVcKxDZWyENIUH8D0AACAASURBVNXW1ir/qPUbNmTIyljf44kEq9a2jD34M/53otF65rkXWbhgHqFg6C3lFAxMO6hImpC3i8+FF9ZtxfVtYjkxbFy2rV8CF5xBcTRK/YJzufOJlSQdj2g0Rk4kyMblz/Ps8ha6d21VhG7Jhr18+1/+gc997tNceMYsNuw7mGBZTGbZRRE5FfR3YIoFTMlH7LkEcgq4GHhz2058O6wSPeOk2PDCvVx0mZjmoH1fF81zz+Mv/vYfWRMf5bGf/Rf/8IVP0TmSVL5GcTHZh4KEAkFFolXC13QazxMzl0nr+te4878foifhEQmHVNTu0PAow45owmzCeUVMKI6wYetuHFc0oGGc5CitG7ZRUSt+diFKispgsJM9ezsRAiIvQ33729jdk6A0P59Q0KS4qpKBvj4GRlOK1IRDQUYHBukfThENBCVUkl2bW6ltnkJuKKD8DWVchGiJdjYcCiMJpkMBk52tb/LSslamzZ5KyDJoXbuauoWX8pUvfJaP3nEzF58zF8uXtDSZs1tdqYapIiIlp2Jfexub9o3wsU99hs9/8iMsvuEqJtUUkUh5ioD19fTghgpYdNW1fPGLf8Zffv5DjOzbwvY97SpaVfy+1EvA2AtD5v4heeIMOvbtY89IkM998bN8+o8/zI1XX0ZNUZRE2iNgmxSWFrFv9y4GU556UQnZBj09HYz6lppmIz8aUf6DU+cs5Pbbbua2Wxdz/RUXMn1qIyVFBSpxsHI9GLtwhSgFQ/nKDPrGilW0dfYTiYZVxK5oqzt3b+Jf/+07rNvWjkSHikbpwCLyj9Uj2+Ql62DQQvaoTAoPiQiN5ccI+nH2d3VhBMNEIxI1O0Lrzn2UlBYqwi3EW/CQj+AUyM0lJ+ixc7/kJgxkynhJdot/XCCThFq1NE4U9bo3LvWIkk3kHCerqBUDoYBK2pyXH1Um/olTZ3PLrTdz2y03c9O1i5g+rZma4iKlFVapi+Q+Tya5dV/HviOeA/GEg2HbxAqL6W7fTftggkg4TCQSprt9J0tfe50hiVoW1wkRSUW+BzOpb7KQ6bVGQCNwShA4+KQ8ieoyN+YMsTvnrDNYtmIly1eu5KyFC9WMEkK01I3qJOo+3iKH1y9tCvGRGSWWv75SvSGfdcYC5VN3quTJ9ltyvX3iYx/m7Js+xcS6OhUBq0jXmPDZ7797eSmf+Yd/Z+kDP1TyvB05snUeCx+RT9IUlJQUc/utN/Gd79/FjOnTqa+vU1qVSQ0N/PP//Fs1NpKYWI5/4smnMuR34YLMw/doDcgdXvl/QapvQKVgsIMxrvnq97l08ceZX1vEJQua2L7qRT7+9e/znYdeJCc3h/lXfJLeb13Pf55dxx1XnkPXjrVccvVt/M1dTzHtrCLV2rJlr1NijbBjzVKu+9hfs+Car2fIpe+zu2vXAX2Jk+xXD7/xD6z0ACSSScKxEr5459+z+JPXUFP8PAuaKnjjxUf52o9X8uCSnxC1U/zTpybxVMmX+Pk//CklUZeWDW/AvMVEJW2JYyLzkDz+4nPUlIZonFxP+46N/OyXj3LeDYu5aP50inIL6dj8BE88V84lZ8yC+ADPPfUcdkkdNZVFmFaEhWfO5wcPvcSSskLmNlezfuUrLNsxwscVsfQoqapjzqRiHn/iKfIj15BnpXji8WcJVzfSXFeBZPZonjmP3736Ex557DmuumA+AXeUV557hh6rgCmT6kgPdfDm7kEunFuNPOvF4itEImC6bNvWCqMdyjTau383zy15jeLGuZw/Zyq24RGMxuhu20vrlq1ErTQrlr5M+1AmvYgMsORcGx3sU4l3mdFMIBzBdhNs27qFXGuUjh2beXrJm4TLpmBaBrs2vc73HlnFhz50M801pbS1tTPqWhlSLclmZaaNcdxIrgGZQUPev4RAu4lBFXBljhSzvXUDL6zZSVVTBa4ZoK5xOv6L9/Cbp1/kogXTGO7YwVMvLCMUyVdkrqh6ArPq83nkkV9jpC6hMGKy7vVXeHZNJ1/5iy9QJql1xDyq2peT11PpSeaddTZvtPyCex54lGsWnUtxbpi+jr08//zz+IW11FWX40qKFKVhy1wQKoXLONN6pg9jHVMEykdmuZAtki4mVlTFwun1PPnUM0Rtn+rCMBvXLGN9p8+f3NBA0BLtb0bDJuez1B+MFDFvViMPPv8CxbEQTdWFrF+znFXbu6ibOkn1Q/AbP5uGSk0znnyq3IkSsCYqM1EGWhhugo0bN5AT9Kktr+LM6RN45De/IWheQVV+hNZ1b/Drl9bx+S/8Gc3lptKuyfkk3VLnVTCI7SV/7xzIKWsi5UD9xCYq7aX8+tGnufqihRjpIZ7+zSO0Bycyfe4cJYeQ+K0bW3mzMIfamkpiIdFGHu1mo7drBDQCJ4qA9c1vfvObJ1po/PFZgiOaAUm6u2/ffhUwIcEDon14OwRmfDvH8z3b1tDwMK8uW64CFE41ocvKIf2Wpaammml1FVz/6a8xoTgH8TOUt3shXzKDxONPPs0f/dn/4Gf/+vdccdkiVSZbNlvXCa3VNEHy8H3rRdqpra1h9+7dijxIFLDMGiHTgk1pnKxu1MPDI/z2scf4v9/+Ad/65t8wtblZ3cyPKqM0PJZ8eOWqldSdcSGT68spq53EuTPr+fpHPsLDS17izu8/w//7+b186Jrz8J00JZUTueqCOfzorz7Cdx98ge98507+6Tu/4PO3X05hcTFnNtfwqU98jIfue5juUA1f+eMbGep3ufKaCzETvaxeu5NLrryJqvwA61cuw6xbwNwp9XhiJvJdVi1fQcmCi5nZ1EDd5OnMqC/mYx/9GC++8Dw/fGYnP/vlfVx1zgylGZly5nW0PX8P//MHP+exX97Fdhq557/+holl+ZhWiPr5k/m3L36RZ7YOcNPia/AHu1m2ah31U6dTW15MjmjiyvNYs2I5K9asUzOoONEybr35KiaU5pFyPEqrKikIJHlxyVJWr93Alr29KsXLGdMb8JwUZjBGTW013Ts38tJrq1j75npSkVI+dPNVKjlsKuUQyyugob6S1tWv88rra1mzbgMjVh4333Qd0yZWsntLCys3t3PRJReQY4saN5MaZPfeNrq6O9nYupWtW7epKMszz7+YaxedT3FOAA+TorJSOre3sGz1etZv3Eb15CkUWB6lEyZSU15EOBAgNdzDqtdfp58CFsybQV7Q4bVlK1jz5gaVCmb6lIkk4i7TZ02lvKSEsDvIa8tWsr6lhZbt7Vxw6SIWzmrCTw2zYdMOlfi2UsyhvoEXH1CazIam6TRNrCTgjvDKaytZu76FlJ3D1MmVxJ0QM5rrKSwqpqowhzffWMmadS1s6xjljIWzSI2O0NA4lbKiPKprahnav4OXl69mw8ZN9CRsbrzxOqZMKCc13MuGrXuY3DydiuKYMlGqKOr8Ihon1rB/ZytLXlvJhpaNrG7ZRunEmdx+3SIqi2NKm7V542ZK6hqoryymbecWetIxlT5HZpvY0rKJnJqJTK4uV6za9xJsaNlKdcMU5T8q8cmiMbeGO3h5mbTRyv4hn+tvvJ7ZjbVK27pneyvpnEpmTKpB6pTUIFVVVQRSvby6YjVr17Vg5ZYzc0oFw/EAc2c24sf7Wde6k4nNM6gty2dH6wac3CpmTKrOOBz4LltaNpJb28CkqlLsUAjbT7J69Vp2tA0we+5MGurqYGA/L7+6ivUbN9HWk+Ca665l9pQJOMkhNsp50TCF2opC9TIYzS8gNwTLXlt+yDkQT3hMn95EQX4B9RMq2dm6nuVvvMn6ls3kVU/lIzdfocipYdjEwjZbWtbxxqZd1DdOoywvdNzJ1N/6jqeP0AhoBE7JNGECY5ZQSdCERKRKbqyqyooD298NqLMySFTulq3bVaTtKfGhO4rw2fZk97LlK/jRT+5m3aatVJQWkROL0t3bT01lBZ/79Cc4Y+F8Vcv4Mkep9i03H+80YdKW+OH09PTxk5/ezW+efI7rr74Uyd8XCYWR6OGXX3mN/R3dfP0v/5xzzj5L+da9pQBjB0iEpsz4IHNQCtGT/HDDQwPEkw52KKKmpvIlua0yL0nUr8Xo0BBDI3ECY1NXSUJZ0dbYtsnwwACS5y2am0vIhnhCEuFK0llPaVokebGYT7PJVWXO1Mzik06llVlHktGKqU2Ok+myko6k5YiSmxs9kGw4EAzgpBIMyHRamOTl5xMOSiRpWjnQi5koGR9V5ENSeogHXDKVUglgsyYk8XcaGRykt79fmchKSkvVA0vN1qB4r+Ru8+ju7GQoniaam09ZsZgCM1ocGRtJQusmhuno6sHxTYpKSpXf4cEZH0Dy0CVHh+jukXZsCouKyImGsHyHpU8/xLLOfL74R1eq36JTEVOemN7FRD323qEgCgZDKnlvJomszBtqkU4lGB6Jq6jc3FhEERgZR4k2FgxFVnFdMK2AGgeJ2B0eHlZjJCZ2SQYtud0kUlppfHEZGR5mNJlWJnZJWyHyuL6nfLRkJojsDCIypiKLbLNk7lvfQ17GhPDlxHKwDE8lSJZpvERzKOM6MjJEIukQjsaIhgIkU0klm8glfoOek2ZwcBDXt5SlIByylRZMVEHKZ2x8++qelcFXTOOdXT0qaEcS/BYVFahpubLz8IrWW/AQGcTXTJRf2XMvuy+b9kTUUVIu21c1zuI64DvKeiDThYUjUWIqeXHG/1Np9sTULdfR2L1U+iMEb3BwWJ2jou2WfI6SNFphIn1yHGUyluAXkcMfV0dWDnUtjAWFiCFVtNk+pnrxlDHDy7QhwUmxnBwiMp+zOkcP7YcSTCK4DY54DmT9/eR8TSfjDA6NgGWrbANynmQSLZtq/FMSROFmNLRjE8qo6vU/jYBG4O0jcMpInYiSJSzZ9dsX7+RryMqQXZ98TccuOb5+cTresXMX/f39yqQpWruGifXKD0hqGX/ssWs99l5JDRMfNy3QsY6WNuWhKw/ndevWKfIphFdISllJCXPnzuasM8+gqqpSPRiOVdfh+zLavIzzujxExIwik44rMjE2ubk8kLPkQmSRmQvUg10mmVfzl2b2y75sWfGNkwenECjZLou0Nf67bMv+zu4/sE38OGXCcyEL6jjJnZjJ75f5LVF3GfIiZVR7auaDjP5TmjSFHI75RWbql5kLDsZKStvSl+zDPGv+OqjhzMgg2MuDS0x4WZIg9ckidWTIQkZOwePwKeBUO0J8xh74qh3fUHN/3v+jn1B51iIuWTBVzaaRbTu7HmtGraR9WbL7sm2L/PJd9qt9CjslnCLqmQARGVvxkc1MIi/jmTVJShkpLx8ZaEUI1bZDMT+k7jHBDmyT8tmyY5jLqGfrlsOlfnXuSN2KKPqHnB+yX4jsAZzkHBo3puPrGmterVQ5RdgyLwjyW52X41xHDsh5GH5Swfh92XoP35aRzVDaeyWHyJ/FO1vHWB+zdRwoI+4R+Egk++GYjG9HfT+sjvH7s/WKWV2uVTGnHtqGvDsdQa7s+TBWQeYaz7wwHH4OyCFSp1gpMm4iGbmVCTd7Exjrr8iWPSezsum1RkAj8PYROKWkTsTJ3ijevmhvv4Z3U5ZjtXWsfSfbS8mfltHIZIjIseqR9uUmK/nxZFoueWjJIjdWMZvLItqv7INBbTjJf9JWdjlifeMeEofvz5Y9fHu2vhNdZ+uTckeqM7v/SPuEoWYfokdtd1xfhNAcaSSybRxNhiwZzrRxkAAf2maGMB96jMfoSEJNzSaJkMfBfmjRt/iVke9o7aoLWsA7UEu2P0fEbOz6VwcfBY8DFR3hy1vVfTxj8pZ1HKFd2ZQtJ9+P1rejFD3uzQfaOAFssmXeKZnG9/142zgemd7yGHVxHTd0+kCNgEbgOBE45aTuONt9Xx4mN7LMwzVLbDIPy+O9WZ4IKPKWK8Qu29LxlBX5RJbx8mTflsdvO5669DHvPQJC1DMatPdeFi2BRkAjoBHQCLz3CGhS996PwUlLIH40MgWUXj6YCGRJ+gez97rXGgGNgEZAI3A4Aiedp+7wivTvdx8B8deSeTrFOJY1d7z7UugW3ysEtHb1vUJet6sR0AhoBE5PBLSm7vQclxOSSkyoEjghPnay6If9CcGnD9YIaAQ0AhoBjcD7AgFN6t4Xw5jphETFSpJViazLes5rgvc+GmDdFY2ARkAjoBHQCBwDAU3qjgHOH+ou0dhJ+gzR4MlnLI7zD7U7Wm6NgEZAI6AR0AhoBI4DAU3qjgOkP/RDtL/dH/oIavk1AhoBjYBGQCPw1gi8rblf37p6fcTpgIA2wZ4Oo6Bl0AhoBDQCGgGNwDuLgK21OO8swLp2jYBGQCOgEdAIaAQ0Au8GArbW4rwbMOs2NAIaAY2ARkAjoBHQCLyzCNh79ux5Z1vQtWsENAIaAY2ARkAjoBHQCLzjCBjx0fiJzDT1jgukG9AIaAQ0AhoBjYBGQCOgEThxBHT064ljpktoBDQCGgGNgEZAI6AROO0Q0NOEnXZDogXSCGgENAIaAY2ARkAjcOIIaFJ34pjpEhoBjYBGQCOgEdAIaAROOwQ0qTvthkQLpBHQCGgENAIaAY2ARuDEEdCk7sQx0yU0AhoBjYBGQCOgEdAInHYIaFJ32g2JFkgjoBHQCGgENAIaAY3AiSOgSd2JY6ZLaAQ0AhoBjYBGQCOgETjtENCk7rQbEi2QRkAjoBHQCGgENAIagRNHQJO6E8dMl9AIaAQ0AhoBjYBGQCNw2iGgSd1pNyRaII2ARkAjoBHQCGgENAInjoAmdSeOmS6hEdAIaAQ0AhoBjYBG4LRDQJO6025ItEAaAY2ARkAjoBHQCGgEThwB+8SL6BIaAY2ARkAj8H5EwPfBl3/I59DFMA3k771cfHx8LyvbeFkMTPk5ftN7KahuWyPwHiFg+Jkr+D1qXjerEdAIaAQ0Au8GAoqrHUbWDOMEWJAPqZQDhvduiDvWRlY+WdsEgyfStE+mz9kyY5Q0W2V2s15rBN5HCGhS9z4aTN0VjYBGQCOQReBwrZtpHtnbJkt8hN8N9cDG1jZSRg+GL8cr3Ri2YdL78gjfffQxRnMGsDwTX6nFhCHJcdm1C4YFvhA/+YzXqmWPy7Iq2S/f5Zjxa+mBB4pwBpR20DVyiRnNfOH2egqn2rieofYbhonjRyjNrWZqo40fOlhdtpUsHpm1jzeOkwqpPRFee2hd+pdG4PRDQJO6029MtEQaAY2ARuC4ERivgTuW5m14xGN0dATT8DFNg8FdCe57cikjzgCmD1bAZ/9mn+//bDWwDoiMEbMsGbOA/DESl1aaMwgAsj1L3tyx3xlCmOmE7JNPdpuwKvkuHykrZeQji+wrAMJAauy3bJP9QtOSYx/5LmVlySWfmXz5a2HIk6ZsPKOYnFgzt19ZSG65r0igSYiiqI0h3TrCIvQ1040MHdRk7wgg6U2nPQKa1J32Q6QF1AhoBDQCBxE4RANnCjUScnRwkf1CSLZvcWjr2IZlJfGxuP9XG/nJ/a/SVCUkx6dvZZI97B0jSVmSFaF2SikBL4JvBsAIge9gYOImQ/SnXHw/iWHmKY2c740ojZ1hhDCMIC4OnhHG9IX0gWWYYATBjIAXB98FMwbeIL4QOC+BYUbHCJpo5yx8pzNDAn0HfCFxBr7vqfoLoyGsgHAvCwwhfgEsRtk1NES6TVqU4xNjpDCXutowhdVhXDuCMVDHl66bzNTrbFyCBM1SZtXGCFdmuNyhmj0f5bqnuKrBUZScB0HX3zQCpwkCmtSdJgOhxdAIaAQ0AkdCIOv2fDQt3FBvkhQJhkfg3odaGBloxbZMfvvAKKvWvwQMAg4QA4RAjWnV8sNUFEYxzHJFqgwziOMVMJAM4BvugcMMM4Jp5YARwHGGFMHDFpWYgWnlYnoJXKF2wTKi7igBP0F6zJw6bOVj+Gk8Zwi8EXW8aMTcdBe4g5iBElWvn+5DTKlGoFhB4Kd7Fdnz3UGwwvjOIPLd8Ibx3QFEVt9PY/guHgYRyyYvp5iD7EuYrcn+njj0jo7BOjymXRQyKJ8Gvro4l4J5DiNeAVcUT2POdVHSsQBRO0RO7u+Pxnhtntbk/T4+est7j4Amde/9GGgJNAIaAY3AQQR88JRNNWMmzZgdD+5OJGHtm3tJ+32ww+Def3+VB/x1FA4F2LKpHRDyYkHAYVJDCa4bxjBsBkdN4k4I04xiBCrBKifp+bhmDNdPETbzcN2+DJnyUrjib2YXYHgJnEAp+GkiBIgbJiEzjGXFSHijTLPy6HYHacei1jAoNKMkDEMRsa1mmJjTj20GSXppJUeOn8QywySdfnxnCMuKMOwMQLqL0Bh5FMOrFajA9YUwmpiumI0NbG+UlBEiJCTS6cXxEgT8JEl3BNsdIOWNKkOtZ+UqbWI0EAArAmYY3x0hz0oSDA1h+HEcI4fdm8S0KyS3XuE2f26c7aGJfHR6PbffHsTLC1MYKGfKjDD2YUEayttQuf5pv7yDZ6f+9l4joEndez0Cun2NgEbgg42AcuXKaM8y2jj5njEGilN/T68QGo8nn11Py4ZW0o7Pv/7zWmC7Mj+CsA3JTuVSMUHWGbPpSKqUuBPAtCJ4ZgwnNIGQVYBh5SDaN9F0CaWJYRDzUrQZJkEvjhUowjSCxIwA5XY+vck9BM0w+VY+632POaZFu+8RN0OKVNUbAfYDxd4IjpBHz0X+TLcfzyog5A4RsPJIeQlSfpIuIVFSP4by5xOtX4XTy2iqi0F3GNJifnXAKsCyC8m1C+lPtYNdoLR6eKMZU66VMQHjjCgtYIac+QS8YZzkHmJ+gmHRDmJj20VKFjvcgGDsiznY6Sff6MQMeoroDroDDHVEIB2AhPjriRlXzNIl/M3fRYiEarhy+gQaLrKx/Qj5RWICPvTUzWhVNck7FBX9691EQJO6dxNt3ZZGQCOgERhDQGnj/N/XxqVSsG7dIK63h3Utg/zJ/3iBuVVdrF7RBwwpLVxtfQ6BYBAvlWZw2CVhRjCMXBJ+MUawGhMD1wjhmkEMK5+YXUBxuJ4ubxTfDGOIv5qXIO67VHtxmgMlrDds4hiEM/GmVJg5DPkpcgyLs7HYDTzhp6lT3nA+eVK/BF1g0JLu5XYjTJ8VYbczwCZlXk1nnPvEN8/MkE6cJLd7HkljmLVmHpMC+VT6Dj/3hCAluNHwCRomW5xRYm6c3fjsDlQy04qQZ8aoNWw2p7uYbAYImVF60r1UWz6v+wEaPIftvssOp4+gN0qXn6be9+hyBxhJtVOQ3sugl8A3wljhRqxAsdIYiseg76WxkW/9GN4guXaEaDSCaefT44bp2yBEU+yxSRou8DhzsIGP/2UJ0SlRmkoqKKuzDyF4WounL/P3CgFN6t4r5HW7GgGNwAcKgWyUasYX66CKxx+F/vgIAyNx7n/oFbra+/m3/9sBiD+caOHEF86kuCpEKBjAcX16hyVgQeIKSsEuASsPw85EpvpmFMewyLNi5IUmkLALqVY6OZPd7jApM4jnp/C8lNLADTtDiFedbecxgEXA8DC8NJYZYsR3qDAjlBkWg76EQRiKGPZ7KSoNkzbRrFlRSPUwyzIImLmsctJ8ISdMbSjIrlGPuRJskQoSJ0Io4jI3N8Cg5zNiQCwORQHYkvYZHDYYDBvsTyd51IXZoRBlvss6Z4Cl6WHmBHNZYOexJp1mkWWTYxrYvsEE28T0MnG43ULi8LnPdzgDkxQ2Q7i0Of2sTLWTm+7CxCMhpDO+UZmAsXKJ5SzEtfLIxaPfHSGd7gYVsOFRZEUoyikiHSyhx7EY7pG8KaPQ3aWI3ocvy6fxPJPpRc1ccEMhQTHZ5mcjczOneMYXT+fJ+0Bd8O9RZzWpe4+A181qBDQCHwwEhHyJWS4TQZkhc2Jg3djaTm9XF9ufHOE/fvkYPTk97Fkvhsw45RNyyAnnM5q06B8NqijSRCqJ67mZaFEhccEKTKsw4/dmF6vccK43TCxYTixYrfzXKg1bEbNhp59hL05xqJad7jA5Ej/q+/TgE0t3MeIllXkzxy6gyDQJG2GqDIttzjB7/AR4LnjD4I6Ab3G1ij8NUxew6aSEqwtzuKgaOgYsoq5FQY5FrNJkz24oT/qkC31cy6diooHkT+ncDPu7fZW8pKTCIJIDfUOQ9H16+32sEAx1GoTCBlbQ58V4gvsHOunxY9xRVEzfSIIVaZ/aQIACbBp8hyQGTYZF2Id9nk/QNhjCV0R0i28y3YJtrku+b/CsN8IaZ4AFqTY6EtvYM7qJYsOgNO98hu1C5pi57PTTJNMdbEnuBT8F6W4K7RgjsUbyjAC5IYNgKExrWxzaxZDtMHl2D7UVE/irT1VTWl/GjKp8wtUHz3OtwTuIhf72ziCgSd07g6uuVSOgEfgAI6B8qyQ/7jinq+FRiI8O88zz63nzzY387KGtdGzedEATp3zi6sTXLULfaAjXTYEfx3BG8Pw0dqAUL9wIdjGmLWSuUAUv+JImxBvFMAK44QZKQ1WU+tDiDlJo2CQNi1Evgen0YduFFBhByg3Y7HuUu4NsdxNcHCzHMoOs9x3a491gCTkbYJYxwEq/nK/nlTPPSNCT9skN5THJyqGgCMpyhTjBQJdo+sB1oDDqExMFY61PfzeUlBmkeiGnHiwbUnshNZLJQhcogGAR0AZ2tYHb4zO6HWITDcw8n+FdIOL02gY79zjszRlmbSjKhxybgGewrd8nFIQtuHgmDEhfTZ+5EYP84UzISMSAhJchiqGQgT0Er+CTRhIom9yZ6uASP8Wz8c2Q2E5VuosFoVo2x+ZQEq5nkR2mzkzwm5EeHkn30DDaQqfXj4OvgktC0RqC4RhJI0BnbxyGE6J6BUJ85Uaf8nm53LpgKoVn5lIYCWbS/41dG9oH7wN8k3iHuq5J3TsErK5WI6AR+GAhINo3mZdUzKvj0490b3XY3LGT+x9v48d3P81om5hWXcLFxdSU2YwOxhl0QvhmDom0FLaUidAUU6ERJB5pykRw0C9xywAAIABJREFUOiPYEgkaqsYKTwYzgC8+YukuHN/FCk/EkiAE38U0IwTtXOrNCJ1enE5niCIzRMKwyDWjpJ0+JgAVVh5r3EFsN8EO36PKd/lIQZLBeJBFOWXMyLNw3SBhJ0QoarCvF+prPFIJX2UFKSoAmTlswz6fYjGB5hqMeFDuQyyqukKwCob7IRCCSDGkBsEoMbCLIL3KZ8CDqHzKwDGhfT2UnQnBuIplIJ2ClLTTDqlhk7W9LqsHfRaXG1TVGYxK4GwbrIv7bBmGwmKYWG6wboevSGSJD9tdWG+4zLVN8lMGrfg0+5AfhL2+yGDzqpdiyEvwWqoN4tswR1bi2QV8Pv9Czok2kLAKVX2tqRH+ZbQLP7kH0pvBTStvOzdYjh0swg6WUhAKkfTbaNtnQ8cotexjwuwyvnpDLROuLmNGbSnhqoPXhyJ3viFZWPSiEXhbCGhS97bg04U1AhqBDzoCGW3LoUSuvz9O7/IhHly2jFWPtvPg6pczMJmlFFcZBAMeAyOQdGwMSbdhRrAkmlPMm6l9pOTpnnsOxGYy3S4kZeYw20+yKr6VHaMtRCRPnBnDt/JJBMtoNMMq8GGLO6TmgZBgiLQzQHmgFN8w6TZCFEgOOHeYLmdQpQ8J+g6pYCmTfJ/rYxbnhyI4Rh7z88Pk2hCrNxjtgOICn7Y+j65uqCkCO2nQnYRwEeQMQyBf5oTNZFGJVYM5pCy0pPsgmc5YbosWGXj9QniNTJDqsI9fYODu8Ak0GbjdPvRl8hIbORJKa+Du9jGiBk7Kxw4YJId8hnb5hEcN2myfRzpdLr/MZoGoHTt8+pMwMgi9+322DcKKYahwYWEMtqdggg+Dks84B7pM2O/61KcNukSbmvYUTr8T7ajwat/h16n9kNoDg8vVzBaXxc7gQ9XzqCkppLfX5M1UmpGhBN/22vAT28HpVTHIoUizigKuTndihKJsC5qQsiEehl0DKnXKl2/wqTq/klsvn0zBhFwK8rM+ltn5anUE7Qf9vnKy/dek7mSR0+U0AhqBDzQCnneon5zkj1vf0kXnrp383++8xvrnu+gWu2JklIl1lQwOh4mnRkk7Lq6RjyEkTplnJaXIMIOSaFeW6Cxm5J5BXriBCiNIMWmW+D4lRoCQn+ZF8fEaeQOSbQSj0zEijeR4KXx3iEE/SSBQntHipXtIYynfOlt8wuwCEuk+FhgmBVaUXKuUmwpCKknvhFSY2c0GfYMegRGPghKDbUkfqwdKCw1CFthNEM0x6NvpS9o3cisM/H0+ySQ4Ym7dA5EGUPEacaXAYrRHtGtgGxCdDcGIgZfKRPw6W3xEMRkUpaPke8szMjN/SV27/cw8rnkGEkiS2u7jRwxCosmK+Nimwb4Wj+fboabW4JxaCEk8ScggnQYn7rOzBQbFBDvgi7KM/ChMyzcUGbUdSAZ8tu/36fbASkvFBgM+VPsQNQ2WerDESTPodLI1voX9I2+CMcTVBRdwxYSFTAsHqfdN3tjv8kBqlG3pbra4w9R4cVrdYS628hgNlDKDbggEuMvNpZJ+vBB07AvC/r1Ulncw/cx6vvzpaubUVlM9NZKJjVHZ83w1ha5M6aYXjcDxIqBJ3fEi9Q4cJ2/4480070ATukqNgEbgFCKQuWalwsyD1hXtT2eCLa/u4qdLl/H9f5fUF/IpheoeKgP9JEfDDKTFMCrsIQ/TLsGXlCKSwNcdwUjvJ4lFLDqNGbkLWR6q4RozRg0GO7wkQ4apAiWGvTQx32GHn6JBtHtOP6/0P0fQjeOKidaMYab2YYXrMYIl2Ol+LMkd58VJuYNUhuq5KlhJU7CYSWaQGhtqCn16HJ/4iEfdFBhsMxgYhYpCGPZBzKtt+3yivkHJVEh1QbgMcqoNEl0+7hBEK2EoAU4XFEwEs87A2eRjiNJRyNMOsKpU2jmi1YbyfTNUKj4fP2zIjGH43T5mroFVlNHyiebNjxoSOYGZZ+CIxTroY0YNjCgk/n/23gRKsqw+7/zd+96LF/uaGRm5L7Wv3V29QwNa2JdBCMtgYCTNjDSWjmXJHmmkwZLOyB6NrSNbRliWbEnYGgTNJgQCGjcImu6Ght6X6qrqpfYlK6sq98iMPd57d87/vixobODIx2pEd0fWycqqzFjvzXjxve/7f9931uAIBo7gnsOGTAA3vUuRyCsrxUaXDH0HUrOweQXm27DwoCGfg4khcDc1fkKz0YJLAviyASvtNkFbOmYNp6OI+3rzLDbPs6IUj0QtsmGToPU0jdYipK/jmkqFt4z5bC74nAsVp8MWTwcNpr0yl2yMyiL4c/xcZpjZ3D4eaPdwwnWWRf52FVdY4qSUXVxZho0mr8Pnul9xeeub9nLg4BiZ0rdiUmI2eMDe/S2+lF+0NzUAdX8HW2ujCJ5z8jUAd38HmzC4y8EK/HesQPym+i2JVcyip09f4sv3H+c3fvN+VhZkML7CzG4ZJItYWTtB0N2kpwoY5VoAp3QW5YgRwqDDug38XY/qkNzJT+Rv5ZbUDq7gE6iIB43htUpCfkNb7DWhXT7f3+DLhPyokkiQNgUT8FjjCVbbTzOEoakz9IzBKEiJ61WCc4UNTAkVlmPczTCrU+zRmuEQ9hdCyhMZwouwM+uQ6ik6EYQiPwaGSlmx0TK4vmJJ0jsaMD4Vg6TKdZAfgVBMs3NikFDQBEfaxcS/IZSX3FDC0F+J83xVDtJeDOL6fYOuKRwd18E6MgrYAlXbwsuSGiIkYh9MA4wLZnMLSm/lK0cXreEX3YWvfT1k4ZDinXs15kkjecGW6TPrhsQeBRsaEyrqpw1OR3H+Uot71hepNFushBF/2r7IUmeeSKVJRE3O26IJAeEeeMO2ySIK1nFMhE9INlzjcuMUtPIwNArJKbvWnlL0uxchWLEMKGE9do+QZL/n0UqMsiu9lzmV5N6oxdFok0OpabrJiGNyvaNtSCreuTPBu355GztvqbJrthjntQjctLk4A3D33/HSfclddADqvs9bfhXA9XsBp566yOS2ETK5pH2xDli77/NmDO5usALfYwXs+6dtPxAwEZ+Fra51uPPuVY7cdy9/+L4nbSEXo/tI5AsMUWdp/TxR5yzGKcfuVCXhv1nQCYi6mLCBjpr0ZXbOyXNN8dVsS+/jR700BWM4Hil5Tydp2pyP+rSjHov9RZLBMl/sr9PxJ2nJY5F8OJ1kzi1b2fXMxqMQLZFNjOGn97Oi07zFLZJSmnuUS4UWEmRyQSUh7IDqgdRVONLSEPL3U1CLWpT8DPt9n7kpqCah0ldoY7i4AcNV2zxGS0DVBgQdKO4Ct6ykIIJgw7BxEkq3xQBPmsoQrJsUYGYIxfiwCv5BRV9u4xEJ2gMzDEkPjNTS+golBgKZ0bPGE2GyRHIFZNZOElWeMKjtCiXATfKN5bMFD543LBl43T7w5L69WFJtRdBstmkGHT517wLzT89z/1KdB+RKkrEnzRESCSO6sT/FSNSmGPVZddI0lUNfZymGDaKoQ185dEUm711GSXdt9yipMKTrz0H+NnpemWrUJ2FCG4ey2V3A0S5t5dLpLUL7mTgaxsnZ/ZP1P+DmuRwsSvId5aEbWA/SbLZ6cF6oyAb/6jdGeedr9lK8tkJRJOotaVbQu/3fcwgC+8PBXy/pFRiAuu/j9l8FdGE/5O7PPsbHf/E+3vjPD/HGd92Kn5aC6oEc+33cjsFdDVbgO67Ad8qVW764yQNHnuH33n8P93xB3n7nKOydoOhqlusLRN0Fuha1LINl5AQsObZ4XrkVK7ea/iUrt0o/6f7cLRzKv5yhxAjDJuK8iahEHXLhJk+ZPkfCTR7rXo4bGeSt29oiDYmoh+8W0U4GkTDXRf/UCX4mk+XRZpPHgxWmczeyzclzG4b9GDYdw6gf8V82AqYzPjmtub8VsKYUKRTvThuONw3NMLChxVWtaVfg2GbIW2cjbh5KUFLgB7ApzFkSEn14tgFjHlRrYIYUQV+wpsGviWy6BcIKEJ0wdl4u6BgcyZ2rgJghzLpNbLGgzzZyifIp15vdUrBltk4+RKoVkCdfE+KcBam9FSbPftqSDSNmYA5/QREeUFw7rTh7ts9FznHXUwEfePgs+Xabp9eHQYCRX+RtiSp9o7gjqJN0M9ym03gCwpWLb3p82gRsB/rK5ZwJRFBnXPmclHDisE6ViKH+Mo+v34XbPYEuvg6V3EHCLWKcLAeNYT5scEHq1/qLNihZYmXCcBMnWKEdtVFuHumyELOMGzYIexfwtCHlD1PJT7DQXqV9epQR6hx8+zS/9L+PcvOBGkOjya2lGczdxb8kg7+vrsAA1F1dief561XAFgYRd//Vo3zkl76Grhn6pw1v/d0bedN7bsVPDYDd87wNg5sfrMB3XYGrzNxVxrzVh9UjDb58xyP85Tce4Y4vNqF4gNrkDK6Tol5/mlbnss3dUMlZCNaIuudRktWh0/YNW+bebHCt8jGmjRtu0Cu8mgOZ/bxMJbjcX6TdOUNP+9wjzQvtk7Fkmt6BR4J+9zQlAXH+hI0JaQRrXOqcthlr4mb9R16J64uj7PGz3L4Bn2we49lggx9J7+G1bpaSMZSVsr0UJaUo542VWB9cgYksPBqE6Ehxgx+H9rb6hsCBdCDNsoYL3YBJ12VzKOJ1nsPYuOTQKXsZkXmpQ+K6mCqy7lZh5qqg6iqWYUMIRPLcpiwIVZ1YHnYOKsgpzAkT16vWbEsY0YKAPUOUVLjVeGZOZFuzHNfB2irWDJizoLbFNbfmnMGpOkQpOHtfj88e3+Bc7mGePBfylQcMjE0wnhvmYiLLTybSjMqauJqNPkwozarp8oyUnRmPUEX0ESOuokQUy63KxSWkE3W5z4Qklcdp7bEvbLBHuXykdwXdfAJVv5vQG2J35e2seDWWwk208sg6Gfyoa+vKbA2IWyDRu0wi3KDTPiFWFow3BF4V7ZWJuvMY+V3qXSTJCvlskUsyIHheWMUeb3zjMO950wFe+7rtVKZSW9LswDX7XV/YL7EfDEDd92HDvwXoQu7+q8f56C9/FVWRDCVt2bnO8ZAf/7c38wZh7FLegLH7PuzJ4C4GK3B1Bb7FzMXgpNuDw0dX+OhfHOf3f0equhwY3seOuX0stz3am0fpt5/BBCvoxDTanyLqLxC1jtqeVaSHVaeIwnUruUoosOTSudLJmtpLJzGO6V+xIFAua29fOezzqiSV4emoS6g9qmGTCRPxrJNmvb9GJKyc8qhiGEnNccQb4dd0iluG20RtxZObPVyT5OHuGT6rXLYlavyvjs8EcCKCtybjsOCugacjwzZPETYNJ5oQFmGvOFPXDJd7cCWAigdFFIE24t/EdQwnNgN+7JoEwzPgbUIiGzdEKKHyhuL2LKsJCqsm6qcP4glRKYjOiNJpC7NwrhG2DFjfmqFLivs13hEjwM4DVY6lVSSGRLamGd++zOsZH9wdEmWi6T4KF8tNHr84z69/9AzPTm6AZDrPbuNtwwcZDh32KUWBiFRkKCcUY0k40xDmE4xW3B8Ykkn4Rj9iWPL2lEYykVuScUfIiDE8qjRuuMm1src6QcWEnIq6PKhcduFwpf0UR1Y+w4jp0yy9ATe9h1TUp2O6SHxMSftMekM0jeJMuMElE1EMlul3z9PvL2GiLk72OllpuxCGECN1ZVEbr38BTZ1qvs0peW5kuK26nV/9x/sYfdMUB/blbAiz/ERc2UpfHRiI13Tw90tnBQag7nne628Buoh7PvsYt//CV3FG44FrExiUIzZ+Q/dYyI+9/ybe+O6XDYDd87wng5sfrICswH/NzAmYe+LJC3zwIw/wH963ENNEs+OMFrYTBhnWWucx7WeI+itEbgkvMYFOjBJ1zhA27rfSm/SvSiitQBcxRgi28YM1dKKGSe6i7aSYMQF1p4hxi2TdHEr5tE2Pm8TooOBLvSXb5rBL+aQdn6+358kR8g53iJlijZzMY4nbVEdsBIajYY/7tcub6TCmkxTxOdI+xX/qL/GG3DXc4rqkXcMY2jYtjKQhzIAXQl8iP0ZgbcEwmVG26SFVg24JVhZhtAVdB86sQ6puyFTgnm5c8fXmGzU37FS0FwzutMKR6llBQjKyV4jZNQvU6rHRQebwBKA1O4ZkWuHOCqD6ltRq8+lkxk7YPsF60ny2DEqiSnKK6KxBlUCviLSrrWL7+PENPv35E3z10Qs8GGj27NnOfneY3ZkcX+80GTEp3p7z6XUiZpWi50AnhPEENvpEUuPqCrIelCJ4qG04vRVp8qyR1olIiEdbbbZXaXqR5m4TcjOaTTR902EDzUPKZUIpTnTOcWLti9A5STV/G/uLr2YFh0AME8E6E14R362QR3PK9DkbBVyKZM6yg+6cge68lX+1P4tKVFFOGiWDg+JwDtbpB8tU9CraPMsVefCrUiu3n1/7P1/Ou965n537MyStbD1g7uLfopfe3wNQ9zzu+bcBus88xkf+6VdRQ5LRJE4sOdzHH3JWJUFN3RMhP/57t/D6d90ykGKvLs7g62AFnocVeG7GnAVzR5b4yMe+xr/7N09Z5DGzdxsdZz+N1grd9mlCYdaiOOvNT+4km5xkQxg8aXPY+JqVTMUQIQ0PwqwoAV6mh+pdpJ/eB5nr8HWGlJvlkE5yDodLKLYpTUPqpkxAQSdphi2OSHVCIOFuBdIY/mEqxS6nzELoM+s5XOOGgo3oRbAaxieIox5c6CmqTkTKV0Qm4qubp/g/OikmE0VGVMSszjLjKl6TjSiFGteHybSS8TJWNyXWAy5eNBTLirFXKCuD1g+DNwrLl8XzYMg3FEeWDYk83N803DZpuF67JG+ICSYlo4TFGA8bISGFiRNzg2ia3ZiNi8Q3Irl0pdhIYcQoKiYI+bRzcgqzEs/h0TeIHhoZcARdtTTtC/Bko85Hv3KR9z9yAsIUFKf5eX+KfOTb1od9PtzR6FDvr3DIH2LMS7DLGFtjVssqsp7BSytyHpwXBjAFmQ042TMshZDuwDkNC8YwjmJOK/IYTkXwuSjiVV4cojwcaY5jWI0i6ipiU7lc7C2ysPJpLnZOsy17A+8dfjMndY7f6a2T7F2i62T5Fa9E0k1zVodc6vXpYHgCydhbxbRPEvUu4Lg5EsndKL9Gx0S4YZuc6bEqho3eZXR4kSl1CqPOcO6MUJnT/ONffTnvecdNXLO/sMXcGeuHGWTdPQ8HkR/QmxyAuudpY4QFkPD0MIy497OPc/sv3YseFiPEtwM6K0bIhcWtpQ2944a3ve9m3vgPBjN2z9PWDG72JbwC8YmWLICybQdPHNngzz/yMH/0e1+380qlmW0kk0VW26EFRlHvEqa/iJPajU7OYXDxiMj4k9TDOuHaF6z71LonBbmYrnWUqmDJthL086/gJ7LXMe2UOEfIQ8bQjVo2Z66tHC5FfaoKPBOxIUxP6wSjtNmZHuUnnDxziQI1V3M6NHygGfJ6o6gkFPcEEdMGbi467M0aNjfgkSZkXKglYPuQolPS/Nrh85zsaN6UGqamXTYNtPMR45uKI33FmG94eUqRz8Wu27YYJtqK7a9UDJehsWRwUuCsxs0R60uwnIPUaVi9DLevtSnmFL/wzhTZTmyiUMNxnIiVS8VdIKBJzAwVQABcDtF1ETBHBps7Z23EIr9KSrGc8EpGneCmjhwzxXOiCRfh0UstPnjvs/zRVy9At8i1Y9uoJkoUlM8hE7FDKS4LT2qgqDVK9fhoJ+SfDaUZ9Q0pD5oBXFgzTFcgGSl6fdiUbQsNyYKy4udmHVIlOLJmuNiVvGODiCqSWTerYXVLaS4E0NEwquBwGPA1yfMTY0X/El9bvdN2yb6neIDrCm+hpXN8PWrzUG+ZVRPx5kSJYc+lEPpcjsTUEbJsAk6bCC/coNF4wp40pFI72ZbeQ0dpLpmQvTpF34Rs9lc4RYjTPcuQOsxa4xy9K8LcXcPP//LL+Ol33cANB4rWQCIuk/g9KR4xeAkfAl70T935rd/6rd960T/Lv4MnKAeiKIy4Z4uh+28BncA5eaEZIhMShn371RmCox+aJzGpmN09iuu5gxm7v4P9G9zli2sF5HUmc1laSe2U4vDhBv/6Dx/lZ3/ydh6+/xgzu4ZJVOZoMUYzTNqZt0iyxtC4uRtx/DkcbwjtDcukEx3tYxqPEUlEhZXIulvDYw4qWCRw8mwbfifJ9AEmlY9PxJKCy6bLTp1mHwYxia4pj0g5JIIVjmw8zp6+4RcLO3hbcYw9rrQ+GB5tGpwI3p5V1GqKb3QNe3zFcEMxNaQojyvabZjKQLAJo6OKbtawvqgY7yR5n8iNboZrlGKuDy/fr5lEcWINZjzFWBrm6/DMOpzYhI4ToE/ASl5RyoO3Au6obSEjXYJKSuG0YlJtxnf53FKfRtthLq/Q6+A4SqL6xPxro07MvIkjYRJx5pwAOTtvJ7N2VYWRyBKB2TKf1zJbvpK4ecIRKbelOXuhwR985BTv+cSjPPykZuf4dn5qeDc7nQI/5DocigyrKNJaRvsUVRRXFJQ8l04Y4nohN9Zcy171NJRTitaSwk1DPgsZIVZrCiVByCH4KVh1DGdThmuM5nwi4tNN2J5QzEidaw8q2pKIrEosnoLPm8iOFYqcm3cLTCqXM70LHG5dohGsc4s/xbVOhhQOT0SbHO+t8URnhSuEDOGQ1z4V5XJ0y9gynRjjoD/Gxe4ZVOsY6yZkT2KEmpu3XbOzbg7PLeE4JVrODoaTB1EFn+HSae76L1/hA39ynESQo1AuMFQRBlMRyUCAJRwG4O7FdYT71rMZgLpvrcXf+r+EpTt830kOf+QcqRmPSKQERBoJ8d00k4U99KMe9e4iESH9sIcRmn1Ec+ybwG4c13MGwO5vfXcGN/hSWAFhJ+RdTByt8ufy4RYffP9D/NhP/SUP3ncPpfESxYnraXMjYZSj2b2IaR3FmNA2M3iF23C8UZRXohNuyqsXx81biSzY/GqcfivOSSeLVj5OuEw/e4j9xTdRTVRpKs3FqMVp0+OATlJQmm3KsZJdSTLQog5Pd85wprfKu3PjvHtkHzd7GVzPoDsRXlZRSUG6DbmMYk3BvrRiD5qdZcX4BHQksi5UOEYxe52imBLlV5HvwvZRl1eFDrc3ltmj03RcTXlNjBaKMVcxppSVJGfziv01xXABHlgJCdHc9wycPw3jEiIs/a4bkAqhJx2ttbgbNt+D61Meq0rxxOOG3XmNl4FQQF1qy/yQkuC9rbgSue50LNFKJrM0QqitKBOVe87lQmyLRrcJH/3S0/zMb3yGT50/DuUZRsf38Xa3xk1oxhXMaezzWRenbRsmE5DxYWdG4YdwQ9bhXhMyUnCYSCiWe7EwUpmC9BR0TsWMoZ8DT7C5GHILknyiyEmMSwMOJTXTnuKstGxo2CGtaGHsx9gnKrMBLzLUXEXThKzYXt0RzqCpROsc7y/w+WCTv5feRs3xSSmPHV6ZaSfNI71LPBOssql9njLS3tFnw3QsuZmPOsw7BTbDJt2OxLI8a5HyXYSsa88+xoJOUHDLdJ0cJsoSsZ3syE6cxGXuvPOv+I9/fJpKx2GqWiFf8+1rQU5y5L1IyIfBx4trBQag7nnaT3nNOI5memcNZzzi2Acv4tUcTBSScFPMFA9QzUyT9nJsdJdZ71yxck8YhfaF5lQ1x25fwJ9QzO4ZGwC752mfBjf74lyBbwNzSrF8ucVXbj/KL7znE/zpV77O6M4smfI1tBI30zJlwrBBr/0skiUnblYncwCVGEecqwIIZUZtVL5GHdrds0T1eyCS8Nq8nZ/z3Dy9YJUwe4jR4qtx3SKPE7BfOaSUw40qwRWlmFIeF5ThvqCFE3bom6btYf0Rt8ZPlUaZqhlSHcOOWcl9U5xZNswUFRttuKcDhXqsTOqUouxDWqTRLlZ+vRIYhicU5lKsYGam7UQHetklanZ5wGlyYyZjWx8SgaLgGxZEGo2gJdJtSjGSUYx1HXbWFPsrsNiB+S4szcPKKrQ8RVZAz2ac25ssK4rbFJMKHjpl6FYM02lN+5TBqyiU5MlJEoeMl8hs3aJBwJsqxwydamz9/gm4ENNEEfSSmHw1T51c4z/+yf38k2/cxWq5yPaha3hl+iD7VJanTMikoygnNSUDIyVFqaLIj8FUQSg6xVwNSg5MTyk81+FrF3rsLjmMXq+lihW1AIkx8GYVnbPxWp1cNpRSirWOob4IyXWF4ys0hl3DylarFRKKs2L6MNjAYDH+ysjgjrxm3NGWVY2MZk25vN4tcLi/QlYlaPcWMMphrz9Gx/g8pQxJXE7Z+cs+G71lelGPGgE3GMMunWSIiANOmn1uhUxilMNKc6H9NOneAk3lc970CXBsfMrlqMkNiSoj3jBHVI1cci+VygzZ4nH+4o47+cYnnqJWzVOuFcnkJI9v0FDxYjz6DUDd87SrV18wiaTHtn3jmGrAkf/vPJnJDHPFaykna5adk9DLgl9hrX2F9c7leOZOisKV5DQ5HP3QRRJTA2D3PG3T4GZfhCsQz80JC6FYXQ/4zOee5Ld+/dP8iz/4AvOVAiNj29gM9tBy52zoa9g8TL91zAbGSuSIm96JcYeYdIsE2iUSA0SwRloAXftZOo0HbJuAcsrW1eo4Rbqdk+wovIpK/jY2tceM0rxdeVSIeMyEFHWClpIarhafbS5RdS7x2mSBZJTiVpXi3ekM27aFNJbgeB366+B3JWtXkTwA1YrizAnYW4SDuxWlIazkmkqDl4dsBrJ9hSvS4WRsPHC7MbvmlOD66QyPL21y90bEmwopRnNwpg3ZNJQnFJm04vICrEsV2E6o92B9HRLifO3G7lBxd85fMGxIRdoVKaZQ1s3qpBXZGY20cR25GFHepajuVdb5K3KmsHRKnB2roA5qVDEOJ7b9r+J0LQlzJ9Ir6DVNG8XX7jrHzR+6g3sPX2Jq9AaqmevJu1VmlLKM50GlbE5xQsPeipIiCspJmBlTJASsSO2qDzLOshKTLXHNAAAgAElEQVRB6rhipWt4el1xXUlb2bWTgPqFuOwj4cYFG+m8stEm1aKy83W5KhakiVSdFtA7oYg2oVGCUz14TVYYxZiFrBZhtKSoNTVJFEVjOK2TtNwK891zlFA80XgaJz3EXG6EXZ2AW7Kad3g+b09VmHOK7E7mWdRFXuYVeb2X4fV+iZe5OQKnxEVviP2p7WST29jor9DonGCv8rg1MWKz7q5IaLLyeEg67II1+jrDGkU2zTSTY3t5Kljiox/+FEcfOoVOOYxPjJJOud8Ed1fzGV+Eh4SX1FMagLrncbvlRSJvMF7CtfNxDBnCI1WGK2MYyRKS0EnJTnKzFPxh6t0le7amxUwRRfZNyTJ2H76IP6kHjN3zuFeDm37hr4DN57KSkmJjHe796nn+0a99mn/7L+/g5Jku2/deR5/tNKMyxqti+isEmw8gAcFOZj/J7CGy/hR9naATthh38yR0Gtf02AxWWdu4j37vvGVtJF9OexW0StDrnuCG4mvYn38ZjvLYg8P1WpERhnCrkSAyXR7urPFk/QI/PRTwq8Oz5PwkBIaf25ekmpaqLEW6pUhn4HIDZicUuazteqeqIVVT7LhekxgCpwa5gpKMY5wpGRYEIxVc2xTJaxWeSJ5tSFUUhTFFrqqYbTh8cm2dmX6GalZhHEVNKr4CAYSQScDIXkikILEKga/YdOGWsoo7XkPFnqqimFQsR9BPKR570rDciViLAsqbmtSmptGCmiA8maGTuTqJXzlriGoKd5uygcU0t8xhZcmkE6dthFN2WNxs8t5PHOUXP3gPw+XtTE/ciOPN8HK3TEF5fDhscchJcqNlQEH34mgWP6Ool6GYj79ncVZFEWyZR/qeIZ9y+Fy9z3ZlKHsOT5+GZAdyHUiOxIYNCV2WD2Hw/Cz0CnEEynRe4Yli2YWhMlQcxWRBUZeZvzZsuNg5w2IEK92YKW2Eik9heKuTsrE2z5guZZXlMWeB926f45ZUhtM9w7BRtELY7icIU5qPdUIc0+dc1GG1v0GgPR5C8QpleJkQnyrJhl8jcMss9Zfwume5OVHljX6NgnK4yXV5tVPghPK41rQp6wTHwwxFLSc0EzzyyBJ/+Zdf5u4nL7N7qMroUAlXfl+MuIzjUYUX/tHgpfsMBqDued/7eG4h7CpWj/m0T/uSH2pfQHLXAuwiIivDllNjbHSWLbjTSqRaYew0blVz9MPzJKc0M7sHUuzzvmWDO3hBrcBVqdXGNig499Qiv/3bd/KL//RTnDvRYnTnDaSG9rEYTGC8YZROE/WvELVPSrQuOr0PJ70f3yuRkpk5XKpunhtUgtX+MvOdk4TNx3BNB5tDZ+udKmi3hO4cJyy+mnThNlomYFJrPOVwRm5ZKZ4xfU4GlznaXOHltYDfu3aan3TH6LQSfOpEn5+d8pkc05bNatehm4IrizA7q+wY2uYa5Fzwh2EkC8lJRdRSaLFfSmWWFMV2IJyHrpizJKJkOp4R0/KNHIQrhrBnGPITqGXFn/a6RE2fEc+QCxRPzhvLCNa2x+kgIl1KpEeuFoe4VQrQbkFBC9MIyTyk+oqSDxUXUmnF2ccM/VXF5yM4d9wwVo4ZQDWjbFxJJLFNEeiMwnSNNVAIvWZkHjAN2nd48Bt1fv1Dn+dDf32c6d3byfm7KLrjdh98yYVTDq920wwZZbtdhxOwKweJANrSUStVth3wR8HNWuMprQak01DIx8C0sGnsXGC4oRnZFbN53TWQwgZXMvnWoNuA3nLMcoo8PDQHmQx0AgHW0BEGdCkOZ25vQjsAreGxEBY6cVXtWSWGEnizzEBqj+FMli90+ow6irVGm3azw67iBI+tw3pX4Rs41jPUWxG3YChhOBN2+ajp2wiVncqxBp3zKC5GbZIi43rDXHFLPC0xKu0T5MI1Jt0yh9wMOxIOL3c91lSOUbdA1cniOzl6vQyV4VFWC0kWHn6SP/vw/UT1DbbtrlKsZAfzdi+oI993frADUPed1+Vv57tbsSbtesgDH77C8Y83cKR3MPr2m/8WsMtSSFbZ6K48h7GTMyeNM6w5+tF5khOa2QGw+/YFHPzvJbsCcd5cLLWePL3GX3zyLl79hj/goUeXmdm/G5OdoaHn6HuTaDdnXa1h5wT0l20lk0rOIfJpzs2zoiRQ1mFEBtlNwKn2Kc73r9DrnkEHdXRqD0ZCZE0fV1yw3bP0czfzlsIPM27gWu0yhqKH4aIJ2Gb6nA42Ue1N/vXkOL95bZWdGZ9LFyI+s9rjfzqY4NBezdIFw/I56HWg18AyQCkD3QSUR8FrQ6Iq4EchngJhkCT6QwkbJkrmUwa3pkhNKtwERJeI3adTAvpAt5RlyzrrhvS6w3pH8ZCC10040BamDXbOKfwtx6rMtW1cgbQL5Sq0l+Q+lX0sC6eNBZbCiIWtmK0by2l2DDs0Q1irw91+yENn29yaTdiIFQFurrCLM3H+nA3Za4CRkcQSBE3N57+8wrv/37t4eKNFbe46OomDDDs5DimHW13NLSkYDWBU1Fot88oQRFDJKqYmlXWwBr0YjIn66Gy5WLMVbG6dDmChaUjXHYoZzbmeodxTdDYFKENSS/wUeBraGXDEoSvMXDlmApXM6XXA7UN/DeuiXd+AZFJR9BXHm/C7XUMyo3hNUvHKtOKaEYUjl1cSpZLmid46J0PFpBvywPqzbIuG+emqzMBF6D7sTcDNOYddrsOY7zGT8KmGWV6tPXJK835jbDh1UbnUifhs2CCpXCa9InU3z4PtZ0j1L5FzK+z00uSjiFf4UDMO6/jsSBQ46JVY1FUOOaOcLs0wMhJw553f4P1/cA+jE0mKpRqVUjxvZ5nvgZPiBXdsHYC652vLYnMRVwHdsx/fxJGJ2ucCuq3LXH0IEnGS8fIIY1fvLP23jN1QzNj50wPG7uqaDb6+NFdA2DmlYqmo24t4+JFnufbg+7jjc18hPXMThdEfZaE7RNIWrJfods9juucI2yfFq0rfH8P4k/hehYxXYdnJ8HrtcrNK8NVgg0bYYKVzFhVcwVEO2p+xgC7qX8R1h+j1LjKTu5nrCq/AIyBvAgsKNS4ppTgdrPKX7TO8Op3j39VGubWaQQeG8yci/nAj5I17Fa/Y4VJ3DNk8ODJHl4ViDTaXYzflzBwk9iq8WXAFnBUhUVMYaQsTN6nMo4nT9UoMmNQEttUhkqiQbBwpYi8j/w8USmq8Lmuuvc7j2VbE9EWYzGqUFkkvnvYXwCgmiFSFuKu1B611SE+CHhKgp6zMmxqLHabJABpd8NMiB0OuAbdUFI82NNc4DplL2LBgd58mEqlYwJF8SpHOoiLUmv/81Cne/S8fINy3ncnyXip61KKqpDHchMOYE4OuVg9Evd2bV9QkmkYUw4ZiLQG1UWxOsRfEIG5x03CiDUNZMTqAm5PZN0U+Aat1+DKGPRkYSirEUCLNGt3LIM8rv02R2664cjJm7XJjYNbiWl6Vhuxe6NWhLdJuApZacKYFP7VXc1tR4a0aVvrQXI9Zu9Ek5HqajQgeDjqktUdDGR4PLvH3h2bIuAmqCZgdUWRd7O+JlzK8vKDY1jMkfMXJfmSB3WuMzOtpxhIu73AchozHBXrkcJh2S3ytd4m7N46yL9OjlqlRczz8fIjbNwiwTRrFzV6Cca/Cgi7RN9sZqoySSG7y8ds/zx+8/xiv++EJqrUyniBcG7s1cMm+kI6yA1D3fO2WZGj2DQ9/bImjH9zAG5ZvPPfODMozhGFos4NkBsTO4GFIe1mKyZEtxm6FWIqVGbuYsTv2sXmSk85Wjt0g7uS5qzr494t/BWTuR9S8IFA8/nid333fB/mHP/MhUtklxmcOkEj8OH1dYVin6ekEgbhSe+cJuxftsEOYnGMiOUVOJ1hBMeykKeoMG71F1sJNlnUSL1hGd8/bNzUJHVZOirD1lH0tEixTSk6TK72WbcrnmahNQ3v0tMt81OITm6dY6LW553UzvKM8xOwOB12JuHwFHjgNvQnFT0w4eE3wOnGkyOYZQ+2VyjJGy0tQ2ynSoMIpC5ul0EMKJcFrEp0nfaoatIT4SgSHdKW2wZlVNmNNQJ7EhYgRQYKAbTaGVBJqResS1GYV1VrI3Q+FDGdcywS6eUW6gJ3tEkCXGFJocR804sw2v6JwZOYvAckJaF2AhESDzEFWasUugJF6KlfYKZFvHSZugloe2mLAuGK4fKchK7NrYi6oKnt8/PM7nuBnf/8E23eNMOzvYFRXeJeT4N2+5rWeQ8bAfAjDAQwJWyeBcAlwp6B/GbKz8WFVgLpk1F2+DOsd6AeKalohjKeThYQ4dlcgykBlWOFGIa08bJ9yCEfE3Rq7iL0S9OVyPjSXoTy81cS6ETdapHYomuegvgRLCalSM2RLEuIMswlFZ8XYgOduH+q5ODNP3L9RT3F9xuHLzUUueOOMuTmW21cYUZqXTY7ha2ONGAlP8quVdeAmEtBuwJgnUTKK0UgzllFMCZvYNxyVeltckspnQfv8iJvnvFex2/2l9ce4GG5STlXY6KQYM7DTh4mUZpdRdHoRQ47Pmk7xhDNCM7GD7WPbSeg67//3f0a9sUp5eCcj1ZRlRa+aj178R5cX/jMcgLrncQ8FpNWv9Lj8hJTAYM+I5e6i0JAoGco7NIHqs7nctn56ufxVYJfxCpRTNdbbV9jsLaOVa80TdsZOGLuPzZOadJjZJQHFA2D3PG7j4KZ/QFbAzs6p2Bne3VB88GP38JY3fYSHHvg6hWoWvBJr6odIOAVSboW0m2O1fYZe+ziEq7anNZU5QOiJ3Fqmrz12KYdksMFquMEFnbJBsH7rKQJ5Z5deVidl+zcjiaPoX8E1XXpKc93Q29juFEiavo2uGNE+S/06D7cf59dHCvzhq3ZyQzmNW4+QuizlK549angwCPmpSYeRVFy0kNql6F6E5LSy0qvnQnUH+I7CEUAmZoPjxnafIo0L0sogcRw6/uzNx3Nw3kTsROVKnLEmwb9SrdWvx6BGeQqJm2sfg9S1MLHP5eyCprskgcxwJqmohNBdjcGNAEdX5vV6sCoMj3S5rtrmM/oboLvQ1+CPi+NW0TgX581VdmJ/nkwa5pdh6qDMBCv8skJYPUfYxXOazRX400ce5Oe/uMrM5EEa3hBnI5+f1prbfM2IUqQjZQOjrdpspC1DUS3HrOVm33CuC3t3K1oNxR1PGKQFYqqk7NqKqaEXyok1LF00nD8PuR2KfDUGwEN9zbMnNdUhyE4rjDw2AckhuMWYiXPX4xYNVYcz5+O8uqEcbCzD6pohSkLRVdSKIgkr64JNZxTVhGQLigFFIldgQuJbmhA4DvcK2osSuE6SunJpRFcYboyztplgpITNDMwMg5cGyehTbUU2h53PlFgbaSfbkYZEKJE3hpmU4kd8xf5QkTCK3V6G65NDPKCGONNa4Z7VR7jFyTKRKlHxNKOZuAZt26hiqA3bIs3rnATX6CwrTPJUZjvl8iz33vEgH/iTrzE547JtbpxUUkIFB60UPyCHwu/5MAag7nsuz//AD+UsWcPwbAq/qll4qG3BnNyiXzSUtmuSeYUvxdRhm42VBlo71hhxFdhJhl0xWWOjs8KGzAAphyiKYvNESXP09nlS0w4z0jzhDoDd/8BuDa76A74C1i2uZfpUceyxK7zvf7uL9/7+f2ByukeyUKPe6uFlX08ie4iek7bBtQuNxwglWVZpApFPk7N0nDyOctnQaRztkjc9zhCw6Q2R0R5O+xRGp1Fe0dZ/aftOH2G6Z4iCOq6T5tbKj1Pyp5FQop72aJmQL7cukqbHb49P83M3TdjGhr4b4e1XqCZEZ+ELDbhmRLMnVLR7sTwqrWLBPBR2KOorUlsFCcmak6By6UqVrDeZwxVGrgt6ROy08acABS4bvD3KSrPmjLFMnp5SmMsG1Vc4o7ERwRomHJBa2dRehZuBiQw89Li4bkGEhNGysqYCqUWwzFwvbneQWTKRfW2+XE6x9iyELnaGzmnErlx3XFmnbdSAdgT1lmJJWi5uUGR7MbjK7oFESdE6Df/qnz/Mbz70LLOjc2wol1vdGu/Ke4wnNH4PJGKkl4bZlCKRUPREghWThq/s4y30JKdP0TiLnUmspKHtSDqKIp0FNwWpEUhlYGMTlg0MjyhSDiyfgEwZ1pOGhTWYkjPuvMKZVAQnwCsrK2s74gbtxtExvkjUDiy24f5ThqmaYu+UorJbYXxFR1ywEeSlgSMtwcxxXIwAa7t3fcP5DYdC2ONLUZuCW7Qy7LPNM9zUL7F3e5mZsYggoUj50PPj2cm0Lxhe0ajDwSHFyIjisKxnHQ6kHO4PIBfBpAvzEdykYNW4zLpFJpLjPBlF3Fn/OqvteaYzE3R6aap+RNCF8+uwawjmRNpvaHzHUFI+xhlnpbafuUKdD/7nO7njnpPceN0442MlSzoMZu1+sA+WA1D3N92frfk3qf4KepIqLxkC1r76nW9B9FQBdg5U51J4Zc3CIy28QszQyeBxJGeFniJTSNDqt6kvt3AcF0c7W3EnMmNXoPRNxu45UqzWuMLYfVSAncvMrrEBsPvOOzH47gt5BaTVaGt2rrli+Os/foZX/JPf4RuHH2V8ZgeLrQybHYkNmiRdfA09J4lnepjWM/S652yibT61jZI/yYw3xGt1kgCXK6ZDJ1ghdHJcl6jRJmK19bSdj3NTswStY5KAa5FUJDJssITpXUTnXsau7CHapsWm8jkatXm8fYZ3Fn3+byb44V0ZUllDICzPqCJYMjZf7rHViGRP8cYDijC/ZRxIKZpHILc7LhLNFWOWyNbHTinCdWWPH9LegEisdVDlre5UAXyLBi0yq4x19GXATKGEyUvGAEwLxSXY8JRBFxTOJtZlqVNY5i9bVKzX4PwXIS+GDMnEi8AfwV62L6Cxokjk4liSfh/WLgmwgtxo7CqN1iEoQEoCehMQOtCcx5oKFpswu0MhJE+/bfADTZhU/PnTj/He5TqTs3Ocj1JU3SpP4PJ/ZTSvzCm6kWLIkw5WRbkME9tjtmtkTOG6EHhYV2uqB13HcMWxGMy6ceWEuGX7YhVDWexaF1OK2WllHby981A9ZOP1YFnx1TbMphWFjLJGCQlLbi2CWRIJGpaWoNWGi02sSSGrDbuHFdXtoAqyF4rgvLHSs9SLJUS2bcKmD8OjCtWD9pUY6MpJ+dnOKp+P+sx5Bds8EUZNmqrBwf44M1UHhHAWl7PI7hvglqG7DrWqIj8JvRZME+cRroaw34VRHT+fZ2XvpPNWwQ2OIjAurj/MRXeIp9oX+Eb9KDmngNstooc1ozVjTR+uo8hq2J/QHAgcAmE4VZ4japTZ0YMcf/QUn7zjkwShx+6d42SzorMLazeYtftBPLQOQN3fZFeugjcDZx9f4/SDa+RGEvgZ6WWVWbjvciNynL3K2M0kiZIBvVaXZFFy6LauJzMfrkO2kKQTdNhcbVsGQYazhZUQ80TM2I1Y84SUOH9zxk5rHGHsPr4lxYordsDYfZfNGHz7hbYClhEQmTFULDy5yT9738f5lf/nQ+RTLkPDu7nU9FHGIzR9Zos/SiY5w+XOGUzradr9ywjTEmQPMpKcZUx7dJS8eWU4FqwQKc2kN0zWSdPtr3Jh82F0X6oMfMLu6Zii0UkbOmzCDcvU6cw1JFK7eSrqMur4zIubsd3gM3NlfuH6MWoZaK9HpJIKf4+0Txh6Z+DsSsR95/u8Y49Ler9C5tckXDw8Z8jsix2t/Usx0NMSdySzbCOxMcICBwFhUnwvGWqyHiLDSu/oE6B3KbTUOQi4k+uKkWFlyygh8q1E2Mmcm0SbNKE7L8BDYTKgszGI+/p8yI6+rI0iJ7lxwjYNK9Yfj9246T0KZ0KhLmMBWq4iESQKnQd/SHHmSVg/D1LdKkBw4zyIUWFMzB8SbbJdYlYUYUrxoQcf42fvWmU8t8OCqFf5U1zvZpmMQlYDbU+YjWPYea2DWoXOCPgCCi9DIgPrq1CvG3r9uA4tDBX5lGKiqGxVmuTWZZOKM4sGt6hs5p4vkSdtuLRp7JycNX5kFYUkVBagITOGE4recYMzKinGyt5XsAwCTC8aw1+fhOEIds0pTFnh5eT+sE9a5vF8qTiTrDoBwUVIpqVH1thg6HQNehLnsqIIMyGfaPVJORnKgriDBkf653ilnmT3UBpPum8FLDYhbIAnBo9cPFsnAFFk8coYBBcVDQ874yjzhG0fbhKpd0ixIfssnIOCbCTOC59ziSp1DPeu3UU1MjT9SfaPOFQiQ2dDsRYpBLSXcnB9zmF/oKmpBF8wBUbHdqJNjs994pNsNC5Qre1ipJq2s3aDXLsfvKPqANT9TfZEwFlkOHd4neN3rVK/2KW53qM0nsQX3//3+LB4T/I2l9tcObFJ0Am/HQVuMXoyF5cvpWj3RYpt4riudd1dBXbiii0lR1nvLH77jJ0wdmVh7C6QnhHGbiDFfo/tGPzohbACtnPcILlzUaC4/fbDvPcNH+fTD32d8e0FOmYf6/0cyrSsmdxNzpHJvYx67xK9zhlUuIaWwfjMdUSJYRJRn4ST5HzU4UjUITQBFa9CUmmarWe42HqKvpgjJHHW5kO2BB0hYM4OWwVLtgosUXkLSmeodM/yTLDB2900H/qRcQ4Vs0ga7yUMZWk3SIHIkP3jsNgwfKmt+Ac/5lIdUTRXDYmeYuNoXKwukReBhN/uiKuzBJipSNk3b3FaistVZtmkjUHiNWRmTnpYxQQRydydxJpcjCu4rHIgYb/C0EnqsbheBYDJ8asOvccNrkR8yH1tgulAPiNAzNCdj5m2ypSi04eEgJ4hSIxAdwHctJgJJCMvrgfrriibiWeDiudhowdpmUXbhHwahvKK4g5l891SexSBr/nAFx7jZ//oLFPj16C9PHvdKq9SPq9yDLclHApK8astw6wnMq0mXxOHLKweh54AZqCxKnl9ysahrLewocjLG8aaM8KOgBlDL5D5Mlivw/FTsRoihttqIc6qWwwULQkRHoZWoLjr2ZDpFcm7E0AHbgDdFeiswOFnDfM9+J9vU0xUFc065MsQdSxWwnhxRl2/pwhk9lDAXl9AmbIsKrsVjoD8BiQ6cKaj+Iuu4QY3wwqaFR3n3u0aivjh1JANnpbA5qWnwM7WCYMos5Fi/liBdB4bw5LbJj21imYfUom4iU36f7MS3SKMnQB0MUdHcFC7XOMmWEyMk0lM8Nfr91EO5sm3aviFLHO1iJK0guRgYlwRbcDxTcN2z+VQ6HLEpLicnCYa28ejnzvFB/7kC4zPaA5tn7PgXogHyxJ/N3LjhXDMeRE9xgGo+xtsppxxnz+8zrNfWiEMDa6vaS33aax2KYz63x3YbTF8a5dbPPWFJdbOdtEy6Srff+7HFrCzjF1RGLsuGyLFujJj970YO2H8tnLsBNh9ZJ70lGObJ+S6A8fScxd58O8XwgrEv7OxE/zkiQ6f+NO7+Zlf/GPOzy4xWphmsbOT0K1aVCMCnTF9coVX0IyabHbO4wSrVsNU6X2kvBwlnaKthM2LWOuvMKqTpNw8G2Gdpe4CG60nQJoj/CmkJYJICkpDlNUog7jzpXscJ/dy273ZNz3y4SL10Oc9td28+aCLcSMLsoQhSu5R1jkqeLDZgaVlmLtGsUNaGQrgN6El8qSLZcV0CRLiWm0aVDqenRMwpkdik4JEaYgkJ92pNsZEIpGETVvcGux3tmbthKmrxSBOZrus014AYmqrYzUH/U1wxTghM2OioAl7BAwXNYdXDPkFAXkKT2TUc3Fdqzen0DlFrwkd8XMJgMxAS8CdgBdhraRBYQE6TchKJIrclzCJwuztNCTGHT5z/xI/+W+Ocus1N3BWJXmjTnG99qmYiJSnGC0oy1z9vZxmqa6483hko1IkEFiUkIKvbLiyjfzox3LmQt2wKdJjDwpZxXANsr7iUteQSyvyecV5H6o9RWUEpBYs4SoSrRi4yT5InMn2fZorC4rR6xThJhz/ouHcKnxiHUb3Kt6yR1Eew2bQybybWceylE5NEZyBaCleC2HUxCEsa6KEaZXAg3nonYqBcH6f5vxawO0SLp3wuFUnGHeyHOnPc2tvhVfv3oE/YqhLRM1GXPe2Jtl4EqgsTRIHFN52RaKgrHNacvUkskXW+vDTMFSB5LCyvbbjCcVUWtnZuWN9xascj7br8OXIYX9mO19pXOGTq1/lrbUxhp0S2XLEwiZEK8oGKYt0nGrDwYrmYKCYDBLsMwU6o3NkyhEf/k8fo3zMobynxFBNQosHJooflOPrANR9r53YAmWLp5ocvWPJyimSHC6SqoCzzUt9up2Q6lwGLTN2z/3Yuu765RbH7lyiPt/FFVnlvwZ0V68jPzLgeQ65LcauvtLCdd1vA3Z2xi5ZY70jrliRYrdcsU7M2B35iwukZcZu56gFhQNgd3WBB19/0Ffgm7+rgeKhR85y8Jo/5/NfuZfpuSKqWWRN7cZJzsWDYlHPWjGVP2YZ7V73Atpm7gtzMU4mNUffyZJ1sqz3F2kEK7huEd+0iYI1GggYuYB2M+jEOEbaImTYIVjFyFBb2LS9sEZ00eQ0xfxtrJiAIFihlDrITeltnAkiXiUzTJHw6ZC80cJMWo9AN4D6KixWFHtnYe1+keQg7EEqB35NsS6gZ0YAgbJVWZbh8ZV1ylqJVd7J6zGos4ydGCQKIqdC54whuALuUiyxqppUGAiVFc+D2dk6OSStx8tl/K1w3UkVg0ORcgXDSplFATILcGwJ23ggnantDZBIDQGA4YahfzK+H1+WPwfdZXDWscc0XYnBqAA6aT6LRmLzRRAaTNJlZaPBT/zGg0wd2EfJqXEbHj/kJrm1aEhGirUATvagJJEpAg4CmMorRtrwYBOO9cXhq5hyoCxxHhG22SKTg9KUIuEpwbxWtY6KMDYuJgooFuNIkWIZLsmWJmLgI5JmWpZyDvQypJOSCxi3XLTnYbkH9Xm45W2KQ46uryMAACAASURBVDVljROiYioBsdLiURRQF887Bpkt2VXWUmJlaopwBfqXDIHIxmvQEIZNovd8WFt1+MgijDmKWXvCbni4+xQPLAa8+ZYJ/OMJUlXIjEPQgmAIMhK0vBnPKQoQF5awcyV27eoWZFPKdvmeEBaxomwHsADs8XJc9VYLoZqS2UDF69wEZafAA/6Y/R1//PyXONaosS01TKFvUC1DUuJTZmNpudM2uIHihizsNg5OlOVBPYNf289f3ftF/v0ff5zXv+EAtdoQjqMGNWM/AAfZAaj7XpuwRSfLmWKr3qex2EM58XCoxJKI9Dp9Q4HCaMq6gr55U1uAzjJ0X1yifqGLI3Ms3w3QbV3RzuYZ7FxcrpiyM3YbK2KeuMrYaTtjZxkIf8SGE2/2V781Yyc5dsV4xi495VpX7ICx++auDP7xg7oCRoil2NW9stLlz/7FEX78f/kw5M4xM1nm3JoiTO5HCwIyPWyjeti08qhUdfWFXjF9y6456V30vQoVDL7pcKF3iaRbwtFJwmCZlvZpy2tGTBT+CE5i3M6dWdk1uEIk9tCoh9KJOGy4d46R3K0s/v/svQmUZPdd3/v536X2rfd9mZ6Znn20S8irbBGbeMM+POAFCGAIL8Ah5OQFcvJe3gnJexhCCBgcSDA2YAPGS2xjgw3YRl6wvGixpdFIo5E0e89M73t17ff+3/n+b7UtZNmxiMzhQPfRTGmqq6tu/2/Vvd/7/X0X2+G1cZ0fzx/mdeUyL8ay3TCMh4Z+5clp1iUwNJe4WWvX4UtzluECjKUNhRuNC6y9eL+lr8+wFUG8loTXyjDVqSaORm/U4GnkKpWGjheaOeo4JJNEX/LruyaGlsFXDZicsXIFi3mTQ1ZqEEWPnNeVZzKmRCL/NngZQ+qgSfS8oybJVLmeaPFyBw1f3u4wXjW0rhp2mpCVRm8JAgFJC2mZLUYMwbhx2jEWnITPAVWRg9K+ySAgvOhaJFY91qM6//Hh+/iMmeam3DhZazjiBbwoncR9qHZMobxD7QQIi30SyBq1hmJsGPLhYBkeFbvWkObLUC3D9QXLgRlDvgb94zjHaxRDdhpS3XVoCdAKtJYScNW+Ysn3eKSOG+eMjTelHYTNi/An52LWlmGqYsh7hsmXGmZuNy702WR1TE5Cl82QWDiD7hPIUyCzwqAFzIOZpJZNdWWx6saUitOGTD+E0kRuGE5fbvN71SYvDdJs2JhT1qc/brDkXeUFKyMcmi6THrVufevnoEeKgB7jxuZiZ5tPQmcdUjOKWzHOJSxnbuWg57bjwSctB8tGeJ1LMczcBJOeOoINw03pPMXiRvR7PrPZ/fyVNTy8fg+LK0W+faSX/omQcCbGqxvnQo4aBtM0DGQMWx2YMIY7gwxxXCIePkA9zPPff/VP2akVufnmXgr5FFaCcSnCu+fPv6uHnb+v27UH6p5lzwp8feUNaVVw7Tv9XHWzzc6SsgiMy4Y78so+Jk9WnhXQOYbuL5bYmGu5q1n3Pn+W1/qau/RBELDrauxqrcQVG4S7jJ0uvy35VJmezDDrjfm/xtj5XfPE6fddJT8V7jF2X7PAe3f8XVqB5LOWdBwvX9/mZ/7ju/nFX/9j+qdShOEAS9WAVPEW181qZFyI61hVbyldlzZeahijcalfxKSG8Pw8cdhLTkiofh4v1S+4SBQ3GQs1t2zTqD9BkBrDCwcTVk5hxmEvcXPJfe4doJODtnGecmqQZa+P1+byfHfxEPk4TSWIKFmPgrHcH7c5GQQuCmTrLJx7GKIMnKla0oNwU8GQOpR0itYfgsExw851hxspakQ3aPCyieheWiqxXnK4ajzrTBHdrDrtMwEJq+w2jefUlzqWiOF1vwBM/ZGEVVEEieJOzJjBV51XFefQzGgUqyda65ouBBRlumjjRq79yx5PfRlGpwxitxiFmhyYsZtQu8DgnUVLe8ug/nen29NYuAixWERd/EoDd1+i19tMWX7mc/fy7vUKL+uZZSVq8GIvZDTlE9Utw5OG1Z0kWkXhvX3jGoUamDfkVW81AfuOGobyhv4lXETK52PLE5uWg5HnMuvEavkhBC2ccUFdrGEZ5+C00pvJvBBAswjhlCHYgfXzkBYw7U00aWuPW7YCw1UTc7DPMHaDR/GkScKUxYTWLOyTC8QkWaNasyY0NQIXcydQrTGxOszElsqb0IZ0yRAIpJcMG4/JLexxtafOh56AF2TSSBF6IfBoxTHr1cv8y6kxjk/3YLOJezpzQM0bBtOj5zRUlyHMJLVmAnaoA1hAX2P7HPRZGFfAtBhDpbL0JBrExhquy1fg9uoOVNuGKauOYsNIasS9j+/Z/DNquSajxREGDmYpZKzT1qVGoJxLunHdBUYIhzPWdRxfbvVyvHyEYn8/7/vQ77G2dY4bbzhBuZxxn6OvMO9/lw44/wC2ZQ/UPX0ndxm2pwM6Hah04gkzPj3jWerbbZpbEYdf0cfkCSmWuyPVLhjT47/C0F19joBud1u6rykna7Esxk7miRqBL2DnuWYJiVPF2FUcY7dCtb2a3L+rsRNj9x6NYkOmD4+6CJa9D9nuAu/d/l1YAWlVZYZoW8MDH77Kf/6e9/KOv3yYgwf2sVRL0YhS+Klh/OJtDnSJvrJS6huDwoD9/DGhHGIvg5+bxdDBBL147SU2WsvUMlP4cY06bfywz0kVarVHiU2GVHYWE1cd822CksukU2SJ7ay6YLdMZ9lporatx0/13sqtmQOctz63KFKjY0iHhjFf7QUeMzlD74xhQyPEIdjsWN51MeaFk8Zlv2mUV78I+VkoHDC05NA8bNxJ31EqaVBxvKrAjALj9J/AQlcv5yoCumNUyT/EBEXSdqnRQeBNo0EPgi6IEzPnJgqaMV62dJTIIqZtfwJKxDQpskNAwJkq1ruv1WN4YDVmWk7IGiyswvVrloEJQ37EYDOSK7ogNFotQ6jYlfUkiNiNFp9KmhgGRy1hLuA3rj3J29bn+Mn+24lsirsUxUHAtIWJIWg1DestuEVtGb7aHwzXr1paWSgrZmMM2iJmlTmn+JcmpDYMucjwKWXnFWBYo9cYWuu4LDtFnjSkB/R1YQw76xAVoJDHtUtoTXfqEMj1GsH5c9YFFEsj+Jlrdfb3BOy/yXdGF8eUym3c6kbFKOJFJJTwq7LdpFkUIar6NYFw/VHUjMD3tqG9YLF9yhwEo22P4cp6RHNJIcEh/b7hu3OGh+o7XNua5zXTA/Rt99GYi0nVdDFg3O+hsGnXO5sRSE80jmrJkDtYzKZnDVHNOoOLDDfq2V28YGnWk3VUnZmm+JkRGEkbZlNJ1l+phasRWw6GHTHx8auPsLO1zdHtcfryGRpbMVlFbyn0WEHTnmFTDnKFKW9BoWOpWI9NM0pp4g7+9E8u8Wtv/jj7Dw9y8uiAA7uxVROSTo57X39bK7AH6rorvcvOxR2Va9dJZT380HOAzuE2MXYZn/JIht7pDKOHS8mbtQsE3SWwUR1Ojcf+4mkj1+5B4Lnu0N3XlMZOrthaq8bmM1yxes58quIqxdbq81RbGsV2NXaehyfzxP+YozAdMLWnsXuuu2Dv8d/CFdAFhgCdTnhvf8eneMP3/y4PmQ4jI6NcroaYYMBp3ILCbRhF/NvYjVtte5W4eRnj2h5Kzp2azx0kjmpYN7dsENWfJC2tHZa28TieHnMOwsX6OUxUw0uNYoKCc8Ea28EYMUdPuplnjKFg62w35ugEQ7yk/1Ucyc5Qb8Zounas4HFowjA5CM2acSJ9jbRajcTPkI7hI5fg2240fNsJj/knwcxD+QYI9huiOUtONVVdF6MvcCCWS85LadMEHDQpUFyGhGJyveqkrqw4xeZdANRmsCRwBapJVQSJQKBjjbQkAnV6TvXD9ptkFCr2SBVbel7JEbeS2AyNVxWhUb9snR4rZQwf+aRlKGNo7ljS0q2VjXPaOslJr6FxJrmWVR5eS67RLC5WJTUARbFMmwH3Li7yE499glcMv4KpuMKh4Q6vL4UcyMPspGGqx1CQRq6iMTKsqNosrw5aGB5NNIpuLZSZlsK1RWQyhrWW4USo7ld4+2JMsG7ItaB40qN0NBn5taQ1lBN1FMI+CNTQEEnnBs1tGD5iqDbgsY9YZ7CYGDeuBePQWIpyj8ewfDgyp2ikrlo2gTcBbTFn0uZdSICbAHa8Yh17aqR/rALL0JkDrwGhnK8Ki14GSa79qmEhjPnUnKHkBZQtjEaGR0nzVGeVJfJ8x4kBerMxocDgqly/hlAjXplUYvX8WjeaD2a7ZILWvpno+cTopQ8aFxytwOfNTUMtbZm4w9DcSti6Ul8SipxuGebUFlKPuTsMyaYGuNcLeWxniT+ev0rvToWjk0XHwvoVyMkVrAsDvc9lOqnLCJSMeQdCMbUFiuNHuZJp8cdv//8Ymahww4FDrpVEBER3j34Ljyh7T727AnugTsfQ3XGrhbOf3uDTP7+EzVuGD+Tw/EQL5y42uqPYQm/6q1cfT2PoHKB7uobubwjodnfOLrATY1eq5JzGzgE733ep+AoZSHLsFHcyxEZzmSTHLnHFahTrlT1Ov3uO/HTIPjF2voBqUoS++zp7t3sr8Le1AvqsCbXo6v38uZj3/tY9/MS/eh8HDozj+xmWGiGBxqQSkRsfT4yaaCjbImpcxjYvQXvZMXLGz2PSk/SG/a4Cq9XZRKm/fmY/kTNNxHh+hsj4bNefIGpcwsvsI0iPYeMGGenmsHSko+usE0db5L00W40FTpQmONl7Iwf8feSMx+0DUAoFdDS6NHRysOXDJxbhmBoYlhP2Y2ULlpbg5bd7eDJGPIELjQ1VC3YxGZNmppPoDL8vqfKSicHTmFViNLFz88kByfQr4DYZv7mTomaeWwJrBk9LpOQV3Va6InopM8QUSV8XJW7a9lXrnKnpb+uOdXcZOoE+MXUDybi4tm7IFCDbNpxbtfSloFQ0TEwm4Evsokweig3RqDNdSU7simNJCYl1rGN0bM2wnbL84H0PspTKMZqd5TgeL/Y90m1DuwmZLERbSVTHkF6/juu87SlDRpplhTN3c/n0Ohtyl1bB76iWCyp56C0YbtNIcgc+vBVx2bf05Qw5Rb7kQek0AoWKbdG6Sre4eQ2ys4YLc5bV6xq5wtYWPDgPt0wYDswartQgs5qwUr4YuO4+cQBTx3ONyMe6rJxiDQUkBb71ODF1YlulidR+rFrMRhJJ4wKkKx5/cV+HT7fge3sDhmNoRvAEHR5rXsdLVfiJA30U2zEbgaEg3d5ysu7xtthVi5/XvpdW0hKr2mxALKLkQNJ1Jk0iGtE3z0EpBdc0il+AKTmeVZ1Wg+yNhswk9K4aF2kTNWMKNs2maXMJH88P+MDWGaZblkODg5T2iwNPrjFkrJF2cSgw9AUJo7uvAFN+TLnlcVf+IH7fMd7y9j/Fu7/K0MlBBkby7pyzp7P72znK7oG6LqCT5u2Jz6zzxV9boVGPWfxSg6DPMjiTALtdJs7tll12bncfiaFbqvHoXyyzcSVxuX7TGrrd5/g6t7vA7iuMXbPG5sqOG8UmzRO7GruK09iJsdt2jJ3vBNECcYECij9wjeJU2GXsvspAfp2X3bt7bwWe9xVILibEehgefOgqJ46/j49+6lNMHBjjWjXNTuQ5048X9mNl8QwHCNJDxJ4y46pYBQsrciRO2DYvNY7n5ajaNm0ZH+I6Mk50iOgP+8gHord8VmuP025cIps/Ti414Ng7z7YIvYBIFWI7j2I7CwS2Qy1qkgmq/OaNb2CtVqEQp13O11jZcEvBOAat0zQ0ti3ZuuFKLHYKTtxkXMzHp+Zipu6Ew30e1z9tKQ9Cqpxom6R7q+xPAJmAh0Zo8fUuo1bvGiOk6NCJWqPTvNgf4yI0HGCQaUKgTrBYI1vptwQoxBDJHSnQ0c2zMw1DrNDa07iYFY3/xNJphIom2Pp5SQzF9FSTUGFFp6j9Id6GR67CrcOGVAvHamXEWu1AoMiSdgIOxAgK+Oi5lfvWumyxhwI+sPgIv3PPFW4bP8yMV+RA3mco8thsWzptaKxDb49x2rf1LVipQuUI+Pvkpk3WRKAodzhhQsM6ySi2AeUKVCZguwl+zThw19cwrF2BLz0IvT70TRgXBlx9CjpitzyYU3ix6rdW4MIFS6YFh4YMGxZGXmpIbyRBvlomRZ8MD0FU64r9tcYFsDKVaOypde8kYEnA2DGg57rrqZgYMX0iRR9JzC/+rNbJsLFh+dgpw2tMwBHpG2P4sswH6YCPN5v8SCXP7dkSnonpmfScVs4xhimcOcXFYWnULhZRuXXnEheu2xZtuMwmF5LOXuULZiZgZMpjexGaGSgdh/oFaCnzuAX5LPTljasLW5AxxuT5nJ9jLD1C2Vrede7z7As8hjdG8VIJY6gxeDpIatTERmoEnXOaRsPcjuGAZ7klM0F29ARv+dz7+c23/gmvee0JxkZ792JPnvcj6rM/4R6o04evbTn7mQ2+8OYVOlGcUMbS636+galY+qezLn/oK0v4DInAxkKdMx97/gHd7uvtAruEscvSiJvsbDYdW6f+SunsdETPhiUqmUGqrXVq7U13v2c9Aj/EL/o8+kHl2AVMHhhyRozd59+73VuBb/UKfIUdbhoe+MJT3P79/xXWrzBzaIIr6zoTFvGCAl5qBGsUROthswcY8Mv0mYDl7fsxiviPm3jpacLCrVhnkMhim5eJxcKF/RgBQOORC8pYv8R2aw6qD+IJ0KUnKCujTsnAnS3aanVozbv6LzFhvV6HnYUNfuXbX89U/yCffKLDC/MVCsQulkIZwz29hvlFjY5xLP5f1SIOTsBsyWdhAeYD+PbjnutpFdBKi4myxsVwFG43ZATUJHxX9IhGpgJZS5b2okJqgQoJoBPbJgZMjFsF4mvdPDoZE7SzBPD0sVfLgBi7JY1au1l1ekAdojVhPkM4mzhzHYOkEe6qxWgUq9fQl35eoECAIQXxIjykOi2dsAvGjQ7lFBVLubEOmwtJdImYvYwYII2N5y2l2wMeqW/xul96nNHZHoJgiu8ISvT3RLQ8w1G5gPOGbMpQLsK2hVoMPep5la8lnwCNtvSDcoxKv6jfU05gRcFo2xSse0jMFLQXIV+CqRnjcuxGMTxYg7UtyKwnsSrpseQ2pclpE5fvVg4NmZRhuWkJRgy33gA5GS4ayfac27ZM7/fgSjISdiBN+02ATustBk3rJmAsfV2XqRNr6Wn/KRj6CjBtCI4Y7NkkAmunbvnziy1uaQds1hLn8/3av9bj2k6NO0YDvv17igRbluiK2EyZLwyBNl7dslMJcHPj+R5DSyaK69YZVjRabov5207MNrUF8DVmVwD1EKycwbmvy32JmWXrPOQPJs7gfgx9TcN4weNis8kVU0Cdx4V0mvdcf5LOl1uMlUaZPGRoqw+4ngQr62M7MG2odAxba4apYeN+r6FUzBOdAo2RkyzNL/C2j7+bV994grGhPkyoydfepCj54H1r/tZh4R/ulxsFwfZKi0fetUGzHrvoETm59EHtGMujf7DJhi5jumPaZy6WNHhzpzZZeaqBn0nqv575mOfj3wJ2cRyTyaaZnOnHy8a0ozYWqYDcoYXYRgzkJzjYd1tyn5sra/Rg8XwPbxTe+0Of59yjSj/V2Lm7AM/HBu49x94KfJ0VcHVfxrC01uZNP/8gd73k1+B6hsF9R7i4aghyhx2Y88IxgvQ0HS9HMzVMRXEjfom2xqoCXzJJxE1M/gQtP4/nK7ghaWYI8jclOjtirAmYx7DRWXMjV1Lj+F6WpgmZj6qYqEosMZ96s2zbsX8mrpNW+3pfkRcVBjGLTf6YbQJr6VhLXxa2N6CxBTOHDPuPGMda7e8YikIkS3B/bLnphMHTCVkBuZ0kkHfufEy0hGOHYo08tdkaC2rTu7ln6hllQG5UjTKNGxvaay7rxYEajXIdJ6/xlzR2YvTqTmqYFJ8qV0xgsW6cDkyjue0v2GSs260Z0/cVi+E0YmLedpk/PbEcnbmE2RqfMrx8Bh5ZFHAFMwjLMkSIGNyB3hugfEcXsAgIhpAd8dgJLL//4CMw1M9KOMKUCRzW/Nx1Q7YJ9TaUUzBxC2yJLVqA3pJqvaCwCZVtNz1325GWTkvP3U6CffWzsZom1Oe6bN3vqMy8dTFs+2HsgFsS7pw0tK/BBy5bzvkx2yFcfkpjXRgagt6xJH+vKmNBXbVfEC9AmDOk81Dph/c8HLP6cOzMAY7d7EajKPNPUS1uaiMwN5n0v2pN5Xx1LlXtG7GDWreymFOLd9K43+Wep9rsWw0Z6zccHnSwkNd4hopneMrzSLWNC3QWY6ntTh8Bzlkl7Diw78Ckxsp6z0hHqfF9GfyjhnActp5KQpFl3NDYd/s8VC9BZgcmRuH6smVzE4aOQt8grF+GYBjXHHJgBE4MeBwOSqzFLUpeimz2IJnyjfzXkUv8zqn7ePizltSwRzhmXaVY3zRklEvYBz0juDzGFxwzTnO3YppU4kG+79Z/zq2Nl3P7y/49b/r5UyytRY6p1zFh7+tbswL/sJm65HNFKu87cenClxu0BewyctdZwsDjjp/uZ+JEIbGtdx//9F0hO3umHFDbbFFbbruexad///n6fwEwz/NotdpcubhMfTUmE+QIvTDRHOmAa3wXSvzEyn00OlVCP03gZUiF6eRnn7C89s23cOvLjpBKB1/VBT5fG7n3PHsr8PQVECaxMdJ2rqxW+dn/6w9586+9i+LsEJn0BGvNAD8s4YVDzt1q/B6aXsBoZ4OxzARTGsPGW5zd+DihlUB8nSh3ksHcMaa9NPOdNXzZLxFzkCdqr9MMSgz7RQ56aea2v0TKL+Jn9uGnhmgL0HW2nOEiVs9r3CRuXHB6ur7A5/q85V/ftZ/vmLiR33h4jUNhiX1+lpS15H1Df2BYUp8nhkYH1jet04CVUh5rKxHzBcu3TfmOyfJ2wJMBYVugx9B73NC6ZlFYrAuxVTFnV38lRs5Vgo3L1GCTUaqAmMT3AhRifQQAr4rVgfo6Li+uW3qR6Op2jRSSk9RwBgZvwKCiDHf6lBZQgEM6MEVwiHnSqFavISZMeHk7qflSplpuC2rKYWsZZzqIxXKloO8A5GeMK96IdqwbPbr2hB2PT3/mKj/9viWOHzzkqs72+xlu89JMZ2EkgAHl3PUnYcdnzljXj6peVoUIN7XdCmUWGAoSs4eYw+1z0JTpQIG/Yp3ysP4EKIPaj6Cne6vOWWW7i02tKbLFwr9frLNwzeOGgkcpA1WNYPtgU4xcw3D8ToMKfpzhuW0cWPal3ytAr4KQxcBpbQTaNBKXzk/dsGLOFI2ot550dHrMkMGTqUKAK5uM6Z1OUlmC2jdF+IvHDevnDbNqfsgZUsqAs3AK+PSlS/w/3xmyf7SCLVnXGsG2qtuSn9eoNrpk8RR2LONGmPhn1CGsiwRv2JDRHzHI0jl2NZe5/QnTp93cN2BYnrdsqmdYhqCjML+egNpUCeci3mkZLtYDBzSPBEVe4OU43Ym4L77CW89s8xJ/hKmi5/SNZjXJU8weMNRV0WZheAZ6tgwTLR+/aunxMgyWDrE9YHnv+9/C6nqNF77gEPl8mkjudxESz3JeffphZO//n9sK/MNm6rprJRfe7Isr3Pmv+khnfdpVS+D53Pl/9nPkZT0JUNu9sNDB9RkMV3kgw5Fv76c8mXajXDcNfW774Rs+Wi8nQNdpd7h0bpGl8zUyfp6Un8ZzopoE0G01V3l44R4Wdy64rsvQTwCdHwY0z1he8ws38+rvv5N8MfM1v8M33IC9b+6twHNcAfcRMTYBdEvb/Ot/+3be+bZPc+DIMRpbA2y2IsL0CF5q0vWpmnCIdHqQlAlZx3A8HGLWtljafiAJY7MtvFBdTdP4fo5UZwMBM9clYDyiqOqmlSPWkjVpTPXLjm4PMtPEJqDtosZC8rZNZEI8v4CN1ohbV/G8kHY45srZ76rewqWFJp/prFL0y3ieZShlaBhYjGFkyLC2DWevWHfiU8WqdGpnDAykIK/6L4GTDORH1X5g6RmGlGfJSkeXT34dF2Ui7bpAm55ELskFJfp3R6E7ykXr6uoECgRqjiYnwFxfF0CImdPZWifXZbAXExAYLVqCFUtOiS/ab4rWTIFd6cafiEES+5PHZZGh2I8N6Jy3Lsy2s2IpjkJjCGpFS0naq3EI/UQb1rlusQuWeM2w85jFuwjLV9u85aGrcPI4OVNm0c8xakIOBJZXDhsHwNbaluKMob4JNxw0TL7AkLrJkVmOMcpJYyijRwWWVpNGjtLLDJ2hhIAMjxnHHGptzXX1w4JI3AsPKdzX0H+TGEOYGTC8aMLws5msA3BvWo6557pFCTiKdllbtvTqsVMGo0o2rWF/EhqtbMF/NOCxYQxZjag3lWgAC/fKUiP8Yd2oWmNN5x7Qnfr/eftVFs8XukkAj/arsggV3TJyv+FO5RSqcq2pPl3Ybml6bRQeRzCq6odkxCsdpj+pnWehavFSGu2ahGnV+0aTm3M2ibPRSF3ZphrPqjFkShZigz8oraV1kTCxar9GFdRsaMkBHUg/aVwdnCrgwilD3G8YImYohO9Pp9nutKmaHDfmZinlb6M4GHH3+7/EF+da+KM+ZiaJi2lfsE5XWgyTEW91PQlxvrNguKHZoc9keKX/eu688af4vbf9CT/zb3+LlcUqvnrUuvViz/Hwsvfwb7AC/7CZut2FST6t9E9lyY56rJ5rcduP93Dkrt4kbLL7ffdwvQ/1IdTHrWuy0OcuUwgojaTYXm1RX+08b4xdAuiUbdXm4lOLrFxqUMpWyIZ5QpNy2+KbgI3GMg8tfILV+lVyqTKhnyUdZgjCkOYpy+vffDuv+eEXki2k9zQNu/t97/ZbsgLJ58ISxYZHP7vFm/7d23nHH32e6dmTXNyexAqgpUbwQp35ugHCJiZnArbiHSaJGAkHuK9+jkudDVK2QaxQqWhFSQAAIABJREFUtswMlfxRtjsrXG+vkVJehU5wreu0MvsZTo8y7Rc5VTvDfLRFLque2Ap9Xpoo2qBlIxrxDrGQgNyxrXkXYpxPDbPR2uZ7+0b47v7bWL3Q4jPNOTbkKJfmyXgcTkPRNwwOw3IDRgPjkkaebEI+VkCrx419ngNpcoqm5EjdgkrOODH59gLkBcqyBn8scb86Y4OOLWJ/ZJhQXInGoF1GyjlZ5X4tCrDJuqjMMLE7suGbr5okxLh1x6kaAZrIYAqJmN1UklowmRp2I0K66MQdvwTmFM6rEZ9RIK8YQjGfg4ZyAJ99HHzp/eSirCbso4J7BTht0RA0Lf6Az0cfvsIvPLjC4cF9PB7VORk3eIUpMybjcpS4PK9vwsK6ZTAwMvDSrDlZJZWiiwbEBEndl+p7XU3WttZR7JOy0ZK3anMZipMQaAypPL2FZM3SCvlNGXlo8BTfUYDtquFQbEh3PH5ltU3OazK6ENI7ahg+lqQaqGXD7xG4s04zF44Yl+/2xXXLeA+E1YRtVV+snKdiWeX+VO+tu54WQNL+EmMnZk86Rcl39A67lGybwqUff9SyeB/c0W/I+saBzasNy452a2j4k7ltfvS1OSaypaSOUqYc/YprxgFdT/VxAqBqmJAhRuBf+rQy2GvdUa90d9oWbYcaOMaSOBuFXQdF49jG3KChN0jMIquPWfYNG+fAbaj2raNuXA+2DHeUVSFneCLu0LYhM0EPZ1RXUq7xwSvLvDwsMTGYwStYxxIGfYaW3sOLSatHaQTGe5PftbxqGQ4CRswMduQkH/zou9l4/CrjozcwMJnG99Qbu9dA8XwdjPdAXXIMSy5pdawdTzP5bTlGj+bx5fLpArpdcefi3BrnH7tK31D5rwf6WkUChBSHQ7aWm9TXu8BOP/83/NoFdJ125Bi65Yt1SrkKueDpgM5n2zF0n2CldoVcqoQYOgfoUilaj8Nrf/lWXvtDe4Dub7gb9n7sOazA7ucktoZ3/MFZXvX63+ehs+cYmZ7m2k4/vs54oYJJVa2Xdo0PxjZpqb7LyxBENfwgz5rxeLK1RCbadn2r6uGq5Gad27XZXsX3UijSJLYN0nGDwfQoy16Ky615/Noj5HLH2PECZmzb4aT5uE7TtondCHYN29H4VWHFFQb9DBub1/i5Iy/kRHmYT9Sr/G67Q2/Q4yafJ43PVjPRlK2vK8Q26bjcigxXY7i3ETGxbdh/yLg6rdVlyM3A1ipkhiA1bEhbdX8mjkeNVJuXLJ2qQdluqv1yY1YBAunqdCv2yIGDpAUintPo0eDNdEeTGutJnyeXrICYTubj4FrOrltCNSJoTFcwjqGzVxJXpowYjqXTaFWiflVq1a3bNo2LtViuZUKYoQWVK4ZMHXLjUDma1JntrOL6UGXiSKV85uI2/+Kjj+FXpjnu5ziD4VV+mu/qCd3I8swyTPTgsu/ynmFcbW/txBXs2B2NANfc7nDbL62gWEGNjsVCBvq9Nf7MGDr5ZIy6pZJ7GSKOG3Izho5GijXLzjZEApJNaDQsvb7hSB5uDgz/9/U2a9c63HVXyunwpDt0bRhaf8WDKIPvqqVgDRsrHVa3Y/b1+oQlyN2pAKmEiZp/3JKuKQOva2A4bpzD1o2zxbxq3TWqFbsmr8s8PPoBy76cZIyGnQjHhs13kraOJ63hU17ED42nmarknP5Oo2a3fwXyBXgaEGt0rveHnNLSVKrqTa8nIKu16gJv25OAZF0HkTdkFLuipXzSEipIvycB7x3PEO6HviFDe6474s7CY+egz4PZtEex43YBw8bjJSbNWFDhoc4mbz/1BC9Pldk3VcDmYxS5Eip/UZl6VWhegYbWyIPxMcNg2jrTS7M5zP3jJ3ngrz7IW995jsn9Y9x0g5yxyQRslzB5DoecvYc+YwX2QN3uguhdr2tUz5AtBs7Zltzx1Tfb8rUN3v0r9/D+/+OLVG7MMHVoxI1Fd09kAoDZQvi8MHa7gK7dSkauCUNXJufnCZ7B0D28mDB02V1AF2Tww5DWY/Cdv3z7HqBLdu3e39/iFdj9HOi9+wd/9AV+5Ad/k8mDDdKlWZZqYdKG4icOV1VxRWLsTOgMDON+nrKfY6O9QtM2WWmvkiOmVX8cL6hgwyGGwn6qcZ2Wl6FtIwITEzeu6CxFpD6Jzjqm/iReZhaTGnQu2HRcp2YCtvV4nWniKl5nmVg5eKkhUp11ljs7HCnAG0q3s7hR4MO1KzzqD/KPgzwvsz53FGEwlQQMr0vzJS9CC+qhx7urO8SVDj9wMkOmAfNLaj/oMjcTScG6wJVYlc6WdaO+dUVMXIfCoSR3zi5ZvP0J8+b0ZBrHaiKg8apiSKSpSyWMnXNgygZVdb+K64nViV71VNJSVR+ydHJJ5py7nhSTJeZGzQwCBYrDEDoR6NAIT7owAdaDJhn9iRlMJY0BKmiXBlCVVKXRJPtNrFNDo88YUmIQsx6/dfoif/jZbU6OHaSIz+OtTX4oV+LGgk9l0NCoCisaiqrsUoiuQmz7wFeacwm21pKcZDUeKEdNsSkaCTa1be2k8kt6Pps3BNvGje0C9a2q7SEPkRo7TkGkWA61XvRLCwgrm4a5bcuBUUOfWhfUsJAP2LpuaHgxY1NeAnAdkE1+r6DX4J/Q7vIwVwyVfQajUaaMc3WBZuuMBQKaXLNJ0LNAoZ9k4SkzzihHTmSwljoHT8zD1kPW9ac2Uhpxq182MTeXfMPnozZffHydn/neXgaPhljPJvpH7TetV69iXqwzXyAGTzl4AnCqeYuT/ajXdyBPQFXrIrAvI4zY1aEEYKYyhui0wqoNRbGT83CpBsO9hnQapEtcFxO5CvsEpJsw0PQ4hOGU05WG9JuQxaDCWjriHV8+zSvGi0xOlLHzccIQKouwYQgGDU0Zi3YsxTFpRg19eehpx4xWhxgfOcZC8UO893c/y76Dg9xwYsJNnXaPId/iQ9Xf66ffA3XPtnu77Jy+tfsmW7m+wbt+9S958G3nyd0c8OiH5ijPZJjYr75JXRomB2LdOsZuqMvYbTx3xm4X0Imhu3xuEcfQZSpkg8LTAJ0YujUeXvhLlmtXeDqgC8TQnYHX/NItvPaHX0iukGjo9q6Cnm1n7933fKzA7udE7923/s4n+LE3vpXJ/SOsNY+y1UlhbAsT9OGlJxxLh5elFPQShP10vDRFI/sBbMZbBK15Z3Bo13QGivGy+/D8Eg0smaDXCdOG/KKLZzPS1pkUHcUINy9i/DJ+bsadIDw/RxS36Hh5Teqcg7bVukYnPUWYmXHZd2VidpqbvLFviLuLt7PZjrnUnsNQ5mSQ57ixrj/08HByolbTwmoHhnJwumF4c6vGj1q4ozdL1IlJDxhKxWREV6wYN2EVwFAuXCRyUKNSje/kHBTbphOwmgqUPyeGR1UJQlsapSkPTWDmsoJmE6bI7StFn8gNq+/ruKMTeyZ5fo0q1aQgbZbu12jOKyUsmDvxL9mklULSEY1mlSaj19d2aCYqwCcmSIAil7BRn122Tvyu5od4CcpHoaHojILH2atNvuvjp2BoP7f6Pax2Nrls2/xAWGQy7TlWKZ0ztEqwsWUdQxdHxu2PoAGd60kVWHYqAXJyUyoGRqNi9dvaAZh7xFKZ9fCWBXYMyqYOyglg2lDBveI78jCvDLwUhDIxhDKUWPqnjOtLvRjB4Vmf230PgerffspSWU66ZQWG44vWBRSH00kPb9A2bGsM3G8IxCSeEZsJqRFDqDU7pABg43SYfi/snIfmRciKMROIF8uqsXZoeM8Hmuxb8xgWY5szFLtdql9YgHLa54urVV720mX+yd1jGIXIaV/IBS1dZbXLnBbVIGIwyxYvZ4jnwRtPmEYZbpxuUu9xuYaVmaf3hJg8sbnKMByAQCNzvYEyBptLPB7l+aQuLR7AAc3LlxPQPTZhyPXDyD6N5w1jbdXfSQFgKKn6y8/TX/T4Tx89zatuLDNxpEx81TrmMBg1BH2QGzHkeqB+DlczVy4asjXD0UpMtt7Pp6IjBOWHeM/vfITh8T5uuenAHrB7Hg7Ge6Du2Raxy9rpBCUgtHxtnT/6lXv40rsuUJhNYVXpk7Gc/shl8uNJoK+aJ9yXbsTYFUOKTmPXpLbaSUa5z/Zaz7hvF9AlDN0Sq5eaFLNlsoEYOjldjcun23yahi4BdGnSgfL0Eobudb98G6/74RftAbpnrO/eP5//FXg6oHvnH36SH/uRdzI9O8NiY5KWzWLcfNHgC9AFvRi1PNjYuVS9lHonU6x6AdtEpJqXsa5U0ydunMeEPfipMdoYRjP7yHg5+ozHfpPiifpZMn4Ro0y6tjLs6viZcYxEXDZyRMW4ohlsm+X6BaLmVSb8PIeKd7BqO0Sta6Sx1KJ17s7eyIncmAtmvXfnMmU/GRF3bMx8bCi2jeLk6Bkz5GKBGsNcHBHEEf9sqOBOiNeVPbdpuPCkdZ2Z0sDJVJAd03jQorolOevl+MwKFIhNkUuxx7gzrICZY9VU3bWaZM6pHspIR9cNnZWJwI3exAjpxK0xrfRyq1CfA3+qO3rTOGtD+XbWOTNdBIaYGwG5QUO8njB/aq3Q6NHlmUlwL3Apx65Gi2LW+o0zoJz5RMxI2bgstM0NhQFbTN7nQx+7zEdXN/n2niNk8HigvcQLJa7PlIlsxLUdGCsahvukPTRkSolDUzq+QDVeeUNBFWYVQ1vu0VFDekjjbeNG0MLwBdWGyVEqzZ2MJtKzSV+mpCll/Q24PGoXd9Jcwxk7hKs6PYaJKThz3XKmCS8bMs7BO95jGKp6zF+GxyPLWBFyApEKVpYDWdj2MjyyA4MHjMvCU90aJUNnIXEnW7V6aF8o4mQIp0WUg1R4XCHFAl3CxTLUmIcNtx72yCqSRUyaRquS8AnLx4b/PLfGLcc63H3HIDQkskvG4m7s2mV59TNyDnfmjdPcBQcT16vmu+79I/Anlm2rGyitfS1gOQbR+a75Rm8v7VddM2xCZswQFg3z5+DqWUsaw1zdqoXOOY0VvZIZMhQHDOGapbYJbQ8XTBzaOsdTE9iegP9w74O8erqf8QNFWk/FdDQOVj2dGGCN+b0EeNeXIJ+H4pC6aiPKzSEeMlNUyk/w7nd+gun9g9xwMrkg2z2mPP9Hq7//z7jnfv2G+9hd17A0v8HZj18nPZ60NNjIEmQ8tp5qcPGRBVoNHWW6V8267QK7nqEcR//RAJWJriu2i/uSB3/t37uArtOJuHx+mZWLdYrp3ZHrLqBLNHSnFu9huTZHJigSegmgcwzdo/DqX7yZ1/zTF+yZIr52iffueZ5XYPfgq3HQW//7n/HGH/xtJvdPsrAzQjsOsfEGJujBS09h3ZBUbSaxi1SYMwFr7TWUtpi2bYL6edq1J52jNWpIRBZiwkGixgVC49M0KRfbMxhUyEWbzqnkedkkZ641T5AadaPaFjG9XhqNdM+3l7lSv0haIcTGo1C+m5Gwn0J7CdvZZLmzxYRveEX5EPmUx+nmGu/qpJn38wxbyxSJkPydm5aHLCw2YbsBaxFs1Nqc3LfNkWMebQnqFwyXq5Z8rlsaLx9HBCsP2wQohMblmbHPuLotnXTtFUu8bh3T5mgUnfSnjXPUun5WHTN0cpbObjBhoOyawGPXbSlWR2Nag3N3un5QgTwxfWsWf9JgH7MJ8JPZQOPXDPj6Gem1dCswqfsVjyEkIiCxqSva5Hl6pw3tg8k4VCYENTOkOz5b8y0+GF3ieHGaaWmKbUQUt3ld6DPtWUod49ZPJxmBgdQsFMRk5RJjRvXz0FyybGmseckSyJFZgaYYn8g6YFv7DKRThuiiJVJjhbL5BJwCaFzsjqbXIDucAFz1wArYKYR3IAMblwSkDC+RuWBJukYojBvG0zBZNiw9AH/0AFzYTManyrjTHzlPR1sQnLdu3dWxa4YhFJhSFfGS+kyTtbLnBQiNYxbjqxDPWfwyRNvw4U/GXPNBGsD75iyKCBUDmCvByQnDthdzjRazPUOJWUNAURVjqnNT2apzKmteDVaAbKabcahgYY3Wlfe2YZP3RwWCCZxOjoPdD7reCykLWlOxvtL5DUBObmt97VgOHEv6eHVR0NAoXI7fVdjQv+esqydLx4bjg3CDhZdmQ46GFT4XNTmpPrb8EW5/+xkefHSBTC7AthRVkriYlTUYzhgyB3AXMVtLUL2u8a7HT013+O2+o6TCnyY9c5A3/uB/462/+WdOYyryQseWva/nvgJ7TN03WDO9sfQ1MNLDyPEyj33yCo2dNqmSz/YDTe76d0f5rp+8i1Kvuu2Sq7SvPJ0OPF3GrjSSZnslYew8mS+e5WsX0LXbHS6fW2L14i5Dl8P/awzdStflOtcduaZJh9nE5XoaXvtfbuV1b3wRudLeyPVZlnnvrudxBb4C6LbhnW/7OD/203/I9KGbWaz10vGGUVSQsueMjBHGx3ZWMGE/vkLHoi38oJcormNcM3qTTu0s1tbVxuqU7klDRN4BuWb+BDPKPDMBCvLpjbZ52HbIGo9m9SEyqUEUXxJ5KXwvQxQ3aHZWse1VvOYcbeNzrHIXQXqEYdtipf4Uq0II0TbHcmPcWjjKpXrI9WiVT5Jjf9BDL4o/8RxJMywsorHpOqxH8CCWX29t85JCkWMd1ZhZ+vuNY9UGU4a+QmIquHbdkooTV6bGieF+49ygIiP15Rg4zYaVXyeNmogUNQn4STZy7RFL4xSkbpC7M2HkXHDxvGI0kraA+FIinncO2DUnw8MIKOpQI5esGL1hdYQmxygBTQFIIxfu7vhWjJ96rtWdKk1gWXloBrtj3ciwb9Cw/UXoPQqFCWXmGT752AJvmltjtDDJXLTDXwKvt22+Iz9Mfz5GzGXPgGF0HNJFlyNNQzW7IuH0+6tVYTYZ8ar9wFe3rbbHU1ivoeMlsR9yp6oLVtq/WGsTGdp5Q3MO95imjBsCt3WoX4PccKK3CzK4XLbWGkweTphR6RiFg8qT6neFygYsLsJnroL0gvvkou06W8shbHwpyeXTayc7rJtJ182qc8aWvq7ZRRl1Mgqov3YZnli0nP0y3BQbLl6DagdyHmzHsOPrj+H0iuUj2VV+8wf6KfaGWIW9SftYSlzDYiddAJ96XesWjTZ9vU80UtV+1LbKEKFRu0b7QlNi7bQeWmOBOF1MiNnVPhYb2j0FOcfzFsR5yO6HuswNDTi2T+9fw8KKpVA0jl1TprWc0hND0khatvD4bNThfiyvzA3jG8Ob7n+QV985xNRYnup2nJhpljXOFStsCDUSX4HSUcj2Q2rL0JO19G0N4ZkK54qP8ad/9EWmMxVuOLrfGYPivfaJ7hvvm7/ZA3XfxFrp4Dg63c/A/iJnP3eVlc/v8OJ/c4j//V/eTe9Q6Su6u2c+lTuofkVjl6K63qSxqUTtv/7IXUAnhu6KGLoLii0pk/W/Cug8lZI31zi12NXQhXK5CtBlCNJpmg/Dq37pZjdyze8Bur++wHv/et5XYBfQNS/B7/zyR/mxn3s3kwePs1gfpk0OIwGZ7WBMiNXZQAFkCiI2HnFYoWHbjn0bCCr4UY3t5iWMqI24noxSjY8fVGjFVfZnD/Dy9CQPAQdMQMvGvD/eoayuVnW+djaJ05PEXjqJKXHnuQ7t5hXizgZt2+KOnrsppIa40FmjGe8wVztLyka0O6u8svhiNr1xbg0VcFzlA5GiHUouOq6GEbHG4RDEgl304aI1nI+bjAWWw1sF4g3LQCk5+SmeI3UclNaSFvPkG5oerDXFbkH7StJvKtF/ooVLxmEu90z1UtcgetzijxvXQiH7bTDdHZttJO5H1/MqBmYsGdNFp5MqMU+CdE3vTlkndpdhwI3wJqTPgs1TSS2WG/PKKCHXbDXRXQm02CfVHtE1ZKjxQoBBMR5yba7D5U9bBo8mTt6dmsdvzX+BBzZzTBUmGTIB59oLvDzs41gqTT4FK+vQOAQzg4b5OWiXoL+IG8/VWlA4AqtbkG9DtQ5ZMYeq3pImrmVcGG5q0jizgYkTsb0iSARuZNbwpW0TFhWblXaYlHg/5FREUoTlC7BZh4lDxmXSSWe3vWzI9OFGvT2FRLenOf3RNDz8GFwP4fBhgz8sjZ8R7qR0zDgtoQO8Gilq3QQ+BZrEoipbTgynAKeqxGTyCOD9pyz9Abx8xlDKQW8ertThqap6a+GBDZ8Pba3xfeOWV97a69y7Yugc2FaEiZKR9QvuJFpCZ5rRaF7gvjt69wR29RZSW4hG9NqvMlLIqKF/a9SruBX90ZcYQAVRjybtIp6iTvqS/Lu5p+DCDuwbM5QUR5KBxcuWaNAwMJz0z66uwKlVXCxK2wZcsDEPxG1ekO6nhwI/d+U0dxd6mMrm8XpiVylmxCQeUPyMce0h2qZIGYNLkFIwdcOyzwxxplFkvnyJD//xfQxezXL82Cypgb1ase6e+6Zv9kDdN7FUuyzc2L4B+qYL9BzJ8oYffwl93wDQ7T6tPvsCcUHKY/6JOusXmqQketU39OWChQ1i6K6cW2blcpNSpuzChQPXFpFo6LaaKzw8/wlW6nPd2JKvjlybD1te+yu38p0/8mL2AF13XfduvmUrsAvoqo/CL/3Tj/Gz738X4zPjLFULdKzBtq8mTlcvjwYxXtCPSQ3jhX00wz7KNuJgVGU+rlH2Mmy25ug0LjszhJVgyuXYjbnt97wMa5n9jNo2036OtrVcievsNC/TjhvEtccxYS9eUML4GsXKFtpCBgo/bpCPtrmzfBeF1Aifal5n1MtwqbNKrXGenJem2ZrnQO44J1JDnEwZ1rwV3luPuDnVx+fo8ELjsU9ALsYxDcuxmx6yErXZF4bsywdcFRDoMSydxZkU5BSNAkPtXFJrJTdnz5TYCeOYpuYSbD0CaQXQaqy2Iuwllk9GCktLpezDCTulJrNArQBibtT6oH5RCeSVRydzxKNJAK3TMAljPCUhfcIIuqw6MUxdEBKExmWRtZ4AX45IARExchLkC71IyyUWSIBBgq+qwYg58p22no0rSfBuFp9TtRo/fmadqcIYa16OtbjGVnuFN/eMcHAg4ELdMia7iTX420mLhLLjFMGxsgLVGPJpiMWy1SA1hXNM1h5LsvA6IUSdpFtWTReKXhHD1lLQcibpo5WZOSXTgX5HZazlYbMJrXXlA8KVazA4bRg4auhcsfhq7ti0pHoStk7gWEHO4brquQw3HjQszVkembfsU+9t2ZAV0C4ZF47sAJPWXQBJYEnGFnXfOsCUuIa1rXL4PrJs2XoIXlgyZNXykDdUY0XaGE6MGMoxlAKfn7u8wP/7PSEHVst0/Ng5j10YtRzQ+r0uJc0VbWnl1FKi4OWhBLyro7cl57QAr8a2+p5OLWLxNMYXqNe/5SJWMLU0f7Ie6ytImN+23MN6H26CWlDCjqHetPQNQaGQaEcX161j1cQUi1E9UDLsbMFYCCf9gN445vGoza25AQpVw3/48zO84TVDjPVm3JWQY32XEkOMIl7qej2x+GI21wz5Pou3FXAgNcH5esxa5Sp/eu9pWvdmmH7hfhe5snvMSTZ+7+9vtAJ7oO4brU73e7uMmz4gI9N9HLttH0UdNbpGiq/7FN2RrLpXH/v4Gg+/fYNW0xIWDKEyqxygg04nZu78CisXG05Dl/FzBJ6OFAYxdFUxdAv3sFS7RO6ZDN0ZeNUv3uxGrvlS9uuyhl93G/e+sbcCz2EFdg+ui6fhv73+I/yHR9/F2IFZVrfLdIIKnowNXpp2apTYL5PyC8SBLlIKNOhwIjPDTGqQndQwpbCHnqjK5a3PEcRVF0FiXYu5xXgFYmOpBBWqomIMLspk1HikiDjb2SLduJBEk6QnsIoz8fJ4DkhGZE1AtXWN0fwJDuWOcCmuMU/E8aBCMdpiob1ENqrSSJf4rtKLuDmTolSwfKLV5jPNmJyf5qXGp2ySoNxLMUz6sBwbtm2HM3GDyWyeu4qGCY1m04a4YClL/7UGG2pvkDC+F+w85JOJsztmKPs4lK5pv6FzDTpXoCZApwg/VWnl5Sw1zrygOBGZsFxjxJwOGBLGJ6NG/axVf+fBJP+ufdnSeSpxhgbKTpPzUSyS9m9TJo1kZBeonUKaLLE5Ag8S+3cF/+5WjGAWrMJzZeZQ7VYR2gILev4TPh+bu86fnd0k2zNKwcZcjDb430ye2VQfzVbswNX+kqG/pLaCpApMgE6sX74A/dOG1iIUsqAxqXLpIs/QkkPYQm0H8gJ2KYOnYOA6rGu8KCCmPLk6LrNOrR3RjJyWEMghvAxpafTWcP2kzoQxaggF7OatppHuSyNJsWB16dtSCZG8FlvGPZifhw//VczYQePATjuCrGJKdEQWMyaHsoJ2xZYKGKtPVwYTTT/V/qHx/LstN41DX8vQ1Gg0BdkemNTviiVvPP5ys8nH0jV+4sgAYzMBViHM2l9iLHf3ncC8D637Lc3HErCmwgrHuGl95iCQU1avKwZODKu2SeydbvU9mSZ0OtG/1VOsb6rHdsnSvAapkmHFGBYfhZlxyA4Yx6BWxHr2QE/NUN2AVhWK+yAtcJk3PNWEkxiOBR5RHLMZx5SyA5wuNvntL13h+2fH6M95tOYsgUboHevGsBnl5I1DbWW3/k0hxWpoCTnAFOfqm1R7r/Kps6eY+0DIXd9x0Jkrdo89yUrv/f31VmAP1H29lXnm/d2RqXOfBv7Xauie+Xh3BFCUQMzpj63xwFvWiX2VMxsaW5aUgF3O0GnFXL2wnrhc0yXX57oL6NQUkQQLa+R6udsU0WXo0ikaD1pe919u4zt/VAzdHqB75i7Y+/fzuwK7B9VLp+Gff9+f89tP/AHD+2ZZa85gC8fxTMp1ObZyJziYO0JfapCloEw2HGQq7CMf9lPxMtSM57LjSn6ORlRlOaoRhj1Y9bF2Nlw3q/GLBEEv25kp7koPM+nn+bRGrkadxT7XWvNErev46SkX2mVMFi8oOjNGqrPFdusad6XsZG/dAAAgAElEQVTHmCzcTBWNDCMuEjnAud5aJNV4ivX6PP+ibx/fW7kJ047ZLrb43a0WV0wfPSbiJtQNKoNGso5znQQUCNuUg4BXGt9lnw2PGKIq9Kn9oAybF2D7IpQPgJ/CPUYap2YWVFWrZomMNG6jKkUw+CXYWYRr9ycaKGWYKdJCGihFoEiTtPmAdSxNSqaFs9a1FcgoEMgZugnLZ63TgqV7IJA5QsYAjeEENmSaEEjTdahO/Mo023VuarzXBXsO8UinpRGeWDXpr/Q80pktJj+/eMGjvVznex88zdHKIMquGMNyub3BTyoY3eZpNGPXAtGnXLQUNBWHcck68BCKUQuhI7ZGgC8PyxehoxN8CCtLSVhveQQ66qIVjs0b2k9Ba6M7Cq1aWh0oy03cY8hMGfeY1nnrqtnUFbuzKUMEZKPEsBAUjGtVsALfYv7UctFMsuc0Dgw1Yo4NW5sJDrp1zPDn2zJ4GGZGPEJp5aRJU2agYJGAdQTxkwpvTrLgtOYCdffdF9FYNNx23NDpwbl7iwOQbcHiqnXgbWfL8L6rTW67ocYbXzeQNGQof1CCQzGlmuZITyc2MCvwo2BfsbcGr5DE1wikpU8amo9D6z6ZS3ANG74yDWXmkE5T4cQaacsQI6OEgLX2rVi7rHEmBrWIdB6U+1WALtknO1VctZlAcbqCizfJChjKqawqvDKk2obVJcvkgEcxsryn3aQPn9l0P5fWqixcu87J0SEGCoZYQHU5GeVrveKqddl4eu/7Qwa9b5dXYwbbaQ7k9vOuq+tMjaxx37UnuHZvijtetJ/y0N4oNnn3feO/90DdN16fr/tdx9593e8mF0MST5/++Br3//qa+2BJOC5+XCOUVs0SZA2bC03WL3bIpnIukiTQSWuXoWtJQ3cPSzu7DF23KSKddsHC//g/7TF032gX7H3v+VuBXUCnk+svv/ojvOPJD7Jv9sUsNSQeG8fGHQI/h+fnXfxIr2aAtsOan2ef2Do/y5JtMx83WbURA8ZnIdrk4s6jZMNerAKG2xIMSdzVxBCQ9rO0crO8zNcYV/2fPocxfLK9RFR/Et/vIdZj1S4RlDFeGj/aYqd2lqNBmZtKL2bAz7BqYj4R7RBE25TlrY02qTbP0bINZhnipX3HGRnyWFqI+WRDF19FQtuhZA0lAqYCGNQ51hFlhrfYJndHISMFw4EbjOvW3FiwrjA9PQYp1ViVoXYpaRQo7BeLZ8hMGJcV5t+QVHgJNHnDGksZdCIuDib5djrpNlaFbw1NZXxNQThhXONB+6qlczZh9qJeXOemHJ7BIJRmNUroOlil9VJ7hKSN3RO7Ol4F6FzUieSLymdbBV/VUzJRiL0SAyVgp3GsAJ2YH/lWxFYZw6XI8I6HqlznIqX0YfJBmXsb53lFtMk/GT7EYhwx4Rkmc4b8CGxvwdCgYi2MY880wpMZpDWfOGDDDKyvwMKypZCSlksjV/BDbZdRqxorD0N1CdLqzlWUiYCONGdNWL9m8NVGoVFeE+o1uHpKXamG8rFkxJfpTarVzIgmH0nchirRNNIsKK6jAGEA+TIEW8aF5fYK8KwbfvXLEcOdiP1H/URDp/XQYVxjao04VdOl++QoTsPVi/D+3495wx2eUq8IUrApYC12q5KMIJc2YWnH8smdRX70SC+HjqWIpcXT89QtrvVDWscN+xVdo55fmk6N6MXKxQL+YvM0Pi0ZrNgwaePS4M0lI9h4I8nO81ymYFJx5ng6GWEE6vUPAeezFnsGiuqO7YPaIgxOGbbVMlGErJjgHVCn6+IaDI0n57C+FnTS8OUlOJH3GbOGx9sNyibFLZUB3rG4SufiVe6+ecQFLq+uWULpPpU+pFoysZDa9i1Lpw3ljKF3IGZ/X5a7Mgf4jWuLTA/VuPeJhwn/zPDiV8wSDu4Bu//ZUX0P1P3PVuh/4fu6crt8f43rn6vjy72kg6RGsp5xOob6qqWlUYyfIvRSToekI8azM3QZUkGWMJ2m8eWYV//SrXznP9tj6P4Xds/ej36TK7AL6OSy/INf+Bj/5lMfYHL2FhaaE5jMFDbedhmK9VQ/naCfgp9jgYg123IntkygboiYFQyzxqePDvPWstFeJdXZdE7VpLZrJwF0XgIONa5VTEYYDrFmDPtsi4Nemp2dU1xtr7gg4ihuJgxdVHMBZiZap2Msk+WX0vCL1IgYwqOOR9tPO8C5XnuUdlu5IA2eTM8y6/m0olWebNT4eHOHuSDHzdbyPfkcgbEstpMs35nAI+9HzLUbrgljf8rjqGc4fcUyWTH0jeIcqsY3pLPGNRBkx4zzG4TStan0XU5EnZg1CpMRYdHSPgPZY4mYP6cAYN/QXjc0Llp22tIfJXonheSKVfImEm1VJkpE+6qxyqo/VoG8Gg+qbUDeFI0H5YKVU1OvJUZOYE86sRRUL/z/7L15sCXXXef5OScz737fffv+Xr3aSypJVZIsyZI3GeOdpj1gAgigh80MTMRM9AQddNDbMD3QYAM9zDQ902CMNzCrm8UYgy3L2GBZ1mLJKqmk2tdX9fb17jczz8T35C1Dd/DH/NESMdZ7EdKrt92bNzNvnm9+f98F8oeyv3HXMrDyDUG9aBtto7Zb2i4xUxOGcAN+7ZGv8WRBZogJ1lyPnWSX10dV5tJB5m02Sh2oGEaHM3zY0gi3Dlurjm7bUN/OANTKiqNQM95Qcu1GNqLurhtvpCj3+1jl0FzXds5BRSNg6b5qWcaa2NFYgFNMVhfqV7J6qlJkGD6pIOB+lp/2yZQhPe+In8vAovSDYr+0W7zWcNLgc9REaEV4t2pjx3Fn3vCrL7SYiSz750KcYmG0H+Uwze7RPSC2QRb/8aVPO+6KrK8Bi+VwNZm7dnwadnbwwcX5esBj23W+7Hb56feNURi2HjR7kCVgnZfOUdEJ2u9CwM5n34lpUxWcNHM+U0/uZTmEBSjVKRyAVbCzzC360ZB6h7OKNa8BFNsqxlYgX3Eoerwd6J6DVH+rOJmioaR4loZDYHB7NWPv2ovQ3IWolZkb1CYi061kB4UODFYMhwcCz0b+m16PIRdwW2WET5w5z362OXFgkrDfwKEKO62HMgVJgiTmVHl5fl1s67WlVDolas05/nj1OW6bLvAnl19g5kzIPQ8f9JmGt65J/x8vYa+qX9sDdS/X4RZ4CwxjB4vE+ZSlJ9reMn/r6XxcSppp5qSby+JT+hq67mafobuUmSJsgXxY8ICu87zjnT9/D//4h19PZW/kemt37n1+mfbArYun4hJ+7V98ih/78O8xffQEq50DmOJ+r9cKky1ahQXeFo0zEZS4gPMGiILN0XZdFkzEqos5HuQp6C7e5tl2XW62XiT8hu1T0SZF0t4yNhwhTuscqL2RfG6c050bOJsjLyV4ustnG89Szc3SkzHCF1wqJkRK8g7d7g1mB15HGE1QJia1ljEcq/EWSz7iv0WzcYo0Xkedsy43zp+1d/ntXfhM8wJbzed4c/EYQ9EQR23KkZJlugyNtqNtAn6zs81dwGhQYC6fUGkEbFk4eJshp3otheR2ob3lKEyqXcbQTiBed2h0mhQcVoHCcnEuOeKL/j4vE/FLnKcgW43VulCsQWm/YfMJ2LyhQHMoFGGxBcEiFHvQi+S0NT4bzDtX++O6eNP5RVNOTeXeec3XqstCfWU6uOQoHM/aGfwhkMheTGRGlPrJgmfqJEITaydZY5ABo1Mvnear0RDDQZXUD9xTHs5Nk08D70g9OGQoa6TXhJp6SIXVBeQqBmHp8bdkI2drjK8Fa6zAlrri81lbg16/xtgbN6Rvc4wPG6qTkBYNyoETQ7XRhVIbmv3GDLU8KLpQ+361oygViG7LAoMRYND7Q6yQ2C0xlqFRY5wHvjoevabzQbvqXFVEykBBiQcCpoZzLcvFs4YxDFNyLYvJ1H7ta9/cebB1eOw5qCzBMZknajA4CnWBphAGC7By2pErGobuhV94bov/4d0D3HtC/cUuA/qKtum5jMoU2lSgfbtfRab9L4ZO7lWBPY1QNV4VUSDWTuNhsa8amattREaXYlYxpreN/1117no3rQKSHXKuazzeXfFNe76jOFTvrsb2C4Z8Aq3rcPWcQ5E22qeKollagoYaKdJsU6LUcG3JUc0bSr2AqZ7hwzi+RSz32DTvP3uFqXib+++bIqf8Pp1PGyqh7Qddy4sjEKr3xKhhdxmSyymvO1FlX26Ujz77NAcPDPNbX3uKqS3HPQ8dyTSnqWJ7+tqIl+n69//Hh90DdS/XUdO5Jvo9b5g4WiIOU5a/8l8Cu+xKk22AxktBP7bEV381pKEbILJ/h6F7SgzdvbznfW/cA3Qv13Hbe9xv7IFvALoWfPRXP8uP/Xvl0N3NWusATs0NIndclzgaphSNUQgHUMrcurEI0MWknuSZMgGztozaqBLneFFmvM4V0p5W1zxprHA0jfu0SuW8SzbtrVDKTTFduo0ujuXOTTomYLlzla2kgVMzhSJMki1cbzmLQkm2CYoHGS4d8c8lVm/RpVwBXkgbxIo+cT1c+zx0r5KYkMHBtzNUOszx0gID0Qirm3/ho1FsYYHRnuOANeR6sJgYbqSOq6QcsTlWTcihkvGhw5XXGuYEYlYhVIL/WpbsL32SejTFakgLFddhdxEKGiHtOHbPZGPPwkGDXK7S3fkPUSB+RGY8AFT7ghyji0vw+UVHeNaw/7WG6KjASh+w6LP+TgySAJo3JagnNNMx+YV92JBc77slxfKNGL+tYv8ElMR8NRTWW+3nnQkgCGPIgSmzwjXLUxc2+PEXLjBfmaAaDrFNwkFb5EhQYDYjlZiIDNU8bLczglDjUpF+Sd9BrDF26wqUShLpy/iQjWeffsF59/DshB4oAw+7mzByu/FOWBXFVyYgycPuOajVoNuFlXNQVMDvDmz3nB95Vq2howV/OWPU0jOOIDYEisjQ9wU2o2xsbYahdQaUjqORrIwk1cPGZ9/5jtUteMdUwHMXHTdfTJiXgUJMk8wGvSw+5cKLKS9+2fGmQ9azpcEI7CxlDt/RIegJlOc1+je8uNHmX+2u8O/ePc1QLauS8+ytnw33GVO5a8WW6vhpBq1zQ2NcvWUETAXkVJwiIKsmkP6x8p6ifgagGEWvzRMQ7fRdu0t9TV3O+PN17TzeyapzUsd956bzBgY5eHsBVI8b6huwfQkGFYPSzLSNZywsTGcmH3XTlicNZ9ecP8ePFgPf9XshcdQIGa+M8B+e2uDAeMKdxRruUoI5bAk1xlWXbVUgX9mE6taFfGKo1ByVEcNkfZyBKM8f73yNQ9NzfPzzT7F/JuLE3Qe9XODWNar/ztn7JBzxMz/zMz+ztydepj0gO7nXVRgmDhXpBQlL/zWw89dNl7lcPUP3KMuNixSjKrlAgK5ArpCn83XHO3/hbr79R15PpbZniniZjtjew/b3wK2Lpc7fX/vgp3nfT/4O84duZ6VzkLSoePgO1qX0bJGqrXBnNMawsZRwdMkcqupqHTYh+xUaZgxlrAdmLybbbLcvkcORakbowZz6k7s+qNhHmqR1tm2FTlDBGEs1p8T9FRbrT3Jv5R7vlK2HAwQY0s5lkjSmagKma29UCgRLSRNHSN0l3Izr7MZb5BSBohy8zkVMvEkaVnlT7S28KRzgXgxnXY/13iobacy78kO8Oz/IxW5KJ1YQruGmi9lKWpSCCpdJuSeyXo91rJKVn/u0/h3oNLPxmL9RE7koWVrReE2eUbemxnbCsGpmUNeoHJzKA7upwGKD3Xa+t1MkZHzNeQNBbRwaHeeBiIBAU1l6Mh3IHSqWRk+iUWvbEGukq+mywFgn0/aFBcP2eceNszCisaTATR8EcsH5XLjOqqP1dWXrZU5KjYr1OIpd0dgvWbScWrzO7652GM7XuIalHlR4wBQoY3w5hdq0FA9SyovdhJzMBKq4krhePaICIwIM0piJbepCbxtqiiYpwqklgQXD4iJ0pLvaZxgRnnwx20bp1mSE6GxDcRx2m1kG4IVrjoub8NVtx7HU+rgMOWHVDatRs5g0Fc4mt0aAGNqrAhHOVziGAkbqjVVl2FDmRF4/B+uLjqHQEOYMx+fha9ccy13D3CFDThq3DvRC+PTjjuM7htEJ49sawgQ2+4BL0+9WX5NWz1k+dP0Gb70j4j0PDZEq9KeknSYgnwU++1GpQJ3YOrF3YgYF1MTwagwu961OIo3FBdbUlav9WgF30WEUrKh/q7O2P7L3v6MxuEavygGc/tv2iInD+OO8cT4zSTRXwAdalwXwDeFNuJI4Gh3DSDF7LeMWLnm2zXit4/i9xlemXbgK86OwkMsY5cQ4hkyBrcGIj3zpFK8rj3LkWJ/maxo6S87f0Mg0qNfS0k1L2fiomq6F7XOOhw4usN3I88iV5zh26Agf+q3HmJxMufe+I56pu3Wt2rt4Z3tgD9S9zGeCbgo9sMtbpo6VPP2//EwnS/vWXTDOM3SKLXm2b4ooRgPfAHTS0LWeSnnXL97DfyeGrlbaiy15mY/Zq/3hb10kdd5+9Lce4X3v+wP2H32A5fYoSW7a13jJpSphT5Qb4925URpY1nHcSNss06OqVoekTslY7g5Kvt3oKdflRUGt3jKd3ipGBgpldnhQp1XN+Uw7T0OEQ4QCiDZHnYABNVK0L9JSLZjcrS72bJ3GtYE3WLSxg2/BRaMYEnZNxIQNPdB7PtmholFtUPZ9smnz+exWKqgyWj5J3hZZJuZz7UuUnaGjBgxb5GBQoxoYzjl8/2liDKHJUVAftIv51jBg+oBhbMb4mij1qsYaX6mfVEzZaAbeZAxQ56sWa8WQiMWR489czVoMNPYMagYrgOdHZsa7XrnsPCt1cdHRWzaUtgy3HzBUhn3Zg2fxBBwF3OSyDeYMTs7IqxCLVZmA3FwWZaLHVvRJ5YhBfaoegQkISDCvbRFrNWIU8ZdNGMLMtStmpnPTYa1lZ6fLTz7xNJfjCtZGNGzAwaDKHaK8HCwU4UDRkC84cmqFkDGg6DDNLHOvJ5OEmLAANlZhuwErW440wveK5jqGHbUudA3DsyBz7fAgFEYM1QG8kSSnSI1DhuKQIeoZ78wcfa1hXKHPywY7YTj+WujK2KFx9f6s57QrDdguNC/IeQk3rmc1ZDnFtSj/bht2FLGibV7K9mlBgLyQgZjTXdg/ZZhKLfWn4ImeY2Eh67F94UuO0RuGu45amvLEFWFlHZJi5t6NpKUTUEzg6bWY/2V7g9/8jmmGpsPsWn4LXGsGuaMjKjNEn50T2JHrdDNjybwmUsybxrR9V7PuEGSA8YYWaSmvi+XLdHU+aFjAT9VsitmRkaIFTek1X1QcDkirKdNCaQH0mhUPo27iQs7Qvu7Y3IH5OUM+hbykAWWIFS5dh/WO8QyeMv2ibfjcTch1Ya0DY6mhbbIqtFFTYmQgx5+fv8ZbjtQYTPKkGwm2bsgpfkdS0CVHSUaKJYdCkWWuKQw6bGJxZoY/3L7KWuMqB+cX+PjHHmHhYHWvK/bvWaz2QN3fs1P+W3+rP4n1d3urGzc5/cgSxYJupXTDZWn2dnhh5cssNy5TCCvfAHSeoXvW8Y6fP+mDhSuDe4Duv/Wx2Xu8/3IP/F1A92sf+hzv++GPeFPESneSJBjHSICVNLBhjSQcZDo3rht/cqSYZIcXXJdD4SA30zZzBJxzMQ8by5SJOI8iQSzb7ct0TOjZaadYeT9bEu0hOkKi8V2Cgqx40gw1yRcPsNs6R719zf97Oxym13zO0xBW7F9vhbR8F7lolC0T+hixYyZHbAzWxTQI/Mgt7lwnbr3g7+6TZIe5cIjjlZOsqidWz9W+wsVolJlohDO2zGtzlhqRz00TsPt4vM33BnmuuoAThYB7c8ZHZqSKKuniheVWYcJqn5jIoh807ks08trEGwTyYqe0SA9nIydVUim7ckPhr2oa0JhtO2P1Ui3o16FasgxFUL7D+ODcqBtQudNQuNMQDVnC84buoqFTtj6rLK8qKZdp/IJ91ueYSb8W1jIGRBo3ewPUZ6osWjkmzVSW7VbQaFEs0JwhLhh6F/GuW3fZcv1sm1+6cY60fIBhW6DVXWIoGiYg5buCiNdr4VcuX9t4L0GpBusaX4uJG4MLZxztxHhdYGEIgh0YPWoohcb35mq/VY1huAgjR/rATfty2JBTAPNwttCLlZTGM1IX65whJx2gfDKrcPz1hupJgxzH0X5D/oihvGAoFwz50axiK1acyWHj93WzDfVGBvhkkNi6DhWB7sRPbn2kSVKCr68k3FaxnmFU3t/F5+EryltpOP7qy/DaMUNNo+EubKzBxaZjVvo5AWcLqxcc7V7IR7+2wQ9/V4G3vbaGK6bexZvZgbNRq8CYRqne3OLBWz9aRmYPLSQKEda4XOeQtHJCQ6qM01tHXwoMyp1xKfsdLznTGFYu3T5jmkqbdxFKZSg9rFxEQ6AQZkkEFEo8ZIjVbatzdD7bv7UZePElxxXVfu3C6HHDwoSheB1Krzce7Gm7hgcM0QoMapQawOl8ytlurJQc3l4Y5qOmzo3Hz/Hg/ChDkwXarazPN9JzatvbECk3UBJZ/zp1zjom4pCZnXHOtE+x2wsYm5rno7/xF0zOlrj37kN7jF12+P3/dWrsfbzce6A/bjn1+Hk+8f7Psbvd5WD+BCOFOZq9bc6uP8lmZ4lKbtAvdKHcsPkczScS3v3L9/CeH3kj5b2R68t9lF71jy9mTouAwl8//omv8BPv+20Wjt3PcnuC2NQyY4GSc3MTtAv72CehfNqlbPNUTcQpv84Ycs5RcD26AmmmwNNpj2tBxIzJcz1eZz1teK13rNleopmSzYRJWqFEO+lrm+9r5nZJusuY3gqhOmODig8yTopHSetP0gmGuL90jBv5Ba531xjKWazNk7eGF8hxJ4aTQZGNtMuz3UUirYYSmCV1dsJhAlv2pfM5Y31vrNT2Ys9FaZRrhpFmNuJTZusBckh/LmG7TArV6SzTbe15WG04Jm43zA/iHaulsQyQ9XYd5oDBnXPeaZiKZZMTc9FhVclVkiDdMGwdGgH2pI1SHEoXnviCI5hIsfkUtxYQzsV0N1KSmy0l75FIj7WReEbz4rOO80+njJVLDJV7jAxnRor8SkprJ/TuQi3Wk7sRbhS2z4J614bbMJpC7Z3GBwCbG9lkQQG33uhxdxZsy72GU59OWP7rYY7dOYXpaTa8zUsu5a0mR63i+OQG7EtBo7nrigvR4u8dpg7jDFMThuadUNBY96rzucZqW2i1YOadhjOXHe0/AXcPbJ9xVB7IdG2dZxy5N2RsIgIb+w2hQJz2ldo18rDzKCzrGKgIXkJ8oXS1T2iMKRDYyzR0JY1Hj8OwikuOZBNMRYl0zsGzTznKd0KtCesydwj0bEKxZKjKQa2QaOfYjOGBWcMnvpbwunaLvzhUYHZ/RKPhaKmKbAhOTFrKVdg65TwzmCtYznTanD3c4GfeOAM15zV5Vm86BQSr6kzsm5yv+qj3/63vCbDptB0B7uiPT/W2EVLSiLY/ahUwV1yIN3DI2apGCbGjGtPqLTacOV2VUZd/ULrOfmWYnNNy2Gr6K7OOnLEawVfkDMaD427bMTVt+OzzjoEB2BDL3MwiWrb/zNHWSxoy3F5xXB81PLcM9++DoQHLB58IyeUNj3ZbHA1H+cNSk4HTz/GBk/dQKkZ0nkl9pIx0ir5dcNzQ7TqCDiiLUc7iXiPhu6fmqea/m/dd/ggj0REWbn89P/G+P6BYHOAHvu9BP5G+dQ17NV/M90DdK3T0T331Ah/9N4+wfqnhQzcvb75Ar9pls72MKsBK4YAURn7MEeZD2mLoPnB31hSxB+heoaP0Kn4aD+gyd+RHPnaaH/nBD7Hv8BFWercTW2UnSvPWwQUlKvk5BmyZorE+puSahxiWQRMRG8sVF7NjQtT+cMLkuOC6fCne4q5wkMXeOtYlpCbnHauqBTPKtJOL0lhcvI2Ch436kPScaZekdQYXr2Nz87ikhUs7/r+IgF68wp2Db/bZHtdtRM7k2DU5TsW73JEbY5yAlbTt9XBhsoO75TQ31ocij9jI69EvKhIsVCHllo9muzOs8LOrlu+1KWM2YDuuc2dkqJuIr7gWF5KQg9shb5iAjTIst2ChB5uX+mBvFxpixJYh6jk6KrLfgdqsoXvV0G1Y1top8WCPwjhsn7ecvg7XyzGrtsfWjQ6PXgs5dTMHzR3eXBzkC086HhxI+dacY/s5R1nNC86xax1xMZNfDbYdCo79g0uxNwwc/ZpAboqLLPMduDbgeHoXFtTr6eBSCCeHHM0/7ZIWU+bzBbrNJjPlAscOhozNOx+RUUwczyk8b2yetnSFJmIyP02qcXtY5S/rjps9w0vCHQGcKEIxl0WOTO0zBCUoyIhxXnmDzgNigZT4irSCELZhrgO5H8gAw8qyY7hpPOsZCrgJoPWNAfZGVldmVXHWgd3nIVeH294m8X2WG2dWspGjGD0PbmT+6GvJzKAhkfasA4WJjPXLjcG97zUoJqZ5U+HOjnjDcN45Duwa0tRw3ToOVwy5BhQGoJl2eHsz4LEtuK2ZMhMZrj4DtRNQLjuEe8PJzP3bSw0///kV3vttOaZqEalNvduTxMF2f77fzkavZrp/ZyVAJ/ZKhgn9aEXdxNkoFgG27ay/1d8HCdzpe5N9wlss6wXnW0TUNOH0dqrJrOO88zQ3Y0j0N9IaDjrvnFUYtkCiHk/jfJ9xKCo3b+gtZhKEN1YNK3mHAos3z8JGyxGsZCYQAVO5sYsxzGnTx+BkxXDfiOXyJmzYNqsmz8Hqfn7zmVOYylP88mteS2ne0pPBJQ+hCHrpHnV3GTrfvNLeguohRc4kPFw+zPe038rvvvQFZvefZP7o3fzg9/8+Lp3hv/+++awGTei0j41fjVf0PVD3Chz155+8wG/+9GfZWW2QGwpIexKTt7m09Zzn1GWI0NhLJ6JyhRpPxB29zhwAACAASURBVLzrl+7ec7m+Asdm7ymyPSD2SSO605/b4Ud+8I+ZPTLFZnKSJBgjsAXS7nWvM6NwgCioMGEcF9IeLRuxRsqQ61Ixea/1XnQyQDRxJiIJ8ww5y1a8y7XuCr32JUyyS9oHclmmRMH3voomdLJgooqsDNy5pEHaOo/JTWGjUd8Y4XrruGSbmITCwJv443jLp9yPm5C1pEHBdGgYyxCOFRKe7q3SjbeIEgVijWGErkioWuXnaa1LiZzhcrxLqNcaVLjLRCy4LnUnoGJ9HMo/LsKjnYQnem3+ebXI4E7E2ZuOag7mLFy+6RgLZQ6wGEWOaIE9ZujecNTbhlNJTLiVkjZ6PHm5xf9jqxxttRmqx+QaNcJyCm31d+U4VAn5kQFL11lmB4cYLgb804VsNDfcKzFfdpTH4dwK3NhVjyiMlQwNtTkEEfN5OJc6DtUst0cwqFBfo8luym3bma5sp59BNzFj+Oz1yDNUyxuGo2MFLu5anj8Tc7oUczIIGVkL+MLa80wVRrjcGuAd+Skik+NTvSXeHo34sOj3lh1XE8dBYQPFuNRh+gSkZehEhsJRGL7oMlep2iaWnBfyt9Vn+4JjsCgi1jCyD0piacgqFTtnHLld942WDNWWpS3YrTraK1BpQvXOv+3DFcPmA4HF5CnodjWrp0JaQmnTEuedwmY+exyF+QrMlMcyRq9SMJQjQzoEs94s4mh/ETYa0sQ59pcNjyUJZ5s57sgF2Djlf3uyw/88WmB8DIZ6fe1kmslFw8Dw1HKX8lDK97xuHPIa24JbdH3Tg8LaHHL3apxr9vXbJFrqBdbPxNRlLmeR36kUC4o2mfnbWBWxefp7z97JcNFxXrspQCewJYLanXXklVMop6kkAWNgVLEswChGT+znYkZkKwBZ5p3gqMF1HLnpbCwq80W9ZVg5kwUJl0vGyx2WlBn4ovNyhOI+GN0P1y5kLSHzMWyW4GB3gHba5LO9Osdnp/nQpxf51sY13nVsgV43oarQaTVhpIa89IIyaaxBqHia2PlauKlCwE8fe4jqyGU++Pwlpof2MXasyw/9kz/iwakf4ui3DngmXdeyV+vHHqh7BY58u9GleaNDUBHbkT1h/y3af3YFL2bBpc0XYt7+/pN8+14O3StwZPaeQnsgTSWEN6wt7fIr//Y3kJCrl7yBthvA2gJOynIXExQOYqNxRkk9G6fKruNYBonZdoYdgSgH+03IhC3zR0bTIusDaksuwbiUzd4SkdwAAn2xaBTNeULSnkpIlSTbb5X3t9p6v+h7og/aOBd7Fs8ZrSC7dPMHyRcWyNsyQeeKZwHzwYB3tRZshUtplzWX+rgV07lK4lJv8ohdlyjtcE79sA7uc47PuZhOvE4lHGTLFrjiupSTFstByBIpJRkp6kUedy2GwhwDnYD1fEq0i88ua4cW2zQMTcN6rkeUpFw5a/lcs0Ojs0szCPmPG2UevhYwFzjWUvi1YXVeDnCt4AircFvO8kw7ZCEwjITwfBuqgeNkPsKFzmeMBQuQz6de9N6pO1ptqGl3ahoqPFiERivLD3uoDMMu5WYKS9swmzO+smu6YGgkcHwWwiLs1uGNhYBjE7A+Dt1cxIMHHa5s+fqZgLlGwF/aDb7S3mbCXacYDnPTWEI9aXeJxbDEfDjCWmq4PSz6Roldk3C+41i/5LxBYGpfNsrTSHRuHOrPKFolK00Q6zV5GG6+CNuLjgVVUN0FvSsQXHfkFWlyVhWLoKBlsUutXlYFZq9DOglbbbxL1ml0KOJXtVhi9gQSLikyxhEqU0d4XmBWbRv9SbzaEvS7kh0IC7iprFkhaENZp+O44d0a0ZYgaBl+/3di/mWpx4dyOQqxY9NY2kHAD1/Z4DPvGCS1Ac0r2TizpDiaDcfPf2mJD/zrQUZGQxJNZFLjj6nAlEfbPQG8fp/sqsv6ZYVL9HM5VhUpo7eD3i7z2fe9C1aVbh0vzvZtEHJA+9e9mQVTy1EdLmTmmVQuZwUSC9hJ66cgYgVV9w0UuhZo5CqzT30ZdI9hBSoV3aLXHhiGTjqKW/C5z8L9h2DqsPEGlMIpR6cC29dheAyqBXwG4/oFODlomK84vrJlGDE5ikGVr/cCJm6b4HtPP8MTtwXc99o5P2KtfwXMnKGikbKcvho5f9353ll5QZLBhNsHSrw7fi8f7H6IsLeFy98B0ZN84Oc/yvvv+EFGJyULya5pr8Yr/J5R4hU46lP7xhg7VuHZP7vs2YUgstld563n7jN09Sd6vP3nTvIdP/YmqnumiFt7Z+/zy7gHxBB7QLfe4Cf/9Yf5yB8/y8yhB1lJJoiURScWubtMqPlUMEjLdZkxEUNBxZsg5Ertpi1apFRsgXWd4WmTDddjV+DOtT2AkvxnIN5guX2ZIK2TKNRNzJmAmsauWlljUSsOIwdrUMBIkiBKRt+zZdLujawOLO3Q7a1z++AbWE3qTJFyOBrjfOs0YdrC5ibpph1W420GTOTzyuLeEi7e9I5bE29gukuk5ZN8d2GBB60kTjHj3SVeNDCZmyAkZE21ZDbHZtom53oMBgWKGO62Ie90EcfGDEcOGzYTx1c2mqwnOzze2eV/XGvw6Fqb/2vdcmk7IuhElFtF9uUi/lkpZN7kuTsqc7Jg2eg5LyJPwqxqVV2lImb2lQzXexkgClLHSgtqMh1osjSIX/Q6G4aJYcNsKctN09p/bMYwUTUMYqgpL27AMCCkEhjOrjhfRzY0o5gRRyEwiCWLGzAzA81N2EkcOwHMTsLimqO07SiVA1bdMn+6dZZKcYH56mu5HpS5aENGTY6vJy0e6yzyyW7PA72iabObGhYKEdORYXbU8Nizju41GC5Ap268lHKzDUsbjvG84cYGXF1zHBgzviItUhCxOkN3Ia+2B+WjbahCTAHPWVdrpKw9nUJyl07jjRDppoT3BiswIKZsXU5jsGK/VIEmvZ0cvxpxCvRJbybppwDeZHYPwfn+75SztgMPPgNDrQHFAnzo8ZTvGwjolWOSxHIpFp6y1GzEtYYhv2uYEChJYDsx/N6lFrmTMT/6bcM+sNc/n/pZdaB1QHWj7wVtGROrMbHCkX3OnH4uBs8H9vYd1eoNFuso0Ke/S5SZ4nxnLzJXCPj1NLrOOmBbL0J6zZGqveio8c0RRqNXmXquZ5mKejxf4XbVeY2i6rgihSgLAHYg0XmZN759w9SycXp4I7spUD6i2FgZWIZGIDqsr42/4RH7VliAstzQHUfeBRzPB3yqE3hnfMsa/tPzX+Mf3T3N7PEKdsoRyQUcZ/u+3YDeNcl4jQe45RlL3ExJXygxN5jnP+88RjEtMjQ+wZeeusDy7jne8NAdlMs53Ks0nHgP1L2MC6Z/6P7d38zCGCOHyrzwyFVilxCEGbDT9VY9ha1TMW/92Tt99dfA0J7L9eU+LHuPL0F8lsi+st7lp/7F7/HRX3+ahSMnWOxMkisewqhztbtCEI3SDGvMm9DLdgZtgS1Sysayn4AbJkKjT7VHjLiYZ5UBR0pdo1QXs5k0uB5vstw67Vk6Fw7hOiqo1GqwQ9pRBoNWDy3E+hutONLTdUg7FzCRLyb1LgLXu0EubdIbeJA3FxY45Jp8Od7FaQQYlGntPE4nGuT2aJQJHI2giMTozeYZn03n8yviVZJ4k9zAgxzMz7DpDIM2R7e3wmksx/JzSOv3QrLLs+piNqEHl6WwQuwcXZfwXccMT/Q6fOL8Kh85f5Zf7Dg+meQ50zQc6A5wMq1yNMjx5lzACZsjKgQ8ZGBhEMbUFNFMOZs4nogdTWP4ZMcx1zYcDg03E7jSgocGDRNDUOgYhnIql4d2bPyI7Fzdcc7CoDpVpfEqwfk6FJp4UFeYgetdL6PyrQ0K6R2swHjZcF5xKwXQ2HX9MsQa6YWgBTRtG0bzyr+DqAktjXar6und4pPrFzky+BBBOEiFlPFwiMthlbLN0QlHeDA/wp8kXf6w6/i9bpO0t8VziifJBwyrx7UbMJo35HOg3lqNq4fL+CaCXQFLZdIpSqULOVVsNWH1In4Ep27WwhiEim1RFZZ0XMOGrR1YXXEMTxiiY/2KLQHf7QzUWRkQZCqQ61inlb5W/62YKp1vYvb0b4FABfsKaInJ0wxLRLIAoJo7Nhw7l+HJJcexnmF/JeAdoyFbdcOaauFSCJzl8o5hp2M4ljNExvC1eswPLa/xGz8+zmwx9D3Baub4hinilp5OuSrNzDgjwOkuOR8u7LN0+kYHvfn8dgrkZQlAWV6dtlnAUHo7GS02+uHVWnvk9JW2LoLC7SZjMfV9aQkFHisCxtIXyiWhRg7jAWTo2y20jwwyyQRTWQ2ZwJYdhzGFM0s3WoLqQUN+ylAMjX8evX1DY8jLfVyCxhqUEignho2uY7po+MHJHINpicddjY5b4dd/9wo/8NBhRvcFOJdmVWCb0Dud9d9K2heMZ/mJzR0YO+C4e2SSrXXLl1YvUsmPMzo1wBf/8iyrW6s89LrjVEp6rFdf68QeqHu5V3efUyemwTB3cIKRAyWe+8xlT8GLsdMdWf2JLm/7ubv4zh9/mD1A93IfkL3H1x74hkusZ/iVD3yZf/9//BmHbj/ApZ0JHxtiQoVRbfoLbC8scsLmuWjzHAkG2GdDrrvEkwJ3mohhEj7luj6MuKlA4qDMti1wRzTEvWGNHRv47taCtHRi5zT70Tj2lsNVQcZBERONewAZ5Of9qDVtX8D1ljAm77PnjAlIO9ewtsg9gw8zKrAVVIl6y7wkNs2WiOVeFWA0eeo2ZCNpYGxE3Lnqx8jegKEQ4u4ilYHXsZCfo6w4FmO50bnKxbRFaotcMgFh2mRWDRcm5EBQYiypM9S5yp+7Ds82tvji+jqfahQ5UprkDbkR3hlWORiWORIE3gGqPvaqwscNjFhYSX0XO6OJ4bJYqsRossQRDN9dk5bN0EphQIHCAzCaM/S2DdUS1AYNY5OGTQMra3B12zGxD47WDH8j44A0WglMlgxpDx/EK2PxUJx1gyoMOK9sN7FbHf1nWGs64jADd6s3Mzfj2MEsVqWx4thJ1FVqfGbYp9ev8OVOh7srxymbgLoJCV3KXb5X13FHWCVvIt4elHk4rPh91qDMf44tv7nb4IuNDkPG0Sk7qmMBo9KDLVrf+9oedcweMtirWfRG6WBG4koPpvgVjZVzEtHrtImUaebYfSoLL95qZgaU6jHr97Nv1RDgEVBT/ptGsatZxZZ3hSqUeRDim/34GLljpakTSSw2bLMfFdJn8HzBaf+x3v8nPQ7uGN42Z33f6faW82DlvnHD6EDMl1uxL037YpzSrLcZbhb4wHNr/NSbLO9+qEZacn78qwYPmRYEkPw4WGBTDJvApICZdHV6Dfpa8SZySavqTKYJ/Y1ib7wNu98RK7GkGDsfSKy6jj7oU9aQ4iRzhp3zeMewRsqKzNENlLbDg9vxLKzY31cpuHgg0+bdYi89sNX+EAguZU/lx9wVQ+uvlCMI7mYWQ6K4Fx8bKeA8mvWdtF/KMKjG24Mdg901TI06Rgrwh0sRd9fmuNq9zKHWMq+ZmcO2jddMKs8xf5shtcZHvOhcSNpZaHKo86gUMhqP8yHzDN2tLTpMs2+uxqOfeYxaMsUbH5rxrKNz/cy+V8mlfw/UvQIHWnc7t+4YZvaPM3KwzAufu0YSJLTPJ3zr/3oH7/nRNzAwVP7G770Cm7X3FK/qPZDdwT75lXP86A9+hIEjU2zVJ3Amj83PYcSymYB2bpx39MehR0yerrE+KmvQOT/WG/RBvF2kmZuQvs7mOW9Cvj8o8g6b87VaTVtiLqlztbdMJI93dylzt9qCZ9cU9Ou0msnwYEKCyj2Y3CROszcjUVHPl3S6dAebNAjKx0mjcaZs3o99G0mTm52rhFoJZahIdtnpXKUTjZKzed8j69pXQGNXPafrkSYbzEUzTBcPs5xBRi73FllxPZLcBDmbI3E9CkmTUv0ZH6nyF/EWz7cvUQuqXIxHmQ9GeEOhxptsiZYxHBaLR8IzLmFQRg3je+TpGbhPvaUO6l3Y6cH1FJ4wcMQYXhPCsXnjGyYEWZTle9dkFgRcDkEtCqFyv7bg+rJj3cEbJi3jqWF3F7Z6UHGGhbLxY7K6dVTqGpWBncx0YcvSoSn3SyBIkrNcxqBcjiEfZ4uonkPAUDluowuGQdU3JZbzG23ed/0JDlWP8kB+mhdcTGwCTmA5EuS8vk5HqWYCHvbTP8eIiThoIw7LtOEsb+jleDZO+d+3LDtJg1PnYkZDR6+jG96A8XtUh4XvjpU7VXlrkTY0gOVNGKhAUIbumvFjRDVUBAr0XYPagqF2v8HJLSr9V9fgtvS4yt/LQIpn6QTY+qeTXJa+hUOaOYEkxSSKxVMFl4CggJHGkzJxlOHyDXjyGcd7H7AUrfFhyTfr6ndV3AlMTQQ8MBxwZsmxmdvh11vbXGzvMLov4Z++e9LXaDlVfvlxqV5XNk70zJqeS2NWuVl3dZJkINQDN41WtT23AJ+Aqv7TSSUtnVi5KPu361eiidET4Z08m5lHrIwGIsBlqi0av9/838kkISwonaLYvv5rV8yOB5fKHNRzab9JiyjGU+HYOo8UkTKiwOosX1GtKAKd/jnGDTdeyurw8uMZUC3fZlB+YhSosg/EzB6vGI4FYqVDbh+c5Ze//FVmJ9qcuGMOeyOFEYMCvRWKLcJQ2j+xi4G0foHB5VKGOnnuaY/xuy89xdDgMBudkIE5+MyfPMm73rKfmQOiYF9dbN0eqHuFFvZbwE76pblDEwwfLPH0xy7y5p+6g/d6hm4P0L1Ch+JV/zS3bjCe+tpl7v+eX6VXz1MoztBkmGjgNX782VNHan6aO23EdFBm1AQcVKsDhnkMEybk7TYTfX847XGHLVJTC4MJuNdYKkZNKYaLTmAmYSvZYb17jdCWccR904NmW4Efb+qqLUZOBfEyUKTtiz62JCgsYMMxXNr0dWCpKsfyB9jqrXE6rNJLE87ZPLm0Q88lPszbGyq6V7CqHVNMyS1ThtorBBKNHIsbrOG4o3I399s8GwY+37lKURRm2iSJ16D5EtuNp1lPWyzmpqjaso9M6RQOcEdU9exjLm6y3+Y8ieHjvki56TocMWqzMFSDTOp0RcJ+jVWzZAqvZ1e7k2LgchbWFIArt2PBMJEYtncdRWNQIG5HfgQ5CJWZZjO2TcXwAjY3NhwHBXaqasKAQY1OI8PGhkMMhWI11rYdq004IHZIwK3nKKo2rCn9oGFty7EucbwVAWSoDWUYY33JkbeWxXaLj29dol2a52hQ9a9t2zk/fr8nMFxxCY86+NkgpBwafieB91jjzabyHLw1FzJqA9o24kRgaZiE/9QL+Fh9mSfjJjO5Cs3FLGMu6mQtBtZmGrr1NmzuOCa1uIslGjVemycNmEweAqGV+0zm+myZbIRaAzuTjWh9FqBAk0aWGl8OZh2vYgHFRHnApHsHATopAMReCcSItRN4KMLZF+FrjzhmBwLvMBY4HhuTPjDLy6vNQmMF6iuG2wctSSFivd7jsdNf5b0P7OedDw3g5IwRgBRIElvmn1uZepm71X9f2yR2SzVgGn8KbEqMqu0XwBWwi/WzvtNV26oqMj2Wtl8sn0aranI4K62cI5KWUA81bkheygKx7bTJXK8Csmqe0N2Hwn6VR6N9JMONAo312GIMFXgswKcrZwSta87nK7aVA2RB1ct+vL0KjZuwcUHRPg4zDhWBr2tQr8NOKzOjFLRPL6lxAvYrp7GR8lWfC5njQ5/5Ou+57wDTFElrYtDBKShbNyiqttM2FMCpaq5j6MaO2doo7WnHF04/zejw3XS7JVrRLh/84mN824OHmJkeelWRJXug7hVc4m8BO32eXhjl2FumuedNR6mNVF5VJ90ruMv3nuq/2gO3XGHnX+xw54mPwdYm4/uPsZmMEqhKy+QZIGG+uMDt4RBHbIk2hiNyOyrw3aU8iWOfURmXpkMGZyNfnXWElB2MOtQ9aOlIn4a0RZZH2xco9tZJPWutVaOHlQ4urGQFlVoy1HpQuh0TjpH2VnCdC6RK43UdYlJGohGO1t7Ebtqi11tmvHuNgaDABimj0SAlBQrH29i07eNU0q4glPWOWa0CMlpodOuMw/SWSG0BW7mPayZiI21xvfE8tnkab6rQz9Omb0vIFY/6TLt69zpFW6QZDfMttsCECbyWcNxYWjhPZlwEHrARB4xF+cMqr1dFpsawwglam0VALBt4x4KhFGTF9wrQbYeGlRuO/bOGrXGD1UirmnWSLl53dBLDQDWrIXt2TQG6xjtLbQgHle5vYWvDUS0augXDgIrqVcs1AMMjhgGBoh7UA8P6kLRoMJ0z1KYNAzXDpSa0A8OYGLMYtrbh8rblr5t1vtC9zoH8PFVpDzUpNBYFFN9nrD8XTqcJs0IXaeZHkIRNEckymYr4ERGl6aNi2d47HPFOG3LSDhC5iF/sGD6/3mLt+hpVcgwUc1k7gwKDlc3bMbQSUBtH3IbWTej1YOcC5A5DoWYyIKZR5ZDBiN1ShrUAi4XO5awGy8oRK82ewMqmw0p/pmBfIQXpDCf6rlKBqMMG24WVG/DEIykH27ATwVzJ+Lw9/ax6IgNT689BU20NBbljHVv1gNVeh+95zRBfP5Oj3K5zdHIAN+R8c4cflQt8idZSJdnFfjSJcub0JhOQ0nbpDab/BGI0y/f3JBqVGg+2ejcyMOVNE3o8gUZFk6idRHVeJ4zXA/rXpgw6MWmXFBHSD5RWNIqcr8qwE/umsasfxfZHwdovk9m+1n5cO+Vo34D4JZ3XBj2/Om4bDp56NDMUV4Xio+x8i9Q9rO3agZs3YfVZjYYN1QpcX4Xnm9l4fdhaltMen5LkI+riNq7wwAMLVHVcVRvdb8tQyLKAXSjAvg6dG9DedUztN9wxNMaXn7jCmaqjYCeplQZovHSDD/76S3z/d9/DyFjeO2K19n6zf+yBulf4CPuTSjc4gWVseohiOb8H6F7hY/BqfTpdYMUUa6T0+//uUT79xCkO3HaCm+1xcsUFomiclksYzU3SC2o8bALGyEJuh1FemyFvrGdoKj5wWMkRCTed4y7pyl2bYRPygks54v82A3hLaZsLrTO+Siz25gndeQvQjfm8uYyt04gkJBp4vR/JJs2vY0K5YMuZrq57jU5+gcHy3eTy02zhaHcusdK+7EeqlZyU313qvXVCoRyBRDF03RVsYRaXxvgRrKgH/SjZ8WLupdwM1+IV1nefwjWfJXUxVs0V0vt5oBmC8uyCMgGGB4pHmPRsldhIfSfQGk3dGCnzEGe3P7Bo6HPTwSMOlrLJmp+0ac284uB7BwxlsXcNmBdQE/jpwV1jhsW6I40NwzIv9CTAh04HNuUXaRtqfUAtgNRzhrhnfAOCgJjYi3weRuaheMDQWYH5g4bRaVi5DEkzMyE0dwU0DaPDWT9swSriJPtaozFFvYzPGgqx4QvtTR5vxxwoLnABw98kdaomZMpYasZi04Q/SRo4WyDyQM4wJacucFWaQmEDZY8FhiWX5dcJ5I4adanmuM1Y7o8insLwL7fXmY4gllvSheRyGgGra9dRmzGZe1T45iI0c47x26yvYbOJI5jVjnb0VGXVhFSxHhL4qwZNM1J9CPhpdNtnsDwj1eyPXGWMEI5SXlsMrRX4q0/GHN4xDA0ZLuQNpZZjVAB6KGNQ1TrS2lAIdBYtc2HTcMN1+cXdJv9hYZY3j+X5wOl1FvIJ84UyrtQfC4sd23DemGGO9jV20sZp/CrkL2CqkagQZ5BpylCgstonxFAKCFYyQGYEQvW+1t9rRJrLGDmfeydgplM+ADF0PoBZ90/S24kJFJC93s+3UzyKWEQlCK1BfMH5mJ6NC458ITuegWrU7lXHbuZAFnhWfV1+u98LW4VyDYKiodOC6zehfL9h6i7DSADDc/je3sEB2BE7KINJAGvOECYpK8Uyj3/tCoUgxxvvGMHmsmBkvy9uMZHqKR7UGFnjbGg1U8qrBUYqc3zy2qcZLoywYSZZmBlic+Ust+UCXvPwIZ8B+w0tcXY2fFP+fw/U/UMc1v7Nwq0T7NVw9/APsZv3nvNv98Ctc02yo9/42Of4iX/1SZ/Gfq1RIYiqBMEAnaDMXG6E14SD3GdCTpEwYgz7rfGtS8s4PgO8EXg8TdlvLLfrFt4zVTBkIh/tUBCLg/F67LOk/FW8TbF1lhgBpjE/Xg0K8wS5SZ/+6noyZHQI8nPEMju0L2DTJjan3iOxP5qPib4IWG2fZzfZoVA84gOLbXcJ171GK6jSkqmht0qarCFThV+11CLhOn6xdr0VjM9tUEbDjm9DyLsOuXibnkpb07rXd0nj58e2mu25nu+hTW1ETRAuN+m1fAK5N13CCWP8GryC44AxXFCkikuITMQFBzIc6kMx40rimzNwl5oWAuMZtBmM17oVQsOARlUlVSsZwnVYTBW6Cqs7zrtGbcEwOAg5+SJa8JdyEso8kMKFHszlDcM12NQodxWKYkKSrIZL4G6743DDhrVVHSuYmMvAyMoSHtytqgYrzqZ8X3rR8WQTFrfhha0zfD2o0Aor7KQt5m2RJWM5ZKw3cW5jPdgXCC6ElnUZQ9JM+y9z5zExQWX4fCsDePr5VxPHvtQwHMCcX9Qt+4KIo0HAz+50+fBLdS40Aq9/nD8Io03NoR1rVxyF8WwUq0iW2u2GQHquQY0RM62aRnM+ClHZn/sz04nGq9IT+pGiRqA6pdReIMAnECWGSWZsGUYmIL0Kf/OZlLjjuP8tFpFIpy4532s6fdwwMJexpOFcpv1bueS883nHpvzQlRW+8O4Kd8/mGa9aHqzW+PDHlxg83mZuZgBXdhgBMJ2V2jfSxOnDj1Gz/lOvAVSIsECWqFEBM6FkmSHk7N0Cla747+mOQGN10cRyi85m+kT/2gT8bmnmdND1dlWMSVemA0egMe9wppUTCk+Vi6OA5puZWUPJYgrXmAAAIABJREFUDOUB9RRDNJr16fp9qrdX12APGtTwUamq2iszr+itJGq2tZUZjuMmDCnMuWC4dgZ21zPW1dZl9MjkhatYbriEs71d7p+c4BOff5a3Hx5gvlYjbac+Iw8VMLeMvyQoEUmGn6JyFbsa2afMTJcpO8unl15kvDDMjbjM7MQAn/jUXzK1kOPk3Qe8JOTWtTDb6d98/98Ddf+Ax/RVwAT/A+7dvaf+u3tAFXS6efjKJ6/wnT/wf3Po4BGWgtdh89OeDcslu7TCGm/OjfhRWYLhIGJLDCfycDRvuConJXBXAHcVrR/l6WZ/TaM1o0VcNVrGC+mV6C4ThaQ5i2kb23weJ7AUVHDxEkbR8RrFKnRYoC1tQPkuxnLjFFun2XWO0FZxyS4u2aBY+xbylbsyhXZnkbj5gp9NmYIUfoa08RxGqm6t5rdCjfW1YvJjNVBoMBxnjRVyyKZqgY9x+X0oYsWayP+OVgwP6rRipg2fi2eLB7HxDnWbZ8lm8S2DNqLnOn469rRxXE7bHDYhN3A+FuUQ1hMhmnRJzz5rnGetDorRi0FMhTRZ2zuOsw0YDw3VGqjl4dqiY0ehsG3DrnGsJpBoAZMr0Dle3MmCYadM5pYVyzYQwNcbGVsiVi+JYLWkURTcXIRzObi2klXtbrWgLSDYhBurWhTlxM0E79vLsLgOSyF8pe18T+uN+hmed8YHMzddj5mg7LX6Gjlb45jUvnMpv58mfFcxYr9GynJ0GvjLOOU1Vbg5aTm/AgM2k2i9LjTMBnAtBRFIAqcbqWEgl+f+qMjBQoFHEssfLm7z1c0NBuqWEbWTBJbibJYXl5vLenQ9WhZQEyBRfXBFOXXGAzOdmB7fS68mXZgYO2GoJNPXCZH7KjGBK41l9fMaPPOEI3/DcN+k9S0HaQArN+GBtxpGqoZUnagFQ9p25Fdg18K5JOBPd9Z56DbH+946Rn42JQmg1hNgHeD9Z9aYGzfMlUq+pcFvg8CdtkcuGu0zGSBkopAxQWBTQLQ/ftTp7IGpwJy2VwaKvsnDnQExdua4yfSBcnzKUSumzjN/2e968FvM4lz0p34ELEZwB9JzzncPhwJqYwJPxmvsfPSJLiZrDlMxuFqmczMtdfoqEDnLyetuQyeG0h1ZG4e7AqMKLd6F5VT6OtiR2SGGQkvmElhqKZZHrC1ej7tsIlKq5EfL/NKza7zbDTHTCtledx64K6RboN0fZ7m4pw35pqQgjl7DkmtN8Nv15ynTI42maSQ59o3k+fhHvsCbbrudA3eoZya7Fv7d6+M307/3QN0309Hcey17e+Dv2QMyRlhjWV3c5ece/n1OjZRw5QdJogMe6AVpS8QF3108wJZL/VjuXmVsOcfxIBudSf6jEdNDoWHcySiRXZQXtQ614FBgfdzGZ9OEZRdrUsS9yAFqacbb3OxcopCb9KYJtUdo/CqGLu0te1bNqM81N01J7RKtc34ts74uTIHBZfK1N3ptXKIVOtGKp6yvFdKW1N85D8SczA2+CDPMAJoqCLyqvOeBoQd2fmXUKim1eYLJjWej1kBBx5rZdfzzZT+XkjwgCIe9CeNYNEJgQnKBlgyHJaGB5c/TJrd7AGv5tEt4r+Z9OB9UfF7rsx/NOm6zaogwft3WiHNJI9WuwEy27kb6egsP6AaV5QY80XKMjxnuGDfUclnR/LUmRA5mioaoBCPVDCypa1VgqiSGLwcvrisKxHK1CU+te1KEcgy1vDRxIPNGGGT6uevKqIuhEWXuxHoKG6HhO6qGM5vneLq3SSUoeGauK1bVGMZMyIyxqGxNx7tJTD4JuNZLiU3KEIYxa3wkydkVmAqhl4P5AK4mcG8F5qIsE283gQNqIUhgJ4YFEyhZgzsLRdaaBf5jc5NT6y3KNcuRakT3msWUHPE2hFN94KZAXp0epx12VAydQeUjYiv9jFxgRoyWdGT97lTP2MmhqTeADBbtDNDtPO0YHjdURqF+Ex5/2vnswNkjhsFb8SPSHX4Jbui+Yczy10tNloea/NLtowxa60Odk0W1fDgqu5aJ3Spv+qNt3nWiwOxsgNP2yfygD73BBPCEtMSC6UMsncSIYvLkiNVpK+PELTNH3+whI4IHgdKfyTChv9EpqN/XgRao02MIIOp1Km7kegbGdjYc8XXItcXYZVl/wpgCch5kSpuo+69bjysAqLeVYmKkSdRrEFheUkAwFGcys4rupXJinOUeKkJzCYbVkFZUL24WUVMqZS0jZ1czYLcdG06rcSLtUgwK3M0gX95Y5VtmB8npOeVE1vh5KMO9GhWb/Zl5Ij9qqAwoyDrHzIUaf7j5JcarczS8UDBHM5+y/uELvOVHj1GtZbWc36wTsj1Q13//7H3a2wPfjHvgltN1fSPmn/3b3+Pjj7/IvrlvoRQeZjeqESR16u0L1Ir7GQkH/UL6ehNQwPDasmFfZPibruOzccK3VyyNGIbEqqiE3cBtRXiy4eiZ1IM+daYM28hXa8360azh0dYZrAwM0ZjvkBVLJlAn2OLUIiHwJZNENEw9rVNvvUhkK6RpHRevEBQOeFNDEm9lOrm06QGeiYZxierG1vxI14gxEkuX6uqfwyWbPmsPffZ1Y6JnFCYs+kIUQ8fr9mw0iZFJRAFb8TpGs6a0ixE406zGhhi5b7F0bN5XiF1MWryUNJgKK35yJ5iZeiASc1Sja5cwZC2nxDimiXeyqqar4mCfxPoOVruZyPxAHna6MFzM1mMBpmrJYMuGwRIcVf9oB65tO77cggkHQwWBIePF8R2l/cuVKNZr1HBzC25IwN42rPVgPe84HBiOVQ2jFmYGjM+nE4syncc7Y+8sGSYKhm4Cs6Vs+75z2jBbNXxp8QxPKXDY59MpqHiAus0x6TqsYris9g6T54W0zaSxFE3E806je8eUCf0YWt24BxUorEOQwqiDRYFITbgF5DQmTuGwcuiEYQTESNh2KfdGRY6EFf7IJfz1ToPmyg5rNywL+/MM5tWcIAGgyViszSz6RCyTJu3K4UtuOkI5YxWuK7OIdHUatwrkCETpMEtntw3Lzzqef8xxdMQwPAWdQVi84NgvUK2aMpcxdV7LteRI1E1q4MlV+J+e3+JXHqxxYqJAvJu1dwi8a/StTtZjxwPeNlfkHz3W5Z8cjKiOWq9v9UyiwJ0YKLF1Oj01ZhVo0vkngKXRqV6r2Ei5TeWSHTK4U/2rliq15rN9qZ/51yV9nnamAJ3AmEBjx9+30Pmi8yaaqJK1dMgQ4Qakuc2ctd6cIY2hLNr9bTLzJmMI1Rur/dcHmB40quVC26k0lLMKCs7G06lMNzLtNKAtDZ1crD18vZ1Yw4IY4YuOqjWccSDvyYAJecbByUKFDy9vM5ok3D9ZJb+WZrVmGD96Zcf5kbsfXou9DA3FccfEbI2N5Rv8zc4KI8UZtskxlg95Zv0l1trXeMNrj1MqiV3+5ow62QN134wr+d5r2tsD2gNaHDTdcYZH/8/n+ee/8AjHjr+RrWTO3wmrq1U1XLGxvCs/w4LJcVIEmDFerz1YgMFQIn3jmZlxm3WGahSlWqtzqvBpqcze8JKE7KpKMqqlgndFASupQSaJF1vnKdqCv/YrUFihwz78qx9fIoODxrImHMF2F72Oyv+8p1qwKrZwxKfRuu5NDwo9m6ZVVaBRH6Iw5P5IBPr0uZGBxWQDMYB+vKVf0+/7rzWw1b91q28w0RBOSb29Va+/c3osbVuyi4lqHoyKLcxFI2wGVb7dRuyzOUZtjjfbAhVlspmIHeN4k8n9v+y9B5BkWX3m+zvXpM/y1VVdpqurvR3fPYbBjIEZEEIILUjY9wQyCBEb0mq1Eu9F7NOuVivpLUYRb1di3yLgrYQQVoCE98MMw/jpnp7pnvamurrLu/R57z0vvnOzhkGCWFAIGIjJ6C6TlZn35s1z7/nO//8ZtnuGq/Ie32gnjGEo5Cx1a3iynapBBWAqWOqBQRU5cdVbsoaIYKjL0FuA+RAeXbDO9mFtxXAmtpxcgfEibM0I/BlGiin4ObGSVugeqkN1GYb64MGGlK1wsM+wJTaa/9DnNqIuXxtEUXNuGWp/lQz9GctUJTVIvhDBWDdcP2g4NBPz1sWzlLw8UdBHt/5jnJ3LG/0iV3tZBsnwC1nDDV7EX0Rqx1oKSZ0njc9Za51yWmISuRHuyxl2ZEEFnZUIZjuANEmg1gb58uk+3QQQF43vWvhdxuMFfoERk+NCI+FP/QYrU00m8iF9faHzP1QAfVumvAK7Gh7zUrpaYkWFCazI060i0UHHYFgb0c/q2k+n2OnkSYFuw8gA+C1YPgcbew0jW6VANajl27XZ0D5tqc/IOgWqMx6fnlnl9Xt9fmFv2alcm6dSs+S6zHiXYOCAkjAso3WfK2cT7p6uc+3+0BkDS9BgpExRqoPaqvpZQgnt2zrIk8+cYs50Pm/qyKo1dvU+VaEUKNSwTXVAqfWJLGxkGizOhOTIDUiOp6eLn4FgtWNELe6hLHI0EFUlVIVPJs2XLUYecWpJS9Iuyxftn15v3VZFFTRJnLWq0c75xiWYeOJjiKYq0H7Suv2TvU5pp6FVgAsXYHhCudMwmTVUm8ZV43bmYTkx7DYBsTV0ZfO8u3KGV1KiZzTrqoF+wRDPW2zG0JIYZiAFo2q/x1lL346AIX+M9375YQb6u2nECWuRZXIoy1c/9wSb7DAHb9mYtuUFeh0qdEPup+LLs6Dup+JjfPZNPHsE/ukRkPO+WgwPPHiGN//S/0e8/SqWog2sBCWWvCLlpELGLzHiBbwo6KFgfLeodx0OdVFaaUVFnSoR+s+0cerFoxFMenB11khsyFllbMrA18BRXfutJZ/EPGxjHkzq9Lcuu/xUX9534tB5Ycqtiyso9kvgysuMIvFEVH0kDaF0/Lc1/PxuvGAgrcapPeqEDlH6u1q3ySomqdMTxHTnmhT9KsWwRinTpBD6aayWqnS6OdOr2E2EDhAqekG2JnKxlWI2WUtbwX4RE8h5tYkfbsTLjWNMhiDsp2489np5Nvsl2qS5qipQaJ7bYGL2GjVlDWEi+xfYjeElAz5bY8MVUkiq3enBp6KYOeu50HRlh8qx4qKFlTqcrMLFBuz2U2NhtSEfqlkWMzCZ0QQIX2lZKkqMSOCSKE0x7NSErOJOG8YyuCqr4sSUG5tpQ3Mtbek2ulNPvJJKTP0pb0/ZnnEEh/W6PtxRNnhRwLHWMh+cOcmYUkG8DJulWiahbSNXSKqZkHkMN/kw4WV5X7vBL3ghw17GRajea9uI+yeHw3sS6ziapcRL0zVUjSnBmCw6EpwHXFaxVq00gUMqYYG/6wPDo3GbPptwVmpbz2PMLzMfe/z+xRZbiw3G/IBS1gO1UsX9UrVKGMMaZ1RrRG7MGhrHrBNGeAJ/ymeVDckKrNYsn/iAZUT5rllDvgR1tad7ICvhhJShFk7XYCybHitf4HgBPni2wXRPzO++rExWdhsr0DivkhMUtxhn/OuPij9qiE5btm4LuP+4KAoJuwYyqWmyLEUEnno67VINKrV51/lwArp6S1L4RtZFxTk7ElUZZcciOxIBOg1EtVwF8lSdq1iswOKMdr6DXwbB2+zROm1dHJcDca6V22nRCoGrmilgJiNjvb5eT/uk7wLI8o7b1AHFerzEG8Ppe44vpLYpwZbUWkVcP5kFZ0YNccWS2ah8XJg7bN39q7PwzYplLEix6EQZ2qrotixdYYbZWouod5Xru/pdeoXMktvKtpXViwBjYvC1X61O3FmUMNRbZGQ54m/OPURvvkTLJlSTMr2DXXzk77/OS146yehon0sn+Wlrwz4L6v7pXPjsPc8egZ/4I7DuR3fqZML+K/6Whe4Wpe7rqFPiBZkhPOMzbjwGvByHSYhd5Slwc6Kuj/I/FQ9bQGM1dUKg7KVRmVskikhSgCCRxONKIBDZ3aTg5LKxPJk0CQQE4zVqrVnCpIqVebFapwIGXjkVR8QVx2FzLU/XjpVGNHCgzXgBXm4rNl52bVVV0KRqtVKpJmsU/DpD+To9+WWmLx+nsnSRytoKlbWK+19dW2G036e3ELqKgbGJ46lZcemcSkkzR4iXGcJ4XW47rnLnFTF+N1bRZ0rXECTx8rSMzza/hJSuOZeqEHIOT5RsVyna4nlk8ZxlyUOJ5WrPOAHJckUCw9STbrJonKq1YD0OhLA1SNvZ4n21dKyt5Yk4Yb6d4Cc+cx4csQny/BvXPvvw9Wb62WyWGjY0RAmc7dDG1F5UMsTptTQLtmcFtmbhlA9nc5YjrYQnE49vFVOALk7avAcXpcJtRnxyZY2hvhxzUq8u+pxdnuXvVu4mozSOpEItWmItWnaJGQ8mDR5O6uwIikwan8VYKsWA7Zr1ZULt+zyHEE2dAvux5xErBSO2nE8MV3dBSUkbqgyJRxektiebew0NZdXm4Qrxp+qG6ThyMa1Z4zNnW9xhAn4pm2WnZ/nXRyvMXPDYnvMZDjx8SYwFbkSqVyVHtEnlhspMWG3LQfBr8ldMuVlq9X7yEThQE09UGaZpBbr3CujqAb8AkUDaMFyat2yoKrpK/EfDkSl4e2aW//KaHka2ZogdsJCIIk288NYg0KrIN8RqOzZhtQL7t2b43DdjJpXcsUfHq1NpE4gTmBK4U+WtI/5w7WWBPlWWXFUsFUo4xa/uFwjTiSsTRBWwV1MenNXA0CklgKcTWm1TBwINLlJZAoSCwSoOTlYpAm5qr+ozmUxbrErt8FSN074IOOpkVwVRN21LnDtVC3Wf2qsN45IoBLp0mjsbFv08a1lQS7doCNsgxfdSCIsL1tEC5iTwET0hNGwppTmyzYqMsrv4szNVdhcs+/vymLJF1TqB6iBv5D5EctFiRg1mRBjPEM7D5skNfOvJkxxtx5RzE7RM1qnyG+EK7/mzY7zudVfR35/5qfOv09D/8dzUGnr67YdZAn36tn6Y23n6+/l+fu7sV6dL1llGfT9P/CE/5unHS5t6Jh2zH/Jb/6l4+U5nUsT3L3z868gOduvY1Zxql7k6P8GECViMVhgNengiqTPqFciZLL0pNUmFDpcipG7VEzGM+uniX39/gQ+PSOEWwHELatWpkrTRc9xpWtayzfhUvRxHo1UGrCXrF7mYVMj6ZRLXAhXnTeQekZo0i4nBbolrj7uWp5spZDqcGcXlv6r0pMcqwitp0pdt4Xs1ZqdXWHMU/Rxv2HkHW+4skMSJ4+34oWF1tc073/0AIKv8bujroiebZaUSpxFRmqFkeaKkCQFJVTYl2LCaNf1OK7fiKm8Fv0DZeTUoWcOyqBxYv8wEARU8ajbm67FlnsSJB84YjwOA5mvNp3ptASWBtq+1YJuFUgbmm7AWpR0vzb9qiQpsK57znUmLTDNxfnq/HGY5HyVsrhue5xv3WlNtuLueuM9L0WDzC/AzxuOhfssVJUNfGQryBVs13DKZZr/G08aR4pMthrXzlpI4aJOwugCNxwMO3NJFz/W+EzqUj8ElORI/VqNvy1UcdRXNNK9LghpJJUO/yH2N00yGGyDsZq+JuC+ybDJZlmLL/hAG8RlXUTSAh2LLocRySVW3mmFb1eNAnEahftG37CsbJkXy96HRgHLR0Mpbbo0yDtQ9aGNmjVr5Hn4m4SZC3tY7wNEY3nhPlbdesLzwYJfL1zVbO+3DbohVoRsTPE+zYlWZNLOWKG/46l0J2dMwJFK/KqzCP+NpZUqj0wrItGDprBx4DLE+0zp8eS7mDy4t8947ethTDombCUZDSS1KKYnkOp0xjs5pxLHLGjJ1CHuhuA1edDzg3m/A8JB1IEpUTjdYZqyrtrqsV7VQBZz0X2pxGf1J6KGKZBdYVckEwARgA1XqOrNJEdqPWTwZEY904r50cZNC1oFA7ZsjsGEU1ZYTsLMkKrdHlmBHul07B8Fkqjh15syqbi7Z9DXWJy59Xtq2Itpi8HdA817r2tyyRBGbwVUQJTwpWmYeswx3Q30eV23esMGwOGdZrMOeLjixYtnUmxZdZ7tgO5brugb4D6eneM7NebbbPPUTMV6fISNTZVVRN3YMqJWbq4VOPmEwk+PfPu/5fOP9X2dwf5mzjUUqUcDk4E7OzDzAFz95N1t+60XOqHmdqvLTcP3/8YG67wEUfigeMt9jWz/2D7CzX8+43XvG7dCP/ZP6idoBB04wfPAjR3nL732cyX0v4kxrnD1BiQ1els8kLUa9DJMmwxHTduDjNi/lXT1s4aXdIDWkRHQ/V4bMCs4aQyBCFBrRcwTHMpqYLez0PKZjy8ds7IoBWWtZsi0Ox6vsCrqxIowlalZKdicrkdABJmdX4qxFLDYuOi6cE03oCusXXEXPzXaqsJiQbOAxlK9w6pQIPnmee+MW3vaWt9Jb6Gb7wFb6tmawak3JniOERjXhVXe+kJXlGu//0Bk++NkPscwKWzaXmFqIaMXi1rWxEmA4cUQz5dFpNyXgEIizEYlEHl6Rut/NFHVC41PwSnSppUOLh6XSNCElLMdt4lI3XuLereFInKo+u8WRq8O9y7DZs/T6hsdramGnlieH1Ha0cLXnIa7jySThJAl3qgNmfEqxpeB5PIxhOI44HcPfRjEvL0b0t/K8dBC2v8LQX/DwHrds6IaMnI6lR5lzlESXHSvLs/IgJEctPTsVuyWXZJg5A+UbDCPXip+YghFuh8JdcqJ9LpN9L+BVvqWnWWM+MXzJBKwEZVo25nR7jXdEK+yMVtjol/mabfOicJArBOjdfJ96+Skz9LUZj4ux5XA7ZqGdVik3SdUrP742zDYsc7JmUfSnD8fFixOo8mAz8M0YNvvW4aZDa3D7mMfPtS27lwyjG4v8x4tVPv/hNX5nJM+V27L4/QnRKTDb0zxTYVEroKQ2eM7jrv8ZMZ1Y/vfn+zTnobGAs0zJ5NLweG03knVGqEqjZUnqzVHDpSc8vjlb5dU3hdx4Y8nFZaglKWpmKECoDN1DEKpFGVua5yC4Iv2bEhba05Zdox6z91imtlo2b07bic6/TgR+AS5V3nTTQVS1Tf18aYoE6CSk0D8pb1RBK6ZaIK1xTM46hyBn0aJ8V52wAoLahs6PS9/2pZOy1Z6QFYrarAKieq+KL0sX8+LEGVUB1f5VpU/2Jb5N/y5Qp/tVaJQARWhY+61rxm5D67B142/1rCWoGYojEC7C3ExKH1wOoVvReC0YnzD0yCD7QcvQJpxi2uh6Y+HaQcMT7TwnVnr40F/P8O/eOk643xCp9drh7rnKoqgHMtheSjl8DCTctGuCV724lw/fd5HJ8R2cb8wyVU3YvGcvb/m3H6d7wwZe89qrsJ4CEH86Jr4fefs1iRLajYRmNaZZi2jVE5LYOtWN8dLgXjeQ11cB6bD+Z3+trbaprbRo1SIa1ciVWsOs1ms/3pvee3W5+dR+xe2EjKRAP+ZbuxlTWersVy2i3YgJMuKx/HQM+B/z4f2hb16rbZ1Hs2cq/KfXfYQTmWGC0kHCoJ+bw36mkypVIkKTpWw8niRhPwGFMGaT75OP4WRiWYktkSoDAdS0uhcfLMSFycvA+MkE7ooj9nk+D2K5YGMnFjhPRAZL1bYo47Mj6OHe5nky7RnsejqECbASKQg4qRXqhBMhVl51ujrLHy6zES875gBGQJOyP8/K9BmWlmq85M7n8Yf/5k38H2++jSt2bGFscpBCnwfyzYpkbyHVH4ShYWyij62jG3jhgR28+fabWWus8pX776ZQ7iIfqM2XpHoLVw2Ur4eqiCq1eKlRsZfHs02aJkvby5C3TaZImHFxaAknkzoV4/OQjeg2qtrB63yf/Zm0RacP/HAChxJ43Fqu9QwlVF1LEyi2+yAAoTxRWcY8QcK9ScQZGvysn2eLCZ3QQNs7Fjf5b7UaA1nLZG/ErasZ3rQj4I4xjyuHPTaNGpf92q23IJ+2tnXCXyHw1nzqRxf2GzwpFVXNldVEHRbuh9IY9N9k8Doh7tGCKkse547P875vrXBiaIB/lXhMegXyXsjWwOf5ZLnZy3B1vsgum7Bo8nyzvcJE/Tg9fpnHDWxOfKLEQyLNsniaCQy7y69HtzXUs7Di4xTWN/vpMZNQYkvB0JcxXGxbl6ih6NSeELLW57YBj+dkjUupiCNLM0l96zaFHndkcsIU/J+NCvlqm11+hsyaIXJAyeBLAKHEAw8OHYtZPgLPH/VI1FIUFihAcbNAV2qkK35fpMzShjz9LPcRcV0x5BuXqpwwbf79i3vo2mud4tMrplUxARyBQPH5XMu3xxAPGJqPpPOcv0vcstSrTfzL0xdhpAsy4i+orCswJ8WqqnPqR15KLUhcRUpDU+BKQE6ecaqsKRFD2qMGNE9CfDrtPrvWZ59es2P5ImD6RHoJ0r7K1kSFcvm+xVOpYlU+eFaqUvHyEmgqbWK1IzQRgNN2py1WGbRKxBDvT2DuskilnaqhBBVao8nkuGjIKf1B76WWihuSVbi8DFtDyKsyGkBPGVdZbhYM/Z5hdNDQrTzjJoRraerEY0mR9y+tcDATs3tbCU9VSa1SZBo9ZZ36Vdc+CU70mQlClMseY1Eff3nP1ygPbHJq9ziWO3eTWn9I/b0neeEr91PakHWVyp8Gft2PDtRZOH7PMg9/ZIETX1vj+JdWOfnVNU7dtcaZeytMH61RXW4TZgz5LtVyv63e+2fPhBYe/ftF7vt/Fzh7X5Vjn1lzJoWje1Sv/We/6r/IE8/ct8pd75zjzLfS/VqZabP5QNlxuf9FNvCDvkgHRF9+ss5X/3SGU3dXOPHlNS4errNxf4GcnCP/hYD2D7przz7++zsCrsrtGUeI/73//Hd88AuXGN9+O6tx2XmJRbKkcIKGgOcYn5rxuS9psssPeTj22W/SEPmPRqCsbnGONmJoRLh2YCmAr8Uxx62hRswJzzg/sQXbcDxpVZSkpF0xPm3bpteF2gccrz1O0J52WapK/xZ/znha2tcdcDJBD164AduSV4NaUDX8/E4o7uCAAAAgAElEQVQ3UxWic3Rxhrnp8zzvpi381z/8Vd76mts5uL+bQjlD0oidAawmFq20nb2F2lEaq1L72cSFpueL0LO5zE037eV5V17DN7/0JNNLiwz2ZZ3wQOpbd/K5coPevCpULbefJhwgk9mI795f4KqdsW0im9Rer+iycZWusIThahL6PN/xD+VQoTgsFQ5E+p+yLa6Tu53x8WSNYVTRNHyg3mYpsTyYrDqD4n4v4AV+jr3GcLjd5hG7xqNEbOr2+b+uSXhlucAVUYYbdxgmBjznwxbuTSfh9jHrjIedEEC9XFmjaGIXR3JckVtpfllLOarqStdxApiBg1AUt0pVIYH4IfASjwvNBd73xSX2jQwT2gwN47lEiysFzBQB5vkczPrckC1yS67A1V0F/rbV5kxrictJjcc6RsXKx1WEmAQhV+RlZSIxjeFnhgwZr8P/UxXIpN51g4GhLY/EfvjyimVvwbBWTys3/RYaTWjKnkXWI3lDQQWBRMISy/ZywPNNyF88UOeBRoNtG3OOBxfNKXvV4C3A8bssD38mYeeAx2pk8MUhLEFmDBpn1KI0hKPGFRsyWn8kUF2UWbTP7Pk2f31+iT99a5mhsdDl0QrseLvS3NXklFSxxvHMWkctSWDIjQskGufnlnRB60lYUe6r9AuqVG0zTvlrZc2iVuiCFLAC3gZ7Hsyw/huX66r2omvvCgAuWwfgpfB98puW7HHI7U1BvWsbKz9VrWcNaXn2qdInfpzOEVXYlK5XN7QPg69tyPNPf+9OTYxl9uvW86rA6dxU3UF8vYxxaR2uZC/rFKWhiM1wymKUbKHfm4b4jHWfr5VgRQIIAcApmOiDpWnIDcDI81JDY6mze8YNlWOKGYOWfMR1/VmEfROGuNtydCnH4ZmL3DFZpscPac1YvKU0D1cLEicWEaVS3MnL0KhZeoyq6nU+cfE0G7t2sqjXbK8wmh/j4bnjrPjz3Hr9fjJF4z6Pn/RQgB8ZqFNL5PDfL3Lkj1eptyLWpiNq8zH1hYS1C20WHm9x/ms1ph6r45WgbzSHJ3XW9wISuv/pt+8B0k7es8rZT9VotS1rT0T0X5Fh4lqxN79P0Ph9bufpu+J+/l7P67yf6SdqHPvrNVrNhLUTEfmNPjtv6XZVAzc560W+x3v6vrf1Tx74Pe5ICxtuTls83+Tw+5dprCVUZXYFbLu1nAJt/fKP9+np7/Mf/+17bO6pu9ef+4M+76kXePaH7zgCCqrH8MAnj/OW3/0cQ3v2shxtwheJB+tyO18TdrPPiHSfZc02WYlXuNZ086tlQzYwrDZhwDMsYxhIDCdbSoywrBmPd7cafKq9wCE/z6p8343PDAlZLBOasI3R9ZqCbTETrXIpqfOYvN8aJ1x0lmtxircW9Hf6WTKuUhxRqdOOVeVO5Kss1uQoJhdZNceoXKrx3//4DfzH334NV+4ZpShyfd1i1ZvJauJViHuH1yObB2EDjS217SLPiSJUcUxallJPjh07xrj+um2850P3UatW6e7yaSYyMFbrMVXEunGu3pttkkRLjhcYeFmqGK6QP55fcDmxfTJXThrOomQvES/1c86+RPFeU8q17PDkBPBuMz7nJB5xHbWE4zbmlDWctK2nWtrDft6JDB9tx1SjmFy4wmCY40+25bnRzzC5lGNwFLq3G0p5iIWNlyFQeL2I5rLHEO9JFRgpBDXpq/ITgYxwpURUtun8YRxoOt2yTLzEY0CqTfHBBAT1McjuouVxfrbN+75+hr7BDQwHXVyXhaW2oUXicm7PBbisVvnp9SWWcS9gezjIbJyhL17mTHuNQT/HhCfOos9VShVILEdNTE/iMRYYirFhPIIFmRJHMOjJmkJgIhVMdKtNHkM7TJMn8vk0eWOmAlsHDfUWLt+1rwQ1VZalYI0DBrozHLnU4kPzHjt82HwwQC3g09+yfOVRy01DHt1F47Y1LDGEqJuLqWI1VtVXeakR+PK364PHjlnuN/DhqMavvKSL62/IuyKBpwLzZlnjdLzw5FmtcmEx9cqT95sAlArInh6nKtgyFK9KVdIn5izbD0JelTc9VykYKmuqZaptC7xpIOU6nax1fz0Bqy5oXIDDX7CO2zosv7q+jm2LgJsqebI1UfVMYpGtqdecPN2culV/k7hB71XgX9sR70LnjwatYtS03nEVv9TqBEmSFcsmKqxUt+XUgFgrO1FfnVBFhtrnLW2B3SxovRbkIcgZcgKIyvcdhOwADrSGG9KuQJDA0mWLzIRlGSNBlux78mXYWTbMX/J5NLSYpVWeu7XbpX2ofaAcWIFW5dUaUQ5kphxBa9XizXps6Cvy7qMPQa7ISKDE6oQFWQ6ND/DVv3+UGzduZMcNg+7C8ZPehv2RgTqVcs8fqrBwvklmwHflaV14lBCkRXugfLmCobESc/5LNfx+w/B2xQp9jxlfdz/9v2a37wISLhyqsvBkk7BbvlMwdE2Oias1Sr8LQPmOGbLzy/rj1r9rM2pd6M/uy9Oe9PTtrz9+/XvnYXqIJp25M3Uu3l8j0H6ptbAlw47np6DOrRTWX/vpr/m0TX3Hj+vbWP++vi3t5/rrfMcTOu9h/W+aBDs/r1xuceauirMJ00Ut1+uz/Zau7wR1TwOBTx3H9ddaPz5P+12b1jHTbX076S/f5RimD3v26w94BBQfpUnwxOFFrvrF91LqG8T4u1gNh7jBy7q81E1BmbxR5DzOZqJo2yy1l+gLemn4yiOF/iI8WLfOrkIL4Eu0eIDYpSaIQ/YqL8drS1n2ELA9MQzjc40XuirWVFx11agnaseptOcoZPq5PejH2og5ExL6JZcg4QV9qeFwLDMszWIBtqWyhbxR1rBeiR6vziqPMX7pAO9912/w6ldcQ7GUIZbXgZ4i8CY/Ml289Y40sFyFLuVOuTep39Umkl1E5/G2mTgKxvjmDbzktj3cdegIlSWF0cszS/YIAnWZjiGX55S5qiJGSYN2e57ANqkkDepewcVmRYoniivMuEKZjzFZdqsbpdgo10EzXMa6tuyUwK9Rhc7wQfER4yp3xRW2GM+BxGHP53OtCnJde1kRXrsr4FavzLVJhp4Bj3IB2q2E0l5DVmHmXYZYk7MaG1IeLoonBqFI+pqQ9RjhicBQX4L2pdQzTVmc8rN7/Ixl8GrD5nHFLKUkfUf0F2FfLcQlj/Pzq7zv2CIj3RvoNUXiyDiz3aq4lKrsydKmlVrezLbSFu+OHByNQ14SDPAz2Tz/vZVwPFljAMuqybkFwJDn0Ygh9FOvvYI17CkaNkZwf2w5UU8zVS+uwTaFv0tY4kNgDT1dhv4Jw4XTKggZcgn0jsPaMmSrxnH1Cj0wGMF1XXkeXU34naM1tudjio0sJx+1XLsN+taMMxfu2QElRV9VQHqPy1IIzUG2ndI8lbMbLMKFpuEtJy7yR7cGvPKOLlhJHECTXYppW1fxciBMfm8CXjqoomuGOHATL1pnxCv/vHDYkMyLc5jm9dZnoWe/SdMu1GLVcxVbpuvm+spAtiECROLBiTeawBNn4OPvSMhXYWLSUOqFuFfG1an6VYBQ7WCnThXwkdpV6g/dygar4rhAnfZXvD1x1ATstQ4U4NMcIl8836IEB9fb1uupBdxp3WoVovavHhvrZwG+DhhVG1SfU6EItVXIKmJN9jkF6B6EQN6KlyC/OU2i8BZhQWBWYF9xv9qVMlyWeiuGQY0xW+AdX5vlxbeEjO/JY31LLJWRwGg3KBfWF9Asp23fzICyk0swU+ULlUuM5DeDrnmK+YssxXKL9979VV5359X0DxZdFN9Pchv2R0vi6nQ1dCUQoXTwmix9W0IWn2wzd6iZGiBmDUkAj7x7kcEtWTZdVUpBVAck6LoftWKilnXycY3NIAthXnE+WhJ9J2DROaETQFwjmT9+vzfx/Nx2tPpty08nXQ1n8r7jLKUbehoo0Ut39lHPbdVjFwatgS6jx0zOx8+oatDZAx0LnbixDGAtSWQdf00kd2ewH1r3HK/znhxgXX/u+puwnWPR1L6m783PGMKc53hwT21r/fH63tlP9zdREpoxcTty3k/ZYphWNXQhSFOUngJj6y/hAG3aJXH73azFRI3UbyhUbFFOE9v6o7/9/en36XOIFCWkTCC1cEI/3d9neXvfPmA/yE+qAAhUVOCL//dDMJ8wuO8apthC6Be4EK1xMejjxV6ObixrGB60sA2fgaDbtbEO110OOOcylpKndqFlPx4HvJDPJ/DzNmDR91w+40AzoZWktlWXHBetydslNohqZDXYs+O8LciwPehhzRouOa+EJn4Sax7omAa7k9IBPqMB716n5nzM+vwGi8Esey/ewkf+/LXsfrGs7hMX7O1JeqiJTSBGM55c972UQJ5ykNLJxh0QjXXNCppwNCYd39BzSsaYmINX7+Idv/3L/Oyv/Bu2bN7CmUVVKGKMODeaSDtOq6ndiia/AedXt9g4ixctUfFKtDMD3JwZ5nkmT4+JGDWWTySWu5M2VxByLqnzIBH7ZZOSNMl4GWfv0SshiY3Z4hdYSJrc05pWnhR/0pfnl3aGLmJMc3u9nfCkzIrPw54JQ/Ga1NKiJfFwxuLlDJlNhvYJS7wCSyL1XbSuspFr6u9QW7IE0qf0Qm0G1hbhpG8Zvtqwe3da4VDymsBh0Jvi7XZFE6NmVPWPe6iMLPBFm6eoZAnFfxnDfAw5ceE9ebvBtIBFMW2njlvhGsM2v8xHBw1/V2nwZGONaVo0vS4OkCHnJ3ylijO8lknxjbk0D/fAEjzSgs+2cDFl4lU1WtaphHuyli3WEK7C9oJh1beMDBqOP2EZ6zJkijBSNnR7sLpiOJe1vLDsM+GVecPnq9zSPMFvjY+RW8xhu2NGRowLqm9ftISDElHIFDBtRYY9huZJS2Xe0h7wuKta5U3XNHnVlhH8BUuidp8+pIvWaX9M3jqQ5BQvqtrp+LVSTztbsQTizKkaWNfF3zhQ2y5B1wZ4cBUmxHBZhFiVR9mESN0qbpgAkiptWqSoyqeqWs7SeASWP5HwnCHDhr40SYMdBl9jXm1b2XzoXNHrqpUbWQnhQWpYB5wkNNB7MFgB0XU0IGDmxn9H/CAFj8bVAs7IWbYogRIopFdSpU6nr6q8JQgV23XOEishZtJQLMPoUCrcyKmdLVBZNPh1S0t5s2uQ3QCJxqiA2BBsCOH8ly3+fkNRFc5C+rjaHEwWYe+azxX7x3jz383x1aEivZsCmmpXz1sncnHz6pwucIag25DMWsIly6smruI/PfZpphpn2FW4ijCzgdOVOSbzQyyeXeCLf/kgW/7ozvQzfdp8/oNcip8Jj13/GH+k+6LqW7SQsOXmEvte1Et9Jebwpxd5/AOrjiQqAKUV6ROfX2ZoW55syaexFnPqvhXmTzWpzsQ0VxNizSwaizmP0kjAyP68A4Glfo3+77wJDPolw+zjDR748FyKbTQRxtA1HLLtxi4HhlSpOvWtVZanWtRmY1qV5ClQFxZ9eiZCxq8qMLq3SKbY4ZlpU+Jr1hOmDlc4/3CVlfNt2jUho3T/8gM+PeMhE9eVGN6uZU56c/uli9NUm6+/+zK1udiBVYGz0saAkSvyTFxVptAbPAVuo2bCmYfWuHy0TnU2prEkYJUeC8nmi4MBQ3tVkSzTI1v59W11KncCqZeO1Tj/aIXl021a1cTZDfRMZtz+6kKl/frHNzfOdV1qJEw/XuXcw1WWTrfS58v2oq+z3WtKDEwotyV9Bb1WZaHF0nQLtXeXLjSpzye0OscnU/TI9XmMXlFg6w1dDuD9420/+/v3PgK203b92394nN/8wBeY3Hc7U/EQ/YFxCtWCl6PXC3nAthnFZ28nV3MJj7yXZbPM7H3YJE52y/KkSZgkYL9indSqVWZp1lKTu4VNmImlZjQcTyp8pqXUT8uNJuSF+U1cJGTZN7wo6zHViDmNR+SkgxXnU6fltlWPUPw3B+TkTCqVacsJJbqCHCv5LJPnbuAv/1sK6FTpsy0tFqR+0HGwqW2EBqTGqcaZhrkWSZlOMLqqCus3xUgJTaoVpcfkZMqccu6uv2ovb3z9K3nvX32d7sE+1uqJ86czIvhoe7bh9ksGyG442zZ+dhO9UvW2F1hqnOWRaJWN+a1sC7qYMJYLxNzikmFjjtgmEkBeTBoERLxXZTL53BmPIb/IQmuaQWv533ID3JQUuDGbY8SzNOYstgTlqwxXnpKhc1oNCaSqlFJ21hLnYWnVMjHq8KBTBBbFAT9nWXgc8v3Q3wsXz6fVIqUk1CvwnvmEWzYartqT2k1UH7Fky+DvS3lRitFiSkG/qszogBXpCvu4FMlIJGEV37l17I3d+oHhDHSNQpcqPtXUyPdALuC/1i1z7Yg3mYBXBXkeyBq+0Frj69E0vjfIQa/ADJYjipuTT+4qjG5IY8yaS5Y+EfUjiXZSTFExMCwTZcV1xbBhzJLMpQkZg6peNVPT2tpaahtSHoAtVeMiqMZIeKSnxZcbFQaqZ6ie28QV+wr0+5baSUumBxqzECzByPZ0IbC6bJ3au4XHZ6di/ujiEh+5aZzMjpDYT1zr08lwxUNbscj+Yx3IOfGAKm0T6QLEO91p6glEyYNOgG1W4MqiBLNq1WKHU6+46KIlvgTZq2VY3BnEjY5lSTs12F27z7J8Pxy8ORVKxGpTbjUk3Uqn6FARBFAlV9c5on0bBXOlHt85RwTwtOAZEEcuXVfFp8Df1olS2/o0wNYRVfj9ajV3Xl/gUNoivb5Kt41UkJF0GyfacM5AXZCrwdIl6N4P3pBJbViaBu+yBELWxbupYufvSUGizKC9JK3o9fSl7fShkqEZQDmnBWbCQH+e19+f4zP3rvJLN/aT08Vrm6EiMYqOxQ6DIspcJFy3IZ6O2XVNN+9q7uK3HzjG6Na9HPHL5HOjTEncsy/gLe/8LN3Xj/KaV+1P7Y3WJ7D168hPyPcfC6hzxyaxqAoVZHzKgz7XvmKQ5Qttzn+xStjnEZQMF++us/BzTUZ2FaitRDzyV8usHm/hywVVF1wV5jrVP3sfnPjwGmO3Vbj5zRvoHc1+Z5VJD88YFo62mHtYS5l0NRItWja9vMDENSUH6pammtz7e/P4yrnTh6oqgLajzlALLn4Djn5glT2v7eLAqwfJlbWkwYHOBz4yz9G/WiGOU98ht0rSHzXodR5dtAR/6H0b1Gnf1R4JjeMXnvq0lsedSUql7RhOfKTCxIur3PTGDXRvTAGaQNmRjy0z9aU6mWHxflQOdLvhLgK6EJz6RIXj16xx829uYOPuwlOAUEDq0KcXeez9SzRXktRfSNxFLNPfrLtjpP12k5527mk33VdZaPPwxxY4/rE1B1oFAJ0HkfBru8m5z1Z5cvsqB3+9n603drvXqa9G3POeOc5+uqr+ieP7aFWo9+1unmhXMa1fTdhyQL2f9Hj9hJ5T6f7/iL6u5xfOnqvyN3/4ZSiMsdYyGC+mWx0LGcaG/Uzgc862OWRr3OgXGTcB92C4RIbdIvxbwwqGh5AHmscNinHC8nA74eeKPpcG4OS5hJJneCCu8/WowgtMwmszJTabHjZ7GXX/OJK0eJk1ZBLDo2lBggNBL4e8DKFXpNW+7AyIU9KXzsOkk9dawfN8coUtrJ6L+IO3v5TrX9SFFQhs+phsZ4yv9/HdiaNRa9PWo84xnaf6r0WJgJs8vBK1XzscIf1NVzypEltqXcYMjpb497/2Gt77V4doter4foYo6QA7AU9lPMl82Egd18DzJWYKqSctEpMl8DPENuYjteOcD3t5dUbVvCI7vCynkho9JsN1fp6/bpyjmTQYDHpdy3AtbjCbRLzMK/DL3f2MBTn6xD1qWi4fF5gBccdYgvIQhMraFc7cbgjPSZVoCYfSFAa10xqnLP4GS0HFFx9am1MhweoiXGpYKk0oh4ZH1iw7Bgx7Rw0rK5aeQUP2KuMI9Dp2dto6Arw4TqrMMOvBYC/dpkw9nmHRL7Jocsi1o2w9wgBK4qmdtYz3Gi6pItWCQgaeI+GIF7C0qoJQwvMLAfn8AM21PH/TWOZUJmSDCdnnWzZrAdGGr5+zbBpVdBn0VA331C0ZH7Zl4ZyuiSV3SChmoCdjmL5sWV2yjPcZIrXelEwxIlsL+PIJy1BG79vjEE36bYZ/VdjPh6I5nlw4xd/078ErBNh6Qr0mUj3kBJplW7LREK1a7p+yZIzlaHuWT7+iTH8948Q3ntqTwru67kktKgXqerVLIgGZFQtsC4x7oHQFK4aBhryGchcUFDFsYaEF3bqOCiQlEJYMyaX0tb0Rg5UhpMCXtqlCxEXr1i5Ddxo8gf88qG0ZL1g8tTediFy5qJ1qnbhv4sEF6b6qJazf9Vm7tr283bSw64bZM5axfeLMdTpJAnM6tySeaFiMjJ3VohUIddXy1MZFIEzChMS3zn8uv88QCLiq6jdkKIv3pq6UKAPzoOOX3Q3JIVj8ODQELA9buq4wFIYNAzdbFh6wlK9LBTDdBTjfgOoc5CNDfw3u3NzF6/5yimu2F9h9ZZ4kZ1FntfVVy1LW0nONwQqEViAeMmSrljt37uS3jx7igcox/Py467fngj7nb6kS5t/8+Ve4/eZJNoyoQ/iTmQ37I+XUnXu4wuwjTcJeD4GpidsLDG0ruDaeWobig5y/t+aAjsq/8apl4IosQ1vzbrBPH6uyfLTtThK1CzR4NbD8LkOgkm7JsPRoi0YSM3ZlkUCS/IcrrrXrq2y9DlJErBYXRX484rNtD5m8vuxAXbuecPrBimspivSplpa2o+eqlOsXU4LopS81yG3yGN5RcMDlyOeXePidi/gDKTdQfEFRhpwCT/wWzTe+4YrX9tIzkmXmZJ0L36i5lrNOGgerRBfSQl5YS3wAF+oN899q0s7EjF9ZdPwY3/e4eKzKwmFlZqY8BncsZGVQ7Dy3ZFg71WZtqc34NQUyBd+1eB/91CIPvn3BkV/FO1BlL2nYNBIzo7J+CrQE4PQest0+216QCiUaDpzNcvR/ruL3pcdCJpyOGiVCsT6HbkN9MWb6gTo9W0MHrgXqnvjMCrXliIxWcXmBjg7ZXXwnfQ7A1peU2CS+o24dvJf+8uzX73YE0lZ4Kvv//T/6PB/8+wcZ23Y1860cYVB2oGxelTa/xKQXuDiwbaqcOc9RnyUrSBSzzQ8cZ+kYlnPAKzzDtHNwU5XDY1nVkRWPZRPzntYs55pTvCXo4trsEDf4RZZQxmfCcSyHUNvRZ6Or8CnAXu04j69Vn6CthIZ4KV1syZxKBBzH6Fc+q8/G7n4uXyjwvv/nubzuF/ZiBK60tBIQ03joTKJpDzc9Ik4goQqcFgia+LS46SzCHM/HVfY6g2n9+ZqMxEMT6EwsPWGJ4T6fT3zxi4wOdrPaUr9LVAn1fjpmYVLFelmMYsXCfrcrUbSMnxt19xUxnKqf4q7mNA0vZMnoeIXcKgVr0ia2irIqsZTUWYiWuMr43JQZ5meK/VyXDUiyCdkC1JRDqn0LYXnRsqJ8UV2ros79Piw9kBrzikLUVGVOLSwlfIifVIP6IpQU1K5JPAuFvHFeg2cbkEngzpcbel+gVqVxVZZgm0jy6XEz8nDTdVV5oGWP89M13veY5daeAUpRlZwJqXpZlw2/zTPUTMLWrKGr17A6L0EN7Nht6PLh6Iray4bhvGG6BbnQcFufYbGWcSrZpUzEX1Yr7lhciiWkSEUXZ5UC4cFYLlXrDmTg8aa8+yy7wjSBokstw/k0Rk3dfeXRhsOQzaT2G4r5UkSXX4OPVVp8LcnwYj/LPhWvTIFv5EqcOTdNIeOxIygQVxK6xw2BeFxSu67BhcuWTATvra7w8vEMz+nv4YnYsvM6z4kW3IJBqmsJHFQdkx+eLl96IwJlqj0IDAmIVSF60uJLqNBps9I2rniRZOD8PEzq2qvKmtIXcqk62VMVUMNXAFs8OVWgFyDYbQgGQal7+l37i4CUaKFnLZ7AV5dx1iKuui0wKeWr/uucaBqndBUHzVUYNUdpf2W1Ip6q9kPb1GMFCsWzq4vH2al4i4KjlnA7VbT6OkcbohJBThVHnZ4ax3q8igEPp4A20zbEy1qASMQDftUgT+vz56wbQ5lWau0igUX1cpoHm9+a8vECqfArlnK3oVa1jGcyPODXWI0XuXWwn1DgVMdKiSCx/DWNE126/dfirm0pl7MEKzk+NXeM/uwwZb+LwAREpoveoTwP3f8gVa/MC2+ZcIsoqXbda363C/Az9D59FD/+W+eobdhSoLQpYPlU261W4qZl7VJaf86q9Tke0rotw4ZdOQfA1MZtVmKm7q2zdi5CwC0c9jj7uSp7X9xwLVI3g3TAnBtwfR4FVbc01jyDttG1MXCAS/dlyz5DB7JucugeCVNVmQeVuZipb9SdkMPXxWYIjn9hhe03pyKCs3dXHKdAFyNdXDdcl2PiYJGw4NGux6xebruWbLeWjk+7aQy6QlvOsPH6LKWhkKiecOnBBvX5GE8u5CMe575aZea2OrJj0UWhbzLDxtuz9G/Lki15biIQt+3yo03mDzcd+Mxs8Lj8rQYXHquw6/m9zJ6uc+SDy/gS+cgwv25d63rijjz5Hp/qQuQAcKtiHXB2u6lj1/l8jt+9wskPV8hu8hyYlm/sxAsLDO7I0liNOfeNGpWpiLDsUZuLOPSxRTbuKiB3f82PbhLVRacFvTtCeiYDtx/NSsLKyYi+ifTYpGDlaQfp2R+/xxFwHw6P3H+Kd7/rHjZMbma2HhBmB6l5isTxSCROMD6ah2ZsxP0YPgv8FpbxwLLYlmGwx92uJQi/IB5UAJdi2KVCtW94Mo64O6rxufYqbwy72BH2synIcS5JmFKpQR0cYzibRPyayTJmUuJ9j4UTWD5kLd02YSVeJtGgCQrYSMvnqlPBWi/HxqLPxXMJv/qL23jd7QfxrXJCO20hTSCdardDU+unkCYqTTryDemo3dYXA+7x2rH0VE9LIk6E7RcAACAASURBVIpY0k1VPwHAphZ1iRNq3XbTtdCzjbmFZXK5As1kPeMoiyxXdJLa9mWSpIUJulzVLn01lc8MrbCPHr9ALV7lSOMM2cqj7Cjs5i6/zKzxeG7QwyeiFcZMlpeJw0MOAexcFHNkFXb1G5cAdSmxDCkyTG8razgbWGfQ2ifVaB1qF6AZwoo8xxYhOmHpf4naf6nCsC3K3TQMDEBJJsNSw2bh+Flo5+D2V3tkt6TFFyNjWnHtBEA66kgdv+jJTpyUjpGbta3b193hIP8urvPrwM3GZyqJeazRZqyYZbeC2mehvQYD41DoMwxc9DiRGHrbIJeqpSbMLFpXdWs1PUomx/tNnf/RmuFnMyNMJT67I8vuEpxrKq3EuvSGId84D+W2SchFMNoS4R8q2bRaVxrGtUKrZ1I1pT77qApbxwwfCBL+uG749RxMGkvdg+fGhhdnyvxFo8orvnyCr1y/l+cMFWidj2nHhmwOLl2yeKsBh2tLJIU6z90wSutiwqhaj7lOq1WATsdIp4A+MFWm1HIVz+xaiVXSc8Px2QSSc5CcV+cnBT2x/AnPWOakzJaNjNYPMoPWYmMNkhVL4yuW3DZl8KXxXNGRlHqgKDhFe4lvZgfS7fgXcZUpb49JRQMCmp1FulTQRu1YLXocQLOdzNg0Nk3bDCNLzy5D8mhHFKGUhjmLJwGHhkIXKFZO4C0QD24GkqZ1c7UcK5zw4px1mawSP6qyqDWb3lf5WuOyZnXtzxxMla6JzAtlJTNu2KBWdITjRNbPWTIXoVqwrgiQnUl5dWqxNxaNqyju6zJMVCxvrA/zH+8+x+ufs8z1YQ+JnzheX+8y1B6xZEqpaMRxAkdSDujrf2aC//Bn32K5fo7uUg+tuMaqbZP1trJh65X8xds/xxteOcYNB7d0PtzOdSO9ejzjvz4zQF3nMMkLLd/vs3xcV5i0itZYTiN9BAyu+vl+hy/y5cABBS0BtNKee0Gdr71rhqUTLQfslL+3NN1MQV0HkLiK1pJl7MV5DvySUE26UZVYg9AjW0iv/sXekOe+ediBxmzBTyNEFGEdWU5dvco975pznDK1cldPRFSXIrcv4qU5BwSN07plz53d7Hy+lmnpTSa+an3myukhfwr9y7G8bunfleH5bx4h3+27su+FR6vc9Y4ZWnXrhBa1szFzpxsO1OkM2/WCHnbc3O1eL9CS3d0Mi7c1uec9M1z8Rp2wx6O9FLM01XKtXKmPq2cjcps94gZke3xufMsgE9eW3PtUoPeJb6xw7zskz9JZnLYDtApbm21z5GMrrhKpE1Hvcf8berjm5QPkulIF7+YDNe565wxr020y/R4zdzVZeF2D3nGF83XOD6kMVy37Xt7L9pvKrlWtOCuZUTt/wm9jyM57evbbdzsCGvdalKydTfiT530FBnOY8GpCf1SFFvbbBD/spmyyTNmI+ViZnLDmsjNjLlnLmPEZ9XMsWstW4GpjmE5gXpU5DO+1LarNOodtxO8EIS8MB9mXyTnv02qSuPmiHeA8vlQhUL7nFSoQBLAc47axgMehuIGxDcKkQaSBoPKCKl9+DuOVCE2bhcoi27s382uvvtGdj0kcOwK3QJxT62mIa2bRpKRTSONJ39Va1WpB5UeBPd3/tLEm4OZ6Ve60Uz9avDqVuFJeked7rp22dfsIf/yrt/C2//J+hnvyzFQamEDVOrf6wYRjzlNPObTR2n2uOheUb9QVCBtX3a41TEDGZIlM6HJiH129BzIb6clv51Oty7wiM8BkfoyrEssRL+LnJzzaDcuylJVrlv61tDq1hmWlBrXEulaq3vP5OpxethwYMWwUUX0G6ssweKshEI/tfMotKw2AtzWdfNVGXJiC88uWvqsM28bBXLZIBGHkGybOVOd4CiSpjScvtGCvJRE6lNGsyqS52GGNLqmCkybfSpp0+Xl24zHhZ1lswPknFLGWcpTNkXQiL+YM9y7BLQKY+rjm0uSsfgxVT5FUlrdn+zjftjzenGJfdpRWEvBkK3F8/m3KS41hqeWy6NlslfvbKZ6ml1Hy3TA3bZldgp09hvochJtTE+HP3h/xD3HM/+gJ6WmmGEtYY6PEHUnEmzKDHBsMufWRWT5xcJhbunLMnYkJi4aVxOPiSo2He1u8/YZ+Sp6luh36xUNTwoKqcwJ3Ul2q7boxzTQVkHMiBJEgZS0j/xqBKI3JccPaFy2lPalfnpHzgw8L91jG9prUiqYb2k/gck3DnYbaYUtbBryqbUhkIHNotXb12rL2EI/RtURTYOnUrGrTSjin88Vdx9PukuP56fOW6FwdITELtqT0IqtzR7y67vTucj7lLUpwkci3T5XcDESdlq2U9i56TWufoRTcrkrFGhoyEylIdUBXp+1SKiiZXoGh7VDW9nXaCiAvpCKI4X2wKpAnr0ANswMwUjUc+5SleY1hWzesHodiLyydSxc19aZlcxgy2jfE+z88xdVvKJLZHrjOk14/kNWlPPQ0kQlQL4hOEbNluMBfPO9qfuOjp9iwa5llk0O029j6+PmrYWCKd1z/Jd539lcoTWhuS6+13+06/Ey8b30d+4zYNw1wVbZ04uqmio34Y25waiHQn6HUl3HVg2Y1obrcor4WOU+70QN517oUsJIrQW1JvZb080x/UKvWOnFDeTBDeSD93zWYpdATpopWcR1CQ9dgBgFHnYgSaFSWWrQasQNUqjBFKvO67ViXUqHEBQkb3H6qupCFM9+qMP1E1YkntH2pQot9YQeMru9Rii2lBpV7e6E7cC1jPXbsiiLDB3KuBa3JW8ekthA5cKYxWugOKfVn3AQosCjjZh2P7g0hEwdLjl+hfdRNwgZx4WYeb+CrvWIM7bmEyVuLbH9ON6qCapsC1U8XVrgndybT+XN1Vk+0HaddHk7dW0P23dFLrksgNG2nju0rMn5TgXgtPQniRgquBYjXKybu5SIcFzHI+mTFs+kK6NmYdfuR7vGzX/9XR2B9AfHRr9zHJznM+PBVrPpjILuQaIXYLzIW9DBpAvKuLarWmM8YcNB4Dhs9Ii+0JGLMwPM7oe4SW5/BcJGIpURpD1X+IJ/nRV093JTPca6dcLSd8JCoAsqK1Tli4B+SxJnqql36QAT3JfBJm/ANm9CrwHt5Q4lQpJKEWq/qHXklErUlc02ay2u881+/lOv2DZIEUqCmViQyNF1vq8przu24AJyu1ev/dbBUfdNz9BhVT+Q0L2Ciq7VaL3qsbjIL0y+6T9JNdzVP7SF+8SW3smFyjMuX6mRVCWnPOdGEZk4bzbqfvfz2NOnC+CTtS8Sy3FdMuMQeSZV2exarNAr59wVdlNtzLK/cxUa/yB3lTQxh2TaQ8IZx3wkoRiLDngHDSg4ebME3q5blhnFvZboJTypjdNZypG15zmbj2ovqDhf6QTyjaBWWpGTUfDgMMllWK1a0lPnH4G8vWHoOGK7bZ+jebEiKEql1JmIBA4mQtYbek7bYpFx0XKmidWDFiJh5qclpEzuXlN8NCpxXYohSMWTVIspdSy1fy2zbMl23TJ1LvfBu32W4pgCXV2F6AUZ3GK7cashrUdpl2J+F2zIJN2zqp7vQzbvq0zxiWxxtG3p9cc7ShIknlIwh02LZn7RhpZB+1DLSzXnQUqyY45FCLAVmD1wO4eM25mVtn51JqqKUIlag7kopQ51SH14U9PNr3f28/LHz3FWpM7TFZ24a7r3Q4j8zw6+NdzHg5YgKFrtqnJedA0dabKgtLu61uvX6ALSTstlRRUw3ra6kKpYKVt37GIpXGgca27OW5sWU9jI0YJjYiYvUUktS2Fl9VLVuC9cbzDlLctI6Pp63CsFmcUZToYVD2/m0YqYUEPHy3PhXO1U2kAJucnwWuFOKhPZJ3nPqnGiVp3lL1TRrnNJW76NQ0r6mLhQCWa6FrFN3WvOnIdR5JVHJPoO/RfMJ2IK8AS3+Ri0OOp2ZDn/QtT8zErZA87x187rWYbJCsVXrrF0Uz3a+BJVzaftcFcgRH3YfMG7RM/cYzEo8UoSiHJE83HG8OWP5ja4C7z6d5xPH17DLKVVBdjDhdoPjJJ63bj506z1VUbOWWzfuccqepfaqi/1TsodHzLLtZmz0Oj7KE3z0K4fcZeOpAkznY32mf+usd54Zu+lAnACATpBOxWYdmOj3lUstpo5UmTlepzYTOwsQPThb9KgsRI5T5wChWiXfzb7EVZ7Svv/6NtItdTbp/t7xkDtSZeFkk9qi1KiJA5JSv65ejJzX0DrwFGAJ8x5D+/NcvqtBKN5d3nD2s1Uu399g48E8mw4WGd1ToOsftV7Xt52eFVrlyp4znYD80KM8FJJIVaTzRl5HzYQ4Tgh8j+pCm6nHq1w+1qAynSptVXXMlgLEYQvlPt45lrpwqz26cq7tAKeOTVAwDGzPuZW6ViK6KmnwOgC2vmPp3Q5QSrmq2CHtjD6TqGl57LNL7rjoc9NNz1843SRwlTvrtlVZbLvXdCDkaY9z29T02tm2e/5TX9zLPfvlexyBdQLv2ZMLvPFNn8dM7Ge50Y/NZMmp3+YViL0sZSyhF7DDely0EYesT7fxkCDva0mD+4zHi5M2UxI5GEM1aXLSwKfkYtFe5k/zvWwJRun2ZGwaMxNBfzfsCuGIOEdaaIvjE2jO8LjDKFoMhAfuszEXjKFmI+oY+oMeFuNVWibACza4VImkuUqvaXOqucCdXMnBm7anEUkdzpsbazqPtTjRpC0ApiuWrs56jO6XkEm/6+f1ip1+Fxh0/X4dxA6I02M1BuX/ou96jPs9fQ89fV3sHp9k9szDBCahqT/KqsUrYYJBB/ISRInYjJFKwUbEzQtO6CED5bh10VmeCClZEhIMzbCf7swmpk2WD1XO8nN+H4Nhma4woRbDSjutwlwTGO4zls+rglqF24s4w9xpA1vFrTWwVoH5mmVtFkZHDa2apVA39Iso34TafKqMbfTAZy9bNpXhZS817NpsiBTj1ITsHhm0GuoPWXL7U86R+HM61nYmrXjKvNWWnMkezpQurrvjMWMMV8U+r/U8J/pY9HKuUhkoQD2AvQOGtRZk+6Fv2HC5kjDnyRvOZ7JoNLczJy89UVQkHghx1ihdvqWV6eXarOVtzWmelxlnpO1zbMU6wUivgQ0xjGWgb8BwbMrSlcDYgOHSMqxYmF22xCOGvRthdsbyu/dGbC+GPHfQcHbJUglhryqEqrr4MNgBiNkk5jlBES/ZyG8emuf3B/rY6BX4zfYqfz66gX0jOVqlxIGZzKxlSLYkWlTIN00gSbYeEpZJFNFIExUc6FIlTMdc1SjVF+T7V8alfczfbeWB65Ir6ro2iwcoFemJFByJHqMYLrUtVVFTyzPIQaCQEw1ZzY8Ci7JT0eLjEgT9HZ9GfZQCS/pMdf0WwFdLeLjjnafzdVDvoSM8ckIPVWx1bqWdr8JVhtrXUk+9zHi6TaPX1/vUQkDbVGtX4+QYGK0Uy+BawjNgi50KppgJG8BeSEFkadAQX7JYcQNlXyJBoYyWR6DYgiEtYj6ScHCnIvpAi5eBUShULbIzyedgeQ2mlyz9m42LGJP34s2hx6/0jvP6+6vctCt21jZWlisClaIXiCMqTmobx71PWglbN2b445t7eNtD57l6YJxZ8lyMVugJuqnFu2DiDL/8xk/z/Jsnmdze8xMlmnhGgTpZgjQW47Ql0JnQMiUZgOKqXt989xxzD4mT465BaZlViigphpQvpxw6jfr/n733gJbsqs98f/uEirfq5tR9u/t2DlKrlRNCCCEyGEzGjA02Y7yM7Zl59nueBfPG2NjDrOHZxo+ZsT228Tg8bMABm2RMEBJBKKdWaHWON/dNlavOOfutb5+qpgWSmVmzlsDvqdaqG6rqpF17n/3t7///f9/3eTjk3QWOF39U+zl05yr3/bfz1M5EbnWj11RIoIlD4CUzar5TtanDdY+376X9LDzeYPYbTTfAwiFDp5Nw/B+qnPhcjdGrM1z2pkG2XlNKJTue6Tx7S4KUTHOA6eLrkTiqWLvlMy3u+qMFzn61ng5esYattKpUnVfFCiq00Ln1DiMZlMiFiFPWLyh532HGesdVY3x3u3SbtF2P3cBwN4rAOLmXR/5wNV2hdhtRx/PlfyjtMLWXhFnl99cFhxe39YUDdcHk0997/r9nbYFeW9bhk39wp0Mzk4VtnO94DucMYSjmNmFNwPEkZsZGaDLeisdO5wAR88c2cgnvL5Sdly9nCctn4xbn4jVOxE1eEvTz+uwQN2ZDVqPEMSt+W0nrll2RcZWHkjlZ61iaHhyKLedtTM4EHJMeJAJzbR6xMRuTNq24wmxrBcINZI1HR4k1SR3PeAQqNzxnedvPvpCxnX3O9stlHGlSEuAXa6F8JfVLsQlu711AJlDmlt/d9y/ocen/7qTVvVe4cKtmQwE7vaeJVttqrImUT2IGBwv88k+8nDu//mWGiyXqqhCxYunS+JFy61w1bDie2pqR4DmGTguexS6zpzQMgR0IMxuwuR2Uc9tcgcSD0XnqnVn6l7Lcms1wvhY7t4dzkuWI4JqSYVPR8HkTc2LJYzgGl0undpYU3zqci6GvAVsHYGTSsDJjObYK5SGDl7XUanBqRdjYcPWNMLQtZbt8hc/E7sh9YkNakBUdtkiiQicrFsmFX2Wq4YCATYUIhYb6A2qRZYNv2CFgsp7lb4nQVa5jOSHAJUH1RcuoKjcHobNmyTSN8xdelAcslsZpyMvPc1gFWGmBR71uaZ2H6zOW7MAID56K+avmKd5SmKbR9lxu5fEEsnm4dswwISxdNBw8ZmnJIaQJ20qGbyzJqN4y1jZ84TGIsj4vCtOpolwy7JO0y7ylbg1LHctGzzDVhFXPMm4TbsyX+HTL8t6ZY9yYL/A7pTxvv26EcLvFpbeowlhuCRqYCkt32ThXCCDfWAEV1dXsUj+96MYrYlqxZyHaONV3q0o+Zh1GXm546uGEuYOw8xbP7dOJP0+nLGDl7sRV/oZiw9rKgRMhYfCkkaMOpu4pVlrFCKspIHSiwGLp9L1q7Ki7Z9LcOPe/gKLzdiUNq6pKXOFhVe/6Gl1qX7CbjCsWzKyrykGh8lTMWH1Evneah5ydnMLOqppWUYUcL/T+WUtSTydHFfvpnMTImS2p00X8DYsnuRzZ+C102yyEQVmxlQzRKYu3xdAWqJ+DpeX0UoYncOH4uKRiPEuxDINDhlIDXlT2+KMn1/n8kQ7veeEoFBLHJur7iXcYkiOWQHqwJ7RCSL1k33LgEt73+BeYby4xmp+i1TmPbZ4jzG1msrCTWY7yid+7k/d98HXfKYD57vnxWW/UP7g3fjhAXXfWXznXdGBKoEAMjqq+yhszVBY73PV7iyw81ETJ/5oTAt/Qd2ngbEfELkmvrS2JDjfq/okGfYYvpQc6zjxa5e7fWqIdxaikP65bsv2pBp7OReHEyqxW5xopOkZXewgY2pTjxb84yROXrnDyzhrrhyPXwUMtMy0sPtbia/fNU39fzIFXD6Xb9xDXs53uRe+7Kt/QuGKGu/9kwUmHZDemBQuao/r2hAgA64asIgXZr6mK7sJD59u7dmGudip4fOH97/PHRafirkfSY7kR73sAm47v7gyORDFpoUkXGH6fQzz/9v9AC6SgxnDf/bP82v/1EOHu7ay2fQbDYWfFlUlaVE2WUjDg/Fj3GsPLjcc2ozC57rcx7zBIlpg/t21mkialpM6MtRS8DL8clsn7fezwAjdpqgspPWfNgw1jhkrTEjlXAcsO5zWeJlErCKn7f8HG/FIi2yvDhqTJ2doRdtl13jN6A1+sHucrlQcp2CatOCKkyYLKNSnxktde5kCHA1o6YC9HTr/1UN9VJ1T/0uPiv/VaF+AZUYX6X9oevc/pt7bv7VNhWunZKd9OD+VGdbM1RsqaIQewCj07VbYEIw09mWvaVpdhPEPSOY8XjmKkV5esOpkWT+ydHs771tKxEb7tcKY9y5SX5bLsNIfjFd60fIqP2EGmgyHHJhTixJm558pQnIVfGvdZasLftixt6bHVU9AkqRLlg/VtSatf836Ku5Z9mKtZHp+BbQHsHIcX78CBqvgpUtukkTR3qb0MWeXkZSAW46LqeS2KFcKVIbv0zHQJSgeuQf0xD0ZCVx09GcIR1aTYVFdObG/ZZN33rsIaFUQoH+rkCejPwvSVcKPx+G+HLJeX4XwGpkupfIeqHXPKxQvgjPB3B640EW8eHOXB1Yjfq5/mdcVNhMYwpFChJ79OWFyAkSz0b4ZjNcvlWkDkYaoEtQWPfz/bJGt8XmIDTGI5s24Zd/dyONuGUsYSNqGcgf4SDLS1SIeDrYh5r8iOTMhdlaP8+b4XEa5AM7JkJSsyA9K/KypysiVN9HdMmVt8pH66DtQIPIsZE1CSF6zAkICyK3xI80Mni4aGFiwxVOZh5CUGX9+PYsMK1SqXcM6mxRMZg79bLGDq/iBAxta037p9C9hJZUEsmkSAj4ARsNRrKtxQBarORcCyq8LgvmN1c537aFcqRf1fBMmxlHGUmLQs0yRgLFCq6I5R2FXslxxLxJTr2lTYoYIRAUeFAM6lY1dad45eVmhfIPR0ak/WY/mkRSeRZVdMcQqX55kZhtIVMP91GG5YcmNymoCFFmRCyEjO5hQo3+/MDM7HWLmAbaU0BIZbRwp86BtrvPLKATYPBsSK88slRCHX6wwqMkk8iy8APJIwWh/il8ub+PDiETblN7Ihs5HT7bO043V8O0p4yTV84CMPcNtrr+aaF2/sLii794x0mPxQ/rx42n/uT1D3XeEjyVtEcOyuigurZkZTSy+FMUc2ZTnzeNVpywnQCdxkyx5Xv3vISXwon00A8KHPLPPYH68RiiZ+tnbXAqo7KWiC6z1EVCm0ePyeCrW5iMK0T2c9YfjSDFe9Y4jRrXn8IK20vfO/zDN7d9PZml1AL7qMxDr5jhvfOc7uW5uceaTKqbvrLNyjmEeqPdQxlsf/etVp4ml/vYmkdx7/5G+1kw+LxxucuaNBdtJzzJlCH5e9c4Dt15fI5FOfyye+uso9H1om2NJ1sFBKRc4jU/Zorabh4+ZKwvq8RqYe3ZWlGuKidum9pZeVd6ccErWtAKFy6q768SEH2twq7qI2d22rXSUCu5KBSCfaiwnB9LjP//yfaYF08WFYWUv4mQ8/QJMcI20BroCajQjIsjsz7qQ0pkzIVixXmgCJtq4mlvMkTHshut8qz24xbnAyWuOYKlK9HP/BH+LFYYHIjzkeW7kgMebDOWCmbblqFRSVW7CWMx0YDCVVETmg+C+lQxfA5zqWzYhhaXKiftTFj94/fCW3Dk1yrHGaryRtijamZTwKoaVxvskv/MQ1DGzLpuyZAFePUbvQOMqV63bMbhfVQsZNSgJzSupT46ifCSPqPU1aPQCoyUOfE5BTrl3vf32um1OnnCKBmP37t/OL7305v/27f8/Y5AYWa5rpZCuj3Sp3roDxSw7EJU35VJXS95OGO4Ugvxu/sINpf4CRpMFsvE4+qXLceHyhdZIwaeMHBf631XO8xsxxa3EfLwx9FlcSDq6lyvuNCHYOGl6bNZyKLfOLKfDpdCzbfI9qxXI4smxbh0dXLB/14d2B4Q23GMpVyO+D/LiU9NOQk0KvAg1iczTBqV3kPxoewElL1U9CaTMEmrzVTnrqofy8HSHcNcTQoI/nnGtkuBCz0fiEcZ2659OOQ2c8oRTF+XlLTQnx0wY5vzVqqaH7ehvKLTjdtgx3/T8H+g0SS7YV63TadHvZNWT5SGcDn27P8PHGLG/PTbEzqwUorMxbcoOGlQgGQqjWDbVYi1jrwtj/GNc4SchtxmMuiZhIfK5U9WNJ+n8wtmqQ1Ekkx6iOdfp5yoiZUXRDSVrRKkejJm8rX8WHjqzwq28aZyrnOTUDMWjzAUyPGYyqOru5aWpLVYi6h4R09afAlYCdgLFurYtpakrnNJil9FqyFTj8bcvhdXiTUmUUxpU3qxw9jqfVp33ycF232HrX21gWyafBHk1146QZp7ZzjLbTGu0C89mutEoJksekE2NTq69TNgV7KupUloZAvDTnRFJo7IiQljCxgOmcGFntz7jrkCCyxpSrtu1K7lhV24qh0yKgnYa1UZhXElUah2I01QSSXVG/E3i3htaEZbkGgxtSmR41mdZMsWRNyvCU5pc+Q07nsQqTg9D25SoCxZ0pu3fpLsOZBZhRf4tgT9Hy8yNDvOFbHf7xplX+ZWYUz0gXLwXZAq8qAukcT3NIY99S2uvx4+zl//nEt1lKagz6g1wS9POtaJ2SV2DIbmOeY/zMb9/NV694A4MDafTrh30e+8GBuiTVX+oBsCPfWOXo56oEg+JycayYNOokIjxzqE7sbg4enUrC9G15J9GhoobeIz8QpOHB3gvd3zK6vvAItJqTLZYqSi96XUx2K3HODNJQ0zl15iw7/3WZrVf1sl4hWwgJVMihVY3b/Dv7cMBFL3uGkS1599z9ooijd61x/x8sO485gdT6bMTavFb8aRi0ezu4cIrP+Ic+1A0t1VeiC+FpFWyMXJnhspcNEWoZ332ogteFjHV6Cs3G1hUmDO/OsvZklBpN9xlO3VVl1wv6KcmXpXspOq8Lj+5cKXpd34Nj/gQuQ0PrfOK07+R28f0eFZVT6nHRrr/fNs+//0wtoI5g+OI/3M9Dn7+TqV0HmG+VyeTGqHt9jPlFpEt3s/HdxKt5XK4RUlY4gXUkgOaZUzbiUTlDSFjXdPj14hRXB1mijsehqMN4YpxenciucQuyZVT+l/LnFpVPZ62zrroSj7+Mm9xoI055WU60EzZLmJc2t1cf5epwjA9NXcsWApqthGJxGlYeIrYNPBtRysWcZ423vvxm+op54nrs7OocgFPoSONMbIJmSvUd5egIqAmU6aH39RQwU/d3lHW6gLiQc6cm61Zyu+IIgUP9L3Cn37oDOqZBkVZV9Ib0FxVDCvuLVAAAIABJREFUknVeF0FqdSJg58vpVZOopFj6nM5e0jqOFwxjMhvw/QGaXo5M0mEkU+QSv49N4RDreNyY1CjZNr8T1dgiPcBggM9VH6ekPNjCTlZasgaDfSEIAK3WU5ZnWMUoZcPdTctHmwnvTCKWqr7LV7xqALxd8Pv9hj2ThnHVyGSMW3Q5VqYO7VnrEvVtRbpkYAQwlGguUCLHqVHDgEJ1qg4U3SpD+KW0jsThO52UbdE2Ccux4aoiDLd9Z+P1qrCPv4nlhRu4vEkVBPRnDQMbJVKe2pLt2GMYijscmvO5rN+wfYehswC186nahhbD+qp3jxjOLKbFA3uzCWFugifaJ7irfpROsplXBBlyOcuawKEHEw3DgcSijJtax+Pvopg/k9xKELKK5HVAAZGhDXDsDBxatBzYaFhas1QEStqQD+BsDHdHhpNRzI3eCv9qaiNn1kb4Yr3Brxxs8duNHKWiceBgVesOAQ0xh04IuBvOFyhyjSlmC4xAkxwglHSqQZiBzuNpDZMDT4uWc9VUB+/WWwyhbLAEimr67iRSbPEGVKjQFRYWENxoqB2zhKp2VX6ewIqYZ+nOSXj5LHTW0zzJ5Ix1lZ7hMHhawen0ZKOlh1g8rczExKmgTdW6Os8Jiz2bVuO6sOou8JQ3tzPN3cuUu56wCjN3i9ddKGBrN29NuXrL3WvXeelwaiNVC4sllG7rCvibIJ6B1qMWdqYMqMBvRgUea0rKthRHJfzcXadlJLUCy6ct/fIazsPAXlBGhD8vWznIK5fRM4Rty9uvGOI9n53htivLbN2cxa4nKZuo3NVdxuUmqq9rwdFeidm7e5x/cVk/H350gesn+rk56KcdVbgvkMj4MJv2zvHQ5z7DP355K29785XdC/vhnsh+IKBO90ivz7B6ps05JfsfrvPYJ9YcsHJgSwBmDfa8skxRll9aSLj8AOuqLyWZITsvsUCu8+je+12FET0mrjgYpKts5YGWPRYeaXLs3nU27e9z+WDyLm23Yoql0AnxKoFU5ydh3pUzbRprEfn+tJkEjtzqX9+pOm1KnbixcurBCkHeY3Aii2yvpO+latYNe4tk+lepL0TuNUmJqPDCdYvuXJWOtu/zU5/VPUQrMndsJ/FFYyXh/NkWExJxFh7VnNZLLtetRooMHdw1TF1R4NjHdTe3hGXDwgMtZ5l22WuHKA6Errxe15ueXPd83L3AMLI5R2lHSG1OFbCGxkrMNz+6SP3HYyb35HGeuCpzj2Ia67FjAQc2ZhjepESe73Ntz7/9fVugVxyxdK7Gn//fdzkdhZYtkc1to4rPdZ5Pn9/HZpTHZLnWSgg4BWXCPa9Q0QLwl3GLLwoJ6a4Y1fjp0gBvHCrQcd7DlpIxDBQMpbplXabh8vAMDbv7LE+twIkoNVdX1aGKLw5Ga1wWDvKlpMNBLJcnHb5ZeZyf90f4mS17GSsaHlqJabZ9JhlysiDLcRtJzQbaOVU62Yv6nKpYNSnIO0kA7OK+owvQfeDC+OuOQwE/baP8IgfYxEZpIKQbuxCUwGBvX2oQHccBxC6zr0RxvZRYl37h/nFjTuHlGCdEnDSx0VLKzmlQOQHlOtYvE2amaIYjzhu332T5duss05lhNnt5TljLtLTtbMx7vDIHjOeA8VJ/mb9snOAvG6d4S3EbI9ZzYbotAZyoWzIxnCkmnIvg/qbHzUmHWVPDFAq8t6/I5KaE7C7DNum2KS1qyRInlkC5SkdTjTd91TklpWvYa22lBHI5CmjCFsJXIvlxxXSNAyDxSScj6AzSwyDBKp6aW+OSIGYtCjlSt8way4PA263Hbfj8o43YZDIUAhdlY1SuFlndcyCXNewaDDgshkm2YqcsQ4FhqGyorKb5anYEHqxaRuPU3F4Cwptr8LPlKd61dIxG+zRX+ds42zbcNJgKEzdUZBEZTtQsR/0GTxm4xeScW8e0Z7ihGDCzYFnJWSZyhm0ZOFS17NA9qi6nCUOpH768ZKnGMX9RneGvxga5KRjmqSBiUznLX9zZ5vNJh3e8MetUCHzFnXWDFWBzGonWsbtmMg1HKsTqFhdd/TrX7zoQr4KqXf1JaflZMtLa7k+ZsOGthrZC3xKcl6SMuuP29B7vG4PkHDsPWuIl64CogFHjHkv2aoOvymT1eRUBDEFG0iYRJKWUfbM1k0qRaLxofAhcadEu4Hja0n7AEu5IxX9jMYVaNKkQRGyd5svtqY2cvGfduNJnBPJldaa8u97CS8O3W1DhWEmxdWLpFE6WU4TCvxrW6ndNyO80+IcsyZzkWMCKTZTX7X4VoBj61y2Nk9Av0FeG/JhlImtYOgeldWi0Uq3G8QF4fNY43cMnK5bxLLw8DPjL5Qx/c+csv/i2aWd/5kLGInd0vvIyVpGOvssGtGZgn6wt2nfRjKY4TMjJYJCNcZ2m79PKyLvsJH/6B1/jtpt3MzJe/KEvmnhuQZ06rMtBsASDhsOfqfDUX1VcLlwghwIJJao/nknY/oY+tl+r5QgMbc4S9Kf5YmK7lh5u87XfmmXs0hyBdHRimHukkUZCNO4uYpsE/JSs6YBkAK1KzF2/s0j/1lVX8CDfVDFQL/93G524caxSqg1aVRmOfq7iNNeGtkuAVHpyCSvH2m5/7i6qJH8POg3LfR9bZu1E24Vs5YGaLXnuvFZPtmksxu5YCiFkhz0nbSI7MgEuXa8G8jM+9HrvqevyjKugdS4QNq1Oqs5E3PGb80xemSfUYE5g8VCLUB57mry6BJ6OIVZt4sXrzD3QdOchK7Yjf1th5t6G887VZ1Tt6+zNNEvo3AQKI0v/xgyXvXWAb/7Koqv0CoqG2kyHb/z6Av27QrIDaahXGoGN8zGVR2Ne8B9HHKhL02+7V9i7nme84OdffLYWcIAduPsrT/IPd59lx/4Xcqq1xamhK1+r7sR9K0z4RTp4mnecZIa0Qvcbw6JNuDdpcFySDnGVybjGjX07yZGj3opZV0jLGPqNJfStA3ZK3ZmvWrLG8ihwn28px5Ylo0Ckx9G4yW6bcJaAJ2zMRDTPN5vrfGzrNG/aNI7XMJxbSJytT6WTsMEUuLZ8CfcsHWMw8Di6sMZbNr6cvf1bU8ZBHVAsnHJ3FCrVQ3/3WDiBMD01yahv6yPKodNnta2Ai6r6dB8RSFPOXLYbEhPIc0US3QlO24ut0757oFCLTRESRc1Cmi11MNWjK3qgMKxmTGnSadJTkVCAl92ICUaIW6ex4RDGH0zfj5ZZitaphVk2Gx95UByxMcOe7+TBxPm9KuznpeHlfDVqstZp8qatBfZv9dA9KFOCwQIMnjNMD8PNvmFbMU/zTJ5vfs2SbSdOzHdKIr0tS6EEYT4Nobp1nzNJT+9jjsGU1YNAg+QwNLmr7dQ51HRiYcIuAalEehU7KTQntwHl1JFwvBVTMCH3JJYFm/AjsndUE8YBd0cRtwUJT9QNk01DZgxKunUr32nZklu3fCZocwlZduXkYAMnjKXjweayZV/B8OQsrHcso7LOMnIhsUxlsvyfg5v5jfVz/H0c8bowpBDDqXXL1BiU8x6VuMG/Wa3wq5khkjDL4cSy2U9ttPrHobMIjdgyrNChGFCJEMvkvSHW2TK00eeBp1b59ekiN5aHOLsQMZ1N86ffNpnlM2cr7DgecV2uyPICxL2KVvEJyj8UQJO9V1fAWW2cHAVPbN0gNO/VWOlWjYpdkiZnDKMxjO8zGIXE5blW6+apKQQu8KNq2qW06jU+YDCHLFkViYzr2zAkj1uCm7resJIg0bhR95xPJU8EzqxCjdJ+k+frkHx9U2bOAbt13Pfk7M588AXwNSZECMgiLrAkcmb4qnxxjduH+8IFXDX+ZP+lkGbX0cIxvAJ6An1iQgXsFFrvFpXod3aqaxsmcCVW+XGLyzlXXp+Gl/bZ0XyTikc3T1saZ2Ho1QKDBu+wJRD72BWyV05j2bNEvuHSccOyLPWK8LbNw/wff3uS113fZOfGHHE5cX6wjq3s5hbGWWieg75swit3jXDj8SGqjTl2F7cxYDN8LFph1MuxagvsvOIKvviVT7p772vecbUbOs82ZT/b/fu5fP05BXVi00QvXwALap6cbkbpQJctmG7q299Y5IZ3jV5gyKR/NvXCPMf/roaKA/w+OP9Um4WHWykgUvRFNmHSH9ItSJWgXeZubHue6VcVOPznVbJbPPy8RxQlLDyq5Va3A3YMzWrM9uvKHLmmyuqxtvOfVeecubvBua93M0yF8J21ViqyqZCwHuuLLScjEpmE+YMtkmYzLSXXfJFTHoLkRaB1KmHbTxYZ2ZRj6WTTJdC6CUX3zWfrJd1JR8m2cWQZ3pJlx4/0cfA/r5HZ7Ln9r890WD0sHj4lD3W8oE87Fa2f5vtp277hkGveNcydiwusH4/cgFK4u7EaU53VUi71x3VsqfYl+7COqlfT69z9wn6Wf6rF43+YihC7yqYirJ3pkBxLPyMQqNfjTML6nL5szbfdi3N3t+7k6t55/sf/SAuo/SX2eezhKv/+/bfD5j2st0YZyG5gOenQFw6xaGMGjeHj1nIbMXM2ctEPWVV9o7XIYwpNKv+rNatybrzCdn43inmVt8ZUpY+sB8ueZYtn8CuWiTFYXU/JnaqKVWMo5CyfXY+5xDdUkg4PtBfZmRl3ThVrnUXWamf5D0M7eH3/BioC9o2EcbFEkuKoJOzOh7y1uJl7lgL6sx0WF+vs/8kpxjcNYKMklWhw+XFdoKXGEaMmhkl9yLF03cpVgTkxeXqqXwl/qZv1mGxtK20Vbav3NNgdu9INuaq7a1vl8Gk79XfP4mU83v6yl/D7f/ZNTp9bI182tFyJoYCd+rO0hc6noYPMJoyefj9JtEC2fpjjhb0kJmBbOMIJL8fRpM0H/CyLie8qUvVdhsZyHXJUELMG78rk+VrU4BPHalxzRYlpMTVLlnbeUMCnfdIysEWLSonFwssGDVEblqS0n8DsnGU8byivQKdkCMesw69Zxc0FGFR+qslWocB12a6l+VEK0xmVrQq/1hS+0yRsCRZSD1Xdj+MF0TcdvhW32O/nCUicDIl8WGd8uDTrcXXHYzVnmakZwqZleNmwrALECpRi2FE1vDmfYUygoyMJKMP6Gly7wzAzY1lesuzsU1UqnF639AXWgZQ8MS8mxx1BP76xfL3TYXHZ59Zhn4E8HKkkPFAz/Fg4QIaAgrUIf5xTZMJAn2R35EanrhMaZyyvRPmO3E/qlkfXPQ7aKtODFd6xcRPR2YT+Ya0LLAXlA1rL/tUCH/h0hw/dmjA5ZAhUC7OeLpa93cZV3Lp8OPVBLSqU46Y8s6l0LnK3zkJKkJmOjg2degqoclMmJZM1FSlnU1F9eWJ3q1tjfQ8BZJ2bhIooDMHGdCwmixAftHjXGhIxfQJaEkXWeQhIqThAciJitwXWKtblv2keMptSEWPJlLgcSn1eIXjp7gnczaYahd6wIZR+nvpdz9tWYFbnexaSM+Dp81rj6Nr159k0vG+1jcadKl91zMkU7MpRw99vyN5isKe6cicCWhqnCuvOWvrHYP4YlLWdhO9POg1vSgNqC4PE8GNVs3aki21o9pHK+gxDMbDUj2T5BMPcfs8CO350E54qkKXhp4WKFIoKUNF5OXCYMNQK+Le7Jnjd18+xvziN9ft4QdDkW3GVEb/EYnsENl/Fb/7a17jxyr0M7f3hZuueO+9XiwuzLh9LJ/pozRKtqvRZq2MlRXqMHshy4CcGueL1I/SNZNLopvpa1mN4W45OGDsbqubZxOXYSfRWgsLqNL39ie0Ta7TphiITuwtOPmR4a5Z2JmbtqQ7NcwmxSrTbxoE/HT+T89h0U4HJPUXnV1qrdqjPxDRnEncfVxm5G3RtS6T8hlpqjZIfDdj50pLTbDtxe5XWfOzOy3VuTUIK51StW3lnBz12vbnEVW8ZcWLH80canPpC3YUytc/S1pBdL+6/IE4sZvHsozVmPtskGPKIzlhGr82y/foyYzvy2P6EyrkOrXOxu3ZBKqs7RmSdEXV0PnHhZKmjT15TYOqyPhe67R/PMr4/R9vE1GYiGmdioqpO1KTRpKqlvShRTEtY9BjcnmH7C8uuUMLPeK5NC9s8qksR9dMx7fluFa22l0Bl1TowqHyk0QM5Nl/e5wSYj95eoXo8ZUW12t3x+pLL01Nb9TCfuys8/+NpLdCL8Lfbht/9yL184ktPMDm1j5hJYoVKvRyXBkUuMVr8+vi2xf3RKn10eIHJMW0MG0l4gZejmVjemA/4heIGbg4yvGxLh4ONJttsnkbW8IlWxC4M30I6dJbFZsKn2jFPJfBgFDmJies9w1zc4EvROoN+kXxSd7ZYr2eV949exnXZYbI2xhsyVPXZxXRCmxqGpZrlG1Ges9Es7eQI1UqLm16wnZtvvCJlh3VjF7umqjX1ZTFpvcWOfottEwBxt5B0fF0Ix2oNo6dYerF0ve00MPS6JpjeZ4TPhFX0GU2G+kx3eeskSOOET37xDubn5MeaECmJXtIlbvYSB6rZIJO6TUhoOBhylbBi6wJbo5jb5iaDQeOzi5jPxjUGjE9sPKrEZD2P2LPs7zcMhtBuWwaDkOO1hHevtniF9Zm0HrWWJZa3tUCEQl79CtGlntBVicoupz6v/cOGZSyDBUO+aFg9DRlpck2mocbkmCb2VE4imrO0H9ECMxUuzqi9Gpa2qgk3G7KSbFLVoyZMTfojhi/fd5TdxX72+UUHQuQidg+GnSqEimGz7/OhluX1nuXyMY/V2FJZTvPqJGuiOuJ7FRKW6HgHSjnYNGZcxePyOixVQdExyY0UMoY9Gw1n65DPQCnxmLUhZ/EZsz5fUSpW1lCrwO8uNRktBmwKQ+5rWbb6hkM2YcXAzRmPwZxx3UnFEBuE9iqwLKZPbh1tw1c6LR5pL/Jr5TGasxlqbcvUkCFfMPQXDVtHDEFLhSmGx5oVrh/MMrLbuMphMVJKsXFgRLlkYpvUj5SrrRCvwIt8wDcax9ypOjTcY1g4C60Mbj9iwvR9yFrN0zyoPrjc1cETc1pRAYEWGqlfdrJgiSvGAXatM2rfgGjYOLkX199VfSqSWeNIvxe6kRqFQQ91CzqaaaGHQF9yqqt1J7ZR/UsMt0KvcsVbxsnt6DsTaJTkjcaaPWvTz6mg46RzZk6BoHLtdOlKQddQUfWt2kOMnQpzxOgpY6EjoiMVbF6WjdceQ6DzFiusQpLhdKhGyyn47VdV8FIasj05l0rb1CUto6paDUOxsetpIVB7DVrnxIYaCn2GD9y/wK0TZaYHA8dOCyAagdySFqiGXMe4LBSx/WEzx8EzT/BZ3yf2y1xqAs7TcT7HvskxUCzw+BMPsXcox4GbplPw3AufPO2O/YP/5zlj6rSCOPDaITZfVUTuB9Ivk3WWQopyJZAn6uCG3He007oRlV4TKYx6y3snmXtZHQEiuSuouEGhEN+XdIbv7KqU9K99yaFAD02K+vtF75lk+wuqzD3VcKya83z0PQqDgQNJE7u1lIVNl/UxsjXHzKEaS8ebTrQ36qQ5cH7gOaHh3EB6vhcf57YPTjB/tMH6bNvZgUmcV8AmLBj6RkIm9ubZsKuIgJEeG/YVuOW3JMKa5sAVyrI+c2+5H3p963V99P9x6CpvxbQNTGZcRy4OBbzgXePsfFGJuafqVBcj5xqhfCDf9xwIluddaTR04drByayrUu3tfXxngVt/Psf8K+rMHtL2HQdMNSgV2lWla2HId+2mAonSSPfERKyWfA68eoQtV5WQEPTyqZYLaSvsKj0nAcHicOBC5uPbC+58tb8rfmKQ1hsSd70CyGIc9ejNvb1ze/7301ugl0v30GMr/OpvfZ2t+3Yw1xnDy47Q6azQ8nJ8trNCgMd7Akk8FKlnhtgti6soQATquWCAxbjGjwceL84OkvVjZCLeWi7yEvIMZ+HAgOHycsADnYTVNVhLIv4oVjaZZRu+u9eP0WLNJjyRtMjGa3hmmIfba7wyiXjv8JVcvqFAoRBx3yJsbcFgA7J+is+yRcPybMJkkGdbZgsPtr7tYlduIuh1L3UGhUvFyCkfzoG0Lpjr6csJnOlmqv+1nf53nagrtyBGUq/rtR6QEwjU/9qftpXWnd4Tq6GHqByxV3otEbERp3mpCiOYTurz6kw+hQa75+XlU7eJziw2GMR4ebzMBuLOHO31b7JQvokBv8R2Qk0N3B1XuSkcZL/1HdDOZmGm6pyjeDLGieJuGshy3WyLG87W+NLOPC+aDmnOW7J70vOqLkFnyhIfAe8UKFAweiksP6pJzrAkEvGYxMkNZgdkH7WuolUh5WTOEs3iHBF8XwUV4E0Zl0RuTxhCteeKGDyJwRqCdUvnjGWibHjFRJuP1SOm/IQpLHljmNaxpT8nX2wLP0viLM3WKj7NBLaPGbIFaK3BQt1yOrHUQzgQGAJJ4khPb9nS6sCODdCQr628qOsqxDGuIGcPhhOR5bWZDP+9GbG/LyFoS3jd8BHb4G+8iPd3cq79FDEUKflC46HUr/nIOoHsW/pTMPvAacslmZThzMvCjpi/WF3gjhtG2RYWWD8Tu4lerJYAVqeBK/iYnobygz4fOBORqy3xvuq4C/cr3NrOW4IdBiOxXbFzqop9sFs9quKI4wJkKcEraZCGwpmPwYZ3GKLuAkPbrUvC42WpR6kLXybQeDJlorIbjbtnakzYXYbGOhTWU7Ype6VywiyxNFrFIGrcdLuoUQWJNEJlCaeQ7haIHxfjZl2Y0xUBKY9NKTuqhhIAUz65wssCYhIwFnCV7I2uTft3oeaujJfEkF9k0hC+XIpU+qxFmKqyxLCpSEJTqvarOL0AVRGXduWYvnnrQunxg5JuMU5vT+2o0G/oqQDCOjY3nDCYZUt8RjqEUDmfLgr8XHpfWT4L/dIjlJj0KqznYcS3XB5L42WErz5R4wX7s3guoyK9b4gh1NhXYY8KKW2QMJ0vce34CHcsHGUwM0Hi9ZEzA1SjNaomQ7tdpLD/Rn7y1+/n+tdfyZ4rSyTdCEr3LvJD8+s5A3W6Ytlx6fl9H+oI7kZ90ScVacl6TF3S555ispIowZo0NCUPx+/ZRrvRgl73+bzH1qvL7hl3kgvJjnJueNrDSiAzYPs1/e6pMK5cHBRC1NMVKjxtg/Sfyd1F9HSfj1L2yoGkUDpIFx2je20qIHBFBM+wL72kY/X2+T0f0bwSGCZ2Ft1TAydWW1jrthOwcxPY92yYvuDaI+cxdWmfeyYdOVWkzScw+T1t0ttPOh+6Dw5MZNFT1WuRhDGlYyRwHeom9PQvT1Ir26/9ThVxb3fu99M/+rS3/v/+j8MPnqHVhr/45H3urlr35LjpU2vPk/MCBm3MK/0sl3oFdvk+i8ZyVcFwX004JbVy+sNGjQ9kc1ybzbiq04qSsiNo1FOrqVGtpDuwb8LQv+JzZcWSCbNc31IVYsIKvtOK+nTUZs0ELBE7qarTnOfmYsyruYJaM8PBmQ5bhzyyNcv6KozIYkg5pzXLN9dSaZSitTyZ20lfNEADLc27jFtvyAvrC1zpNqFhI+ZOoMyFXrsshsaQAJl+98CZVuD6XxOJ+pTCq26b7usXAz2BOLddF6B1ZU0cW6d9OOAnakLvi9KQNkgaYcDvT1k728Y4VdWEpH0WOUr4ue0EuS3UG0doVu6F8g0uErfRH+JkZ4mBuMn2oICNE/KxYcbKZ9cwHsDjkeUypfMW8/R3Orzq0ZPcmWzjkq0+x55KmMqnVaOrZ2FiGHI7oDpnKcaGThmWFy123NCqQL1jqTwBQ1no6zMuTaotxQ4l1PfBxPbUOF1MTl2hNunLFSE5h8uj0neWyBvWWFontPjOseYZPhY32G9CXorvXEnUVCqqzFjLG4KAkx042LRMBnBuBsrSHcvCUAYubcNXJWXjB5xZh1rFMumnhJIKr3QOOwuw4Fkeecoy6kG9Dx6L4GVFeNcUfGUmZo8J+BznuZ2YS7x+B4Z/NGd4oq0webo/zd9yjKgo17ppnY2YNNEetJbNOXio4fPRZJF/15/h8myBjJcQ5Qzly6CwwVBfsXTuSMWRJafxo/sM8/MFfuPQGV7+VJmbrsm7hPvqty19fel6QP1V6wkxUQphemVD+27ryOXMINTPQv1u1yWcTqDYOb8OVYVchyBopF1SQEOFCeEcBHtMGiJUxa3SlMSeSaj6nCV7hSF7ADpfsyTz1gFliT5r7aEhpTHgijl0ExVBIlsthUA1j8mvVeydHCmUpqSiGYUjFYad6AIygb2OrMq0QEgty6zObSUtkFD1sEt1EJJWqFW5dmp4Fd6o8lnyLLX0uK4IRzmWAn6aZzTGBwzZ89CUM4f2pcIF5XYqVCowvAma87C+ZOnvh8YcbH4BNB+DcDy1hssEMFFMbxHZOvSNwSNnYXfOctOQz8/ZAh+8fYW3Hyix5yUZNO/rliJ/Yxcm77KDXk25n/DyyX18ePmrFKzn2qlPCjA2ZkpKeMEATW8PdY7zXz91D7956W1kpc33XeTTD8Oc9dyFX7tXq0Z4poeb292P3t39uz7Vfa+3vTqp6G+BOScn0gVvva1cx+7+0/u79wX0tnP0ufr8xV/MRcfRdtq3jqHPuuN0P69dX3zO3zmvFPiJkRM46gGcC8foXWN3cdQ7X7e/3nu9F7/rMxeO1/3chWNqAdJtC3dN/0RbXHyc3jnpHKWbJ6B4cZu4z1740T2p3jl2z01tom3d9SonQz613e+41+7asvdady/pjaf3z/O/n6UFUpD+wMOL/PQ7/4bBrbto+ZupJi222Yjx3DbeFJS4xcszLImcKL3JnVXNQBvmTMK9UYf/Mllga+wTGEM1xmlEyYFgatAgi+OpkkexH2fxpNyZ2BqqMsvOeYwbT/JZDOTgbKNBzbYYocKTp5f4sZFR/usLtrApypBLEo40DbfPwHAONojx8WCpZZlvw3y3jx62lpNejqJdY/n8N7j5hit40TVXugmuvz+6AAAgAElEQVRRYZsLEY0upnINoz6n5ad+d5k2B+D0v9RGNJGpz7liiG5oVaBPwE0PTRYaE2IQdFcXqBMzpc8o5KRjaf/yMDaGZrvFn33uq8zPJmT7R4iSBBsrRpZBrhICcOmBNYOHGKtFn48XDNIOSmzIbmJnvM7R9gxbMlOukHezgXUvx7Bn2eEZHowlBB1xziacTzpMGp9vJUp7i0hMlvu8GttbbYqrRVctOtwH+Q0wIsZEumpjhvKgcRN90YEnjzkxo0VDpWIohqkmWK4/1RbryJ5LLFR/inXXzloqp0BeoqHAd8Yg/83OjFIoQKEsJaT7fYZvL87xzRq8NTvIThPwmBbYrs4ioYjh0SRhX2gcISPHi9BPLaPmq7gFidJXHosSjojBiz33ubIvNjitpFYoOYrTUHTRM+R9w9kIKnomcNVWw2ji89h6h5ONNn+SVBnwy1zj5xwmuWTY0OjAnGQrPMN5jGML93hpX1yopeK1isofsR7/trLIT420eN30BtaVHtAU+IWiAE3ZOHY5mITO2TQsXBoznFu1PKjK7wcDbrgm60LILhS+ClFFqgyG5P60JMwoZHkGwhG5JRjnq6pQqoBY+TLTTTkCo1BhAqcOpV65pTANiauATZWyAl2SNnHFCJKiUSGCcsJGFBo1biEmlYb2UYiXIVB+pBY36udaxChTQMBN2wrEDRv3/UvSRuoSTrdIY0ihUz2V/yrWTXmpooG1bR6i45bWUQi3e64C1yj/T8UcGlMCdAJvSmLUNgWlNSmsnDqXOBTlxlYK/hzQ1IYi0iR6LSFj8THZNJ9P84QdMGT6DdXjcOaEpbxVCwxDsNlw7hjk61DYqbxE4/LlVluwpjB1w7DRGEqSeGlaLg3yHIwamMoi1w0PEEiMWWyj7gkClsrNU38/nOaej2QCOuESHz8fcnV2iL3AXcTs8wquKZ5qrzM66HHHZx7iNT+6n6lJNVB6j05vND8cP59zUOeAUpp24DpF7/+0pb9/o1z4/DN8tPeefj/T43/m9Wf7rPbbO87F53zhNX3gYuDaPZfv2d8ztcF3n/R3febi4z3tPL57u4vO8XuOe9Fnn/Ze75wvOl/3/rO0pc7lwva9bd1JXdQ+33Usfb73vOit5/98hhZIAbeh2YIPf+Sr3H/3AsObrmG1WWM0GOSm3DTDfoECnks/m5CWYKr+w2Jk+EYbZqzhpzMBB0qG9SQVbB3Jw1aZqufSHLfHT8GGgqGzivPZLI+noYwk0mcs55o4bbCDbSh4ebYGWT71+Dne/8Yx3rllgtUjWU50EnYUDFftMey4zHDXrOX2ZozXNM5ZYtYaNkoRPtE5JZw3GbKmw+zC53jpC6/nlhsvd+FWl7gsINYFY+5/gTBNCnqo8yhMqoKHHlMnVxflEKlDCpwpd6wH6NR3xfTp83po8hCDpwmv9xn9r+PpGGIsMSws1/n9j3+FtYrkPUpEel/CWX7RhVmVd2c0a2r20m7DEWy0iG3PEvlFNmSn6M+McaZxktnGkwyGI2wJh1w1ck6SL3FE1Xp8M+m4KuKOTRgxPjnjcdxGLONza2aYb8ceUbPKJlXhrRoaGTi3YGmvwpFahNexrLXg4ROWxSh2Mih/OBfzhLXMRB7SxJ2pxTQaijZ4rup2WW4dVY++Po+1quXYSsLwoEexaJ12V7MCkquJapAbUGGDx9fOrPGt5prT4bsxn2EwsBzsWC5TMYKRBiIcEdObNZTzqUisUtj6AlhuQtWHvVOGVsuw0DZcmjM05aaQTVVV1Ib6as8kYhONi+Bt8kGOGStN2CmP2qphvRbzvvU5tmWGWTEBb8Zjt2doi+WKrCv0KSr9RN9jgmN4TNvQkdRKZGnbgD9tnmckOc97+6fZslPVQZZiv6E8lQITpc205iFqQmFK6Shw9BE4v2h520SRNQzVdoPLChknz6FcMG/EYE8odAjetKFxjzxG0+IH+Y0uHYL1AAZV7aoCBvUnASFVt3ZTQDMDkNtuiBVuVX6eWDM9+lTY0s0j7frHmtiQqKiwofzK1DqrMQth1jh/U8eGiSnrtqtYUNl3SU5EtT7O2kzs2Jbu3xoPOp4KLsTQaRzpHPWUu6+KMlQEs7E7Gag4QuNG40kLAoVoF9JjONHlQmqH1mMAHYBSDoBApkCgckRFpKg4Q8eV/aXAncak2MgmLBxNpcVGNhhmD1n6B3BAt7EMZV1PFRqrkB+CXB4eOAFjGw1DfTArBlTCyc1U9ut9Dz/BTxwYY6gv53xpXXuKURQxpEuSLp6KZPoz2FqHjx87yP7yNNtNhhFZKNoO4ybjWPdypkNl/oQT8r/tlq2p361bDLpv64fiR+92+UNxMv+fOYlu3/9ndT3/K+f8v7LtP6tGeu5OthdKf+TgDL/3kXuY3rudpXqDKBhmKLeFwAsd46EzUnRmLXZzNQdy8NfLCRtzhp8OPQ63Iu44Da/cFjjh0qFRCEbh8COQbRn277A8etpy3Q7DUAHW5lPJhUbOkjOGTmypG8uhJGHe8/jS0ll+8o2W//3GzZSNT3VPzOhDhk8+bLm5AJsahlvw+LZvuLcFK+2INYXnCJ33Yj5JKCnX1GqVO8hjT51j7twaExOy29JsohBryqRponc3ek0geoiOUwGFgJ37X5NDN4TanSjcBKK3lQKg3YmRUxhWYZ+LFx96Tw+BRk0mXRZPK/lPfPphTp+YY2LDKAuaOUwek5nAxmuucth6YvSyLofO2hZJJG7IRxm+Ye0gp02Gp3LbKPTtp7X2LWbrT7CctNjhFziG4QkTspeYbV7OMXTL1vB1DC8OPGbigCiqczyJENj7YOs8d84uU8pvZmLGp91OaPjwV7PraWmkZigZ8Grm3iQTW0n/rUExCzkBT4WaG2QG81zdCXi44fMftzQ4eDCmmmQ47hneuppw1ahPTkypZCEaUPEMu/KwqCIA7We5zgPGst6OnXjyec/ncQw/4nncYq0LGfZlDYsNy+0xvFii1SH05WEsY+hbh2wnoo3PKclWJFBsyPLJEPnWhXufqFuKGYuYl8l+gxdZJiU2e8Zyqgafasg8VdmjoogSPk/Ma2yAcu9Wifh01OI2G7DfZDktNnHKMHPWMuLJCtTwzXbM15MK/z0/Tq7jc/rhmB1y1pCUpmzTVLmqis8xS0Z+rhVcgdr4BhhNPBoe7MsF/Ob9K1wxkWXftgxxnyU5lIZ4tWAQIF4bh2G5Qaja9QSszsGGNyqakSb9d3RuZ8SGWsJ9hsEhqMj0fgjEgnVOQUZpBCVcPpkqXZ2nqgoNVIUqseLYOK25TGIwLblqQD5ScZshkSyNwJLGjYolFBaV9qv6v15TPlkvBClwpddFPovRVl6lQFsPSEqof49xsjSJtPKUZqrxot9iu7r7VJ6cXtc1yFFCoVu3D6H+3jgTey/JIeX4CVSqqKSapKyhAKYW/ZLcWYShtmFWBYRlaPYbJxzsnwG/AfMzlsEx4wCxN6HbgcE0Le0tEKxA+6Blcrfh+FLCcq0AuX389clZfnmsL2UsxSq6VCUwU25IE0n4OGPpP7qJwcJTfKazynhmA6Hzdk4Z2DYey9EAWy/dw+/+5gO88+1Xce2V/RfSntIL/cH/fB7U/eC/g+fP4PkWeFoLOJzicukSPv6Jr7v3Kh2fajDKgfw2V6Eq2YUpY9juwZMqkMHQH8HXlduSg5/abHhkJmY443Fg1GMwn2KXyhL0dWBYIbKymDOQZpPuw3NnYGQT5CR4upSKFStkedazNIj4Uv0Mr8x4/MaGSxnsGOL+hP7YUNoLr/OhMQJ3PmEZbFuuSSwbA5/7LI45+linwSV+lrPG4zJVQ0YJm7YM88m//zL/6jWvZmLkgFPQl+CqmymEU7pYJWUMvgO+3PLazUTSV5M6crfiT8nZmrS6oNBNONqHZFJ64dZeS2sy6uXZdSc6hxVCS1X2X45mkPVCA5PbiPGKJPVDqWhzmE0lTbw8Eh1xs6pmtaRDQpt27SHyNiYJ+skUL2OtcYhk7Q7m+q6GzATDJNwXN7ne+CxgGIkrztbql9qGK0yRsrXcKXowqXNDyXBnZYU3JAW2T0+wr+ixuQD/ZmCQcItxKjWqZPSW8yQrabFSI/HoNGLCVoagELJWC1m4wzJ5pU9uwieqGi67WXl0PnY1YO2+mPMBfGlNumg15zfLmk/+lGGbn+VMRdc2y4ZsibX8bo5Z60L5f2pC5mLDNXjyLaEU97EFwz6T0DbGVZpWWrAQWbYYQ0UC10qRtKn/62GF+huWrcU0TD8VQBDDFt8w37Dc3YEXyOoqMDxQs7zaFrihKFbQ4zovx0GboLqEYSyb/JAfMz7v79R4t2e4OszxyKOJVEAcWTuP5ZOtBX4nHGAqHMCYmBF5rCrcJ+vJTiqOu3yXZfDm1Dq4E6VgKXNegr6WZt0yZkL8sJ+PfmGOD79hivIOkxYjKLQtrbUOjEr6RABuwjilhG37DIFEnwWCTuPM5dmcuiaIVZQuqicXj2oq/OyUFgTe5MsrVkogSmy0KkB0vhPgKVReg6XDluwYDE1B4xCEG8VodcOri6nUiPNe3dBltFdVEKNUBIuRL7nCss4mrAvs3LgzWIVSG91j91li5cRVQena+u3Cu1oYiawW4zXZHU+6kQgzyUlCCyWNU52/tuuxf7oWjTkVVUgW5gsWI/C5vav9l4cwYxlQqHQl1WvkhHV2b6MHDKuqGl6FUx3LlqohH8JUP05MORkybL7EkM/DJYEhcxoqW4Z539davOHqBrs6eSck7SqN+7vajPIC3m+wa5Z9ewq87Yksf9c5RzUzwRKeKw56KKm6vOVl26GKLvZBPv6pRzmw/4VkFSG5OIWrd4/5Af1+HtT9gBr++cM+3wLP3gJCHIaHH53nP//WQfo3T+AHl3MgN82MEu79Em/yS64g7etJwoCF63yPe2PLh9qWz0uKomapVT2uGoPxHFSrMLpLK//UmFwYR6v/TTlDbS71ec0V0xyfxTbMVqzzW1wL4HhieKTT4nVhifdtGGPc8x0AUy5mdNK6e/OGDPQVYHqHYWEq4W9W65SP9PGaQoA02b9Ub/MnUY1V2+FXMqMEyifriEIrcOg8XNvA5WY6pk2TQI9ZUCitNyEozJPrekxqQhE60AQhpOCM0NN2c5OIJhD9q23F7OnGq30pn85NRGn4ytEUugvqs5qYPUMm05uFYvDLTlxY35XJjCJbMIc7hT0lQhk3MH5a/EAinRGxZQ06tUfw87sJCrvIla7DrHyRXO0RmqIo/DKjfh93N5RUleWAP87+oMlL/TUyeUOOft6aL3P5yDDbcjGPnYi57xC8+fWG7S9Sxrsm1q6bgIoMNDlq0lU5qnKxlAWvisXHk3TC3p8jUWjwpKV9zpKZLsIBoXiXxAdbA+JReIm0NE3BNYWdMTRWLY9UfaqVCfjyvSSdGgveGYiraWNmN/F5E/KI8ThrLa+IMuz0fbZ6AQ8llpuylv198EVp4BUsU02Ph21Enw1ZiWF/3jkWcrSSnna5TykBsu6yrqJ13iZc2/H5yLmI+2oeP+cLKGb4lo253Fre1u9xV83y+Y7l50M5Wnm8xe/jhO2w0llj1OS4Isw47+NP1Jb47VKOa8wghzvSUISJZmpJlU2gPgeFMuSkCSfXn6p1RSW54TTU560YMg3DmNGxsvxBO8tn767zxkofuXHLwoPua2U8trRUQTppaB62FC41+GKD2tA5YvFjQ1b9rQWFHYZ4EVfxm1uzdHxcLpkLla7julNnTdXPqead5EISMadnIV60yEtW2+RHlBoB9Rk4fqdlmypp1f9VsCGtIyHXbtdEQK6WFrW5PqT4rxgzjQmFR50ki4o90rw3x2jJFUP9rWwINxmScxYjBk6MnmPDu3cy9cNloc4uaFPBRL/DeGnYVWxgD/AJTOocFYZWG6viVrFQLdIkfxIaMhus+55KWt+0cAVBwsVD8imeVJjW0KxDs2XJ9htWFyyVIegbhvNnLbnQ4IeWS+MAzud5eKbBLoUjlIsi5wwViMiOT2s4OV9EaZ3VVf0H+L3KPdyRq/Aav8/dkv7ahPR5OZ5IKvhRnqHpbXz0P93Nj735Uq67SqtJXXSXlXz2G/tz8s5znlP3nFzV8wd5vgX+mbZAb8XXbBs+/Dt3cP+3DzM8dSObs1fhJy2u9gvsC0psMUr6trSsJz1Z7k0iThvDL4SGy0owYoxT75+SPEUORLZ0xmBgBGqadBTJkgH7EhyswfSIcV6YLSXJx7CovJYEln34YKPBv9jo8QvFASbLPuV9Np2FtZI20nA0ZKQn1UpX7COhz/X9WaY3w6eWLM2K5XLPd6xiLWlzX1xjtfoQ9cZJZ5Z++3zMz/3oleRt6CSKeqDJiSd0AZ7CUgJo7j2t8ntAzIGxrrad/la4UROD/hag07OL0dw+1C80meihe7AAkgS6FcUNobEQ8Q9/dw/fPnSEQq6PuhnAhONOi04hWCe21pnHBBLqamAE7HQwJ8AlkKoseSXsSAql4v42XoZ6OEyjs4TXOU8lKPDqcIz3Fwu8JeNza7HAq4oFXrGlnxeN5rl2yOPaUZgehbDuMW59dm/1+MrRDlNTHsURQyQ/UhWJSB9SYTlJWehM+lPGQQEjWSwldUPStWqKCxbZ/UYlSyLGRvlo91ms8pUkzis3jZpH36hP35DH4LBhz5TH3r6Ef3jA8JLCdVwSZHm1F3JjUOLWTI66H1JrnWK9vcDROOHe2PDtuMpTtsVeL8D3M5yOLdcUYUvG495OQiv2XLpAgKGmPMsYlK7VimAuTud/aeFNGMuXOzGH24bLPEMhgb39PpUgYaWpPM20Depxqt122hrut4lL2zpC7ApRPB++1KlwbSHipzaO4kcJqmrUAsYrSUEAsiJkFTEMoKicUgEfsUELKZPT9OGpRUuuIG9bGNigTpXhXx8+zyuLOaYmfFepmVNhhLrmBNTPp/leThBfROcyRNKFiwyeLlb9Vyyc+qFcKORmIjZdMjNZiMQ+S+xeC5kwFR123+8KqMglq0rXjCN+6Rs2xA3DqjQND0O2T2K9BqvP9AqBBNIUhtUiSKkJugQ9dB46C+W6aZGk93rYxEmSaDwZV9DhqxJaGQiSENHiQZWr2l4Vq1223IVfJV8iwKdiDI1VjVuNsy5wdFp2YgFHjGPfWmcMlSXrbOVcm+g8FUE4lxrE6HqUr6hiCoWlzz8uwX8Y3ZY6SNQrsHYMGgXYPG6YOworq1pkGrI5WF2ByXLC546u8oqbypTHPJdm4PL5BDo1hHXIjakqwOSkz5fPLfB4e4BdYZ9LGVBfHfU8JryAVbJkwxbVxRPkyyEv6ebWfafKK93fD+rn86DuB9Xyzx/3+RZ4hhbo5dI99MgqP/OTn2d6zxaWzOV4XpaOF7JZIEK2X/LKNnBrYDiSGG43lld04LXjHiXfOEFOCXcq+lEYhmwZqqdUkWYIFOaRd/YSzplgOYTtE4aoAk+eszTbqfPWCevxd/UWN5gKH9w1gNfyKe21FMQOiUmTSjvGVfkp0dlKp1ATrVixair3cKkqbJfhYS/GiwLqJqROxNrKl1jTqtf3aK0d5Q0vPsDGjaPO1UGWQE7zKrRuEnSR1h6I0yQhtk0Tj576W4hMOToKuwrQ6SV9RO9pltVEJlCoz/dAntuuO7kKGElzKufz0JNHePev/j6bp4dZqAlk+hi/jJ+ZxMvIEmyQuPawy4JXXp1mLeP5GDczaLJU3K2OTVKJfeXhtcJxXprbzA35HWTDIc76JfZ4fbx+qMwOU2IqC9lBmbiDZBpUidhWiCsvqaKUaNF3eGLe455ZuGarR6DcJEmUCBioSygRXhOlxGW1DyWKj5pUVFaTqyyqthg6qm4cNuQkMLeYVqc6xf0zaeUrQ2mbKY/LjoJ3QouANn+4tMaVlTH2Z/Kc8/q5OjfIDX0FDgQZbgmHsPEIrx/N8dLiOruas9zeavDl6ikearW5j5yTEZmwhpXEo41hwFgX+n84gnHP0JAXcCgNWlWvWo5YQzlp8bnoPO8Oy4wb4/QUZe5+uGMYxSMbw2AG7opiZkhoYjlH5MSOb/SLrBuPzzfOc2XrDL80voVhAsfa1ZuwY69hahrqauepFJgo6i63AqUl/L/svXmQXdd93/k55y5v7fd63wA0dhAgSHCXSGqNbEuW6FjjRR4no5EnmYntJJM4qYzHKc/EM2WPY2cmjuOSnXLsiW15keXY2hdLokQtFDdwEVcABAEQezd67377e/feM/U99yHl8szfpGuqu0g00Hvf7XzP9/ddZJTwDREGFtccn8CxrxQwrWslcVzfDni412T/ngF3hVWv6ZLJSMdcRhZ18kbSrQlYqalDblAFOwtEyAghwCQHqLLiOtBdkYsWH/4slrVzSeH8oB5cfb3BmvL9BEQdgYhSmSyqhrhs6C3m4dGqdxt70BDI6bzf5OYDaduGew3PTOk+kaZMbxfguwm2dJ9IUydgp4/3wd8uZ+80w9am4YbxJg9/62kjITZYv4dn0+XyHbpKBSClOxUY072nYyymT0yc19zl5goP+HS7ys2tSuUDyk3M+2bFoOnZUhIQLBkP0prPQu1+6Vlh9bNQuQtvilFH8uw7jB/htpWxqZ8jgxFnqI4a6mWI2xH/dmmVW6OEuw5XlESJk9kkcjhl9+k4KJsvy6iGBQrLLT631OBYeZ5pHDEZJ8UQ2wK7gzLX0pjq+DUe+fx1fvSHjzM7V/C5ddrkvtkvO6DuzT4DO99/5wjcPAJ6wMl52Df8xW89z1e+8yIjuz9ESoUll/GeoMLtyqjTsmXhHumWjOFS0uEfEPJgJUCZo90WlMegKq2OtHPqoFyHqvovw9wMcW0976RUC0lfmWRy63Vgq5lHOClo+1fX+xy+JePXf7BKpRcQTzsm96vaSOXiBrMnT6f3xd5iNV4VA6R09zxcdaD0+Gsw2jXcMWIZG83Y6vQ520/Y7LxEO+0xM5LRXN2Ux47vf/t9hCbIMw81BtLCKoCnxecms6Bnph7aevFAzy8x+eJ0E82JGdAiJeAmVk4fp9c336bXwmP+4a+vnX9BYy0vvHSOj3/uy4zWptnWgqLm1+IBTKyCT4mm+l5PFvQvkflokwSnIlGtaK6DMQoh3p27ZI2lGM/TL+3nrrDOjxfHuC2ocr27xNdtxFov5qDJ6Kh4PTLsmspxZ38JBhkM+hBa6GewcQNO3Gm5dCFjuZly6I4gb8XQcRHQlZappVYY5x2FHm8qeFeLVRv6rzjfElA4aIgEKvSiCqqDhmCfwahSzBifB1ZQo4C0mdfyurHmbMDPv9yluhlydxxjbUIvzZjN8PVkSRByNwXePVfk/Qt1HijO84HxMe6vRcyrASEccCNb5j9uWU4mDWbZ5khQITCRB2tzeVsuTyaaomv8bGlkfX6td81XNtVtyUu/5gN4InWe2TvvBkQmYEb1dZ6x1kbFcJbMn+rMpaxnfUbtdQ5XDhMnFfYUnI8mqewxjPVB+WRr5+G1k44JaUPnc5I1bUGykl8yPbXkCExFhgdi43VsoWrzerCvVOQry32+Z5dlz1xEaxvar0NB48Fh5pquOZ1PNTIEN6NDBJp0fIc6NYEoOVf7N3R5GUK5R8W2jRusgKHaj9acj7HRaDLaY7xDt+9dqXkzRPEYVG5R9AekpxWMLMPH8PqXaULATPeAQL1uGQF9vdYGTTsgMXVNsXlivYcfo52fPkiyUX2uLnOBVHXJit3S73Hz+lOene4pgUbddxrF6tP1dRU5JOOJcu50rb6WX7Pe/SqHqhQJun7VJav3677T1ylAfzHfeMUHDKXCMLxYv3cKlbFceSH2WQBX+XXb152fOAx60uXh20EKzpH2A5qFmNfXLvP+I/O+MtR2LLZnscZiU4stW4yiUQJLWrP87hfOsW+qzJ6gwtMuo+H61N2AxEZcHKxTImN7dZH7xmrc+Y55z/T/TWDrdkDd8Pm282rnCLzZR0AjM+301s60+Km/+8ds7L2NoHAMa0I6JuJ+G/mKxgtZm3cUI4LM8Q97Xf7bKOaDUwEzo4bNRr55Vn+iHuJ9udJq0OvqAZc7vbqrsJ2oGtUw5hd4xxPXHc1UO1LDBWP4N50Gt45v8es/XGeiFnL5WxnF77GUPcOgvIpc+J1K/6KdeA+CuiGVvm1Ynp6qEknfZxpe8w9bQ5wUeKzT4nzzKaouZavXY2aixLefeJVDhw5x+5F5iJSOPIwfEfumxUFA7K8CO73tJlhTbInAi95/M+hO8zR9jj7Oj2SH7xdjd5P10wkXFhOJQMDaRoN/8q9/l8tX+qRRzECfZ+VylYCnQMEYqs6xHVRJu5cpuBZZMIINJzFh3X9cUL4Vq+byYBQ5Y1OXUnIDXgiqrKcd7gprjIV1vj1YZjYo8Vjq2B1ZCl1Ltp47jgWqlBEXyf3YyllWLVgKa52N4WvPJ9x2wlKasLkbUr+PxmbSJul3Ekun8z8M9RWw06jRnfETY9/36lkVLbhi+aYFznNWLxS7q3N72nlQKZ2W7Ye0T3f4s36Hd0R1jpcM14cjT7VKybS8t+Z8BEpJYbxiHpct+4JR6sEoD0YF/v6uMvfNFSnYLT65do0vk3A57VAzxssGBND0EjvDkjM8lbW4ZAIeiircE0byFbA4gBMFy0Gb8LHOBs2gwAkbMEHA8y7jgLG0TcB1tV64Ns9tX+fv7x3hbaMzZBspX2w5JsuGu+uGngBIBk0Bk6LGt47K7pxl627k2X3yqnSX4GJLHoWUWyLLWtN5oHwtc1zIIj7bW2ePNj+dGrFxlO7MY0Wyy7lxQteeAFAgrZ5GkgJYuk51HSqPXbkzHQEr/EZJVWCR8uYUNn3W0ZHmTjl7iv+Ic2er6psHAkP78Jq9SPl2kSGdhGhRQfDGZ2br2/jDqnGmAJoQ380Nkd5Z1sw7DwT2QCwemo58/3JulPBMm79Php8rplHHTCyjcu1030lzd33ohpUkQ0z/+b4AACAASURBVB3sAo169ugaHs/ZN59FJ5e6rjsR2WLxdB7E+p3K2TI7pSacIahT5nfiiNV1rG/fyYFuJhPZGpQLEItN1qlcy1k7NXfMHAsYWMulJevdsssbARf6LdYGF/i9Z9fYt7dFur3ChbUbXHl9mUurK1xlk4uNLr1qlYmKoX6pTCd5jT9sZNjCGH87CLjoHJfJ/I+tjXWbMr3CGhe/dJYPffgOyqo2/RuQW6fHwc7LzhHYOQJ/E46AX5Hh6198hvO0mBg74ntHt7B8ry1yuwn4RNriViI+3zV8Mmvzw4Hl3eWQiUnYFtOWOeaqhvJCvm5kp/JF16rCSO0OK3n+1sK6IZJuaAhyzhjHxZ7hLdbw8Y0++yf6/Pv7xxlfDrj4TEplzjDawRfQW+2o9eSQw20owE60c0+9ARQrQbjGLFOwddHxTAuebDrq6/BaCEdswLNYurbodWdbSpPH8BP//LO8655D7N03SpqmGC06Wus9YzZcAPRvPfR1rPS/QKSApp7s/svkPcaeOdOi4dmEYQWYPk//lk5JeV36fP1bC1sMX330aR59/BRTs1NsSEilOWbWI+svUbMVNkxAg5A5N6Bev48z2ycJBfoKe3BGqhuNYSOPIUdi1UlNs9k8iZOxINmkYGM+Q8jfKYzzr81ufr7f4YfiEX613eId1QI/ERbQgr1LrtAi1EahX8zT9PU7i7yoGMPfqkV85mMJf+fvRn5RUzWtZ1JU7yABu5L7pXHScRHIVSXYJHAHvn3AXMF3YWoUlj3isPcOx7cTCnjNXZjmVrAqZ1/KGGvBh/a2+ehTKdsVAcnUx4T8dJLxqy7weXUviEnEMVUx9C5DUYL1ckpLPbCvBuytVDgyB0fKCvyd4X1Rn6/1N/gdXZBCBkIftsL/GNZZH6zwiGaRxrKLEkewRCZjU5VkfcfJ1LIZlqgRsJ45T3ht4viTtOUbJ2RG6Q9WeXcB0o09HCikLEwZRmTCcPBCw3E4NAwszO/J21Suns1L5EcVoTIJ/QrobZpGq6Gj18u8xlRRJdsbjnFjeKtYwUKdf9m6wnt6de6+pUywz9H5rvYRhmh/DprFVg9edlgxeGLCdsPgVO7s1Igxket4GgqKa3wF+pecZ9YjjdSrubbP7jGeOe0OoL8No2/Lw739dX/UEF3LR7wSRES3ie13dE4bSncOWTIPpIZa05vXv+5Z3QSSLOhFmwC5rnSf6H+9eMQ01J6qz1VyUR9WLa1fDnB1m2gaIG2nNgk+mUjh1dowinEXYNWr88qjG8aW3BjGn8jhW4SupAerqqjLY1M0gpbfZ6BqvG1HFBg6ioGpQXQMxpfyEbSAubSI0itWpgLsTMDr5R5bWY/Pra9xY/BdItfjY/0N6FxkYibhJz+qm0MOVv1yeqDo/5w6veW+Of7rhQBzaRcv9jUbH9B0husmYow+rxvLfmNZCcb9Z8zWpnmR03ztyy/w47c8MNxVDY/dm/Rqh6l7kw78zrfdOQJ/9Qjc1NKtXmvxv//apzi3vMDBybfQJaJiAm4LI/pkjBFwnw34ZNrifS7kn45UOLYH1htQryqA2PjaofJx4yMSJBTWGEKLbHsROpfzCp9GRWYK6HZhYxUu9TPeMxJwptvH3NLg1+4ZZ6oe09rIWHze+PDPikZBAoGqD1LmlHCPdt5j0FRN1UVI1iAQmKgZnnvN8fCZlH1Ny0EM40XDASPB8YDPbz1PMRzxTQ2DNGGsHtBpnafR6/HAncepjhbI+tnQ+aoVIR8P+nwKvxgNQZ1+HoGzm+yb/q3d/03Qp/VKC5Se32IV9DIcxepDVHNns4Abq1v84m99nHOvb1Ks1ukk+oKK6sg8idG1Ie8M6ry1fpSPjN7Jh2u30UpLvNJ+mZItkgU1bDxFYNWxYMms2g7kHCn6jti0e4mS6/N0NM6jqeN9lRIPxo7f6jS4NxphOUsx3cCPXUuZUToKaoOQCFxjdI32BMJLFVD01pcvGLqrhqNKANa50C+Toz4vKHcCZGKCNG7TMRHg2yOto8SUzo/1PIsj96Gq/QTC9ZLk0Rde29TJ2cKgZLlyY4vfv5pQK02gYOByIeUb/YTvI+TQFBSzfEpek/hdQddzuV6tUDNc7TlqA4eqymIsre2A3cUCbyvV+XAwwrwY4qzLxWSLk93zvJT1qQVFemmTTVOjqFqywPDVVE5rw4IN/OkskPl7o5MZ9hUCeoHh8QRu9BeZTjf5X+q3M5NZP73fGsBCAe4dMbyaGs5vZvRbzo8767sg3TasLTpiHcgsD9yemofNZTjdz9iKHPeUAtIunO7A030YV+NGVKC9GlO5u8ODd1bzQNt140Ga+nUlRxDramW+2BK4UVds7rA1el8xP8fegyOT5hO5gUlMV3FKhK/xLQi2ATJLFyag8k7rGTCdY7Uh6NbQ91C3r/X1LxDMGLbOOeIpQ6B7VNe8EKruB2XoCcf4EedQ3nBzs6TRqzZRwjjSxklbp4+TGUJRRxp1KktOTJzeLjZOTlbdc7qfLglI59VnnsEXOylNoZg5RT6qjkwAU58vlli4SbDK5xLm4E/SEh0j/azJBr7rWM0TvtpMZpBpQ7qkCYQjGgvoXTf0uobz5VW+uXyZf/HNp/j9y1/h4eY5XnLbvGD0DA2pFPexZWeYG7FM77qfifl3Mzaxm4np40xMzDM+2ue1632+9dxlvrXxCq+3z1LuX2BEvc6myIFolC2jacmAB0N4KlOjCrRW1mibVd73nhNURkLP/Pv7MT8kb/ifO6DuDT/kO99w5wj8fx8BPfAfefwVfvGXHubQbe8gSabJTMz7bczlLOEzLuMjtsB1Uq5lbX6yMsJ+A5O74NJybvmf2wfBuBZ8jVIVs2C8E693HfpyrWlamkItzheBa8tQ1EMyCfjTrQHXVhr86vvqzExEJMtZXmnUgvG3GzSac3JNKgagPRzBDhPlS4cU4mooKPh0wtB4ybF4Ch6YCzg4avAMQ0n9jJaX1lp8o/WyN384mQrIaPcTZsZLPPrEKW5sZrz9xBGqI1q2M4zYOAET2XX9HGZocNBh1L+1LOgX04sWJ8/YDRcyv2IM2TkJuoesn5gtfZiCbDUi/rlf/xSf+PTDzO2eZ6WZeY2ftTGJLXAkmuJD4w/xA7W7fdODWjy8U7cwy9cbL1JLVukXFvIKMcm79eO6AZnLsIo7MYF3Ll/vr1KJxlmyNS6Hjg8Glvud5Qllu1HiU6TMp5aDFcP+WehIPyW2TqGrWvxiaG9AYzPvi/3MmZQ79lvGdplcX+ejKXIBvIwTHsxqtKXoBo2zxzwZRv9s3oLgNVLSfmnRHa73AhrS5zmNcM8JCBrsjOH1rYQ/eKLFfeOjtNQQElqOpqFf1OQNCZJ8mlfZKw2iobORV3Mp8f/yIpxfhXrJID3aqWKHUstwQKHEWPaYAndFY9wVxLzeX6GUNhjD8LZokvdGVS7R45GsydnMcKsN2GcNzwxafJ6Mu03AAI3EDQ0Dm/0mG53TfLh8G1lQITYZW3JnS3s3qgoyOKLxsnN8WhuhxHjWUYaJmSlDKiZKzNOwKu35dcfHBxnvSUMmY1gSuIrgrlnV5sHp1PBQJeD8Zpe3zFUoy+kaQ/eyY+3zUNbXUnyGALaqvYQCrg0BkdhT6emGGXeNZ/MQY+koFRCujZMYMKPCewHsBwQWdUHn59aPyXVfrDsyGScEAAXOh8xtecz4PDsxtkYjXYEpmRd0vwi06XqRYeKvjmU38+5VzRi9bk73le4p9QHXDYPn8mtS2Xcarfrvpd9NgE2bg93Ga2r952ojpbgQMXhi65TVJ3CnX0FAU3+VmWclz90TYeuNQauOQBekYk5UXTkCYiqtoprCvPauHwWsXbJ0ux0+deo6H998hr/3py/z2aunWU4bNIO9TBdnKRX2UynsYxCO05dUwkQ0By3aZozleC89U2LZVlmzdRrMUB/Zw8TEHqrFOllQpNu7yPL2t+j0L/nNRhBNUfBa2i6G2LeUTO+KOfnoc7zrvYc4clAn4M2tDtsBdfm1tfPnzhF4047ATZZufbPHP/ufP8XFK3vpjd/PQiDZuOGa6/OsS/mILbLlk/MH/J/zFd49GnBjPR+JTs3k8QuB+i8j6GmqpTHeALrLsLKEr1MS0FP4sFvPp469nuOpluEz/R5fbXf5Dz9e55YoIullbGoRuAaVe6F4m7LvNBETgBpWF2kxktZHXZCLzo9Lpde7rkVrEWYmczBXm4LtnmFETtyB5ffXrnO2f4FqMEJPSatZ2wOhZidl73yZbz/2XW5stXn78SNUKyVcIfOJ/97NKrZNrIMYBC0aevir1FNv08tff78WpSHgyyNN8nGkDAW+JqgYcPLZ0/yjn/8DJucqrLXkZi16C2QQzZFGs37cMl+7n4Ytcsz3RhiSKONIWKBtpjnZfJwxE9G1Va+j8/2wcsXq5wlzNKYsu8h1aA/W2Z2u8hoznO42eV+pQpKq6s1wFI2lM64PUmZDw2zR0NgadqlWIZqAqm8ZyJmZfaHh+euw94ihIgZEi2qOKPMMLi3GWjj1drEp0j9qyrlgMWJUruSLsp2Bnq6jdm7MECgxGo/pN5BJQF6LYoHPnrrGqBlhzBTZ45zX1S+qZaJl5NGgoxDXxDA6DkEfugqrVb9ny3gwdGjKeBzxSsvSWjdM27xxQiPDftbnWY1Xg9304ykmcZztXuXuoMr7qnW+t1bgQrPLMil9W+JjKrNNGzxlC1zAsZg6nupvU+pf4mhxP4fiWR6REccF7I/gVAf27jNMV6CzBbtqhrftNqwtOb6zBTUH8weMn0CqYmqj7ej3cpBxupjw3n2Wsbb1rRCFguFFjUxTuFNAMgz4T8sDDo6nHJwsEhQy5B4O5/McuVA/qoKDNZKUoaCUv11sk0J1Mzk9q4bivPE9vCouKYuZkg5PLLMudTlnhwYXH+yrNyuWRff6cyDnq0a7gyt5nlwgNltAbM15xt4K1MmBKvORJBO6j2XUuNm8IkdtN3ezmwmT14Vp16P3S1uqq8E60ucd6pvVBkHAzP98ug2Vnyc8o/G/NghDTZ3+LvDpP26ov/NO7SHrL1bO6dhICnIuZ5DVe+vlBHq+iJnDMFhxBFuO7sWAdmC5mDb5o6+f5n96/GF+b/0yTy+PMDNnGS8UcKUFivEs3WCEgY3pYXGq+HMJzoorVv3i63ygdJh2OM4eW+JdQYUVMprE9KI5evE8zutiBwSuy0bWpr31DbaTde9mD+0Ex9QT7Qa+Baa7vM5yM+EH33+EUjHvP3+z2LodUJc/93b+3DkCb9oR8Iu/MXzqS8v8u19+gukjx2mFB5gwIRsMuJAN+IGgRMmG/N+DFr9Yc7y/XsGMOGq354t+ccxQnzFEw1GIJ6w0qlNAaQ9GZyFWdEAMlWN4sHDhsqMTGc71Mv79Myt84u+UeM9byiTdDBsbWmtw5YyjdJ+y0YY6OrFdKsZWAKlGSnrwK46gYmhfxpeLj2tB0eZdu3fJ1Qx8Y92x2IFyYni+dYmXuxeoRtP0spYfwQrVGQZsdgL2zo3w7cdeYn1jmxPj+6nPVgli5bHlYcNeqK1FSUhCTIVYKaFVjYNuMnkCekIjMlHcBHz6cfW/GjhKBhsHnHzmVR74oV/BlvvYwgKD6Ag2msIUDuBMROgGXLNVnrZVxqIJXxheK8DMjOW11YyVaJKm63O58RTFaIxMIE7sos0pTGnuAq2w4Sg9LSj9Jbax1C1cC0doZkWPQ/84a/MhG7HLGv5tmvD4RsaHDoSMDqC0Fw8mpClS/EVch5GCYRTD0iIkHdhzxPjfTWDXL4haLGVg0fHR2E3HS8dDi6xYO71frFTqsLdbn2WoH13HJ31ZUiKHX9zFvDZgasHSXD/H774yyg+NVigMMs6nN6N1Mt5uDUFgaG3BVNkQj8Hmeo6xp+ZywJfdkAnTcKMrps2xxxrafWgkGV8ddPnooMdRG/jmkeMSgdYn+J7ZjD29EmNBwO6gyG5ippUZlqU8ZyPuxDLmEoqkXO68znaScSSW+skyZQy3BwHNAdx/wtBU2HDDUIrhyqLzl878jKHSg5UMXtiGTTlYe3BjAK90YSFWO4FhesRyZd0wMmu4suV4te0zm6ko9DezFEyPL652+dDxMmWNMKX1stDrgMwj6of15gABa5XTK+jbt7wYulvQuQHlo3K5QmdNzRDGG050f3kwp3tqVToyh1nIzRcyY3jWPDM5k6Uxb2gIlOkmp6nOv0DWtowTeU5dvjEbumHF1ulBoU2PZI3a7A1HpP6auWkw0phawFO/1yXnXa6+I1ZsnjL3pH3V56tu7UKuz/RMoPpgNQ4O8WNryTOsrlPdp2qQ0PhVo2htImJYl87QQGF2uIFUdZuCoF/JKIzILGZ5drHJbz7+Ij/7F9/gLzcusFaZZ7x0nLGRORppStOOkgi8xTN0w1FvVCrGM+LOPINWMAGhiej1r3EgnvO1iz9mLFM4voAlCWIOklE1EathhaKJ6XcvECjWSHFG3dd8Y0w5GOEb0RQLtsCYy7g+2mLpsef4yIO3MXGLHsJvHlu3A+ry9Wfnz50j8OYdAWN8RMKf/v6XeezJFhNzRzhspvym+rzrc4sJedAWWU5afKTu+NFKndqUo3aLQQXstg1WVaBFiIcaHVUOJVXjQ0zFnMgQ0ZfuP4XNNTi35KiHhhe7cNb0+Il3F/mBUtnn2WkxkJutLCftbkPtiPGxA0bjIrF3eghrlKSoAv/4go1rju2LEPehuZEn6Cd9MQ74kex4C1RXakLDenKOx3o9RqIROskmThVbfnXJPNO02YbZ6RqPPrXOb/z5k8zNxdRLu5ishfnCYh2ZrIkdgxGwE2ARC6AVQSAzGwI6p6AUh/EuVr3ZeTATELC6OuCRx17ip3/ho6ytOcZnT9AqvxcbjnsNnDLpNDdScHAsM4Rx/EhxPxMpHN9tqAzg2YZjV1mZa7vYHCyz2F+iGI6TSfzj+jhboByU/JrZcqpNGyFNt72IvpO0mMrabNgSJ71OJ+FK2mTellmwMYEZcHGtx6Eopuxge92xvmS8eWJL57Odp1RsJPD6Y3DsDrE8w+YBgdiywQhN67ho0RUoWBsCO7E/0yqFV5pLPpbzJe9lg3sN+tJF6vOlkxKbo2OawMNnVti8PsG7SyXmlaUXQ61vaBsdA6O2K0YVLRE6r3Ea25MD0vUbClE2dKuG+WP4MOo/3Riw14SMxoYvuxYf7fU8I3rIwI9HBaayjHdVI+4dK9Ncz3zLxB4ZHxJoicGJA76RJJwICtwXlLmSdXlX1Odfzh7jpXaDz5kCJfUjW8OfdOAjewzToaG77mhVDWFqUFzJ+KRhvGe8AWWlAb+9mTLnLOUKbPRUupFxpge3bVg2neNSI4+aubcsbajh/EAEnFzLIR/rXeEf7hqh3paYTlVqUJYxQpFBE9K7GdQb3L/sEBunGI9k3eWZdgodFthuGK9JSwYQa6OionudA4Fz/Vuax2EUijZMAkM+/1quXIH3DYdYOZ1WmQ3EnomR7nzDEUrvqNYR3bUaQa/nRgcvGvWMdq6106hX7C7qwtU1dFOzqftdZoqrhlQAUvsWgUZt9LwMQp+XPyMEEM1+g1sc5iiqe1nubKUCncxZfc/sidETBkogqhmCa3l8jA/WDhwhAf2C5dnWFr/x8vP89Mmvc/JGm3D6MDP1E0yUjmLiMXpJewjkapTjSUwwwn5bpGxDVjBU5VgPK4wFdS/7yFyPM/1ljpUOExjLr6Ut7rJlDuAom4ASKasuox2OEg1WydItXDjuWfwsa7PSfoEgKBMWDyAjWzXssnHjIrP7E97x7iP5RONNWlF2QN2bdOB3vu3OEdARyLJ8R3f+3A1+8sOPYo7dRs0expoCp9Qt5BLeZwssYai4Pj+zq8aorPwTICH2uWu5QH381lzD1L6am9l8i4SiEORU7UKhDAU9UFXWvpKPxUqx4T+3Em6byfipD5bpref6ldDmox7FrxWklTs2zC3Tzl76mFUwc5DpgbyYA8XXvuKYnDOUBCqXcxZi5mjOKvX6sLRsODwJh+fVKrDGt1fb1IsjtFKJaxJ8kK8Ce43FmIRmP2J0PCKcnOTTn3yaj37lArfEo0xMVKlU1L9pvdaI0hDA6XcTcBEbJRQyRJue1dNbQkdQVgae5fLlZX7h3/4pP/srf8jAxVSm3sK6PURY3I1JNpDOT4HDorIMIXHWp20j3lXcwzohtaqjqoqiAdxZgrdWSyzEu/nc+uPUSUnDMZyNsLbIIG155iC1MQ8oqCwYYTFrU7RlthWYm2wxEo0yYSu8nGx6GdE+vxhZvtjKGCt3Odwv+KL2UfXyFnM5ksoqxnUOLJxUCXzm2L1gfPyFcgP94ipGRC8Cvfq7xuRasIfsjPovxbhq5ugBnDpOhQO1YOtcC9fqUArclS0nX2nyZy82uXNskvtH8v7g9Q7MRxqfSn5nyIwj6xsC5YQlENegXZYzETYa0Eyg2obnmwHHQ8tqMuBfDQb8dFTjeFBkv405VoSrmcKHobvlyKbypH/Vd+m83rPfIp7tExvbZNZyOWuw1Vvkn1cPMl4o8rV+kbfYiB+0If85yXhn7Ni7YVjN4BmNVHs5YEu8KdmwGjjOtUE5ePuc5c/ThNUu3G4sz2cDDtuMeRvSs3hmbq+aJzTJFFOZ5S0YUWg4v2bYXR9wIi7TvAoFdbBqtLzlvOHBAx3hPRkmghzU6fwJ0EjnGFQNfUWStKC612B0zjTq1CUuFk2gSOdOx0HnRGBqWGaicadNjWdX/bmWBEIYXuxrBJlMFrfk43RVYvkNkXeXD9lu5fpIbyomV7o/GSMEyAQCBSi1IdOHKI8ydRTUKKGP030mo7J+JoFObSik25PeThs/vU1gUKDOX0diHi39s47B1WHUjmQiis7pQecCRAchPGywrYBrlxp87OHn+LFPP8NTp1pUa/uYqM6xUNAHjflpxqTLWHQDDgYVerbApC1xry1y2IQctiVeJ6CoIGOXcD1t0LUF1A6T9i5xW1DjSDjGQ4GlZSypCZg0AedlSjOWA4SUkjVWB8tE0ZR3uatJJpDltv0S6wzoFQ8J1mErTb7y+Kv8+A8fZ3Ki9F+e7cO78A17tQPq3rBDvfONdo7A//sIeAlUZvjkF17mk599lb2776LKJJfTbQYm5m1hlVkT8LXM8a/qZeacdByOcs346i894GdGIakZOg3oXcmfq3pAa/Hw8QPSNXXxYvrrK7CttofY8JnVjOqo45++tUA0r3owR7xhCI9Ii2fY+C4UdhkCxRX4sdEQ1GkxmQRl1AUHcgdtUcxeXY49Q6YYhmYu4VFFT7CV98LW51QHBN9YPM9T3SuU41l6QYmsfxWj1cqH+AppDDxj108zFI68e36e5vYZ/uKrj/LvPnGOvRMh3VaTqFpkpFIgU/WSxrfNnJ0yJYsJLEZ0hc2rpIR+zr2+xF985VG+77/5T3z31UUO7p9kK52iYXcxHU8TR2M0ByvE0SSEdVy6jSEjkvYvbbAvHGVfPE55I/V5foenDBcbMJpmlNwI1oU81niUkgkZ2CI2GsMN1umTMquqMBOybQLqQZV+2qIcTeDCEXqDdeSutSbgTP+aByxvjUaZDwO+1Ruw3e3T70ZU24aG6qYa+Si9dgiqqWG6Y/jjk44Dhw0zBy2ZhPGesZQQfgh0BcAnh0BBrIzQm5yxcjArW0zjus4w+kSLuACyAEQ8lC2WDHuCkE89/xr7CrtpFiwlmU26sC/O49fWujC1C257i6Wo0ylfSxN6G7DWgPkqPPWyY3rEcC01XOtlfNq1mbZF9tqQtweGGUWOZD69g9EURgPF7xlvShgvqCUAPtbKmO4E9AcNNkg53XiFt5f2c0swznor9RllBWd5VwTHxZT1DRUZIhTOm8B0F5ZkHlFjh+pxhXcdLDbhRqhEGEPJGv7IGa5lLQKXYgULQkdcNR6czpVhwxie6sGDMh05y8ty5bot3lapU5LWVGzcuCH0OW25hlP3q9ol5Ex2gRj0/P1ZZIgKuSEgltlEIEnZgWJlG9DW2F3jcrGrYrZkhpiQlCA/17atJheNRg3JZYfdmze7DFqgMGWWYDOF+KjFboB0ed6JKmOMHKwaxwucKaRYZgztr3QNyDQjvazA2YiCpCWzGzLiigXStSRsJ8AmplAHU+yeB6uapQ7/rp9Bo3yliejjg7w9Qn244e58M5l2HUFXP2PAqjU88s0L/OM/+Qq/d2aN+YVbGZk+zlQ8zUUK3G9jTgRlXsNxOW3z9rDG3dr8upQrRs0lHf/vsrG0PNJ0nqF7MCiyS5stE7E1WGI8WWG+dJijQYAzBkX1FTSONSEF5R3aIhthjfXOWawOfKAbpY/TMyrejW2fymWJ0Sy7ykXWLzzHQ/cc5PAdw3Ycj3rz3/mN+nMH1L1RR3rn++wcgb92BG4aJLbXu/zwT3yJduEgYWGBrbTDZn+FQlCkFFb4irH8SjXkvTPWLwbNdh6ZsPKcY37U+NDX5iJsn4PpeyFZB5lTFdVQreT9hwJ4Gwos7UExsnxuJeHT8ym/9N6YiSpkY1q/DYNunlyfPOl8unvhLoORcFkPeuly9BDXQt/NOy6lodr8kqOkzsxR6Cb55l3BuY2nHNXDhpEZWL/guLQEy014uXuJp/sbVArz9Ahw3df9yNPPoL0tVYhEyMKSuS5bWpSrR6iNxrT7S3zuLx/jP33y21xdX+LshfN88VunWd6MmQsj2t02rX6PbrPH8tY2v/2fH+bhbz/Nd54+x4/8r5/ki5/9OuWJGarju1lq9jDl+zDFvZSLexkJqnQHKyTGeMAZ2oiBCVBZWM9WmQ7LfLgyQSV0WiPZHRjEQirYuRo55ip7eLSf+dFMJZoji2cw0Rg2bdHOutyQ/sbG7AqqrJGy5brEQY0pE9BNNqhGY0yEo2x2FLlvuKc4xtUUfrfTYLezvM3GjI0Mg6S13qpCTI0hpePoqwAAIABJREFUXejKDHA14egJS1nNEgJrAhNi36Stkn5JC7MWXAnldS4F/rT4yjChENxhmKyE7X5RFlunkaBGaYkhG035o1c3OLA94QOBJxw0FDw8krPAryS5mVLAbFUVV+oOVv1t37Fvwvhrr7sNl2PDyUGfl1prvEZCy4QUjWUflpPOcT4znM56HC6kzJmI9cR5F+lYEQYRjKncvW94xbX4TmeRf1Wb5F2lvfzmoMeazWgoL0+5ZorPCQ0LcutOwMSW8TVfYo6OzBkPFqenDFOxQTrJokapcslmhtc08jawnGzxlzjqpsjbqsaTZIpvUQ+ttHpqGCtFOTNtSPmtS+t86ECZvQdiui3nAXOqkeka9M/nwEiMslyc0twVpI9r5NExCo9Ws4tG5nLPml15rVg2kjdciG1Hkk0BPrFoynzTrRLA4LLz51NVcIEcpWL2Gvm42p+/MjS+ASURXGLo5FjVLaaqL7GxuhZU/SeQqNgbGSQUaaJmCLF0GvmKeRNouziMyDk8rAYTwPTPBYG4oURD15V+Tsk19Fr/1jUoZlFyETlaFcqsn2Xb4QoQKThw1PDtk+v8H1/7Lj//x5uUFu7hgelbmTU16sEIRVvlblUNypnuBiy7jCkbM4LlMdej4VLvqL6atbwkIDGRD7e+x8S+fUc6Zck0ui5hS6y76xBHU6y4mAWnaE3j86EvYnjRpUybgFETstq9SNeWCNUhJ1FHog2fTFfTpOov04GLDtIebBKcWuZ9P36cuBS8KfEmQ5z911abnX/uHIGdI/CGHYGzL19j5dw2xaP3ejeoHhe3xFOMRuM0U8tPjRg+uM/6ntaqKnsa0D3vGL3dUPKJ7VDXDnP47EwCWFfDwxWIxh3VCqxr4d92voLsRuJ4JW7zGz9QZO9dlkTUyBJ0CnBVjI/0d22ITygJPy+I1zPZj1h0VCT61thFo9wn8pT30V1aGQz9wNFX2GoDZu42nmHYupYvehdajmrJEMf6aqFnwpxEfnrQqndJT3lZAW1hqOJXz1gVk67T6hbo2TKl0gij9Yhq3OLPPvMUf+YLLSUS+0uO3zmC3cpHgIHYoz5cvnAzP8JSqteZ3DvFjWaPXvO8r5LIepcIXJ8bWYs7wlHaLuOGMuaMJQvqmKxHolUpGOFsd4W1+gK3VItE7YzXB46RDEqjhizIqAwiPlR5K4/0vsup/mVCoajSfv+5wUDoqkfZpWzbIpPxLAu967zYu04hrBEGJZZ613i3GMLSPj7dOc+BcISjQZW/jCosZTEvJo771XO5Lwd2Gq1l16FSgTva8EcXDU9/0vHe/ypvJHDLedhs/3JeXZWPZJ0vM/f6yJbztVDmiCN7Lq+ySiTGVyae4iRGckAo3KBTM9gOONOxDOjzAcpMxo52Aq+38vX/9rJCpmH+rOPgFPQLUKsablwyNMpQr8HEAlw6F7CveoPf769yR+UwDQzrMn042G2Mn+qdcxEHxg0FuTJlutFmpJZrJLe2DNet5UWJ/+IW37vrdqY6hp9OQjWb8Yzpc3sh4IY6VcVUBVDrwFgJbrQd44EhUVSMHJxdKJbzurONJhwNNNg1tFoZB03KJ2yRn7RlHggND2/DF3rwCxUoJqDElymbt2IVO3BXpQBjFdZFWV6rIH/MjUvQajv23Z03uQUCZorpGIW0g9c1iuXyI09JCK44n1G4cRVqkWOkYghDGClBJr2amC7dPlIHXB6ybSmk63m1mAChkwmhBnI1F3TwFEFUM8w96Og/D9nblA+X38NGhiKvkx1q44ZB3D7qREy3MJDEkrrF1JWq8bVcshrRC97oOTAc2/pngj5OLxrDipkT4ydxYcn5DDt9LW+0ag3Zf3XZ6rxespxtD/j666/wj/7wLCzcyvffe4B3ZiW2si62EHtAOClUGBR5xjlC51hJEvbaEqsuZd4ETJuiB2KfMwHKfr4m04ODE2poyVLGjWoWLS+YmPF4lpeSDT6QbnB3ocYzgwHzoUHejl2Jzm3EE1mH+2yZo+VjnGw+SxCKXZ/ExB1c7xpGrtrCAmnjUVqBZWRiP3/44pf5h6ffw/0PDvULbzBbt8PUDa/BnVc7R+CNPAKSsGiK0e0ZfvW3vsozT/aY3X2MdlogVWm0jSnYEs9lLf63apGDGilsOWJrqNUgLhi/69aOf/sCrMucIAOoAkkHECq8dk9eaxoXYG3b+f5QrfO/c3qbn/z+mAfuU8kkWOlZtMkeNYQaW163XlxfuSMfE6mBQKUPnTjP6JKWrNV3rF8YsPRsDzOS0BJDsjXgsQsZn1uC0+sNvtPZ4pFzHZ5spDyZXeOVdpNX+y2+sfkCa8kiSbJFmqzhBqqm0AqhFUG0kP6e+hGH3uaX8t4VL1ZOszzTbn0wwdio/p+mOrpAcWSaS2cHLDc3WNnMuLECW+0Cs7smGalNMVKfoWdG2eiNoNZbl/UwQi4uIe2dIxissti7SEtgLOuSKspeu/J0C2NiqoU93Bhs87dKdUZdmSudzDNB+0cMdbFnqWEm1hi2yuPMcKX5GBOmTCtLCQvTPlxrkHWYMAG1cIQDGObDGg36XmOXmIAHgjpPpk2cen6jSX5PbK0tMRmUfOj004OE9UbKLeWAkTL0+6Bx+tYm7J02FMcNp0469t5vqC1ocKzuX0OoMeurQ4ej2LqiwSTDeigJ3DUy13hMbQUicJqO+KDJ8+2US6hrNXIUs5i9tPjYyQ7vmp5kKsxYHRjPcL0qtsbB3QXolWBlG2YDk7NYilGRGzTIjTRPbif83HaXMWv4UDTNLcOYiTEy745VMoYcrlHPckcZugPD03JOh3CtAZ8ZKBol5bOt1/jl2lHm+nUqLmN/wTLqDOtZwmddjwUXe7LpdOpop4a9kWG0khsSFq85uhnMG8NmS3lz+KaIQgE6PcWVWK5nCTWa/HflGrsdHJqAh/YbHrnheMbBjDVM1uBY2fDFLce/6ym5LKOxvM3tpkq0nkd/VMVCxVCQgcE5nFOdHgTSu8rgoO5dadAEwJxBHiAx4WUBuIHxcSGKmcmz4IZPKQE8jZA1Oo/zRgrf8CKUJUZMbKDGtZJgCC1rtPpWnfdc5eAbRrxJYbhBE4Mr5k63oEawcrqKYfP34tBJqw5mvU9OWH3ta2IL8+w9McLSzXlAqWtBQE+OWIFYmSn0tfRvvdZFJkA9cKhjujdhOfnSVe76v17mi0sJxw4d573VXexzMXcGmXcyLxjHZADHvHLAcSrDA7mXk23Gbclr63bbkHPAPcCdVgDPMotjxgQc8L+e4TvO+Wo67UtVkLGRNYiSFd5ZmifCsGrgXp2rFLZcRh2ZQopczJo02mfzZ4IOuKKYvHamgQnH/a4naZ9iqj7D9mqfiTHH9777cM5g6pzoOL5BLztM3Rt0oHe+zc4R+KtH4Obo9eXTW/z2ry+y/8RtrCQ1Bukm04V51kl5Ptni5+Mx3nE0IFt0zMshJ2eqVusRSM9AdFzxBo6Vpxxz32N8r3vcUIOEHF/SNBlWbziaK9AJDSc3Orz9R0Pe8RbZ+xwrLzlai4ZFEs/QtKMB/UaTqFhmcqVCesFRecXxmvRG3Xw8VTcZr3ZTXthuUHZN9q/B84UiI50BX2gUGQtKLJiEF5Tqq7TXrMttm2VGosAH7apAO8h6Hrz6p560KmLo/ItWNs0O9fTWytHCqaUh3utXqaD2dkw05e21jXSL7azrRyE2a1CdlVCp5DVpAhiZKbDc6uWmB/EKtowtTZB1XoVoGlPQ1+xjwjEP8Ao2JO1fJ+1dx9gNsq6SU1Nc4QBlb+II+NrGNe6bm2R32TJfcbkHoQ/r247zgSqOUr6vcoAk+SDPNb7EhLIFOxeolo8SBWWW0o53s152bfbZMlPxPEn/umeqlkk4FE1xOllnOhvwblvmm4MbHMwqjEczbJHxcub46nbKBxKJoAxbPZiVZtvAW0uGK3MZ3zmV8WMHVVmmUVdegabrReYWgXh3zuVxF3PSZTmvOdPYb/CI8+ysyuuzM2JklUOYNxqkKpTfAyf0NdZWWUoOsjaAdurYX4dWAs0tWA9hYZehsOZQm0lbY0CxSl1tPBxXtgyPir7BUg7HWMNwyGXcagxfz1KRO0jBFlhYGjhWxBhbwyCGM5uwEMI7Q8dPSW9Q2EXNTeB6GQNtchQhYh0/srfIB4OU33wNxgK4zxo/XV5zjrgDE84QTRk2O45Opuva8e1L8D37cg1fNIAXEsdyEvGj1UmlnKEotQltlLYc31cxfMNkfK3lmN52fF85YD4w/GQEL/ZCvmk6XF1PGTkQUO7lzFv3u8AB5/WOCvBWTI1rKItuOCrVIdF9LSJq3FCRq1WmFiHcjt4GMscIEDllUCrXbeDy3Lvm8DxJx6cXOVsFzgTkNKJdMP5zhGKUm9t6Pc+yi9TRKvClUWvT5Saamw0k+vybO0+BNN1Qehk4L33VmNcT6vqZNLrNk3xyHZ5QhcCbGEABOLF0kgAIrOp62HDeNRtUAm50O/ziH13gP3zuMsHx2/iBYJJNF/JDxrJQynP4+pu5GeVgpPib3BAy7wx3EBAHEYeM5bq0r7L5+qFojqBWnOFeE/j9zKtOOcgZu6z1YE4M5jEbsRHP8s2ti/yca7M/GmGtk7AuTaXITmc5j2EXjk1b80HkhUy6kzGfe6dg8ax3GZe1MNE0Lm2xvfIs3LKPJ3/tClc+3GRBLSNy3etmfINedpi6N+hA73ybnSPw14+AbvQXTr7Kn3ziFOO730JjoOUjITIh1qWM4fgH4SgzVcURQBBC+QBstXIGLuppR2kolgOmxlWiblAafUfdkrMprfOOM4tdHt9yPJ71+Fr7Ir/ZWuWhvTO8vOg4czLltWcNZ+Sc2874/JfgzEsp/2y1wx2rEfuXYxqvOWKl7Xdh9wAOp4Zj85bRhmWjXeLOQp0HTZ2kXea42jhLJXZHIUfjMu+v1amYEh8Oytwbj2CyIm8Jq1xONmh5N9lcHg/vMowejFpx5PgVkEo7PpbARrJ4NrGV2zDxHK5/GRtNY+N5FOjr0obvEsqSZQZZSmInyeJDDJwjGWxh4hlsYZ8Hivo8z18lq1hFGZSOYTRHktNVD12jkbCcr2UUaaIduBNjmGyS9C4QZE2WkmXuKU5yrF7DRiHX+o5q4MjK8LQck5OGHzsRcHRrinNLS7yavsxoUCMReDQFBmoESbe5nja5aCwztsCUKXClv0xmAgrhKHdHU2wnW6wIu2d9X2xfsJYDzlEMih5gd7cMt80ZFk5A3UdNgEwpy13Dd07D8RnH2IjBqZFA79eCL+CgZU+jOa3R0lUNX7yrsmbYfHIIBiTal45SQEMjPY0Ma4Zz6wl/8LVtinMTlLOQ+2LjJXqXu3DfhCq18HEnBwuGtW148bJjsAX1KUMzsnxpLeGjacSDQYTJOpyyIXts6ImbokwkJg8l/ibw9hjmi4ZxjRE78OVBRmQCvt7d4B2VTe6tTLM9CLljIvCmjKkCXJPJIoYHxgKOleDb3cz3xZaMZbkLumcELsQ6j4r5GxbDz0ewmOFB5Eho6Chiw1rKiWW0ArcfNHRWDC8v5y7ZWwPDwchQzyxfVtuDQJ+AloOHW+t8ZM8YJ94WkTWdzxGMjkBnO3eDaywr4Oa1iwmsv5CXpfjAYGEpOdgFwnXcZVaZyNtFwn4OhvxIdAIyZS3qspUp6qLDpcb3oGr0KiOGH3OqLk7AUHsmzYt7hjA2JMqbU56djDRSO+h6kGtablcxgHK56JcSOye2VsBvGDQ+eN15faSPUBGLK92HHKwir2S6GG46vR5EoFEaTmEakb7GEdQt6Zblz17f4pd/70X+5Fsd7jh8Fz8TTnMYi3KSJ8SixjBZM2z14UoKuzK4KC1j/qU4Np6P/6eyiAPAbYqYcRmjOArDbyjMeTq/hBkodliGKueQX1Xe9mkT81raYi5tcks8zX4pQFSdCNw1bTibOKoCkkGRaHCNZTJiq4Prh89+/OoGyiaQOatGb3CD/aUBz6+v8L0PHeDILXli9A6ou/mk2Xm9cwT+f3gE8g2wYXMr4Wf+5be4uH4YU5lmMNj0JfcSyycu5cfCCvcXC3ScY7QWUBkXuzCMilBY5yY8t5nxylKDxcBwrt3jO9ctv7SS8dmLPb55I+GjW1e5lpT5XK/PyWaDT88ucPxSkeoy3HE0YG5gOHQLHJwJWNiwdCoRR8sV/tHRIodGoW4NY3uNH5VlDYPaIaZjuN4CLeSqKRPbsV9xFkqwzwx7RUpgeGvN5V8fy3NZn+PKocrgS9svESTLBMqCEwtnC2SDRS829iBNu+1oBlu+FVs64oGbGyxiiwdyHV7vMiYQIjFk7Vd8waZGqT40WO60tOkf3ra4Px+NuISguF/hZugB7CMJCgcwojuCGiaaIFBGjMYqPt5fLFeI8T+bZkYWF46hIOGtrMWnWg02um36JqU+W6beC4lDw8q64/YRS1hzBO2A9+3bx+nNS5zOKowUZmhlLR9Vo0DjOOv5MNXYxpRtiRtZmyTrccNEnkGYiKZYTre4mKyz25Z8YPFjyZrWc+6OSl57dkVT8rahtQjFApQm81DptRXHmTjlnttDQlWASfyuRV2f3M9rrHqbebWXJxBEiOpQTMONTVhfVt4cWBkuZK4RJlY23JZjMCjz1NIFP/KcCCvUXMY1OWAt1MfhQj/vqZVjVZmJ/QkYC6FQt/zHCw3+zWDAT9iIvgl4It3kLhMwY8u0FP6h6A5pz60iWuBEDFdlXmjDYtfRsgHfHfS5PNjklxZ20aXM7oZlWk0QET47cLUHi2swp8iQTm7USAPDGZf5X/8OyQtK+LaIlS1oF3LcWh32vV7yJgxYDSUzHfCuuuFI1fLMiuNyy/kRru7fTQH4Y4bdmUGpPN+yjocHA7ou4/ygzWiQ8OD0iGfbdG4URSNTtfRlsRobdPlq5KpT0pd5wWFlVFG8yRYMzudZdu1Xob8CK9dgTCNY3RsCSGm+yUuXc62aN0fM5uNO/7WHo1Bp9LymTWNQgXT9NwutcwJ9hkjk9qbDTNk8c84jJpku8jGrD/PWSfExJy5neq/l7lc5aMXqCpFrzNtTuLBG7b7+bPgx2qOJ9ZO/QDrfcsDSWsrH//wCf/93BpytzfDOXXt5yFT43lC5dHAqM162q+iYSOP0BJ4YwL4AlKSiCfBjLiVLU04lhr024qiAqpMm0rDpjG87aSpOhnzTUSNDE4KzLlP6pOLB1YBHzVieyhwzLHJrYYFxEdt1/PfUzxImhlsdfrz78kDRJjcohHVSaVy8LkGTiD4u1Q1VwpiI/mCVfjNguTnNDz60/w1vmNhh6vL7aufPnSPwBh6BnI5fOr3Or/zsw7T23kcxGqeVdjwYisI6DwUVTtgKaeAYs45Wt8v1dMBHH+/xQpLy+EbCn7/e5l9sh5xu9Dl1I+R6O6NFwG1tw+3dmLOK7gzHOBzEnEu3+I13TvK3F0aZ3O2Yu8MwUTP0l/NwYAUHiyQrVRy3R7CnCsFhsQGworT7E1CsQCH0EVe8sAL9FszV4NYyLKpI3cI7RuFIaHzB+3oT9pbgBRncBgF3FuGLvTZXOlfJkjUPwjT/yXoXPANnjGIjRjDBKLa414M9MXICWwJ1brCSs25ZA9c9h0tWfO2PkWu4eABjSx4c+j14OOG/hh62Yt2c4vv1GO5f8yxfUJULpOhdl0Gg5HhLIZ6nm3UJovF8vqXMPK/tk9goIwtGKdgy9eIcT9pJPr+xyB8s3WDEBCTNkNl6gf37HdtnDeGcY99UiXTjEJ/fesqnh7hoxveUxm5ALawz6C+xaSvcFVSZCsqcI+NWMpaM4Qk34AfCEY7bIo/3LrMnKFIo7OK7vaucIuOh0iiPL2WsFRz7R2HpIgRdw8yY4TVjeOzbljvvMEzvkoljGPqqMZocjooaue71+rn2SrEWYndWobgGqrVSM4nZhkxAXKM1jTbPOSYnLUuDc/zmYyHvmZmk652msJjCS1v4BfAZAbIp45mrKwGk24ZLa11+odmlEo7wwSDkjiDhC/32sArPeceqzKFKztlATRWOIDFe23TIGS5mKU+kKZ8ONvk3e0aYblX507UW9xQDmsuGZg/WevBwDw4U4ETFsLjtPCi4b8L4eIpHVL8WGKadJQ3h6QTf+yqDZz/JK1kLQV6LKnCwkqYsFC1xz/Dt9bwEYX/Z+HtChqWttsK9nQcTwmgPKMuOkGcDx7616wRLs77NJRI4UQesIkUOGZyOpxCDmDGZLuRglshLY0z16Eprp2o39fyuQeUWSJYhGYHKmPFRRdmKI7uitxkPotT9qmxAuZv7r+XnUgyeZ9PEpIk9E7mkS1t6vhAa34WiIk40khfjlph8XC9mV0BOgOwmxaZoEn0NVQTeyE0Odt8w1FpvT/GtJL79QgCuMTRC6J6TmVZS2cjy9NlFHvonr3DWjvL+/fPsCUs8SMitBhYUhZTBpQzuHdEhMlzoOB7va1SaS/NkTJHv4oixrCSWs1nGB+I8jkQ/xpHYcCkVoDR05OYn86NU2aUqPiDb+iaWw0OcKrnhVBDxhcYS/8OeOreOlVnpOaXJcLoHjT7sjZ13SvdMgee6rzNVUueEIbExRptAMfpKk3YpJqx5t/uAFdpXO3z4R+5hVLvgN7BhYgfUvYFL+c632jkC/jE51Ff8zh98k898/QK7d5+gOUhJXZ9KNEUta3OP65DYhI/3B/zcdp9eZ5E/vjrgsqkysh1xsBEwH2txDPhIucw7w4B7SzF/q2LZ46x3zInynzCWx7td/vuZEn/v9gqVQ/nDUnEYqs9UxMmIohTkYCxBLTCMKG+qrwDUPIS2NMA3TWgmoRYC7jKcuebY3cxDjXfNGWak4duG2+ti48QawnQALYm4+7AncPza1hYtU6RGwo3+ojchJJ2XPVsXyGkq9kyp7T4LSjOPgh9vBPGs16xkPoRPQqMQN7iCcwn1sQ9Qqt5BP2vj+oseKA7CMWysQtMr3j2r3kfpt4LBMr3OWcKRewgLe/xIdNSWiIICTSm5xMhJ9GNLuW5Pr/34dcOPTpU831WGlUv4x9U7mS9MMEbKH/dS/mLjKpNjlpmx/4e9846zqyr3/nftctqcc6a3zGTSe0hIQiBAQijSpKPYQAxFwYJdLFe9r6IiF72KBUVR4IooIoh0pYm0AIGE9J5Jm2R6O32X9X6etSeC9/pevfcV//J8PskkZ87Ze5+199nrt57nV5I0a4vmBYqSCsiqFPOTMX65+n7qatopuY242jeu9Tkni6crpAlIymeWDFPhGyqbUX+ASlCi1m1kshXjWa8HsWeoizXRHXp4/iizquPs9DRNSc3kBoeir6nORkkJJnnBD5gzyzLRcGJIHIgVhVTeYpF5tWFJycQtfmMCNMR4OFQms9WABKneSYyTVGNSESYgbvG79X08u26USyaNpynUJq3Ek5ACsTizMdw25SlaHIgNKbbmfT7iDVPl1HK1YxnF6BTHoTfM0+N75KW1BRxmWVShjYnx0zrgKFsxO2YZ4CixX1/ND/C+5pCL0nVs6goZcG0ygWJytaK6FnYMy6QOyRJUVysDgAToCKg4LGtRXVZ8q+RTI1XVsjJGz7EwUrB2CqgRzGJDnaso+x4PBj7HxV1SCWWybMXqRKqIjgABRzFS1FQCyAVSBItybXM65OmwTKPjcFo2iz0K27sgO6gI2sd866TqJV51EoEl0ccSzTUGvJSAuxHM91eqpimJ9ZujiFcp3BGF3ie0AvAdZXQ8lrRH5fxIJJd8AOkbSkVM7EbEx05a7qJgFX6bnHDhtO0GawiKBfD6NMnJyvD+5b3S8jWdRelbyoiIsEMAmSApqfYKuBPAn1NYsuiTa0q+MuLtIq1XafUKhpFjEg6d1sYovL9i8bsnullxVydzx7exqGo82QBmakXZ0SySNrsRgSkEWG8bFU4l9DqQdzRN4jPoQkUAsAuuqLE9i60SnxuGVAcW9Tb0S+KIHLNSRjQhPMoWSyHnd4mIbwToSQY1mgIW28My25XFgD/K0SmPadkG9vUG1EmlUsOwrTloaXaHmsOUzZO5NYy4DdhOhjAsRn518iWRQQpGUUqyKBTNNSUOHhyibdosjj26ITo35gp74+fAf4K6N36M/7mHf47An0Ygar3CyI6AX3zpBV6xJpB0MhRKB7C8PoYrezjgD/OUleAVYmQtl1PdBPN1A46d4RzXZoGjmJaxqLZtE1uVECaetAgsbXzTZJKR+2+DpdhQrjDB8fnwrBT1Ryj8YY2/K1qxW20YY+GgN7IgkfldYqfaWqLUiYSsxIWsLJOEcE2E7OxC/wF4pQ+OzkKbZGOmoc6HVKDodjQlmVwsaBsHe/owysGXdJE7wxLtlsu+0k4K+ZdI+wO0ZZYQ1p5m2laO9GmMUZf0C6WslDCxXZaAOruGsLKfsLguMidOTEGrGK7wWZRN2eszQoRyvJ1lqRmElYMMiq+VkJqDUWxvAOHdSdqDW3WYAYbK76fk1iJpD46VMHnnMeUQhEUcRwLAK4TSZhHWuXxwcZF3qrG8bvywxGi8jaVuAzGnij6nwlM9u3hlb8hAaJO2YjTXSSBZQN1oE4tmJbhj9WOMq55AYGcNR0f6cePCEhv9IYp2AkdW/tIy8oeN4fF+J43oFS5xapltZ3k69zJNbh1NdoZnCtupdwLm+dVsHHYoJTW1hUhNLe27eFLxyO4cOqWY0+ZGbVT5GFLOEGWkTNmH0iWkYiTJZi74ByHRoAirIsWmMXWWFwvoyUhL0yJdsPhJbxcTVQcdYxPViA19ki08Jp60LYVcl8lxFg+oEvf3lrgkHucYMVkW4JaySPqKzd4IM916qrCYgSKuFHEd0q8rnOg4DCrY7Fvc5xfYSp4vOS20+ordsiNfMUdsdOSfIvCswEseHDlOBBJRtN2ERsWegmKoomkuWZyaUdzm+Wwoh8xwYjwfeowSMkGSBMQvWEOrY7ElGKUlDJmgkyR7yCmuAAAgAElEQVRtTUtW4Tsg7d3ZrYodeY34Gsv3pRhA2oKpaWixLLKBw6+8IiuaHaak4iTGa9JNCmEYqB1glRR92zXFFCSEByeetgKKpGUqkkw5R56AFkhIFNeAIjYxGlsB5yJSiItORkBYlTKXp7c7AlRhk8IZS6EIxPi7Bco7tTASsKoVpae0EYkYHYRURrdoqmcqLFnIHYh4l8LFMx9OgKaAOXmxXCuCjuUaEh/D3kiwIRU5S6p/AuQE/IlhsaRRSCqN3AZcm92bCnzxnu18+tE8l7bM4KxYI/Wy6hClaZViWdpiVO4xSUVLE8bQW/ifc+Q+V6VIOvBCSdSvFj3iC+nCjCwcHJTugEWVJD9YsGvsvrdZQLdSJmFtpqVMnvIZNYqSZGArxdwM3BkETNWW4XD+UbmIxva+3q2cozuY3+4QpDQjsoAVdbSvouvcstng5+j2h4i5TQQGxAmXLoP2B6UkPcbNdUkqn/xIJ3Ma0xx/3DyclCwYx4D3n2aDN+Yf/wR1b8y4/nOr/xyBvzgCh5RQqzfv5oNffZJJkxvpGTpobDVk5hU81Fo1l9HEBD5tp5lrxY1nmmlhaFhcpWgWawit6BEys0wAwh9SMFFIxQGslfaKOCBUYFc9XHV0nClTLYJ64U0pKv0KhhUxUcWF4EuIeBmGJW1iAkyerSgMQ96NDFkFzMUbFX5SURyGXE4zsU/RIm751WCJIlfBwV7xxVIMj8J4V/FUOSBXtNiiK3zFG6XFqWZ/pZOh/ofMqnZiy2XMyB7DOCvDiN/HaHELjqQsSGyYtGHlLijgSvSshU3o8p4IWIlwIjYeOzGZYmkbpfw63MRkKulF2NJLCgoM20kqbj0VW1q6dfheN5XCq8SqT8ByGwi9XtOaqbeztFox5loJiso1xyHtWBEsWGEFXxotRkUQmtxHNz6OvPbZqUsk4h0cVDaWcpljpxhxq0jH6+gqunyrs8KE9pC5dQ4NI4p23cG2oTIvlTdRH5/APiyqwyJT3VriWCb3VXiU0+wqYlaCLu2zABvhAr2qPQ63YtRZcVYWNlPtNpGMN/F8SVrxeaZbVcR7LQPAxTeuMVQkZEYtKrb7ZZZ2CGCMKjbmohS8LBUiqdZu09hiSyHEfFkJiDVZ01gmrFxHHWNB9GIQ2wNWi0VaLDz+uJVOt9EoeGtcaHUjn9pZaVhXhOEUjM8oHhkO+d5wyLlWiholNs7CL1OmrWlpn5tygzTH6kR7TaA0fRo2yL+1xzRiJn5srw65ubCN78aqOa0pSyqrOVgdcE9PyBLXNibH1X5EopfDnSbqXWN8rCkL91M+VxFGYtBfUSyN27QkNd8e7SduxflxAKdZEggnlW2p4ige9S0Wxl2mWBHgqEoqOvMYtbH4QD4rViQlZTycJT5sVrUikVGG+7fDs3khGOKkwGLatDTt4zSZRkVWrEukKjpe4crxPRNRG4T754ufm4gzpMLVHlk1JvujhBSvRxuAJGkRtuS3ioBCzpe0XgWQy6BJW91V5Ac0SQFoDpT6oGJa0MYZxXzPU9KOF21QnTL7kutE+8qkOphqrRyftHFDCMUeSWwjBVjKDUX4dvJT/oiyVrJjJRdWosjkwhLFax7CtdpwZ/0qxQOrBjjzi7tIOg2c0TqeHmKsCjxmKMVMpYy/Zmt1ZL/zTD8MDkO2gKEMSPVWOgddHsyVtBG5xHWk7xGvwZitWO9rQ9ebZCnEck+sS9qUYkAAlANT41CViqqxXSVlFgnSKq8PJDYsOueDhDTi4qgdnN5WTY1bTWdPaFrZkqudSUCPp2kPYzRQ5qn8RqoTE/CUQyhgTntoScORxZ9YMFlxSl5A+7iQ3/2hl3PeOo+2tuw/TAX7vwB14rMjiFOmH3lE/xdEe+iZsV/8DT+0yUeTO4ls7tCE9ze88W96iWxPHq8dq/zvf3O8f36c/93OZZ+y1z/f53/3jr/v78TjSh6y/7/3eP59j/T/tbW/faz/X1v4q89rbbCQvO4feZ7kchQX+XxBccPNW3jhmSdwYhWKoYMV6yDuNFBOzaA+3s47rLhRcE2zYJLwX9BMNzE2Ec+oW4paGqZViQeYMm2cXh+2iX9TAOPi8JsezcXLbI5ut8iLe70F/mo9xqGJ3N/Fjk0qNqK2k4mgoxVik+WGD0LSTjtQFm5VTsxfISX+Zx7k9kRpEXVCtBbFreRZKozNhbTwni8GbCoo2myLG7wBimGRTGk7/Z0PMa1pPrpqHsOJ2fRLPqNSxr+0p7QLR8oW4lsWG2daG9JSDYubDadOygDG7NPOov1uk8sqAM3SJTxd5sjkNMNr2VPZS1O8jXqphKkYrXaG/tK2KNPVSqPdWmJ2lQHCks+at1zmKIe09tge5LAk/NuKmYqd7VSb4G7jBaEtLAkLD3LGlmRyfByuitGFtFkd4pbNSfEkNY7FIsdm22aPl3tL1Doxaodt5jZO5KcDmxnJ76E1lqVoWj8ubU6tUJ3YVzlITvh54nUWVmi3E9Qpiy3KopMKc906s5/VAnCtNI3xBorKZq+uUHRhneew0FFMTmlTkfNjLs9sCpk7yadpkmu4YAa4yUQukRhy7oIxzpxM4p0Q7NXoMEoKCLuFfyULAI0luaESM9avyQ073Le1TFfgMDtWSywMsLSFa1IcYMTXFEKpYPh8rXuUk8MUh9vCZZJWmc/ylMNmT7OjrLnLyxkvvjbLNWMg7MX/8HPMs+JMsGw8y+aaci8rUjbnZsdRkwnIxiwOlEJeGBYAZ5uWnIgy1pdgcgLaHWWSJxprlTFHrhdVrKdZWZFJHGa6AhYcdhZzdIcjXCCGz1rhxWFas2JzTvOLUHGUaxvBhi0VK18h26lLwaPiI20ih+UeC7WmJAVDo/BCQeM6FttzI7xtSshCKSkpTVACrwdiUgGdokxUl3RhpV0pMWr+Tuh7QeL5IJZTBCLgGAK3V+NMjVSpFSEcSmSXeO1JIoiIYLZohA8rrdZENSSlPTopSm2wy5rSlihZQr6f8qeqSSEG00GXJjZJmQzX0ovaGBs7cg8QvZK0XqXFKthOwKb40gkl1Qg1ogpdsF0jViOWfBZ5XoCejFNSAKNNuQ9+/Ie9XPzbUWZP6uD4ZB2HhWIuHamaJXe6bCvaXYXVBHuGMdxQMfMeskSYAE1xRc7SCDe3zpGxV4hDT5ePadVKhXalr5kWynlXTEsreitRwp3wSNcGmjpPkXYxivFREXJ5MCGu2RKGvBiG9Jn7qoNrxXi+WKHFG2K6HmeAoqTjTKyO1OUC7LRls704ynOlLVRJFKDco0y0oY2Sa0iifGRmMVX9GJmwzPDIDk5tXczsJS0Rh1JW5/9zoPRXp7TXv+B/AeoEgEVHZcCLAWQRoIv+//rN/7V/R9s6hA//tN2/4OvyP992NGFH24zefQjkyHPyCf7bbUbIdewDvHacrz39X98d/e7Q+Lz2+0P7/Wuj8f/7e7NH+WwG0L0GWP7S/v/Sc/+/+/9fv/+1QZXbnDn+CORHQOFv2e5ro/1XXj22L3MNHLrw/spb/n6/jgQSw8MF3vOh2/BGbVR2IoHbjuPUUJS4m8REFtspEyY+24KJjqI70OzXAR3KMt5zKUdR7aioFWFDW1xaXcpwf3qKcHxC8dyohhmaS99kmyqNdA6V3HOGIawWn66xDMkDEdB0xadMuq0zhS8jbgaK3ChUkjIRgCuRRVIOrIauNdG9KxuP2nRbd2gDLG0XM1Hv8GC31kyzHEaCAneXtzOptJPde37PD5Ycz9ULT+DR/l4sr4ZT3AzbJZrHchkqd4ot8FgL1kf7PaZ1qiypnUTfWKOKjY03ajNd3oWVnIkVn2ReFxY3MqgcpmaPZJ+dNo4dbdjsIWA0v554ahaulSCuxN+qihiaPivB2VaS+Uqxj5CyrpibsahiS7qMq1yKdgarIuzwIpZTQ1wsR4Ih0/o7PFbHC6FPGp+BMOIGLbAt0r5Ne9zhqWGfZ3OKAcfjqHiC8+ubuXXkUXKlDJMys6gKxWjBM7FH4+PNJt1inzdoDFP36jLD2JHFjRhShxVGnTo6g2F0aRvZ2DhzzUwRT0MxStW+Mewdl3aYPA6qAthYCRhNKBbWOAZ0G2QlE7WIfQUMSBVLJm1xXhhbrFvCsZTqjPkujvGppCUnNClXk+6wGRcW+MEfOlne0MJhrsMeJaIE6JLKjqV4xphMlxh2k5znODQo4URJtddi0wiMVGAkdOjVBdYGOertKjZqjTCSqiwXEdzOcVye8gvMDPt4f/V43LRFLK2JlxWvFn2aKpoFNU6klpVrzous0UbEHlErsilISdszBi96mrynGBdAs/jnaU21leWYOKwq9FOxU7RbNl0VTW8lpCL+h6FrQNuSekV/WdOahjU5zQEPc+6zmYjD1Z6GLVIRyonq0udOHbLVz/PuJoep0pqbplHiCSetV7ERqVHG/8+2IdGiyEyVqD9FXFBvT5QSUtgGdg7EB1xJOz2I0jQkTUbOgzH63a/J94EnPoGiChUkJCkhInIY1GixtImBO+ZbKQrc3DqwamFUDMqllVmGUicM7Ybs7DFVrvDvpFonWE3SJISnJ6BNnpM/Ym58UFrGUTKGPCe3VUlxUVmLrT0j3PbTbj6+yubitjZOUwlqJM3BVWwUg+qkZex2WqWyaGsKZRA1tsmOrlaIobc3BHWNippGRVd/ZBTcF0LWiRT5JTmfFdigtfnIDVrR5MJqqS5HdQ0264BGFWCVbSY6kYC3uQpqU4r9RRHfhDiWjYgopEo824pToZfjmtuwy4pWae0LlzFQTGzCRMnpoRj3BN3ErCrD6Q3FUsmKm0Uo/vAYyJMKlXCkQyq5Evu6S5x/8QKSKUdKSn/CT9EI//3//ptB3SEQsGXt89x9/3NMnzuLhG2xf/N6fnnXQ7TPmkM6Jmd+7GEmz+hkR8AqqsRFFb1oCtblUdat30Qs00S8coD/+NlvqJs0mxoxLDUf/7UbjYGRpgomk/6hnUQ/Xz+hHzpOKYn+8obv0R+vZWJrAzqU9onFyL4t3PHrh2mfNocqE1f0n7b1ukn/0G90Ocf69ZtxUzWkEnJiohue/P7Q/sbeRs+e7XR252iqrzHO8K8HD4dee2i7f+nnn78mch83AEcuBjOx/aV3vQZQd2/dyEjgUpPyue2WnxOrn0STMOBNhXJsYhwDfq8//r+81bFzJp/3dYP+58cYvfM/P/ef//+f9yW/N4/XHYt5jYz1hs04Y2Mdhq99CV6/zUPjfei5Q5fEof9HG3/t/Jhtix+axD+VR3jk3l/zxLO7mDd/pjE6PfT6N/JnNISKe+97gTt/+jj1HVMpW1NIWBlzUz4mFlWXpDpTL35dYHIpexSMCy2qLRgWD11pnWlFJqYM9SVlKyQ2cVspIgWPOJobfZ9rlru0TlSEMkmLk/yQxjosat8EJqRb4YmZ6UHoPQjlcVDfLJU4ZVby1h7jS0v1LCVqfTOZSKljtBOKBxWSVCE+WgVR4UkLJxlZnUiTbau2GQmK3J9fg1VYy54DL/ODo87mvcecSGxawC0bhhmo1DHeTbBH+zQLoC3v5aDXhWsnCco7UdjG/0nrPGGQx80eix3vICzvi8QU0v4o74iAnRVjqLQFT4e0xDs40akjp8s8pz0c7yCeP0Q5Po5KvJ1EOIqnA4b8IcYplwYxJVY2T3vDdOkS9bF6KipGWVnEtM9UK8VAWepxYLv1WFaMclgmXu7k1ESHiQPbosWIWFHjWTzuhaQszVHVFqEbZ65n8XSpyLcGu1mQqeP9kyezs+cJVvvVqJj4XMTIB6Oc6NQwQcGaUNSxGZywQinMU1Y2FS3AvsweXeFot56CP8J+rwdtx5FYuel2DU8Fgzzh5xj2LWKxGA2+tJ5sPr97lHMyLvV1tngzmxnScOtqInAgk7RRYgp9UCKgJGw9G5nLjr6iCesgVqNMRVZ8yyQ8vjte5Kf3bybWOo4pMn4hRvEsIQWSnftAeQ+rdZn3ZRqY54AXRMpGsQfZIq3HQJN0LLJidxMMU1FJk6IiNKYOyyUQ09jQ57ZyP99oraODpOFdBmllwMjTPQHbA+FS2awVkBrAbol30pG1mnCnuke04VDJkuDFPkVfJXL1kJZjqQSSUjAlniTvl7imPEI9CQI/xno8xinNsZaLa8s6JlIQj+Thh3nNiFIsdxXJODxRgiPi0JrAqGSfD0RFq5jtJmgZTDAlZtOfx4A2AVEiODBiIwHIIwp/IDIQNgCvWZE6TBFrVKTHQaEefvOYpn4AYh6UxHS4GSq7pB2qTEpFIQe9OzEcyCobSkNRy9RpU2jJj/UxPnmiCknVRxV4N6XItopqVlEejsBd334Jr4e45AVLOVOAvpQSRWghQE7EFgLmREghnLphZZ73xRhZ6B958GIWTz/ZyxG3buOFSjVfHNfC3EDRqEJiSugiUNHwchm2lkDS1A5IXFgBUjaMDEPGktQP2C/dAbEcQVEl2c8uiO2Mm4bBMrTVifpUcX8+ZI5cv6FFUI4ijVukyieWT7b4CEK1Y1G0YIrcq5wx3qSWCp5CcnllATFXEnuUzc9z/ZysHDKlLPHakJEBEA/1OrFG6RdhTMyo9Vf5BbJuHb6TRntCbhzr94rJnaTVmIVoiF/tsXfXAS5ccgzNs0R19Bp2eKPmGikc/02PQ3Pwgc5VXPnedxi1nrxxYP9WrvjAJXTnInhcKhUplYUxKdNsVHUJfY9yxYsqMGNASN5b7t/J/AULWN05aKD6isvfw5DUPLWPLy2AsfcL+aMsydkGAMhEHR2y5EYGgYAdeS4k+BMA0AReQKYugyNSJXmIr41cq0PdXHrZxeyT5ZwIv8pl8z4BPUEgk34Encrl0tjzwk3YzrwF81m1XexA5QIKKRaL+LJvg7iilqv86vl7v8bcr9xjuCDG2FTCwMvlCAiOtUTldeZ4g8A8bzYp/x5bKftehZKwOs3nBz8QPY3wnwJ8/7X3mGP2/QhYykaCEtfOmMM9K2U5VmHFZZdyUFLc0XjCS5ULKkIWf/GY5HWhiSuQwOzKn445eo+mVCrh+YHZhhyLtCDk4Y+Nm5xneY2Ms7wnEGmX7D30qXjR+wKv8qfXy2vk3P2XsT58Pq+MjbWs+mU8Kp4/duyHzpMox6LndOBRKIq5QrTfPwHGQ8fhe1TM9Rdd7ptfeJY3X3ARVl3tPwzQmYEYA+Xr14tmq0Ii2UagHKpkYhePJCfLPOWwXGkSEsulI8f+Fg3CGZlZr5iaisasT2m8IFLC7Slo+sqaXrlRhpqr+0a5fIJm+jjLuL+rfZhgdyXvrcW0X4VTIy0U45s1WzHUHy3GZRBFqyB+vFWzFY1twusZa8nI6cyDLZFK1XI9QKEb6saMVDcYHzPYKYTlsMTtxa2sLr9M1/7t/ODyc7jsouOwx0MlEeIEin5KPBeMcoRymYliWGaM8h7C4jYsV8QRaUK/FysYpTorgZUOodeDslyc1Fzs1GFYbhvByDOE/jCx9NE4YY71I8/xUmU/1VacC+w0g5UuY7c1385wnOHLNVAMCiStFMNonglG2C4ZtHYK36k2voA9YYkTVZxWO02TcjgsNRPfSmKHRXytqY23ssHKsNsbYgpwtmUzQQdMioccG7e5ruLRnwhYZCpecIqTps7N8u693azunsD5dW+FoSdRxR3sDQtkCNkhAgkNF8bbGIdNr51C+H2BAEgkMSHNsLLZo0PqU9MlwAi/sI1kWCCHz/l2mnfaHr8Y3s9Pe4bY52gTXzYh5/L03oCgV0e8KA/C3REPy6gDZA09FAF0iX0SIV8oqQNy3mdAYaMpUkbvNQbGIdWjjbROrmWH5/FCJRJJVMLINmJ/WDF5nUeoKo6MKVLCzZRUMrm2NCytg2eBSbZYUwgvtMKo8BSDPKNY9ISeUf6+WOri4/E4k3SWipDkygKCYF235o+BYFOLZwryGSObjjoxrE3CoqyiFGp2BzA0DNsOaJp9eEsSmqQFJ8fhmu4lm3IBb4o3c3e2lt+Q46lwhGf9vDE5FtFGjy/cVE2vBy9XYJplsVTuoXKN5+B4F3pyYPcptmrN9WHIckn3CGM8mXXRizXNjWAf0KYlWdkz9sUWkCuLLFGM5jSB+AeKR95+TW673OMUblxx3Hgky4xh4bjmYdvLUO6BPX/UHNgctXTdemiboYiJJciYbY3Vp0WMacZ8dAAGpWUsYE04tF2afE+UBS3G4vJ0y1TFxpc15XViJRPdCyxB2MJvE3GnVOMkF1iuCQF0sqATTl2/xsqJmhi+9kAvJ3xrLwvjU/hhwzhOF7GU/DGKbphgicI5+nOMCGtElBKHWVlFW0oxq0bR2gwCAUR0U6hAbgh2lTQD4pNoKeocxcJmxf4RbRa+h7s+thewUxT3Y+JcMZaW1mybp9ipLWO5JyplMZQWqFFbgblpxdSEuAQI1cDimdDjTrn56RhdqWFa5mAqvQ3NilpXUZRFQLOiqVbRLaVOy0Ebs3JJk2gY87RMGVslJUHJ2sNTVYyT8ih93L9Grvg3HtDJPv4Hlbpowuzq3MRPb+/jU/+ygrRrM7BvJ9+/9Q4+/Kmr2fbknfzLV2/gV7fdxN5CmiWLZrH9pUf5wPsu4ys33EKqcRLzZk0yK1rtV/jJtV/lwWdXsaXXZ/r4NlxvFxu37OATJ5/KBtXM8csW07/9JT7xgffxiS99k7CqlfnzZhIzqFix6om7uPob93Lm6Ut54pavcsPDuzlt6Vxu/+rHeapLEfQfpKpjOuOrFV//1Hs4/iM/xRrez3OFKj56yUX0r3mQyZPnsvHgKC88dg/FmgnMaE1z83X/wsWXX8HKbcPMnzOLh2/9Ib998nk293rMaGvjyV9/j2/edBu33fxrWucuYFJrnQFMO1Y9x8fOv4r9a39HetJSZjS5fOszl3HSOReypQcWLphHnfQEhFPas40V77ySjoUn0pbNc8XFH6B24jz2v/IQK674CA/dfw8bDobMP2w6d37zGh57aQ0/vOYqfvHr3zKYmMgRszp46o7r+do9Ozn1uMNNi23lwzfz/jseZPPaLtqn1dOQSrJpw0a+/cmT+c16mxOXH0uscIB//9dPseKqT7L5YMiiBfPIiG5ewFmxn+uv+RIPP/oHvnDJqTza6bJ86RK8vp188cNX8sM7f80vb/sNExcvxd+/ia9ffyPPPvRTntjrkBnt5KoPXMY9v32Qx9YeZOmRC1j/+O1c+6O7uPcn13LtXc9TVRngS5cfwTUPdnPS8qU0xIv86BtfYsXl7+X5/zTWj28b4LTjjqN33aO86+0X8M1bH6J96nymtcW57otf5oWXV/Hv372J9skt/OT6a7nll3fw69+vYcGihdRlk4arKaDx1Sfv470XnsfFV36ZphkLmd4EX/vcBazaAouOWsYxR87FsaS1IV+4N+5xqLK4e5vP969bTefQbtPK812RoKYpW3EOszNMVjEaXMvEBNVY0FOOqCwTqpThtwkRXngjE9MKyd6W2Jx6V7GrDNW2ZBkKCAv56LGiuhvzxBJSs3w+uRHvjQCeGI56eaiItYKA/TpobVGE3TpqD01WZgIeeVUTtiriElNU0IZgXSnDMzth4QSFZFL2dGmyScXQCOxFsaa0l58VttIY7mBo91Z++r6zuXTucixZdFVpigWLn60voUuKhliGSTqgr7yXTeUdVIwtibioKnQl8rOrTR1Oa3oRvZX92EEeu2oelsxkYhAs6lY9ii53GtsTOzUTVemit7Sd5lgzmbDEJkNmDlienMYyt54eK84er8eo5kadWs60U2zSAaVghKWhz94wx4niVSfc8LDAFLuKYSrsE2WqXU3WriJtOQyHFRqDQaYnWoxZ6gLl0pKwqI9ZTPFt1o4qppRFFSmChYBW7XJSrIrP5wt0qyYmOUm2Df2OZqeBdHIyLwcj9EmAuF3NkXbMGCBv1B7V/gj9ymaWHed4ZRsLBlH89euQhFtP4HWzW8U5N9bIeDvD48pn2+geqm2pQmQZ59isH/JZ7LhkOiCUrNpNERnfWFUIEJDWn0TwireZDK3MS2JDJvwtAfxpZXh65hryJGrLZsjZzX3P+cxqajZVZbF82eQPc4PXS1GMWK0qzo7HJFrWLPSm1EgogfHe55GSz3xZY2mfh/wcE60YBSvBECEDClaWDzDZHuFf29tonuSYfYs9SUIqRgr+I+/TrmwKoWKBK3nIAZsrIXOVTf14aEop4kVFJgN+JaoS7RPOp4aaGDS5ilZHUQigpUWzuClBa17zncFN5II+znRbmaEcbDGrjQlIxIgcOmyo0ZjKuOcoE7IggDWRVKypaCR1rVrZpqr4mM7zvuku2VYLP6Eo75OKJxTFg03McvthdI/RAiGXqKjMRYBQ6AUnDuKT7QwrGqRSKpU6ifnyIFMDzw1qqsQgOa/YGVNMJsqydWsUowF0rwNRv+58VrNjFBocxcNbNI/vk4NXbOvX/O75kNF+xeo+sHbB+iHIpRR1Rcy9xSsr7F0ad090XJaUwMpR1T9fAH+LJt7q8OyOAv/6fCc3rgk5cvp4PhnL0hSGjIiHcQCVycqIchrKEd8342DsbmQBM7dG4s4UiSTUdEQBM490w4tD0Cym1T4IDpammogjZPbck48yhAUI53xNMm5xYoNFl0QZir5B7nVSjQ01O7QmpxUtUiUUEUUSDlgRYBQwnhPPRnGE0Zqy9lDa5+7SIFe0NZLrtfET0NoU8Yq9isa2LDZ6LqtL+0jHmvFMqk0JJYIsU3jJoYMBQ0IW/m+CgPxwN/M7alm+bOE/RAX7N1fqDk13oUimRLtspkEpWsnyfQKlXDeXnP0eag5bzsc/+RG8/DA7d2zgvCNPYfHbr+bW6z7EpW85iWde3Rdtyolx6lvPNf++6vKLWTCzhR//4o80Tz2GWx79JT/80gd48YVXuO6SI7Hnv4O7b7mej196Hg8+szb6VssCpqGRX/zwC6zespc/3PtFvv+F37Kzc554OcUAACAASURBVCv/9oUbmDCulc2//Ci79u/hjw/dyue/28Wrv/0Os8fHYX2MXO82PrD0XD73o9/yuSvP5Prv/IhSYYTHf/tj3ve5F/nFr+9jYveXOeWauzj2zWea4/zMR6+kamgzl37ya7zzso9w2VuPZPfurrGRgI6Zs7j0CxfA2/6Vi05bwJ1ffhtfGJjBlq1bmD5yLaf9y23GI0c2lqwbR93z9/C7tVvZuWEzP/7l7TQ3u7y8ajUf+9INfPZD7+aaj7+H9du7Gdl6P1d/8h5WfO57nPOmRbz/w7cxlBvklgs/zZy5U4yvj5yP6QtO5u3Ae666nBMWL6b717cRb5nO9bc8yn03fZHVqzfzwC1f4uqVDnffdRd9136Mz932UHQ+5G/l8+BXrqc33cFtT6yl93uf56b7HmPP/p1Umo/hm1//N6YN3sW3fvM0udE813/rWvLjjuPty2fzwosvcsFVX+Grn/swP7n246zb0Gm81W785gtc+rGvszT3Yy78zjN85gfP4z5wEy+sXcPjD97KFZ99njv+wlh/418+QXVhO3OXn8PFX7yJb151MueeMI/VnT10Pvl1PnX3Zj718Y+w9sHr+PIN2/jwRz/DUePjdPULcSxa1Y3uXcvhJ57D6R+9kQ2rfsUH33UKf9g0wIWXXWvIZaecejRxkey/wYBOtn9IwLK/dxdPvrySKZOmkisNoPOrkTpuzKml1U4ZG6guX2JsQGhOTQ7Mr1PEU/Byv+bAqI6Ui8WoLTQQwrZipACzlWbDcJH3nZEiM9/GF9R3MFpdG/mcFLuldZuEokwiMUhNFe8mw28mn9eoScpkhEpbVSb19FHiqaVN7qSKKRMOnkjBlDrFgUEYOSBeTrBrWLMbzY2Vfn5e3MrEygY6d67m8oXHceHMZditGt2uCRugvs7hE/XVdAtCCEa4e+iP3Ov1Uq5aYMySjXVJebcxG3bTR5KzHPbl1xFLzoD0IvN9k5xFLaUNAQnxiYZXF+ZfQVtxmquXklEuK0dW8sDIH0mK1YDbwupQGDMBMyyH1vh4yqbkH9CgXOZJZJckUYRl2pwasjhMUC6jVoo12mfA1DMU0/xhU5lrVTEmOBnuFU9AXTZ2D6vCgIy0c2Iwy1G807F4PtTsU5qUshCVXVK5fC+eNrFGL6TmMi0xke7hx+n2eplri3iixLPBCCu1pk7ZLFVxBvCxKwfo02UGlMvhAn50gC8RBbJAdGqpquznp+UDPBeGnG/XkE5O4Lt9Q1zct42E0tSVYJf0ygYVuk8TkzZ8rYq4dH3aRFOFI5pA4qakQiOkdynyixFto0JsBcXOItwg1b4I3E+MT4J8CUuqjGIQrH0e8YfIao95Th37rYCuXEDnqGK9B8+NQndcM+DBkozCCTQxQ9CSfK0BHOUaD8MWIZ/7/Xww0czkmiT50ZD+oibIRNYhO4IQR4ccbVnsl0xPS9NabbG8yqYqCaVBbVpxcqEM5WGtr4146EXxTnQivp18lmJJUxeDg6Pwhy4Z4yren2w3isaXgyF6tMd6WRh62iyu6qwoHUMylbu0cAIjY9wDMXi6oGn0FZNsm0ZpGaqQ6oqojGRlFlmHdI85XgT9UFmvzXe6ZjLE2yExR0mh2DyqD4fYREWiLlKDjuyN7E5yrkYvECECJgJtsKh4aCDg4ZdL/ObJkM88rbnnoZDHHtCsGtZ0t0H8DMX8sxTNxyjedKziHYdbzErBkomKi0+xWbRQsXiuouVkxeQlsH2jZvMIbMjBnq2aPZ3wUi/sXKfZ91RIuSeq8IbDMNpn8+v1wyx7Yi/37ctyaXMLZ4cJbD9kpxfZvch9pLNHG+6miBfmVhu9janGdbgR2B8oa/bmNXt7NU++rNm9T1MnCSEhbFWwoEYxLaUYjWu6CprhgibjKoZ9jG/k9kBRGYkKyQLcxf9OOhqTU4q20EJogUIHbkrCqFhwGpoBLHCNpoQqDcuVw6lWFafFamgb6cItVnBiFuVBzUgBGqbCoGQd52CBqF0qQ3ih2Bvb6CCH9iVr0TWJEspKm8gwiRwcLMfpmN7Kt372FOt2SJb0a3NBdLb//n9L5fV/9LC1fNv3UxFLbvEYdOTnbmJVzdz+yK/51s0/55rfD3Pyio8yvKfL3Aq3r32Byh4447g3MzgkEp52UznomDLJbGPWzFlkq+XKhTPefCZzs53QsZTRg/s4kB6PvetVHn5yNxe86WS8ijB7o0fbzMO58pSj+cmN36ZQfxHvfEfA97/5PTInXcgR86ayuroDV2v2rn+a993wOebNmkG7fgt88yUGuw7yEvBvp53C3PEJrmqDkqygd+6C+VkeuP8BSo3v5qSMorq12azIJkyYzBFNzXzvy5/mlu98HTue4T3vW24+i3C/3HQtEyZ1cJQ3i7aWOp6/aSW3Pn4j06dN57zLbubLJ97OyA8/TL24K7pp3vGjG7jkhp9hLyvz7i/fztTWZjZNm8Lj995B2hZ26hFYsoIohHzn3ms56bhjyE1MctlHzuLnt85n1cxj+eyyhWYwxOG6btxE2qfB9NnTqK9t5NUBuOq00zl8fprFxKnkc+zc2s3c2gwPPPAw4y59B/7gkPmCHVoBF6nionPPZ8bMRj7zww/zmZV7eNeiJdTbv+cnN9/Kdg/Gx2No8RnicD79qStosDTdGyfxwKP3szkjz0flez8MaLrqIo4++ki2zVvIx2a+nSOPXMLpR8i6oMyBHX95rOX9c2fPQu+XBEhYv+oP7PFHOOXsyxju6cZLHc7dX/w0Ry+ZRUfD/2HfwI+47ktfJts2nWVJKXdHj74umYngzDNPYkIaPr8Mto9WmNMu112aGbOi6+9QFW3sbW/ID+PQHsDwepkxXfxyGitea24Czshz5JIz6KibQ1pX85ynmOtE/CBZjZc8KHvaVO8abWV4KZVQG56cZFYGUmnxFf8+6nP6MsWCCVFVjR1SfdU4bcKijk6K6Xb1RCt+q1kqBlE1UJRvqVYVVfPEDV7MTEUg4SvKmzV+o8mxJkwpMnuh1gv5wW7NRY3K7F+6cr8sF9hcOUiDLnBwZB/TD2vjircdTyypCDIBiJWKrY3RcUNtDp5fTalO0283UG1XU/L6CU1UV4+ZxazkbONP51f2UklkmOQ2UtBlBsMEypN+UtGoziyrERWfiJdP4Q08hK47Cye9gER5N5XcarSVoj45g5TbwI1hhY9ZLrOtJM/Fmo26tCessJ+QnNvCw4WN1AWjzE3F6bASHKMcBtHMsaoo+zuYFJbJJtoRR7kqt4nd2ufl8iBnp8exFkkp0BwxCHPrJT9U096nzLUX+oqmkms4kBI/NdN2OY043XVncljfr1jXfy+x2jNojbWwzevnxbDIZKeOZXYVQ4lJrK900VnpZmpiEtPQpg0rGbW7g5zhfGm3ke78OsgcRZ2d5iRl8XTSZqDSw+8LncTjzXx/lc+0WJz6o61I1ZiKVM35nZAUGw25loSfJsKXdoUWUmchqi6VBdjLNSRh8WJ50genTahj7qQ1bPQCLCtk5cgrxq14sghStMc47bDeUpxgaY5zovzOl/fLIkPi1BzW+QEH8VmiNSt1QFqXiAV5the28XZH0ZZuoasUiEMN0gYTs+yD/bAyCOiyHSZWKy72talerdmpmNkCNXGoFCUqSpIJNB2+olr831JwVCAVRpA6hKgpO23YWMaYaJdiouIMSIT1nJFexGpvmH0qzhQ7Y/z0hF62McSAxSVjYESMbmtEO6RF0ARi/9jiw6y45tmKIi/RDQ3aeL41dEJDS2QLkpBbpBvxDF2phArtRqY14TEK6UxMvkc1sXZFbJpGWrbxUJn2Y+4gPLoV5PY7VKM5LquIn6KorbE4bDfU5YyWCXeKtHOlwm5Kh8j3W6quYkviN4tFSsStpVkqlVEJs6OiOHbe2DmXHvIw9GzTjCagSkQV22DtE5rKTMWaAz579qzj+vUFmhvncpQr6s+QKVXRYlRsSURg/ay0QIbF0DkyDX48iEBXowa5JezMQ7ULe0Po747MhRcnMWp94Qq3u5rBoqY2pqhFsQVNfa0YE2tKwk/0ynSEIhxLMcPWpOUeK9GFZc1ABVYqzQUiIotHIprVec3JVcq4s0j5ODsWSSduLfvDMkmniv3U8vs9+3hTZiZNdZHp9rAYIXsittDMjjlMjSfYHuTJ2hnDITW9ELE0satMjrU2dlMSNtyMJSVTNtC7ZQCOlCriG9kTGjMJ/1tmsahBJaXIVgPqfnX3A2zesIZf/uIbYF9IJhzk5Ve3cdUX/p2vXL2CT644m219ORMZt/D4M1hx4Ts4+rgjmdAuuDl6lIWUA6xdv46B3lEcuWHYmqBYhD3rSdQ04uzay+yjTuTS91zI0cuWMsnsX/hUWlidnH3hm7n5+9/m8FPexaUXnsy3b7yRw884n7a6NL2/34PtujRPPYIffeRH7DnYy3PPPQXdBRra2zl7PHz3xz/mvrtu5bv7IZuporqhCnKtXLjivZxxwnLmHz7LlHP7ga2b1/HymrUMJ6bxne9/jyUtr3DOm35gZPPS75dqWaFY5oWnV7J770FmnwH/8fDTDAwN8PRDt8IZx0ZikjHbkbmLl5N8+ib+5Wu3ctFZJzDS+QrnXvRBTn33FVz69rOAVeZaKRaStIg0SO6pbTO54XPH86GrrmLZu69kaou4hoamECNAqZKDl15ZS29fF9nGJozDkS7wEmVi8ThN9ZBvmMWK917KCcccxZELZ5noJ9m2sBObyfPoU0/T27OPe2/+DsvaU9x3wwq+uC3LJz/xIY6eBr3lEKnYNjHPmG6WDqzl5LdczPzT3sUVK95pjlNApnDpegYjEF4o7GS0HPH7BleJUipGbWP6L461bGDN+nWUpC4OLD/9fC5853ksXXQY49tb6B2ukKmWfhFseWUVk5a+hRu/fx1rfvQNbrz/efO8/FXdIAnUsPKl9RzcvZ6vPA0T6mpMaLvc3crCgP0HPIx4QxSeI0Wuu/kp4/w7KLJSuwaVnInOHA1ODXeM7ODm/FYyeoCBggSKK5pTFr05bTxAZfKQKt46TxO4kZO9tDKmNyg2+T6NtQEXHJnE740UdNZMhT1bEQhqFfVig7R0NE4Kw+Hxd0ClS9O5V5MaB3H5/kl7ywGrI7IxkDVcQkyLxZJdCmMyZE2YPFiZHNbkYFfR4pVcmbVekbg/QK3aT2l4kG+eewJHzM3gOQF6M1jC6Soonn+ywIc35kQuyF7SJNEUhfQvBGOTSRXZAhiuipPGjbUa3miHLps2nxafBV/8HmqNxUkYjJg8RzdzNLGqeXQNPW7OaiomkRhxrHCE0fw6PL/fBIZv1QFiMjzZrjKt1U7LoVeJ0XOCmrgokbPcHnp065CZY8vepFJ0WDF+b1cR6pCMxAShqXFqecYfYmelxNLAMXYLvSJsiUfcxXYL3LwsfxST6yFmmA4Omyyfk52Ac5w6JjW8k5hyyY88w/5yFzPsDBPCMjuDQTbqwPDrGq20sQ7ZVelhXVjBsdNMUQ6LpE3sD5tjIT6Brvx6OoMcaRVjohUnbWe4t9LHQ+X93E4vz2wv4w9JwCrkN0H+eR1V+oe0iHuxJiiCmCIUEUVTlP1qeJDNMLwdnCKEkpDuaxJNMWl6sDO/i0ppB67XQ8apY6eW7o0lzhqsCgJCRxEXhbaGkrTRLKga1iafcweaihYGlIMdlrECcbmNc1bVPFqlTVwFPf1RPJ1Yjuz2NP3eKO+SNlwOMiWoC6AkHnQleLVXM1DSZKsi0CBWQMLPmhmDuiTskgk/0KwtarZJQVBMs+MRCN+bCPlxoDg73cy5sVpWeiMMao+kDg3/S4RKR2Ujmx+x1pC7k5m7RUGqZduaqVVQk7DwLJ+8P4i3XchhEgMG+QHwpRrXEO23tA/K3VF1tLhXM7xDU+yK+KsCrsMuGIprnuyBZ/dq9gloScP8BXDiOYp3HKtY3Gxx3IQEhzUo5p+gGL9cUbVQERODYhsq22HgQU3hZU2lE0rDEkOmEBV0ODUCuME+TSDHMRwJEYzvZJdGJ6DpSMXh8xStR1h0nKg4/Eqb9glw36M7uL4iBLUWusMKT5SlRx+ytwJdZcU6NDvF1DnEZKxKEoRQN/eMwobAdJnpF5CdhSltisNiisXjFUvaFHVVyngK7hRQnhSLGegNNY5ktBaj9ndzEha3KJriorxWVJKS3wonxxSNceP8ghl6qcyZmEJMbvbJUuErw6seSFVvflwxRQrFGo5TsthzINHGytH9RhEryRm1TTC8N+pytGZCxmeSTMzOMzdR+UyiiI+sTGLY8fEopwllyQpIkjpKDJTl5tvG9Tfez9BQPmrTmi7BGzP5/M2VukPocuLCo/nNLf/OeZecw6fkmKacxO9XfpmOcc24/kEuveQyFk3KctWXv8+Z557HhN/exvuv/iC/qikz+eQrmTAuahnIWxO17Xz9Y+/gkiu/ziM3Xc6lF1xgLA0sO0bTpA7qxk/nmjse5OyLPsaTP68jO/VNvK1Dlo9muMzf8xedBHyBI+bMZk5aTiucs3S+IdVXn9CMFc+w/KzLufKUH7Ho1Ldx/HibmafMIN0wgS/86iG+ceNd7Go+gjOEK+zZvPOtn+DqRy7kskveQ+fOfn5w16+oaW7iq1e9lQs/dC13XPteBtbezuXPP0lspJU7f/8BUo60oUMsy2LawjPgg6fz7buX8bFvvcr2K9/JiSfcCtWzeOnmy017zVSGgJr2GVzx6Yv47CtJDpveSsaDT19yFt/72v+hI12BOWdga5/s+OkkYmOnyk5ywplvh6/9nLNPWmK+KHJjMbwoN8Gbv3otp196MYsm3c6C8883cTMyS588bhqxTDXnfeR6HnnbO7lsxR/YO5zhP259m7k5mYHTmuYzl7Dq8ft4zy/+lYerLmXXirMY2phg8nu/xuc+e5AtG2DJMskLdDj6rBZjXpus6+Abn7mCn337K7w4ScDnDJQV4CSqOa5V1rhQVbeMpiqp8io6zp2LimU46S0f41OPvPu/jPW1H307F5/yIVZuuIPH7vwOKy68mNbkHs74wI9oa6qhtnWS8cSS7Ta2NnDpGW/l6beeR+OZK/jAGcea/YXSvpqyiEfvvIkVF55B1rG44gvf4/Rj5nDghX3MO+wsbBm0f+BjaLTIpkI/pNKEYsQqbSbxZYu3MT4sscbOcnQwykZvhA3eEF2SC2pl2RvGWRQ4lCU8XQjVNgwUoT4JG6Sj1g8PhxWuXhI3q9hARFZbNb7wpoQYLrmgwskJIL8R6hZL5iP0vQreeuhYHPlVhdJqk/EQP01Ru0hRUYLdhdPTosykL1oGCRGfOFVxVcHmxs2i0Ax4BU1/aTfNeoBtu/bxtpNmsmTWNCM1c4VgHbfp7NT84sAgn38ph07GidsWoRh3qiyWmyWsHECLNNC0LzImmksEE1Z8AuTXUvJ6GHGbqfg5mmOtOFQY0B7NiYm0qBgbwxKl1EwS/jCjUqGTEG/JZfQdM9Ybc6uZnppNEGulm8C4e8zBYrUOOd+xGfBzvKJiZMT/xUqayC7BseLwIDKrdLyNqkov9/hDvM+tZyE2fVacjVaGQX+AN9W0sjkfssmHqUOOmWSGY7BlFOZIS9ZSPORpFscsLD/gwXyZs2Mh9W4NxzW+lcd6bsfKr2VleiG1sQYaKz28pEMWOjUc5zZyt51kU6WP4bBsMmQnWkmq7Cxlpxa/vJ+G1ExW+4Po/BqWZRaTDH1ydgbXrSfj9TLq57hnKM8R2xbSdoQEzGtsiXULNIX9QpyH2pkQiJmsJAZI+14EFCJ0kHSRLZGfmVhl5HcE1C9I8rNzZ/Hu6x4lMeMwvNQc5oppM7BIyUIkpF+qij7stTSPeSGTUxYtbuQ1Jpea3BF22UlcPyeZVfRVupmRmIht1TLc46H3KWMem5NrMg85S3iOGRa7Np2hZrvWzJfqdRgZaIs4QnLru/Mg9DHxh+vWcIQvtiDwZFFzZhyGA8VRMxQjfdq01HaNQFUh4CRGCIJ6llVVk6t43FQaZrGT4ghSnFml6KiGvQfgcAt+FQTUYjHXUkYlW6eVcRN5ZlSI/RY1MZdwVJFfH4kjRSQyPGQwAxURI5WhMAgvbNVMr1ckRZncAsMH4fFBTbXWZA7AtMUwI6aoq8OkQZgw02pM92u4ShMccuQVwYjQUSeMVf4yUas5sU8zuiniwIkliqoGZ5zCqjbFbpQUSMQORWK9BI/XgbcJEssi2oWujhwnrLTFurVFfvZYAd2YZUGigzerBHYwwsGwzIh/kC0qywbi1IkITtvGBFtUyZOD6HwcIe1vK6qQiYg2lYramj0FbbJx5eKRGK+9eVhQC3m5RYoHYQBzXGhrhtUHYGKrfCc1XtHm8XjIeSIuiyna5wp4VvxusyYThxo/lMuWGmXL0Jr5UNiyQm0ZCmFVWZvcYinaNCmbCQScF6vhl6VBph0ocno2Sc+ekCpbGQ7xrq6QwYLibCfJY8GIsZIqWTF0WDGmwyIPVqY6VTI2J2FpB4Erbs8ZNg70MzSSp6Y2mhPfqKlH6ddLBf/aXl7Xp8qPjhpFY7IqQ0KWBGOP4ugIlVCRyWbMBS5PF/N5vCA0z8mk8brNmDZexQtxXNuoJR03ZjgaopZ13LhRJ5aLeUqVgGw2a8DLn9lWiBeM5+M4MSwVUvZ8XPm3pfGkeW7ZuI5N6FXIFSokUnFja2EFI3z+XWfhL3wX5y7tYOnxZ3Hvs5s455iZhJ5HLp/HTVaRjAvjSC72gHLFJyE69jBgZGQUJ5YglXqt3Tc2BHiidhUvrJg4TpcZzZVIpDPExdb99R9e8L3vGaWvvFYeUk4ezZVJpsRYTIiZdsRbHPscG9c8xWfeejz3h++mZ/1tNKYiA8zX8ImmXKxgu/K+ENtxzRiK+tOyHRyZSP2KOaZkVZqYK/ElkdVHUNqPk2znkZf2cMrhzRR9y1i4yHHJ+dY4JJKOcWu3bVG3alxXeAQy6/mMjORxU8kxU1exowgJtCLmOkbBqpVl2vWiZpXxkfMi4zo6OvpnYx0GAV7Fx03EzTWUz+UItEU2E5FORAntuI4hrcqxBZUSo/kyiao0CWOr8+eyB7n+RP2UzUo6gRyqTxCGuDFpov1jHnJE133rST778ftpndJKb6kJJzGV0G1gYmI8We2zC5tjlM1Cy+V3/gjVtkchGCQdtvCWBpeB0OKoskvcVPCguxyyRVR53SVOWhrjjFNcrKQ22Y5iqS7tMktW5aK4rzVzplnCyn1bPKb2/lEzskUz+VJFtYgkqsfoqnKZCm1WCquSeG3MRZVR6elGZVbwpmX0kuL2VTku7h9kXDjEcGEbGbvAwV3Pc817juHz511EcFCz9/+y9x5Qllzlve9vVzy5T+c4PT05KcwojhIjaYQkQMLkYJOxscHGODyndy/WA2Mb8+znZ1+uwWQDJlyRBSiDkISkUZiRJseenumc44kV9l3frm5phO175fWE1/NanLVm+oQ6VXXq1Kn97f/3D/siHhqZ4U8PjTIg6GuhgY1xHRaf4rhVICXGnVI2VcXIK22I/xLRpZxWnOx2lN+BLh+nKhy6zGZe6aWNsvEZqVC9bhpT3VwQRwyJGlTHnKkN4lZPGdNidB07s8OMDG60RMnKcmtqFf1eO2ftHO+URpSucanlcm+4wD+FJTbbaSatNNdbPq064h+iBX7XzpC1UnykOkjeyvAyp4AvymWlmI5CavYiH+1r5OFRmJLjFsNLU4p6LvFPEw6ZtAYldP6xM5oOTzHSEvJ300vous92L8vB0iGemPg6hdwFtBWuNhOgk8RcZWXpshzENkXpgGeqp8nZDSw5OTY7jTREJfaUnmGVnaPitjBVHcQXpW9mk1FXD0YlTtTHKQTTFN2Ya7zNfOCVTWxvhmhOiPpaDhOzMgmoQKN4OItgoi1JmtATGnu9YuaANu4nTgdUT2oK1zp8ee8h3vqxz7N60y0MOuuML5sIHl6jHOZ0jWd0zC/baWMbIfsv0U9CYz2eAFg8o0O+VRsmF84ZXlI1rvDWhmu5xk2xRj5vFvqaYF6sL1A8U4r4h1KV9b5Ld5QYV7dLh9HCnBNiMHuN2FYEcEpO/QiuKyTCiW9Pab4WaN7lKdP6u3yVYmFR8+QMHEHRQMD+YI6dbgs7GyzjuLBrpp9mJ8+008jf+A7XuS6LoeZQCD+INRcKHyulqMQgQ8VcFR7TcLQWMNY0wZ07O6j1u+Q8TdaCoyPaUFiu7VOkhbxfggkpKNOiVNWke5SxDJEJWC4FvvjJTZrTHLGWCevS0RMFy7Kp8P2a9NViNaKgMVEpS/yfuYld2nSiDI6rmnhaCiGFfVYTSrEr5sWdsOpKKexl4rX8+1+A6rjGvkj8LBVKEHbH4rHj0/zJp+tsqzXTnLdoi7QhUjXKuCYcOFXjJ3WLv6+U6SXEcfNcrhyulpe1oizcYWm7egmy2ezAUVHsL8COXoWkkYjpg9SYIuSYdcXTTpFLw9FJTWMBirZiaknTXFTUIs0js/CDWsSbUhap2OKCDpnwavqnEwsd4eEJjFSLFGKvORfCDSKU0NoobkdqsE+KYQFQEz9nTsR1vl7az2eznbwktxqVCk2+78ZNiqODmnDRYf9sP786/zRdmc3MirOGsTERJVoirZaYQonjiKriN7NIW7bC6OlDfPSv38gf/b4w339+txesfjW7IKP3slectPLS6RSOI0WBDFlyDRIZtk8qJYWTLJo8L4On74so/2drGo2ybBzHNiiXbdtJgaAUjuOYAV3WIYXeuesUVeOzN6V43vtW1iEnr22bgV/WIQWNn/Kefc5y0mw4fwdjp4+x78gQf/zhj/Gyqy8wuYTKtvFTqaToWDZCFo87KSTk0iLSGvO66yR2IufujwyYjoNtJ8dFtptKSUzPv57wIJ9fljU32ZYsn/axHXn+ueMiqhu5Lc5MYXVfzkdvey9r2v+16BFl9lNQQ/N+s2/JCAOkrQAAIABJREFUsZAW8cqxkOMphZk8To6nfGGwY/tlbD1vm1GQijP6yuvyfXu+i2X297n9Sj66nMg2fjo5Ziv7nSyb7Le1/F3IZ1i5b9ZtLR9LKfB+5ljLuuUU8nw5f6S4NtaNyfmyfMzlPZbjkpJzcfmYP3d+JOkhcv6lfCngksdm+3KemCP6H/GfRpz5f/T9R3ng4XkKzS2UIx83vQGZ5a22sgzpiKss29AVJlFcYWdMOsFC3GTabA9XQ05rxZRe4qnyPFkJfY8dwtDmcKHGr+50ya1NLExElWa1i3p12VtMOHJCZR3TeN2K0jhUzwrdAIIORdP5CltObRnIZY4ioQ5yfarIoK+wZVY/DMNPY3g+kjZhzcKx/piRfpfVvs+9tSmc+phpsS3VZ6hWhhgYmeehwwPc9kg/nzw7Rj6luDST42pvFZvEOiSY4Ex9FDecJhKzKrExSa0xxVIYL5lYLmlL570uI2ColQ+wxSlyY2Yd4zqkrFJc7YqBsWNUw0tYzCqXjU6epWiJcu0MlvDuJD0ivd5EmXlxmUPBeJLbaBdN3mqTXINQnIoD1lhpjoYzxqQ4b8mEKCmmpuI6g5bPSFzhBsuhbudMS1XQ3pKy+EZQ58KSzcVZz1h4NPqKnyxqmm3FlXKM0zA5D/UirNuuODOu2Zy2uUp77KlNMaUlvqqLnFfg8PQPDOdpyW1J6BOE/DRa5BorY2KtRnSdEmKs6pHREUXbJ+s2MVo9Q1k5tIrFQjDF4WiejV6rMVUdJCa2bCYqk+zwW7jnKYfD4yU2pR2aci7RlLggaeLlLF+5couQgk5ROybtuhOHNXnpmm+xjdjms3tKvPs+j960qHkL2E4Lnu2bCZskYwgH6lBcYp0IO5QywK+gWfLD26dFDCQJD5qHReEbzDJXG2R7brvhPjboGJmCiU2HoDWZCKK0xQNLi3j1GtudnOFPdQntRQo/W9GbSRIFjgeYmKizec3uLsVapZgqaY5VMHmgXqxoseCYwDMSlSYmwFpxUFcQHtf5fhoJgu+vOkzoKqejMsrJMhRZrI5tlrRmMVassxSXuYquFJyN4Gwt8emV64pEnbmWy5t6fVrEYmNW8ngT1eueqSROrzGd2MgIQjoealrS0HKDeNQpioLciZec/F5HYfasJtulDNPfDAVCpTiqTVvXERRVht9l1bIh+snVUqhc0gUUQ2hPclsVzqbEWNrJKipliA5AtiTNcpUYGsvwloZ4Bmb2Q36jYqZk8emvzvC+v1pid1c7u9O2AQuFHiLxgK6tCVKaoOawGChe3ZbiEtvFL2s+GyyxGIX0eQ7jsSiOod2HuQi6swrZrct2KAopyXLVdDeL6h1G64m1SV4rlkqiYE2uV25NkUslfnuZWDFehUGtuMZWbGtKEifE465DCvssPC6TeS3NT4X4feZlqFJiMiJcyMQeao8UkUKvMkZTkfHM3FMdY21acV13K1FV6EbKUCdyGaE+Kp6es7krmqPoFKmJ9VJcRSlJY5GLZtlcPLXMjKW1Ec6R9nxKc9Ocd0E71+66yFipyNi2PIy9qIPQv6+oW960DKTn3mQQNf9M0SYDZ/L6yvPPf/zcO1dWI8tJQSEts5X7y2Fb/2JheT1ZXwIJywLy2LzvZ+6vvPnc11aek6NZbO3k8iuvYvf1u9jQ15Ugi8uVhOyL3J7bXnJ/uUH1vM/47DpX7sh7l4+JKUTE/malYPlffYvmtaTwWNnnlVXKX1ltU1sXl1xyEW1N+WfXee4yyXLPHY+ffS1Zb7KN5/Zx+aJg+2zesoXGnPfsulf249zv8GfX+bxjsuw99y+X+ZfPPH/dyTm0stTyITQn/bnb/tlCzKxjOZrtuc+zspZknc97///q+K+87UX8K4RZ2cfRg0v801t+yuG+GC9spmzlyXgdXOK1mNlzn7KpWA4blYtMtOXCIwqyTVnYlbEI6i6vyNuU6xZ/GYbMVsvm97K/PsvWZp/rL0wZUY0lM/ucQktvQ7wJBX0TekBCX2VxJHHz99tkkITKqLSw5cqcuNmbkVdm9NK2Ff+tHMQ5qJwQIZBwm0TvBUf3a448BvO+IBOaaadAX0ZoAt3M21XOTCke3HeCn5yoExc9WgprWZfayCaVIW25SJ7nT6unTQGgnCKu303VynKt30OLDjhTG8aLFlDpTcTKQYq8UDkmWLtiNxNYKbrsHNfYKbS2+HEcskVZtInVB4pRQfokl8luMNmt0to1rVh/FalwyRgYS97sasvnFr/ZqCjndcSVlovYiLQrMTdNs4Rms7JZslxE514OJvDjMlu8NpbHUmaVxTMyW4/mON/KklMW65oS41RnCcpybZeAe6GyiaJYxmZBSaZhW4PFravSHAwjjlYVb8z3sF77/GjuHmNq7LtNOG4jRR0xHS3R6OTJOQUapT8eLbDVznFAR4gaV1SoU9XT9NppalaaQIccjxYoOgXK4RyTQuQOR7GUxy0N7RxYlOSRWZqcOs68x8yUQ0sDuD2ghcnSotCzQtQ32jba8hbetMXkrOYfHhjm9x6o8zddTaSqI+yNApq9VmIrQ9+yuERaWQ8Rcb3tGc6SJHiI95sIGBLwV1FB8bCgqYuP0e610ZrZzN065ibLNvmgYpotYoRRBT8ai/hGWDWq4ybtmMLiiDF9Ts5pacEeizB+Ze0x7O5QbM4pBkc0+2XwN77LilExHbYU+3SSVHB82RT3majCdXaespagd9grhZadM5514tnweifLaWLzWaTIWOspNjTAoXnosBP1pURTiWHuV3SZV+BywYKH35z8vErTmEQYyb6dDTWTZ6QFqBAFbN+VFs2XiDFxYh2jm6FehUgMb09AbrVCSdqE8N42KFOoLZ0AXyZcPYYfYAQXxvNDJuuCbEqUoMR9ybxauLCCvkvnT1x9RRgVQ7EL7G5FXXwLJ8VqCxalcPLAPmkxGkT88SMl/uqeiP+2rQUVCyKmhRprLhlLESYyT6yWxHfg8qwy16/enMW6lM0tvkOc13ylVOETJZd3Fm3W5RMjaMEyVjerZCI7qY0RchAqZquaVknNUYl1jKDbQh2fSWPyd/tchdBCp+WcmI+5W8VcXrXo26hoXgPWmKLQk2TxHpuDoKJodzF0iE6pMxRM6mS7knMgSN5RrYU0ZFq6jxAbD9Bxd5K39LYTzFqYBlBJuo4wOK5Zsjy+Wz+LbzegRSyh6yZeUJSwMomUsUhHiyhpzdaHjRixp8Pl7h8c55Zbz6enu+nZMfZFHG7Mqv6dRd1zxYoMVCv/ZE3PDcLPPX9uISPLnns79zUxshUkzCxzTkEkb1nZxspfWUdyP1lbsvhz6zbrOHdDy/fP3Z55v9nnxDBZtiPqVSlOzG6e89lW1i/vWVnHc/vw3Haft8nnfdbnjsfPruN573n2QbL8sw/PuSOrlX0w/8zA/69v/986Bs+tanmfnnvi2Xuy7uQ4PH/d5pg973M9+5Zn75hlnn30wu8k637+8udu6n+77ZXv6/mrePbR//b9zy754t9ZOWcODZ7k9z99Nxt7VjFdCgn8Hm7we2i0C6YISCmbG1QyAxYqrRRVomte42HaVVJ3Ha9DMbJ5u5Phaj/PD8KAscVpXuY0cHbW4Ux/jCv+Ua2KlFyUZea+lFiT6DFtlI0S/dVgCCWYQaXjMoUnXW3RswgdUhC7JY1OCzEJY44qnVJRDOZ7FZJE8cwxeOyQps3T3BMpfqxD014bszz6owXi+gQtxe20tF3AfFM75biFedtnzsqw33SDHa53XXodh71B2aQnLFhpNnotdLstnBAUsnLM8FSkZW87jaRNVqvLmWiBDifPmJ3hdbY0+GUQEe+zpLjcIsi9rrG/PklObD+EkKNsYwoaBuMUrTR+qo/FcIJYjMKiJUZJUbfz5JVNXik22AW6rTQPonkJ8lxsoqxeabm02mkO1MdoswvGN26tDPJCLbZsvlabohh5VEKfzTlBhZThPvoOPLUgLvfJb2p+OImcmpYJfaxoXGfTU3E4OK35tq7zmnQfV3g5fjj7AFkscl63EWQcDxc4HskXWiVr5w2yMB1X6bPTpmU05+QMJ3kinCV0Cib9IqsrHAsXWLDzrDfH0ONU5SSvS3XyS9kGzpYc3ntqicrsBEeFFJ5WdLT5OE2WSS4RRNbNW9hVi9KY5qGjS/zBfYN8enGe/7d3NVdbPoesGk+Wj5B2WinZeXzBF02inLAatUnEWK8cBPgTRPJzUYVXey5pW3E0trFF2Vs7S1fDNVSVQ4eyOKUtdrcp066TONMG3yLvB9xVLnGflec0AU3E1JVl1KkyEToidj3LbTShA4rJrCRLCJwk9Iv5GAZMeEZMLVZ05y3DpxqNtBnoKykX6dX0aszzws2Tfem2HWaDGUIZoPE4okOqKqaoLSarijtizXnymw0T5eVaXzEQVLglDVc2p3BrMUFZsTRPIqqLYchOEKr1lyqa1isjUhK6hJIZk4hU0hAOwMITSfu0sF6ZtHrdlGStSls6LbME+cCNyTICggoBNB6F0kPaqM7NGScVtPyuhRwqy4uB8ESCxEqRKDMNV/SPosRtAltU89M236nN8f6PPwMLBW7rbSFbl6IcQmMLA0eCiPNFdZtTTMyBKPFX5xNEq1rRdLYpeptsLsp7bG/yeONWxQOD2vBO/bSiR4S+NWWUrgVH0S7xYAFkWpNotykbKnPQuUaRFiR9KjE+byooxpc0hycStex4bPESKbBbFZNLMCiZt0pRPpm0e6XtKoWtmUxFmqr0bCzTkDBt8wPCuUOxRQ6nUogoaq+yaCqf4TUdHRTsFLl0TLZb4bco4yfY1aCYmhricWIKbiuhVNBycMWQWGbQWhh/QnSJMcIuBU15j9nJo7znPbvp7m59Xj3xYo48L7yoWy62aovTHDx8goaODqaOH+Bb33uUbedELUW1Gf7h7z+N19hLZ6tgvys3KUaSomlloFuYGOPOr9zD5/72AQ6dHKbY1UJrU5ZHvvkDBuoe1twEX/2nn7Llik2UR87ytU/dRS2XJp4a5Ytf+AkN3Z20Coa9fFtZ78rj5O/KdhXzgwN8+1uP0L6xj6yU3suVw8qgv1JIRNVFjhzsx8kUDKfszLETLIQWxfxz2zLrXj4mK9tLCiJ5lHxOuZcsojl7/BTzgTLrWEFvzn2f7MMLuT23ry9s+ReyznOXSdZ/7jO/uP//5QgIJC+RXvf9aJTvfvsofrGBalgg9LrY5HUyrxy2JNdZ0ybqFwhfwRWOos1RLAnCFmBmzsaAMwGQGYw1f+vZ/O2lLVyXcrEnYpryirsnNfsOavSyf1y6rhCeriXxSiLtlwt4I4RFWDwMbTcpHOHiyFkrP1e58Iu1gsQBlaG6pI2tiU4p7AocGYBv36+5uFHxXR2TC+A6y6KGFKQxObmU2SkyTjM3+uu4JdPNK7rauV4381Y7zW47zdW2g43NE9LKqg5QUp5Rb25MraZLeRxQmnJ9klRUohbNkvdX0eS0mMQJSZeQofUit5W1WGaWL/y+UmzxTSs2s+2lOOB49SxuXEY7jVjptQa2sL0uwvoZQl0jthtxdMB0XOZUXDPq0wEUX9MBVR3SYLm0a0VFwToTFxawKHFdls9AbYRDopSTtAcs2m3Y7Dic1ikcvUCPlTecrS6xkqlBqyRuSJt2ShtVc7o9KeaUmLx6MDSrSS8ptjrK2IOMKZv1/hoDujy19JTh0U3IaC+0D8ulhGVqcEeHlHSdReHLGY/DFPNYuPURtBQhStHsdVKIq2Z54StKHFoprtAeTbHBXUWfb/GmTI6hepavVYf4f0bn6FgICRZmODPvMTJmMVCqMjA4xX+7v8Rvj4xRbPL5rYZVXBD75JXFpqLPjxcOM2L10uq1GGuPTZLfqlxORUuEojYWa6I4ILIcJCFF0NgB7dKvAx5beoouQfn81VS1Zixc4ArLNaipqMCHqtBowYO1gNvrVX7dzbHDskwLTTiPO12hqpi4UzosEzzBNmBAUhvEsDiGIBYkTRt7mjEUG5TFVBjwZJQYFi+EJa5rCrlciZuA5t5I0iMgozQVXI6EczwYl+mxC8b8fisOZ3TEkTjmAoP8WKSzgs7G3BFaPFyb4zdbIprtPLNzMYMzmHi03jZF2xpMOowUbmIpJEWYzolvoOn/JVxWC5wqpFoUtrRPpXBLKZyCxIuBklguiWQQEf2aJOHB5MKKt1BO1LbKpMLo8WVurNQcEgGXSQpBiQ9TAlGJPuVYcn0grfF8RVRRfGbfad5z5jjjnW28PF+kIUwydgWRXN2A8WJ8oKxoSCu25RUdEkHmwNyCokFoHSnF/BIGOZPosg5H0SEcbfHsS8V8aUrT5lr0eOCvVbJp5odhvFEbRLs+o5iJlUmVEJFCcdl+R77HuF3Rbik6M4kwZU8Vtmct1lgweFLTvkmRr4ATK5MxvFXB4QjGYs0gmjWeMkWedELE4F0OmVz2HtIhI1omIrZJwXiiOkVj5HOp20AUwPQ0VCvQ0quoLyoenJpiP4s0uq3EdpowrhPJpEtaHMZDUypogQaTalqkB0Kf2nnFOi68cD22zHx+DreE8PQCVrxSsMyeepodF23nzBzMjx7lHe/+loGjw7rI0iUHzqGxIWvQBXlcr1WpnBN5tVJ4BQuT/M2bPsn733OSi25ez9Dt+9m67lOcGpzhzI/2c8djI9heioZGHxXWue+T3+QvPjWPxyJ//dZvcvi0piAMSrON4Nn4Knm8Ek8lhHtTYOnQTFCqc3P87rseYGwu2dfniquYWjWJopL3BzPDvH7HpzjQL6SEgE9s/gx3PXrKbCuUvC1Dthd2pXwpyXtX4rNWiiIRaSTPycI1/nHLZ7l7eR3SipF9q9eTdcl75HEyXicxWGYjv/jvP/URMKAnFtMzNd71kSfBa2ChXkCn1ouMjRHlsQZt7DEELBOehxCZu1CmZScaHAn5XnThZBVjginWCfNKcbwS8KZiyI1Ccu7QbLjZYuMVijd3Kd4ibYgGGP645r47Yz76/ZAnjmjmM5qZosbOJ2q9NnFwDw1YZQYHLWZNQpbOYWbMYo/j5BOyvNi6Tx3VnB7QnNeBiSVrqVvUbWXC2gtoMlgmR7XVW8Wg5dEjvlco3uA73ODbtNmwLqXpcmMGoyqdyqfVyVMN58l47ZQlkYGY19pFbk6vp6zrxk5iodrPiK7hO3nSXidHokWycSVBaoTD7UBbAT6CxR4ds5+IlNtE4Pealq1l53DSG4jiKlF6C4GoW2unwesk5bYbkcpT1UGOVE9ghQucVi5blM3FVvLZViuLN0pCghZV4xJVO8+NFvwkmueHUYljseZ4EHORsvlGVGU2mMXNS6qApi2rmJiHwbo2xqcS1TQ0B1VBPVxo8xXtNchlYW1R8Vu2b3wJvx2H/ErrDbyz8VpmlvbhVvoN3zey0qySpINokUApfMmNVWKlMsMsETk7bVramgC7PsFEbZBpyauNFqjrEN9ppJhaxWdrs9xRmyGsW0xLEoOT4rWZtdyS6+HQWbjqbs01365zxd0DXPONMXb9uMJJ2+cjDeu41e/hVc0ZmrPitRbDQo5WtRmqx0jFVTK6SlpHHNcBJ3XIsPI4iW3adTPGvD7iwSji61ozHy2xyvWp+asYkO9VWtNOnoLlGYuSvnxCEzi5BP3VKu92ijSKoADFNhxySjTQiZJR2q9DUZJiIGIM+f0VpGgOYZ/4oAG3C4JikP2Qe+qBsYu5QNqVlmZqwqJc0wzHRoRqxuOxWLHW0/yROCuEouAss1O5hi/4dFTnobjEU3GdvSiToJDSNpL/Kr5Cw1N1Tk/DQ5OawQJsvlrRuClJR0htVuTWC9dZ4WQTlXokpC65ybAg6mMPo1CVv2JNoqY0WtqvzRAuaSZkeYHwJZF+GYVT4jconpLCiRX1ukg+pSDsxHhMVgSqlAMhgK803tYorMukkNM4gWUmkZ87eZoPDC2wOb2Nm/wNfL9W5mNhhZOh4kwdDs4m4p+bMopO8bIrQM2HjKOYDjVjgu5KURrD1HGYmBC0Vvh9mit2Kd7a5/DpaxxOFuH7UczjD2lG+mFOayP8K6ZElAUNpcQs2hhTpbQ5fzqFJjKgmVlMgDFhIQi6Xg40I6LeLiiCw8JZhN4+QXiTc+N1BbjIUWzUluH9jomVjeRla9gkvD5DG40NPeCQ5EVbApl2Y3lnaVmlRZ9F6/rEB/HhfZqDU1AotmJieQgIdIClQ/J+D8ppMm1XMYOUFqxMpkTkM2fcvdr41d+5g+lpgUjlO1iepZtHL85/LwypEz6YZRHWpvmrD3+Qh/YepYRPNtNAefBhjhx/kptfcStd5+1m+6ZO7rnjLvq27WD2xMP8wQc/xDe//EX2DdW46JILSUsjXcGRnz7Gmz7Uz71H3sPNuy/kyleupc2ZoGV1O+WxEartq9jcanPs6BCL5TE++Wsn2XFLntmz45x6qILd43Pxzg4O3fcgf7zzW3zvR/vIrW1k/aoCX/6Lr7HniaN89u8epnljCz/+x2/yf77vJ8wuTuNYPte/4WLa8hLDo5g6c5q//8Ov8s2v7+W+7+4ntyrHY7fv4dSPI2aCsxw4dIRT/RFHHhki01fmoTv28cSDB/jiJx+hY2MDt//57fzzP+/lvv/xJLmN3azuSHHXF77Dba/7Ht/61tO0by0ycegJvvjlOQYe7KftmnacpVFue/OX+PyfPsRZVWHHZWs4/siDfPYzT/LjL9zFWTvD9s2d6DjJon1xvupfrOU//ggkfDrJQ77323cwNtKO35DGcTroS63mrEq8vMQWwUXRk/DRzTVdHAZEqSUxyqsz0JVSdGSkmJMuieLJUsi711ts7nCSGb7Efp1JvJQKG6AxD6ukldagaJQZ9YLihyciDp7WNM0q+qW1uh2a25V0IM0kLDIZsck+qxlwViWRQDIjErL8nmOarAgk7JiHQs3NnuKk5DeiKGAzhLSlZK5jcY2OeXN7ioy42ocJ/2akguGuPRNq5mzFGzMpDtdDTsZVGlKrkcKwW3nYhCzpgGP1UTwiE9/Tnj0Py/JM7mtNRzxJRJdyWCe4XRa89pixac2FjstQMMlgfYq87ZPDoqY8IreJgtNAFM5giQBBCrzqcazUerTfhRvO4sUBtXCa9ZbHTqeBGsrEPUmRPaEc9umAnCB4lsex+giNEu9leXwuDrjEUnRYPvcHFabSETc15mmd0XietINF1Q4LggqIZ1qjRCApEx2lxfpDMnnalGSSm2WEB9ZUTjIz695q7td1MpVDhPI54iraaaBueaZVFCrfcL7kWuZKYW35VKISKi6DnTKIgWWJwClnuD2SWxvbOVQ4zaOWzzVeM2Vbc6EYzAY248rl5dkiN+SL7Eo5XJFOc3W+gf8j18IrMykuzNnI/L1ah1wIhawmn3VZ31Lm9tFhiqk2grjKYcNvTJmUlOM65FUi3FIep4i4UNBCK8VP4xqD1TO8v6GHkCyRSlMkpBbNcaPTYNSis5L8IEWOirgznKdoZ409xZ9FVXYqMZFVjCzz2CTlrLBs2t1QgEu6BS1STEi0liueeZonteYyW3I768xqm22Wx2RcZ09U4Ra3gfk4Qnh6zUoZdFHMbjsjuKjJw3bLfF8yzp0UncpiXEcckFaslaKHiIW6ZkyLgGmB0WiKzZUWnJYs2zZrruxSxoc4lglVlyIe0giKJqIHbUx0zbBo1L7mOpWM+caipHwWgn5JnlCEQ9q0P8VTbnhIU5TMVykOhLfZZjrzCfQkXjxCv+hIDMXFW1DQPrsnae0alZgcsDaw05hz6MnTIR/+xgB/MR9xQ8MaPOUxpeX78mnQIbe6vmmhPy3Fcj25xjQKN03Uxx6MT2sKlrgdgBgVNDTCrBj6lRKBSLooogqYHAJxJruiXRmxwt1T4gYd0ZS2yc4IB04zGMDDkt2rNed5yvDixLJELG3EIscV0/QQDoufX6zJFOT3IghPghiWy5rAhWGhOPjQhzJiFrFTaY5gs51QNmSSPCifR/h7yianY07rOrGymRPVKhPsaFqDMyVFMbhNirFxTdFXzNo298+cxncyJuYxbaVN3vSCxI0Jt87AqhVzPRTLE9+qU1+M2HGh5u1v2U1GDBLl1H6BXboXOn69MKRueaOO38RrX/MGs+63veVXOH9VA3c+cQ9XvebtfP2/f4hff/2fM1qa48sf/hCnT57hn3/nFTyZv5A//MM/xC+PMTMrduXyds3M5CTbWzpYu67NODjmO3r5nY++m8suWcPS6CK2a1GeXuDzHzjMqk1bueUdKbKbVvOqt10GYzVufPUW5k8d4do3PspvPfhG3ve+9dy86+sc6J+mdmKSD/7JOG/6rSs5fs99/OWHanzsG69kba/Pw4/UE0HE8hEaP3OKWms7v/vnN9MxO8znv3WSy1+2RfIX2PW6nbz9bbtom6lx1ft3cOWlazn02YP84L4F3vqBXQTj/Sw2tPH7H7mJ9ukxvnPXIQ48/Dgve+cTvPf21/Oe31jLpz70ALn1W9m5OeaaD13BBb02f7r5y6x61cV87O5f4qHfu5Pv3nuAahzzXz6yh9YrLuAlF/eavXvxa/jlD/2LP/8hR2Dlt3rXj2bZ91iR1t4McahZkMFUObxNFAzYpt0gKJ1cnEXOL9damZCu82BtCvbOwomqphpp0y6grNm2w+WKSz0z+50Z04QRuH2QvkwRLinKe8FaCzlRWzYrVrXC7sjm1c02lQVYOKPZv0+z5y7N6GnNYh2DVjg7lPGlEwBG1K9CqBZF7KGjmsxsQmwWn7xhFMcDzWbxAktpjqg6U8JHsjxusV0DM3Q2Ki7rUHQ2KMYkqctKVIlLdXhvh434L7dIArnt0yHFFzJzrvP9YILvicGYUyCwsqYwqVdOsE66TUqEDd1caqW4Q7mMW9oEqp88YVG2LMPTWWvY4YpMHNGU6kM7BVIignAaWOuvphIuYbutJlpMEDtfi6fiGoN8qWiRidIh7l58iqEVAVMiAAAgAElEQVS4xsscly2CMIpaU7m04OA6RcbsPDli2qw0v2GlOBKJQrXKa/0WDtddvjtQhU7LiN/Ex3JNTsyLk4Fdvs8FB5ZqgtYp0i0JX2hkLIm3u26N4qK0xdPErMso/qLhWmacDqzSM6SiReZqw6ZwkolAoxRvBgcQ3tgic8E0bnotlr8KO1w06GJYOWU4PpFyKQejBMEEaTuHX5/kH2uT9MaWJAQKhZKMUuSkUAkDQ3jfZqW43HUppGJyjnhxJgO3RCaJp9mUUqxZD809LYmvYTiBY+dNckO/rlGKq2yIa3winGcUxSXKY1Q5iPAgHy1xNTW26XbsMKBJlLJYFKpDZOM6Ev+0v66Zjy2ORTUeUT5nlGOEELcgrc86Oy2L9ZbFfrFaAzY4sM5R5LMJ+nmiZuzXDZLzqPjVEXE8qvHXQWxEScLLO4htyO1i2iR2JCdi4xhieJrymxQqZKykbdcI9QlOxBUeiuvMKIsNls/rBTkVtJCAgbjM3miOblXgpZsaePNLNJdstIxPZE1+5KI0FYPvIthXCkFNDFtFkpn4sQnipueWL08iVhHoUtDHqxXORlCbFaqkKJ2B2iSJy65wYUPQcuFYTuFUq5PWv6BaK6p2cz2SL1kKwC7QGxNBllW3eKI6y6V7j/Igitty6xD1siCa52Ox2/NoN/nUgrLDrS70+CD83J8KeVFEEouammS71jULoSYMYGISpieWDc2l3Sut3jpkm2FuOEHytq+x+C87HSq6zp+MlNgfyRVRnBAkESRmwsJcm1KpxPv8QBlabfG5k+5W4v9XkuuCC41ZRc3RTGdh0oHhCViVSpYvWZpL04KMagZtTT2njXF+n43JJu43g60yQpghralHdXb4Bb46ZvH40KzhdIp4ZPGMZm0Wuj24zrHopsJUtGT8FivKYVQQOzFHN+RI+aLFKiCN5XUYM/2W3iz7Hh/jznskz+rFL+hknTKqvKCbQY0si761SbTSpk29RPsOAjdxy1UXcVZKWX5iLC/WNKVJ5Rt5xz/cx/DHPsmH/+qnbLr0JlISwGZuCied5umpJZaW6rQ0+lCb4UufuJctuy7Ca84SS76fa5MnRXtHJ51dLuOdrWzYvNp4ua/pa8eZGue6Szex69L1sC3NK3mciaFpYsviL791Hbt2nc/I//gBr/nv13Le1vX0WFU2/N4pc7KufOiG1lZy8Vm++fknmah6tDVk6VnbjkNE36a1tPXkKWQCejetprmlndmjdX7z0y/l8svXMHy4Rioc5ltfeIJxnaPTqjPYP8FrXnE+uy7ZAJdsYOcvlclkXUPyXXteH8WUw9Nofu/WS1m3Os/b/0sfTx2boHdTyCsv28q73/tSM7jL/oktSYKRrxy3lb3+xd//HEcg+d4OHXjS9EfEE7DEKhqdBk7okA604YUdlNmhad+IGAduFnsGcRgRvolKLADGyjBVTu4/Ugt4T4dLtgUjXCi2KNPqEI6ODAjRrPRVEm864efJACe8szXnKeqDGv80+BsVrXJhH4P6UThyWBudxLytaRaX/mHI5xX2EszOwMkBzaZmuL+sWRfbbKFuzF8vsXzGakIJtrjWWIPAZY7FnE7xk6EYe8RCZzQbmhTTi5pSHXZ22mxt0syejlnjF+mOlEHQhfR/VzBF2vLY4rWi4w0crw6YVt5EaT99/mqanEaalY+2sibOyyIw3La9KjZ8mCHj4i4XVYeqnWYirnG+UyCrI07omNjO0u13MVUfxLYLVL1uasG4+Z1lrAyeX2AunOH7S08aeO3Kpu14TobVqZi3BA531mEm1nTZRYaxTHvxepSZqed1jaMSNWTZPFIeY0H3GnsTCWjPihVCETZ3Kx45rRmoajYIf0lQDEcTuooJS7PRUtQEBWkBb1BxLIjZ6qT506aX8eHx201hl/P7xCKXFr+XVqeBffVxMtEisVRZ0o+SIjazkTkhdFTPmJZSVB3A8tqM8jWqnqXmFKjXx8hkNvB0ZLE+jgy/yasn9hxlVxFEYgSrDWdQvBGlvSmh86tcGA00MzYsLMDY2Yj6YjO7m13uL5+m1VtDOi4R6oBxZZFXPvVglryVY85OsVdrMlGJxepJ3tG4jfWux+bQ4s56iZpdwPfa6LCUZNtTskTpGPNYFJvkkV3IpEdTVClmBKVTmkASKkJoEmVxHSZcbYx9Dx7RZuCXOkfER69Vmh/HdfbHIbuVR6vwHFGcjpbosbI8o+EUipuWty101g1CnM/BwUWNrtm8zne5L5xnws6YemwSm4M6ZAqbM/UhJiRcOSzwjlWa627wcZo1kvVqdyvsRWPxgJLM3WLiRxcLkmaDI1XhgiY4nYiS/OuVabXq4STuqtORTF6FXdKmjSpo2HoRObWohLIlxaEUJrIeaccKjaIBY1Uj/WlbULy5ZVPieYgWNU6zorxg8dhTp3nHnWchvZ6eQpETUcBjYcR/tXyudS38jOZsNeb0ksUlmxRbV8Pxg4lQYWenYrii+d4oXJ4yGg9T0F3Ylti4tLcrluY0EyLQEq/MFNijcOSYpmseGjthbA66S2le6mv+bKLMLdkanXYDJ4goBQ5fkizsyOhE6IpgyYH6kqYgySeO0cEwOBlRa1D4RRFniDA8EZQMLGpyc4qmdkn6gI5+xQ9DWCzBTjuxypFQQlGO21qzoCy22mkyWs43R3JQCaoLLKkWMram2K4YOwbPLMT4jsfa3GqGtYvx1TTWaRkiK4VlZRKhl6QoiT2T3UiMjydGfWJ5c+gkcP3PZRh7YUjdOZsWI2C5HTh0hjFxWqTbnEtl87w06mMqMwGV8hR7DxzjTb99G39122/zf9/22zzyjKTjJbfNF2zmJQzzj399N4cO9/OdT9/N2373p8RWCicKqEeaKNYsSr86CKiWAqrVujGxrVGjXNdk8o38+ImTHB4Y59Ce43xPEOf2BkrjAU2+zLEgt76Zj//h0wyOjvPEI0d5WJyRjNI1wax/+NFv872zOd71gWtZ3xMzWwmpVWuUqXL4wEnKS/OoZoeDT51iYnyEsC9HxnD5Yr7z4W9yh3nvdWzoCpmJXFo6OvjWD45zeGCMJ753D6+/9vMMzZRQWYdnHj/J+OyCgeoff/QIM2OD3PXnZ1i3ptkYBHsb8mZ2GNUXuP1LX+TJI/2Gt/fz6Lsvfw2/+PNzOgIrVAkZoEMJUhe2nPYNT+yXrCwvVZ4p3ER0JhPtVyllgsC7lTK+SiUXMrYJZOC8vPB6JLgd9gXQsw5MGLYkKgmSIBk3ShMJ4TpYbtO0w8IDkNmgcFYtz+zryfJnT4LOQ3GHcHyUyaS8qHuZ1Pyo+F/BeEXzzOOaUw9FnHkkYpvJMrWo1hQ9ElekxOvLZ00qcbs/D5u1CtZr6CnCBZ7iwUrMg/WYUk0R2nB8Cdb0wNXXKg5X4fsjEYWuPG9vs3h6aZBqbYDmaImKVmxzGrkpvc5oSyWZgLhOOVogb0lDVhgqsSG9/10csmBrVluWQXnaZN+EZOc0kXMaaJLsTuXhWSnz3qxyaXCK5Owi1WiBa+Il3pu7mOtTvcauRDJLlddFU2odTuU4b515mr8tTVLBoq9bmVbOdglztxyOBJMMBLNIdOZLLYvzrRQb7JRpWz0elDlSm0cVbOzmpPCQcBWZuOfzMFtPBrmoSXPotBjQamMNIe328gL09Chu6lCsCxR3U+J8v4mPrXo1sVWnFIyTD+dZqI/QqzwucVso42LHoWnlqLjCXFzj5sxmtnktpn1tSUtJDFFlAHObiIIpY+AgSuUHolkcO2RvHY7H2sR4vbpR0S5mr5kkk95LawSM627FZBCLQkXQ3IvFPqemWe/BFW19hozp1E4xHi0wryzD91vQkbEfOUzM96MK81GJp8oDvCxX5Io2UX/DJU4C/UmrVq7cT6GNcfGAjumMQ0bjMuvsBlNIzxByd1zlJY5FxYOnwySYXUj8qTQcFN7XVBKSINYbAk6NaBhTigkstiuPEeWwiOJkXKekQxrRHNax+U1OoSU2GVFFiiHtgPD5KjGN2qfHLjJXH+NKO8f5ccCcsjkZ13iydoYJO8+VbouppnS+2Zh1i6ihKjswo03BJZMs8Y+z5EOKJ6QrPDltfrfGK7ANHDmhBIAV6HoIGkXVK/MOEdbskxpBk+8CSUo06lWZwImRdAQV2ZYgi7PaeFPqaY0+qInEz0XW15Rk7zpnLcb3hHzw6/3s/noJ1bCRzZlmHoodrJrmY1mb64qKRSl0ynCDsugMNcdcTaAVA3NJ5Jm0XgXl3ZJKlKhiD3Mghu8OwJGFhG5QjZJ823ue0Tx1v+ahJc3eJph1YGZUMl/Fn87itatcvnyzRXMeflSO2GXbvNRJrEdGbBiQCLJCwo98sKbYE9Q5FExzTCYyKjKT3iOD2kxAqxJfOA37jJVNgtbWpLCUSa6geAE8I+f7cmSZ8DKHdUS3tliNY8QTU0Yt1sC+eJxKrJl4RhFJJrYPGzKKRt/iWOxBOGMmtVKdax1iuxKO7KJsHzvVh5Jugw5Rdo5QIEwaCUNNtSZCigS3MXdepP9eGKdueWPS+xX0bOnEj/mnx0fpbe2i6Jd45et3szAywJPHR3njr9zIo3ue5LKbX0MuHODGW/+Y0sIZLtz9K7zzzbeSX04kyDS2cM2trTz0+Se5/Q/28vT+Ch+7/U3svmoDRx/ai+pby4YWh6P9U7zsLZcyfuAI1fbVXHVhM0/95DAbb9zOxTs3sD41xBf+8nEevneYP/r4y3npNet44s5n6LlyE2t7m+hc04E7uJ+vfekgU+MVNm8rsusVF9KcFXmfjZ2NOHVvP8ePDjI0HNG+uZPrdm0Qsxv+7gPHeMlbNrPmfPj4u/bS8xKfwLfYeskGutvy2LmY/vuS9w4PhxQ3dPO6113KecUxPnPbHh57ZJy3/NlL2XlBL6E/zSd+fQ9bXnseb/iNTXz1j+7n/h8epuFV5/O+976E0sgg/ZMWL79hIyqa520XX0X3Ta/l4k2rzbf+YvfdX6Tz5xer+TeOwIog6Kmn53jXW39K73qHuXg9dbtA3m1CW44JlS4Ir04J3iSIhCarFb052NKqpI6RLGzq4nZvcl/hH/2QD13s0ZaziEQQKU7yoUoyKMXiY0mbi45Uir6EUUtQeRXKgxjOlnjYSYRS7yXgpxSxZCiJ6lU4Uk3Q26gQkXcqrUiNK74+GnCgFnM0tHi0XKUgJk+RImXbjEezFLTLlrSDH4jgI7nYCW9MwrpFxvS+KKRHrE2r0JqFLRsV45Pw6YPQ78FmYu6fOUV/bY7mTB91p5H3+EW2mWNisxBMM6JjMnaW4XAWO7WaV1guWyx4GpthYtJhiUts36iHpSC8rzbKoLLZ6jRyrZM3BObjYqmBplV4M8QMiAWBDlhSPludPE1eB09JhmzlBLauEqTWmnzV9voQT9WnmYlStEcF8pFCBq490SK3Wq6x4BBfOuFYDWjFCIqNpkU4z4nxRV7b00R7p2UEL5U6BKMwUlEmEtCvK9KuxGApBgW1SEG5BnMzUKorehuhJkaq2mJYx7y8pZHVqod7Jh6kbrsGmUvrKo7bypidworLxqbF8Xqoo+nTVZSdYUiEG8pCC7YkMUYizArE8T7E2NC4zZSDMjemcsazS9SiUlmV60IBUCZMXXjdQoTPFxNUqadbIe1EUQUKf62jwyKr03xu6DDasXDtPNrypbFJICpmJ89a5ROKoENEG8EMH/T7SIVZY8Q6Wp/m2/WQrV4Tuj7BSauBVZaf8EfjGj8RM19ihpXiMR0YIUuDcnkgrNCkYwKhMkjBJNYlOpkQnQnFVFYbrUCIxYNxzRgyh8piMi5TEpFLdcD4IDrGrgST5jIRhzSj2Gwl6sqaKGEzinlXG6HAU+GcaTGPEDNZHzdWMSWnQK+TZ4eVw9HjXB5k2Bp52Dlt2qQSCeY0Q+oKMVgDfVYQGzH8FVGAsUVPVOepxDbIqiT2IsqDpVFwF8GTCZrw5AJYlPCVLQo3icBOTOKEPzajcdcq6ie1sVhxpXccKUIXJMMXSQeZsxlbWOJPHp3gE2MhWzt6uMRuZJOGN7s2N+QcLl5lGw7ZweEkM1cMlGdsxRcmA3oFyY8T5fnhRVgnsVtSpAp3MRYUFz6zFPJ0SXN0QhkeqdgmLXTC+ZsVm/osLuyA3g1JNFqbpei7SiFc4PYml8sashSpcnCyxo6sz/kF2N2nyNYU67uUoaVYVWV8BO+NtVFgNxJScBzjPycK1Ual2Leo6V+AkRia5qGhSRnBWZucKlaS/LFBWreivUBEURabLYttCk7qiJRykN/XD+aO8cveGjryLvJ9ZMVBwPBjbZ4Kq5wJqjR5XVTiUhIVJjYmwYzZiDZGxMm1Ribe1do8PR1w5x0HuOXW7T8Xv7p/X0zY8gAWBYGJiLIcy5AUPcm4i0LCKDaxUUFdYqqSiK9qqWSiu3KFBoNGJacyJkzdcRTy+uJiGS+ToZCXtquoQo1bp1k+DCXOyUW2mcRMWUY1KqkAsl1hIM/PLUhCPA3FfBLnVQ2wTCKDeFzZBOUyi+UaGemBmIQKSaswjG4spVmcmydULtm0Zz6Xn/JNlJYEvvuS7iCy9kodSzTbsSRGrKQ1/Mv3ShqEisNkn7wUxULWCB7iOKJarmF7PumUy9LCAtVaSL7YgO9aBoUUANF1HZPSYRp34tG1kjbxbxQPv3j6/59HYCUL+OHHD3LN5Z9h3fkXMrzURZRazWWpPtqxGCLmFmUbhE7yDiV4epWosQqKnA/9MsuOMZy043XNobGQzusUb7vANeRmmVEreVMWwgmd8G2GYO7rGmcXFM5TJieWo4m3lBUL905RPqUpblWoVRAd19gbRZWhTd4nGxKTU2tJc2QY4jnN0oTiK4sR+6gyis0G6e4EM3xn4af8QfFa3t3SSnaZI1YJ4PASjFhQlzZdlPgqDgV1LmpwTVvjzJzD95wSm1yP3a5DtnGGb44N85Noo1FhvsFWXKA1Z3G5q3qGH5YO0GRnCYJ5FlO9vD+zlQnLMTFSYnj6g2CGG50sv+Rk2BfW+FB9nG4rRZOd49VWmlliw1PsFHRPa1N4HYrmmAim2ZpagxXXOBiXabDSNNeG6a8cIyuiCqfZOMS36MgUVS/xOnhZposcaR6Iy5yMYxaDKUK3kRvsgpnprxHBgnIQhOnp2jP8XnMLv9y2muaNMbNnQc0LsRvKDpyY0Wxao4xCc/EsHKtqujsTtWyzqBwLwjtK1IQiqJyINUXHZV/5JB+e+DKLktclUkivm1xmC5F4BAZCg4lxnBZKwqZXLqn6mBFYGDjIiN4yptiLK8eMOnhH08vZq2M+knLYqhrpDzS9qcR3rF1axK4go8pEda3txMRbzS7B3VNizAzNKWU8004t1njnwP2cSafIiXeXIYplEkTQKRCJ2lnXGF06xq9ls/z2qovYNx7RmXKZqI3yK3NnOS+zhnpU4nyvA0e5zOiQe8NpLrN8Ouwcg3HdpInI/YJWZHSdXcqi1fboVRoBP/oFnVPaGBFLzTOhIzqUWM4EPF45yZyO2OE00Gvn+a7kocUBXXaKbqeR1yjLnCPTWtFtic0MHAxgrQwd4nOn4Uj5Sb5Rdml3MqZFXPZ7+XU7RZ6YfwoD1qpTfKFnDZn5LE07YoMeVichdVFSJBrCrHDkhQMnMP0cxJMaa0eSUxVVYGlQ425MFLIT98fEh6BlSyK4MfYaZ6B1t8IXyqlM7mR9QqYSD7oFqO9LijvVKoWecPg0Vl1RG7T40eg8n957nG8Ha7jCyXGl1tS0yyYPLi/CqNBBs4qy2MDMa+NXmI4wvnc/rEaIruENnm2SJCphEnc262vuEMWprelMKXISxThp0dNpG9X25iZFQ9vydUiGaylp8xAKWin8Qiu5Hgmn0BnWzI0pPv5AlW+4EZ9rzbBJcmDLMCeqU8mizsPZqZj3LwRsDT3W2JrNHowFypgSi6Jccpdt4QZbcL20XtsVkzOaoXkZw8WAWLMvjNlt2SZR4lCciC3Oam0Q2rNovhIs0rH4IF/puY7VTgMjlSjxNLQ1parL98tDfL56jK7cBcwEU0aEJWBRXD1jUDtRv+pQApSlpnHRtX5WNy1y6sgTPLrnL9l52YXP5sa/WCPZC+bUPbtBrbFdF+FCn3uTOCzjuiznlZ+oOqQFlcpmDcosy66gF1EY86m/uINTR8fIN2dwXMkjjaiWQuO87oghl6i0TAGliOoxlpTGWmYfGnldR5pQHCVtm5QwkWNN1Vi3S1SZRSyGksv/nJRjMk5DIYlIIyyMDXdJ/hMyqJ8WyrFGCkip9cJahHJsXOGV1CNjEOqmLLRpcwnnYdl25F95byChfZa1vE8xVbHalhPYtnB9yaCVojUmlXWNnL1ekfayNq95nsXg4Tl+7f/azeXXnGc+r3nzL/77z3cEVogNA3LHMdboVTTbsRBatXDmduDQJYo9CTWX886CVAQpMfdsgFaxGJHXljR7Ys1Crs47O9LML0BuUZNZr6gLf0a85PLKkLDjGU32anC3KWOJIC0X7WjchoSjU53ROGvBXpuk2VgblGnr6FC81LQhMwsCeHYeDh3TXFpQHKxqVrkWl6ksTwfSpgr4TjiPbeW4t3KK8pzN6xqLxrC+V5ZbLUgTDE7JjNmi3465tsFjzbzCtyzGcjBY1ew2CtYUvWELd4QiY5tju9OBuAYdVXLUIq7z2ugvK45GVXJO3hQoS5ktCD/bRjOsAyInzwEUr1aRaQsLp6wLzRPhHFtdxY2Wb47z3jjgdhFPSAM3mGWntGTNbzNrlLVWMEuj18y0dw0qDihVjhkj3SC9js444MH6EA8uLPG63HquzGSYr9bpifOcFm81LBYkq1lm6aadp8hR5HfPVEmrkItii65FEJFIRjIu8+AMQv+AZnW7wm+HHZK1Ktf/GRiqawqLGNuQLdsVM8MwOglLbo0bCuuJ+GX+ePwrFEUdHIcEkijhdWD7vcRRiaA6YApTmQiH0haSgUWIMmKx4LZhSTEoaFr1FNPBJJelt/Bf5/bxuXyKXjvD8WqERHrJFdOOJNs04XatrihqNgxWtGmHdUn3qa7ptmPcbJoPblnDr+5/hGJzq7HokdSL0Bb+Yw1LBZSl7xgH7CquYUwrrLzEVEUEWswSjzAezLDotbBdwGNdp8sgHXUiq5GhOGBfXGW7lTJcv/26zm/aPqG2KFqaVb609bUxD56MhM6gkTD5GWUzruuE0YJpS+M0mMFXWvm7gho1O8ssilxc54Dy2C7B7kr4dXA6TCZbD5RgtbQaXcV8XSqnCrbVSLfbZo7vpI4MHaA9miftFamJCqgvpqWO4SKmzheFuSaeUpiTUcZPOfnkNy4J81LshxgAQdBc4bG2CAIyrZmZgGwBkxIRVMDtgvQk2HOauiCAXYpQ2q4yCZDjNgT1ZvBbQNL3dE5jLSjCWfjqUD/vfGAMz1sHmSJXaM3ljoVkDPfFyogfxIJnMaNZ7UFfEaYXMXYcwgMtapsxS3NbZZo25bHFzbGpHYIq/GaDoqVZkU9BLu0xPqBZGIeGToUnBeKYFIbgdmPi6IRqkJMYOhdGHtTY+6F5uzIuIePHNTcVfS5VMV+cjXltLTEXnkv9T/beA0qy9CzTfP7/mvAR6V1lZfnqMl3VXuqWabVaahnQMIBABgZGCITb3UF7hjMMs8vMMDvDGZY1nJllDgsrJKTBDhIghEC2u6U2alNtqqrLm8xKbyJN+Gv/Pd9/M6VGCBBCcLR7Ks/pzsqMjIgb90b897vf977PK9rfrNu/7ktBFtPAZ9EY3u443CbGrgE4awxb57O1tA5spPDstOFN/YqlnEF8WIlS7FWaL6US6QZv1pk7X4QJa8bwtBiRnDxLJsefhQ3e6/Yx7GpCZfAkO7ZqeKiU8sEr4qifEYgQY2nPjl9n3QGcYDqLKRSkidF2/EpcIZCoEGrcuKi4956sa/jNPMH97Ys6a6HJ1gcr6Plrtkb+dEcPJi3/nRGidLp+7F+9zcYc2btbAdLLgL3bjyndqp3OnnyXr53f7fxbvr/8OezP8r/tv5Xv9r7yHNvbbrtg2Z/Y/2f3l85d9suv+Rwv+3v5585jfK37yu329/KaX3a/negz+/g7oqvtv9l5fbKxMuK2Xzv7+mWPcfOf3/p7QA6tVprmeszv/LKYiWq0JNndnyTxx7lPILLGIJNTMbc9l8LvGsN/rilGQ+leQyuCNCfib4UTG35pM+Q3D+aZPOTYK8+4nuX0RBIH5GNTITrPZ+O62oPKOlbTq9YvgN6tEL2u6HlkDGpuz04sSuKf5EpW3pCSItEP6YwhqsH0ecGDKDYlDymF4RSeIbYLl45WKXoDdJwy+7XDLlPiD9dTq1Ha5SXsLmkOq5QZAj7QvMILbsIv6z7u33OAeTfh/zi3wdtL8BOFMktBwoDr8Y6BMp9Z6bJHOyRpQlspyyNbVTkGvGHozaH9YXpJm2eDWd5WOIiW1ADl2ZO3aND/W9RhnytuvJSOqM91gaOuw90FxamO4ZjK834n4RMm4Yp2KZmYQeXxJZNySFyZwMPhMhNOjZHcJAtErLRfpNV6HlO4hYHCYZykzR80T1MPSjiSTuHtpSSYERNwzHaWjO1SHnZDzkcF+std/sNmh5/dqHDXQMrYrQpvGnKimaooNhqGvDEcPqDIr8DSAowOw2BV8cfzhilBywxAU4uI23CmC99eiHlF5TA/ybv5L0u/zajvUg9mUMKmcMYpewN0ozqRhIybMLuYlmJKGG0i4E42BZ1pOwcCI15qPEo1vw9yY1xKtrjPKVi4q1wDX48Me1wYGsmcu7k81DcNux3FXSXDrHSVElgP4B7pxHgSSeBipGshRaPbZ9M8JKdOzDpbwSyvzQ+hOv2Mpgn7iorYVzaZQWZiJSdPkPT4Apqazlsen5uGKE/TSlvcKp1HnWc+7TGKZCXLRVLKaqo4GxseSWGfjLkQfZxcPiV0TcJU2uFD3f93eS4AACAASURBVCsc84a5Mz9lUSiPJ11eoVxrcvlSvMFBt0INh7oFDis7ipZGWsMY/ish79AuN2LDE5TsqE/yf+Xi4R4lZomEcUf2cZ73jqVMFRziPQmsQPsybF2B6gOSHGFQciG2rW+Tto9o0iSUVmDUsijkJTZsb7bOpU1jM1oH3exzLky7LdHO1jLnqmhVZb9K5yR62mA5dSMij1A2NcI0DG6fQ2sz5QNPnOf9l7oMlsdpez7fCbypoKhgeEWfZnHdsNSGu44oNlaMXYc2HDgfQ7lmmcb86UZEXbV4zCT8XMnjXccVe2WJ6MH0DWgIYFscyLfLewGLNxmsQiS1vFRXFZi/Cvlxw+AehZmFqGPsxYycLLvSoZQEvNhwuQv392v6D8KvvBjw/UOGA37eagclM/jJNYduWuSzxvDdYjByDVFPUerCK48q+hrGOv0lS/Zpx7CcGspNTU40itsFXIxiVWwSSlvYt3TpzivDWhpwVGIBTcjncFnrXKdRHuJGpNgzIBmKHpNlKf6KcFGKvAhfS+6xgMFTi35R7iANXULF67bpJKPY1MRs2UZPkV/6+S/yhrfewuBQLvuMfpPO93/7om7nnPryamXnd1/j+04h99U3STj8za+be+D/n3tAVllFoDr8vjlnV7LIimfzVE3Cukmsjq6QZtofWcyEgC8LvZgABgqKM11jbfNR0TDdhffsdrnntgzKmZRBi8u1Bb6AloQuv5B12Ryh0udAJFNSDCBFnAizmxCUFavzht3SvZMp3QZ2BCskeiPibDlBONBZNQy1FHEB/nDLcCCn7PRgyji8aEL+TOe536mybCIK4Rz3V3dxKkitWD6WLtuc4vWykMXXrJXPjz3ev7jK1N4prqx1IQq4NzdAVUtWq2FsnyEpFCh+tIdfjXghhdvQdvwlI6A35w/SiNY46/ZTkcSJYI793gCvcAfYbYwdo4gj8gmT8mLYpaTytmsyLJBh5WOi1JoZZJziiytWulZJm+u5KfaZmPuVFAewqnMc9UY5H62xL6pz0h3ms+V76ASzEDdwlEuqfPpMysOdRbuzTuYTxt0hzukcUwruVD5bJuLZSGDLFTaNZjCeplg7RBDmyM1DNJpFHy0+a2ivwmIFDobQ6UJlLEvxELdpn7hNHbh80XBor+LNuxWvqWs+0Y05olJem7+F7si7+eDqf2VA2H7RsHX0BrKd3kDGypIYLeUTJE3ieAW1HQRsxAopukKBGEcrnA2XqLg1/mO4zPfoKt/jFmklhl2uYrUDjVXD7rKyo6uggw2fL8sIOYEhNzMhXFmN8dxh3l3p43faNxgq7qcZO7iex3rSpiRGDdPlePFVNqKr3k1ZWIOLOuYR0Wo6ZRrhMo3cFMe0THsMX0oDJrwBWkkbP2lxyJ+wjlUxPLxOu9QxOErxlACDe1ZeihbyMJqH4y4TMppVOdZMi0FxC/tjNoVDXL1b0TI5f4LLUpE4FQuLHtKKUybliol4QPiLUsiZhIMozidtUp3jzfldvBiusBRvMqgDxnO7EG3sn0ct0kLAcUcYeymzTxt6g4q+47B0GU6fMdyzDwYmVdYpWjBSg2csliuG4CyY3Yq8mB7WDWpE4eXFKGBYa2eT9rLOQLuNNvSPbEOKJUGiCvoeBdeNeAbo1g3lIdADmvOLLX7/CzP827mY2wf2c6cuM6oMd2pY7oFflc66oZvAngJcXjZMJYolx/CSTA6s/EixoFP2kvIup8I75JgbGTXL5E5ZfMnwpHxMMr2ZTP6FZzh2JLuwlMleaRcEy+CsQNyDuQuGsAhjBcgNKsoDxnYckxJsiJE7hJZveMWgZvwunz86EyK5cScLinJVsRUY24n8sYIidFJqfRpP2HaS/PCYYWhScb1ruEUmIBOKhy8nfDHS7HUcuuLAJrXw4Ura4LO9Lf483mRaFmFZBHs3WCFhXsR34SIfW0n5YNO1Mi8bU9FzeNDvZ1z60XqN5WSAvNNvTVIL0RqHdYmUBnXh030ZeO2j3RFMKqIRj3DoRZQIni0Q5Zt3JvxbGSW+eU9785H+yj2Q1QN/5c03b/j/xh6Qi5mNtZg/+OCTrLUUfnWKnDtCRzhgKmfD5wXYKYiq80nCqx1FLlFSZ1mxt7CXujItixSXNiMKd8R87x2C8gAlVi2Ziw3K1btCiytTRjdDkD+gLGVeXHdIJ64AehzbBWidNvRGYPgWZcPkZe2yIyCzLfkQIf+WoXUavDx4ITR9Y8HH/RIOnkBDCYzU55jMTIDnkpYFex7JldjtGTZVwi8mgcVMVJIOlcJe2x3Kx1f48BJ8pg21Qo33FHKM+xLArejI+Ljl0+w22Qjy7Hd8TrhYjlQnNby2kqcVdHg6XKMvP04nWmchXObO3JR1nYmf7AEPhhyfT3aXmUhahMbwNpXiqBIzcWpHLdJBFXfx+XiL40mXB/JTthCTpkleKSITWrSM0gWuJA1cE3LArXDEGyfnVe1zOtE8kVOl5A3gmx6z4QKe8tkwMaNOFYnVFaKMaHWPaofTJqGappyL4Q1+kWbd0KtCuQMD6zBQVkiwjOAqKkOiLRK2l2JZOAs14QsqQhF++8rGXW12su7YBa2425WO2RC/Z1y6zUdtdm2QtIhSKWwd6/gN0h6+P4rRedKkjulNgwkwact27MSRJx01Lw3wvVF6JrAQ4/udMntFF2CE06aYbsKydJTE8DIgcXOw2FB8UQkzDOvYnuwHP3W4FPV4Opwj7wwQJhsY7ZPTBYswES3cq8snOeEZO9oUlu+WxHcpOBvX8VLRQ/cxrVxWpCOsfPJujetJ00ak9akC50yPnGBtlLYGIh9lWY8yMt0tAQupsu/BxbRHTMp5YnY5FZaVz1zS4SWd50LSYCRaJXQKPINDnxbXsphoDDk0m5LOgeFxHB4xIYdJud8t8U+0vPcTnupcoW5S+v0RAu3RJeZz3VXeW2jw1sExYs9QPaqYuWDwK4q+A4pBD2qS0CId8xooadOL6Xcry2KVQihOIXdAJjwKPQjpVcOFlwx7xhWhuK7XMt6jqHxk/fAPZkgUaTOJ+10+o3qvsuuKWnF4urfMHb99nUd6Zd4+dIB3egVeqxSHXXmtsJWIOQpmO5IrDbmcIugYPtmKGfe1jTHzhbOI4lVVzYGcw5TorMrwYysJd2iH40c1LTH3NLCZ0/I2KYxBLYWeMDFFw7YJBdGcxNnfFPdBWsB2BJsRtqNWccArwmpD3OAwPgBeDiTMo1DR3HabyyevpcxGhltHNe1Vw7C8Pz2JMtQM9RTDo+CVoTCp2JyBhzdhKFEcDl2Oll0WWl0+3tziN5tnWQovs5Qs8sftFd45FPKGss+xns+3+3nudIc570/wQH4feW+IGePyf91xF/+yf4Dv6JV415TiTVMK0+7wmc2rVKPp7IIk3iKPw4bO09BF3HjDvk/lgyRQdrk6dtINonZAoeTwfe96kL6aCAu/ecy6m0Wd3Z3fQv/7Ojug30JbfHNTvmoPZF10xe/85gwf+vBzjO/bRS8eoJkb47vcGsclQxS4qwazMXw0gfu1ouBKcHWWIlH2soijCzH8TNrlZ8s+B8sOZtf2uEXIQqIZ7Slc0ZF0FXrfto5uEZxRUBJRcTUbn1qx9BKU5cQr6nY/iwMTkbbo2cXVJUiFIFF4DSiJ7q1puNQ17BEzhoLH44gODvc7vu1qeJJUgOL/CTf43nwfz/RSfiSMCUzMgHRanBK+SdhMGlZHtaswwZZOeEAlLCUFNhLF3mFlhd2yrvUGC3xpU7PfiBvY0DKK1cQw4LqMlOGjW/MMe0Ok/jCNcJE9Ok/bKVnXq4smNg7PmggTLlGX0bBTZlTnGMBQNgpXKWQYfjreZMofsyL8JwUwICwyNCtJk7PKpSajSnuMDGtJh9u8KlWnREu6OdsRXUnaJTRdxN+5lHZxwkVOk3KHP8iIchgV966RAHTHFoafj9ucKIsY3afcMTZDtyhdnL5MOntBoLQlxYBo7oQ3KOki8r1ruByIMTLjfUmH4nC/5ksNw1yaMJHT3JMf41oUsNB5jqpstxQ8MlmXcHGBwsgY1q0g6b9GuBhGWdRCJiyWlyqtX1c2wIKVa06JXBpzUOXoCPJBIpVUZpBYEeBrLL0FhfINn+5FnDSO1VOuxIb+RDFQKPJU8wWWkoS8W7bdzSScR8fr7K+8gpoW/aUwA5VNGxE8zmmbILJko+Bcp8pg2qXTfgmJeKsTsUd59DkVXkq7Fhp8WPvUlOE7PJ2ZI4Appayh5UWJJ4s3bQrCMdFFaZeTFhwt0VA5jmG4FC6xR+c47I+xKCgL5dhUFWGjNcUdnPYsv06AxG/QioqSLFrH6sPXTcQXuzM0cpP2fbZXOcybkIWgzvsnd1sAt3D2ix60I8XKDWNBzrVdijQwxGuKeNbQXYF8VREuGHRf9pmUeCsxNITzhvS6Da+g0VRsrhhyXWW7actrhtqgYlC660LLWDIk1yUtQiFyLfe6ou5pfuvULD/469MEew7xvsIodzmG11UVwsU+kFcsiUPdwKijWO7Cagy/Jagwx1CbTLgz73JbSbFHZBht8MswNqXIN7GayyXJbt7UnNwHEyPZiLkr2jkpLHvGFmahVlRL0NFZolaxoKysxJF0iSb0BYqwrVgTo4hSNPZCtYXVc84YGDGKgkwW5PkFXeMqPr0o89mIUs4l3xHZRUjoK/7xuMbNGVbairSjCQLDtBgmnID59iK/u9Xlsq7z2qlpXlMs8ApzgJNmgv9+YIIHirvZ5YzhOSMMeEMU3SEqbj8jctHj9XGh3eR/vq2PV/cPsneoypGBYQaTIQrd3TzmHWBdYvAEtxQuEHQv2mtut7AHLZ/BROA50hjfwpgecdRmbDhg7soSo7v38pr7pr4sTfuqU8k39OPNou4b2m0373RzD/x1eyCDCP/Rpz7FFx+5Tm3kCOveEfYrh1t0nt1aWzZXp5d14wZ8xbGiotkBuSoeqcGWaOpk/AYciT2++x6P3OHtGCARV0t3R6JrGmJ4UDbv0YpCxFEnHRZpX7WkI6CkDYWJFNemoXwow5VIZ+/Ljjl5LJF9TBvMTCZcDrbgo2twWLJdc4qLAXwyNbxOOVYOJEWGaHE6Tg7PhDzeusyTwpETTlPSZTFaxVUFa5QQWnwbh9tzw9Zt952FHK+p5Tmyx9BchKcXYbUt2JaU/7S5wBZFppVnT2C3lWHQSxl2ylzvbPJ8vMEuXeJobpw/DWZInBpv9gqcTQ0vGc0tRJy1brOAO91+Bpy87d6kFujh8Jm4yQsyTvP7CE3AG5XPbsdlv2t4JFFMuSUK0t9RUq5K4VHjmaRldW/3SPfKHeCaW0MFs6ThKsatkFN52mmHsWCeTbfGsPa5RfkWTtu2jyOSxYjHO1sc1RUmhzTtFGbFJbyWdUlnQ2wKQ4nsxD1chblmdmyNC4tdGPUUgQcHboE+wUasCbQY3uR57PIPspC0mOmco+IULcTXOFVMso5OQzxvGCPFmxGT2abVuant4tVYFEOZVLp0JmFNF3ksuMZD3gAjOivg5W3U5yirrRKjxPq6jEvFUWloac1UQRIyFO3EMNLvcT3pcWnzCgWJlxQThwkIdIFD5ZOUlWM7RJVYsaokxcHwAlj9Upq00CS4SZOu9GFzk+ikxVC8hXL7mCWxb++89nmlI+NWxaZkjFpZWcqZpMUn0ogHnDIl7fEWyZFVPk+ahJrSnDIxJ5SyaJUzbh9zOk9JO/TJ+1m6giagm4ZsoZhOWrScHCd0iW8vOBS14fFIUVCGZ4MljFvmgETXKRBn+XFnkXeNjXGj7eI3jEWP1EYVYwOZ67V1HZqSsFBSNJZA4r8kwsp2aAvKmmhW1sDpZnnwThH0mGJ2yVgMiuTdSnLMUiDFmKJ/TNFdNIivUCLCXMlbXZICM+BfP3mWn39hmeTAPt7s9HOPUowmgvRRSGTsSzFck2ZwAjMOPBOlSOrCbb7mWEHx0AGXchkW58UoYdg9qRgXM5YIthxYaxtbsI6nGY/u9nGN1zOUKoqhqtwO565nGt7D+5U93usXDYNHMvySKwk4wsxzpAsHJU8Rd+DCtLEYG8E7razAkWFFXrqawyBTS8lyHZ/U/PL1NpfaqeS72Iizo6lmd9Ghte7w0kLImbUuj8XL/Pz6WVarMQ+Me9xZ9fm+kQn2laa4J5rgvtsr7BkuEM55XLUZ3Ak1L2W6l9IzCX9kYqvXXJCLheVL/MhtRSZLA7TiiLhfcAUpQVvxfCfiJaR7PUmxdIK93hil9gvocF6kyjaSTxZsiSZUKm9XopLXo7Vxg/vuHeL1r7tzR+7/151Uvu7bbhZ1X/euuvmHN/fA178HZKF/7LEXeeThiNroXhpqlAm3xDEtIeTahn9nyh9jr/4n4uxkPlLJeG5SlAm4+COtiDcdUhyIFXo0Qx7I+FXtlWJA2fxPEUM749uFncw+pBKUDpx080RWJry6LcgLQf1VCkecsPIYRRkBZVmQ6Q1D/Sw2XFw4T5c2DEsu3JbLrv6fiLJx4n4bI5VR3QU+e8hRXE27/HmasqQL1JMtimnP2v8LXj+pJe9rnHiRS4zwfWN57i3mqGwYJg4oiloxYRST+6WZ57BGm0+0EwZUwpSJKOFzSw4OVx2CNM+frb7AWH6I4dwks1Gd1bjOG/MTNs98jZhQGHRy5tFFK2K/xSlb/YyMtcWJezHe4ns8n1d6ZZv1KHFn4vKU6CHBYLyYiGnCpaZ8C5aVuCTpfi2G87YYGdIFSSnlSm4X48pjK5y14d2F/CRNYmY7l+12bboVjFO0HbNXaIe6iXgpjjhU0typ8uyqSeSbYrMLyRAcMooVkXYlkqMLTXE1KoEUGztaF/6yxBSJeSEIlB0j3641L3Xk71JWpTDN77OomfPt5/HtPL5ku3AWq2DfuoLIkaymdbC4Ex+d22V1dSbZsAVUFM7T5/bRkZOUgkkpDCX/1FWsSzbvgKLdheUOlCckGcjhmYZcnKQMKUVdGyrGY1h3+WT9efrdsi3O4mgDUzhA0Rtkl3JtysWdCmbTlN80IXdrnz3aI+5dtzFiPWESujVip2Ljwyr+MAvWORsy5RSYRdsYqfXEwVMpZRNy0QR8wW5tyoB08uT9KNc3JrXj1Bckwkl5uCZgKVyi649QjBvscYp2nz+ctgjTDjpctqkAJ9wKD+kyAwJlManNmFU4nEranAumybn9tCWD18lxurvIj5X6uHXXIEUB23UUpRH5LwMES+6quNsFgFsqYceVrg+bW5Ary3HPYgIlvUK4+eJm9SYgWYWLcxn3URBgYnBRYrIZgQnhHLch6Bnyt2g6Kw6PPrXFL175Ah9ZrfDt/Ye41Rnku8R0hLLYEOm0NlKYlCQGYbQYeDxMubeiefuwY926AzGsdRWzy9jCLj8KvU15/yk72qwW5HZNNZBss5Sz7YRkRQw1moI4W+V1thWHxoWrJykYWRZsPq+It6DUh2Uchi0o9gt2SaFbkJMxRSRyV8WlDix7cGtfVsjKZEO0gq0KjEk3eNnjZ6KQUe1aZJFnNNebdb6wtsAjfptgbJXbkwHeXdnFg6UB/tHBYQ7XSvYqurMAsZ/ihin9acpyKBMJ6CTidlUWZr1ps7gVwmyW9JaVcJ17h8qcrA4wd8nQdRXDVUWSKp6bm+GUaVut7bA3Sje3izGLGNqk3nwSV7kob8gmuhiJ9Ut7lPyI1sYir3/DUV7/OvF6f/O+bhZ137x9efORbu4BO9GSgq5xNeWj/+5ZTgUSE3cbQ6rAiFO1IfTSjZF1WaTgctIRbUufURaI2ethdbiJUTzRjlgebvCek2WGRAcnYGE50zoZqkAPZIZuEVNLgSiSDcmUtBVOUSj0xiIY9AYsnDWYSahKcLzgS2QBzVt5lTVRNC5DIxZNFMzNGK534IF+hSsA13bKmkq5x9FIot9gKesUiJD+Y0nMHxvDW/wxDkqaAg5zynBLbpJ7lEvFJDwTzjOYbLC/cAs/6Gubgzp0UNG+IAWKYfc9iv5hyVNU3D1SZnY+ZiaNuaFTlhIB5Ho2Lm2wz+F0Z42XIo9Zr2w5Y8VwgXNpTM4dYBXF02mHo06Zogk4lTQ56vbxUEmcm4qXkoQ/VJofrRSIYm3hpSIYf5USTIEkD6S4RrOcdrmUdllUOd6qPY45eS7oMldNxJm0yyslYkrnmdN+5rJNuwRJA9+pUUgD6pIyEMwyQcKcU2ZFuFWk3O7XeKS1wTguBzyfdstQcjPzQVOKNum+DAgOAtbrMvGWgj/rsornQ3hkko9eFC3eASW+AjEFc8aeiEQ/mGcwfxA/XqXePU8qhDDJncztIo3knZZa5IkSB2xcB2tGENRCARMt2E6YUh5RtGyLn9PRJu/wR0iUy5prGHIFvGxYCrPouesb0HIMzwTiANWWPVZ1FfUebEQun4tmyUm3K2lz2O1nuHSCCzZSy1gQ8T0C0tXKZrbOxyGiwawHN6ijMum4N4hOGraDV/eGyUUrBNqnX5dopD1bVAsg+KotTlImdY7HgHGT8JIRzIvDS6ZHldR26aK0YzNNl3vTNJwyqejqSNnnDVpN5boJqMcta77YUiV+yqvxjqLiaoLtrGqtOJ7XXOzWOROt0p/fj699rvfWeEepzo8MHmSwphDDwEZbdF6KzjS2u1YbFgZpBm5ur2HH7l4JdBsWLmLH0IJBLRxWNGYyjppEyLWXsBFbW50sYk4CCdqp5PzC+JRCV8FZ1mxcMPzqc9d4z/ULnIsnOF45zLt0idvsciE9Z3jOYC++JJ7wXBTTcA37I4dXlzR3D2rkLTIypZDosvllQ62kxMxLIVW4TcVWz+AGSohdNoZNNLe+p7gaa3622+F4y3A45+P2GwswL/YJ4NyGTlGeEtAy9DYgVwHRBG6dy6YCS5dALlqcHCy1IJDXJ9FmBpY3YWBMMTgO8aKhOaeoSGErzuYe/Hpc52z3GjPBCtIDu3tI87bRMb53fBf7KZB384xoxxoptGcQf5NTEoOaYrWlkE7qhK8sW+8psRErxf6CuLFhwBHcScI5k+C4Lp9ZX+L7C8OMj7lUJYJtS2LzNE83lnii22XM78P3+rlhQvbpHKP5fVxxB/A6Z0iSLbRU6YI8SjZJ4y5hb4XF5d285aE7GBzUXz53/F1PozeLur/rHrx5/5t74GV7QFA2YpJ47vw0P/kLn2DfwREWOMl9/gRR0mCf8pnQjgVYSkdG3HYdoxAnoUQS5YX8PqBYT+Ajyyk//MY8993lWRu/LGUWnC/GBpHKLWGvkJ1BheUvyAxKeLM5bN6kQEgF+2FSxcojkL8VSnsUqUTdbFNzRPiu1qC3AHnRrlSxXCyJKBSIaKQMGx3FqDAZHWUp9dKBEOzAb7UjHiHhrcKBE8G8dAPcIkcdiX5yGEKC22Ouhkvc7VR5gz+CGPSGhhVliZBegcK4oiQ4BnG7jSpqRcWLV5aYjSs86NXYI6Oe0GW2nfAngeJCpOl2L5DoIgfdGnf7I3y6c97qYF4lHZykbdMURpVPg5Qv6gKvRdsYsd+NA/JGsy/JEStDQYpvA3u15CrBDRlN4hIIizlpE5NYTdxR5fKActktcGKk4OxxUrnscor4TonA7eeY28dyOE8ULVnDQc6tcC7aYDFe56pyeb3bZ5EnCyrm1zrLVHpVTo46pKHhoGAlfMUlSeWoYnliEpPdP6yIXeirgtcRM4y4+aAgAnOJn+qCmCfHHai54pZNGfTy9OUOci5cpNE9R14XiXUOrfNfPuiSAYvgFYIZK9C06BETY6IVtL+HJNmwRZW4aOW+iapRLGRRZ+urmRlBTrg6UbzYjHmE2BbU/ZJ8IUgRx1DQRS6GdW4EV8mlbaLiMaq53TY67FZgUuftCHBdNGfGZTkJWEo7bMQbdKRr6NQs7VcKW2k9e5IDKuQP7eMrj4pSDKqEZ+I2z5Iy5pRYiNZp6BxDTpEj2sVXmoG0S92EdAU3IaPltMuWOLMFt6I82k7eaisvJS2CeMsWeQ/mJnh/tZ9btGEuskle7FWKXTKSNYobwRKX0pDydhdxpTPPDw3t5YSM5bspL25ALRReahbdtesILIshoglVGV0vQb1tGOhXbJTE9G6IFxXNRtbBWm4Ym+faFXZdD84k8KR0bLuGjR68ZAxHJJxezFA9h4WG4QPXTvFzczPs7j/I3cU9nMC1jEUR598wimaGsaSshD8I9w+7vG3YIXFhcFjhDMD5OZPl/TYznakY693UkncQRGtXOpYSxRbD9KqxLMkT/bAvEcNKwr/aqvM9R0vsrmjWZcZbgKLo/nxF97IhX8xGriLflHVKLkoiAS9L9q3ErCqsBEWKP+PBGZk6BDAYiaZT0a4LwgRuhCl/1pjnd4Mu3x3fYNRp8nODR3jH3nFefWSQ4dgl7RiW1uFyI+XIEWWTYCS3uOVBbhhEsyzkMDEphdJZTAxHPc3aBjwdQz01PJn2bHLJHsdnOg446l3in57YzUBfgaRlEDJJZwOeC0Mei/OUZG1Qnj2eR5SH4FKuOWV80ft2zpPGayhX8AQJUdRk/0SPK+cv8D3vuJ09U0PfNKzJzaLuZSfkm/+8uQf+rntgp6ibmV/ggx94hMGx+wjUPkSXUVQ57nI8i9eQhfklDQ+VNZM5RSWUaJpsYrqvonhyPWVhV8pPH/UpyuxQumqCYJMrTSnUxB7bA++gUIszkrx1WQibI8BCXXVOQsMVMm4QBMqQgC4nxF6ZjWftGGfdkDYhnc26fzIqWYsVXiQh9bApBokunMhpu+iKnK/egEfi7Gr7dZLkoh07RvutNKBPOUwqF20MPQUSqbSVtCgXd3F/q8Ddr1YM3y6CcXA6WIegkoXfy0C0qqLQ1/P8r1ubvMMpsFe7jMko0NU81OfwyrTGpfYc6/Esy8q16QLjXj8qnOOYN8Z1E3AZTUn7jAqDTecpigtPiebO8P6iy7ERzYVuaqGwNSkYHHjtgLKL9JboC5V0I/IcdIpIGrPonzeTZgAAIABJREFUYqQVKpqmvMojoNk/S1oWhSJFz6pkyPrjrPkjrMvjdc5a4G45N2ldpWHvGnHS47JbpuRWqZqIRSKmkiL1QGDtondXrGtsrFE9lrFn5oYtDEFJdFcOlp+21sjYXwOSdLAOY1XFoT7F9TZcTcUskNCfFjlSPkw9XGC+fZq8kpJI2rsyfm2j3SpK522UkfD1BNci+jqTNGzShFI5krhO1R3keTS3+kPcl8JVgaS7ij06g7m2hfmuNEtpxKMm4TuVh+DVrht4VqANOke7+RSb2ieq3EvRrSAD6WXJ8hJ3a6KsFs1VDnXRGxHR6Z6nmQZ4uUmMtEJMsh29JBDtqt1P7WTD5q2KwzfVBfqU5sW4wXkkGSFkXIwxKJ4XfZ7SlIUnaCL6ZbybbLEq+bLyHnWkt6NpmgAVr1mDz33+OO/0aoy7KS8JVy+CQ7lMM3cjVaylhsebT7Pu9uH7E8wEN6yE4Ae9Q7a4luMj4G7RGE5IdyrEOpk3Lhv6quL0FlgzdLvKaudGxmFI8ByrWA3pF7uGQUeTbjhsdrTtaJ0P4PNGc7t2aCaKG65iV8nhxobDiwvz/Py1R/hQp8Hdg3fScAatJtARxI1SnDFdLsRrPBNtWdPQ65wcw65in6+Yb8KKq6glho2rsF7Gjnd7W1CRCx1xqgsWKY9lu4njWYo7Md4/uQkLwq+Ti0LXUE88uiZmOe3x0HCRXJbaZpfTYGlbJyiwZSnm8nLhmo2bxRC0Lo+RgPgfrsna5GVoGkkHuaWsqLkul1spz640+C/1y/xqo007cfmBcoEfHzuEcsYo+SUODSu6QWIzYMVANjCa4U2K1WwErIegMqzw+xRy0ZSvgO7C5iLI516HMO5jL1hVqhhXhvpOBKBKyTfn+MHJKcrDeXu17IqRaZfi6eWQL6y3KKbzaJXDcUq0TWovCMU0Fgsm3RvEhDNW46ocWXVCyu4Wm+sLvO99b2D37tGbRd3f9eR78/4398Dfxx7YKerm5tb4jQ88zMD4KwnMCFvK4TieHWEFSvF4qnh3VVEI4VQAFYmIA2YNRE3Fw/WY/+6tmhM1h7htULsVyTWTgUYrirQmhYdCjWVxO0hWuyykIioW5/wlucyWCZti5TlDa9QwfJ+I7bYh31KoiEttTNE4Yxs39O2DhUXD+c0s9LobZlfQCwHkHCyzbCCveCKA5yJ4QJASEpah4NOpxFQb28GSYioi4TQx19OAW5M6PhO8adKzUFDJt07b4s4FLZoaMXRso1p6z4kmS7hxDn2xy90Fx3YHpWswnBoGUxfl1vhS9yKD7ghr2kUgsNOd8zwab9IxKQWdYz5u8JCT451uhZm0y6/05nnQFPj+8QKTedjqwXwoYz7FqmMsbuJCJ+tCXRRWoNIs4DCixI3bo6YcJAzgsLgstUNLRrYyajVdLpmIu9wS406RM94IRZ2n0zlvU9uNP2EzSxeCGRaTBkM6zz63xtNpwF3kCSX5oApXW3ByUFGL4WIHnpYCyroTlQ1tF/jrhpMVcF43i1saySk2usa6DVs9+IxJ+VwC7xwz7HVLtDnAs9ENisENAjFsSHvWRCinarEhMn400XJW0KXdDFQsTj3rhvVJxamHpmMCIrePipGyS9HWMOZm7wkZ2X/YIlTEOKMZ8DU5x9BK5fngidYpW1Rqd4i6zpMY8b2KFivP8PbYe0TBhTRi1sR026ctC87xx0ijFevatW1lJYN96aBuEeuiLUw3lWtHt67OcyNpUpNgdRMwG60xm/boaZ9BnWdUudZtez7aYLN3HSUXAdKpEzdiMIdjUvKiI3T7OOn1Wy7cNcldlhO9hrU05WNJxDEnx41onc90zjKQ20WHmLFki9fnD3CLV2ElTNjMK15ZUhTksxnAQsvQ3JCulGJlU9mumnDhap5ifhOemTbMrsLpJsQ1WB81fPZsm0+FdRa7bT4VGJ4I53is12M0DliKA/63YAPdDjm1dYmfX/oo0xIx03eUNV2mqzT3qIimCflkr86VcI454/BthXE87XJuG7ZrxC2fGl5sxNSb2obc7yphx7pStEl+q5uD0fHMwHVqzmRGjghm1uGalgslWBWDUg/2e/BDAyU+sBHjrxhO7vLxJJosASl+BKy8JROBDgQb2Uh6cQ7cMCvgajVY7Rkmc7BLTGM9kQUoysMBn1ne5JeXNrjWW+FAaYQfqAzhq36+PV+iUjN0tUdUTllcNVTWJJ1KAP7KGlWka9rvCa5Fsbob+sqwJvDjPYpCRSEfU1l4ixVoNCDnKptvLXrHJePSSVM+kAbW8DPd3eJed4i9gwU6Ml6XdasFTy0XeaSzRCFZoZibwhXzlEls4oa8B72kyRYpvj9OGi6iJJNZ56nlNtisL/HeH/42pnaP3Czq/j5OyDcf8+Ye+LvsAZskoQUiq/i1X7/B44+9iN93lJbxmVA5IqdomW6bqWJYwUAA9SRLCWppGPBgDpjRKQ8nKf+k6DHRL2r17ZGpUEL2CUpi2x0nEg1REEutJoC0NCvwxMlqVXYiwj8LPQF+9iuqgjxp2KaTzV1Uo8q6Z3uXwBnLXJiNumLvrYr6JowK0DSG4+JsTEQ0rbjQjfnRXsy9jrD6M8ydjESKRvFaLSdyxaZED0lsoHb5EnA8afK9/iD33J2jKiG3kcK1IxdBR2R5qM5gll8r+ceDEw5ekvAjCy3eJuMxOSglqHeU1c3IoONc1GM+ukrNH7fdr4I/gRMu0Oldw3EHrLO1bWL2uFUGSSzDTTk5jjs+oxKb5iqeaGFHsHdqRRRJwQBjGqQeTtDcJh1CY3jAujkVon/3tHAGHYSNNqNzVlM115tld9piwB+hnkaU/REr8u8Ec5hg2i7g+cI+CuEy0+Eio8Rc9AYZ0SHfVclxasvwsV7IUFsMJZlBwXcU1wVv0oB9/YpeIRvPFzxoCcdLIpZSRUm6qQEcH5bOgqLV06yVwesl3GEqFrB8pfU81WiJ0B1CugQmqtvuly1skk2bb5vN4yWnN5QhOkr5JNEKRWIWTIJo2iS7cpOU+SRh2Gg+H2UXCO5252vNOPaiRRrHM7FBBtgXu5comYjQHbSsOtF2xQrKksagHAatCSFhVjJhw2VM63mUU0IpycyWs6Zr95/MIwWcLPdrAid0nlfovHUttkxIIuNyiblLWrYLJ9/zEg+mS+zVBXYlLZ5pPydQF1vQWZetmBzCFXy3Spib5LjSHJWEEqUYkAa4yj6PjoS9S+qAjOZ7V3k63qTqDbPVfJp7c+OcKBy0nZmaoyyi6NiYsuaRF+pYbWrDgWcbCdOdhK0uPLelmA8jPp9b4decgAsVaNVcC/od78JYJeb+oYjDE4q7DvvWMPPeSc29ezV3Tua4zW/wC+ef5Yy/wfdV7+ffHjvJMVVlKXbxCXhmZYalYJHj5ZD+sMhYPubGVsSnu13OEFNIEzbChNM65rxO8UX0rzzOtxXiShWW5loMklbRk27satYpFrzNLnHpasXDocTTGU5WNXty8nqgEkE1dXj/5hZvs+8FH2fUcF1SURoZ4kW0cmL2EH6dJEnkfcXQgcwVnAsU/XtdPn/NcCrqcKG3wk+sLvHQeJ6fOlDh3Qd2c6dboRS4rKSRvcD4WDeln5S37nZ5QUbV6ykTvqaYg+sB5Hsw5ivSIZBYZF8MU8K/c5AEO2Q5Ku2Fjbks77Yt41jpHCqsBvelqEVOovZIaPUCdvc1efDAsCQaIppI0Tk/lQQ8vBbSxwaruSn6tW8vFCSxpk97NgluwyTWoCMXM2nntL2KVWlAt7FJqX8vr33NYXISf/qV4Ktv+FT0DSRKyAc/ez5ZwMU+Y3/+qkisr2yRZLhm4D3751+54S//S/JebeC04ASyx82e4y//6ct/I90Rq7z4G5/g5ff6h/731/96/sG3bPuAyu77WsdK9q98fT3H4pu37d+6++uvfo2ynxTtdpdf//0v2simSA0TeCPs00VejbYU9zPG2CxCu3IIbDQHtSKsC8JCGX6hm/DesYC9u3wk61EWVRFRU800ddEFg3vrNl5AKh6RS4nxQbIem6Dku8QKrSmCyNgCKrcNOpWOmHTwBFemfcPmiwYB/Q9JFNUibMaG4XllTQuuNiyHGZ9sbxmKI3BhU/NrjpJNsSy1REZxMez2DW/IOXyknXC/pygmUEkNhWSVz7dr/LM78hw7bKCl6N0wtjDpu8vYbqPdnm2tUa4li2rKbcfz/ETY5MqlkEPFHKpnWNIwnRiO9yv+0+hh3vbiRRxvEdfto6s80tIJnM3P2gVYRm5Xkw7/e7TBq8QsoH2mTY/fXtf8SOrb/XOfLF/AmOhsinCmDo8aY+tkYTjvss1MxZPtlGIiBYyysW532hQDyRNNuEXlUPkp/rh3nX8UbVrxvjLGujbXSiepRWs2uSAVt7HbR1FpHu3NMBJv8BF3ktuLOb5toEirGfNwnDAYuxx1sNnAlRjO50DVDblV2FNTDAxnWrvlFcMjgeEWBaOlbEwmJJeahsZyynxBipCI9+QnaA+9ndPLH6bQu0RkERw+Se8a2htFeaM20suW+rIAyFpqC722LfyCaNWKNHPRGp93yjwAvFK5VkMnGZiflA5bGthxnUg7r4QektwxLzo7CRtVNaujK8d1qz3MebutU7YarXLDKfGkrPMqx2q4Sj6YtnxCJaxBQSM6Mq/L8mpTk5CTkSfaFmnahHZc10FbvWcbSUQQZIQm1Z7N3xVN4FDaZUUZliTzNlohLUp2i1zyuMQydnb7cJMeBRvzVbJF37IxVh86guKoTU6ARxOHDUHG9K7h5vewHC4w7PUzV7iFMQTOC0/JZwfF5xYN1zqGz4ch05LM0DX8y/0xtXLMarXIvpbi3r0e37VviH/t2wqXXKwwHUWxCXqkgiPi1vNiTDaUCxVOSNKEA6fmW5ybvcb4wCT7K8c4mRa5YwpOLMH4oqw/MZ1XDfFGJe8JB29UYtRiFq8bTnU1s2tbdLrz+CMevzntMxMWecTM8X/7eabcGj/u+oir3dU+TwcO+Wl4fdng5Awv9FLWmopxLa5QxQmtKAoXD9gvRgfPMGlcfqpQ4n+aneE/bExxbKNILk5oCf9yCBIZR/cMo0ZRzSlWN0VPqC2s/Foj4fTpRf55W/H9ap239o/w0/sOkm/mrHu/756Uxqjh6T8xqJLLxQie6BlelVP4cvHTS/nPvR53DhUZ7CouNw1VWesEedOCE3vE2CMuVyhcAX1LBs1urcJQUVEUhIyBTXH31uFBo7noFFkzAYVUMtx6FII6qn3cap9FtuLKaDdyod2EYkRBsoq1z7w45E3CntSlpQS7Itm+dbSghUq3kXbO0FMymynzG7/3JD/3M2+gLO1CW8lsL0x/9Ynmr73lGyjqtsc38vTbovCdWio7pWXan53fyQdIxkTyJbXBV37/1duV3SgAwuzrLz/PV99Dfs4eM3P//ZU75GVPvLPNX+ux/n5/95XX89cdt519+Pe7LX/x0V9erH2tY/Xy23e272W79C8+2PZPO3/3NW/8G365c4x23gp/03P9DQ/3D36zUjF7hudZvzpiNUcVHO5UnoUtXEoTPpoEvF4VcWSkmQdX4o2koEtB3IOvXm/zXa8qM7hH076R4vUplAR69yBdhFjccYey7Ecj64AEVMr6IC0MSSGXxVPOhTKudaGVh4kTgjZRpDLHlO5YITNHpBcEK4DNivVL4M8ri1yo7IE/PZ2lW0jupNOC584bixM45GbydZH1TScwrLFGj/nY8AY0IzkZpWBHcNKVEnT8kOfiFROMp/CFfC9aHS8bw5IzuP2KkiRkrEL+DoU/pXi7N8h3nGvzwdRnIYEXJC/XpLzRS3Aq/byicJjTvTNEUpj4Y5SEDacrxI0nMKXbbVdnPO1xyanYDumgifiVZI3lVpUaRe4gpagdjKMshPVo1bC0qfBVSAnPCuMllONYqik7cCOFmRSeskuUsiLqRWJu8wYJ0g5/EtU56g3YDsKA14cXb7GodzEhGpukRTeQECKXqeJRWsEsfeHz/PPZG5zcezs/vW+Q5zsxvzhrkP06biRsHAsjfiGwaWuUZKwZKVYDQzcSECuM9EFzDS7PwFiKTeJIUwfRYR30lA0ef2XpBPWhdzG//kf47WdJ84ctDNqorCMnJyHbvRMwmvVJyvxeUifkU6xxohWme9OU/BEu4tj8eRnpVZTDJRNTccqMiZ7OpHwWw2tlrJr0WOzM8G/GHmCm8wIfWp+l360TJEMIwe9G0mFMF+y+Ek1bIZI96aGLxzDRItrtJw0FuijdiwTHH7GmDdHeHdA5ZjDUg3kKTomruoCy2bI9ZMwrUXt9MjvUeTaiFdZSsQlP24JQLv9F95Qka5TjTXqFw7SUS146klKsI9rWmH04vFE7dGN4gZTPoNkXzHItaVONm1ZzeH/fG4mcEs/GW3zYahYDvjNeoJtO8aa+Gv+xX+GNODbh4WStYHls0i1ywixai7GcNRMwbbLvuexiTGIykjQl3gd6VswTDiudlE8HF3nf5xehbxf/qnIY4ypeWYk483zKKRTlouKhw4p9xSKR4ELKBq+gCBoeu4dgogrhWo7YHaEyAg8VYtb3pxSXFdcvB/xqq8mf1BokSY/Hlkvk/EFOKFjpFHgej93aI5C3RKK4WysumphrkZiLJD815a1a04dhUhf5uD/Kv2mv8UuNUW497rMkmdJi5JIjqiEpGsyWw+iE4sKc4G9aLPsd3jJueOJAjcbcOOUUxoyMQRO8XdDpKUnkYv84LPqCU4KVOS3XtLZreOukw8/sLfD0xZQjQw79degOweAgtvhsiAlNuoZS6EkHcgFrFAm3oBNiHfBBF6r9WCbf/Aa8Rfs8alKeF06LCmi2Nb05RdTIsnHLw1CU+LBKH0HYJp92cRlit8RBCmvTGnryRMqj5wqDxkX7uzGS1SwCQzwOT0QoG+D7zTlVfd1GiZ0T7cKlc3zoN36P/qmDDNWKzJ4/ze//waeYOnqckrQPbUdHNi47rZugxdmXLuAW+yjm3e1O0E6RlzkFd6q9oFnn7LnL1MbGWLt0ho99/EmO33bE8nXk+Xe+doqMnW1auXGF6eUWI4PCc/gaX7Y6kI9zdrUt27bzcDuPlf3uZYXX9sN8IwXFzna9fEsuvvgYH//sS9x64pAVab78Nvn3zn3sntnuPMrvv7J9OwXsV98z+3nn/juPlf32Za9HHvMvdFO39wdw5osf5/PnmhzbM8C5sy/hFPuzY7VdhJ/67O/yp6fq3Hl8r+2g2ivdbEPtVe/X2qLtI2zfBtl74ivHfOfvv/La5Hhk25qmKVpr5i+c4Q/+6HMcPHmrdV2+/PXt3P9b8bu8pl4Q8aEPP8zyyjDpyK0c1xWWUDRFTWQMfyqaLe1xXIxhJtPRSSD9DSOdIhiIu/zYfXnyTYdeYihK8HWYFXTmvMF/QKHERCWzLLksk8VSbP7CSJFxrBRpcgUtfMx+rDar/GrhEBjsLFOOqwuNp4x1t+UFAtqAJ2cyZIHoWwRzdmUd+h1sTJjkz54RELIPvRTOpjajmkAMCGlKPtLsd7K/F85WORF9jub3TJd3Ox53uRXGdhsSWZgFWSExQXUp6hTupJx4IJ0BiYSWeCQhzo9Oabr1hEevGvYVXFxJaJAuYFextQkTfh+fD1doYahZ6G+Ori5avEgsvDWnTMUbJNE+36Yc7nAKNqz+gvE4ZUIe2YaL7h+Fcqyph4LigDVBEogOSwj3wJwS4TiclXQLGRdux3VOA8+YxOaLnklDm3Bxqy7yVLwmpRA4BYsG2ev1s6g9ernd+II/CRdsEkbeHaAXr/CJ9YvsUUOcyNfYFcf8VpAwprRNuhDdeL+GEVeMEtnJqSo5mRXFHWOK9Q14MsACeA/llS3yBIV1XQwtScS4ZU9rQneMWbefXuc0TtrBiFrdTkVE7R1h4gxILG8qC8Cw+jsJUJPPpSZItuj3x2g5JYtoaYp+L+3yPrfMXhPzuXiDEV1gj0p5Ie2xGS3Tal3ix8fuoufm+eLKcxS8IrFTw9EeAmWW4zIifIlwgWY4h+ePWSOHkfQJf5d14wp6RQl6RQnk18NxKqwrzZD9OSV2q+h40+JJ7HpgUpSkWJiERDk2os4XI0TrGVJdQgvOJW2TxpuMFg6Rd4rkdY44adJyytylXY4pbePiBK2SkPKCSWnHdcLmk4ixRLZVu8Oc8Ue5mKY8mMxyn9b83GA/P3SgxtuqeV5/t8PBPVJMaSYHtB0/yshOyXm8ojCSJJLDjvHkcyyfN4GHiwHKiIFkTFmYcOeU5rnrS/zalS/x788/x221I4x7HheTHoHy+VivwX9LZTDsWE3jc52E2euKzVXDSt3wwlnDw9eN1e79+bzhzLLhRt3QnTHkthSq6XCoV2Z3XCWXFPnHd9d4y9AQJ9cKTDkNtqINPtars9BboC3Rcsrl0TTkTNKkaRJmdZ6nxYZjDJtxxt+8hrFaxo8nAUnSZW9aIdyUdV3SUsThrjlXl3WvxaeWm/x5foO7Blp8h7OL1+wZYnK8QHspsY5UueA8J8aNRFEWltwynAug0ckmCM9tKr73bs3QFHTmYF9eMR2nnFmEW4qKkUhMLuLih8E7sjhF0wEl+bpRxvwTQ4aUK1Fq8HKKtqBKBE0gaB4xpaGZVSK/2OTRXo8HvAkO1zyLeAlbwrZLeTZyCcJ1Vr1+xpyafc/K2F60xcKvDOWiPs5I49qTJ2/hJnOE7R6jY0Xe84MPUijIAv4Xz/nfyDlOhjFf15ecVOVrfeEq/+xf/A/8j//nR+w5pbV4mR/9yX/KfEMQ9tLqTAiC0BZQ8nNv7Qonb7+N566s2tulExRLrEcUf7lg2XnsjasvcMedtzOzCVuLF3jPD3/Mum3k7+VkufOf/NzrBbZIkQd94g//Pbf+Lx+zbjZ7dOyGGBJpfch5LthenOTcJ0TX7eJOHm/nsXYKlTiRJRviOCROUttZjAXjvfNlUl7+c5IkXy4Q5bnlNnlccW31ej1CYTUAC9ef4od/4Hcss0j2kYyZ5StNki8XuiaNs+172WtNopCuwMvswc6KP/m3SdMvP+/Oc6ZxRBBK1b+zr+S19+jJa94u6JJ4Z3sV6fb+uXTqd/i+P3iR7to0J26/nVPbx2rnuOQKVXKFrC2fbbb58n60G7b9vzSVa2S7cXY7ZJ/a593ezzvHfGf7vta+l4JOvlZvXOSH3/f9NILsfWX3KXJMky9v9/bTfst9a8wGbFxMYGLckt+vCg5BKXvy/zCKf+H4vNPNRnlyNepJsLaMMIS51Qv57ttL9Pf59GJjkyeiFwzpPFb4a0Uh8oql2pBFSXaxMMxmDYlFxGcFmxR1SlAFTcGGZH9nD4cYKarCsIPoBkh3zh1TlnRfr2OzE6XL1wyxouUD/XA5NDy+XcRJM3B/TvHm/5e794CypKzzvz9PpZtv5zjTPTM9OcBEwjDkHFUwAiqKYsLIGlZ0DauyKrpG/JtR1GUNCCoroAQRhgwDk2e6J3aYmc7dt2+uquc9v6e6YWDd8+ruvufvee85M/f2vXWr6lZ46lff3zckFUtjMMdSvKzB4pUtsCyjqBeAxIetFcWYEgdlnzkNGWY1hUw+CdVejdUShXVbjQpX+nntEO4TywZzaqJiirBWclAVb10fYzAoEajQ2FgoLc7/IizRLHSyXF+/CnE0lUxTiX4qhnnc2Bxcp57Jch/bwwoNOmApFp0o5tsulzsOSRzasYhpm1/u1/x+GH6cK/FoMGrQp+2I+lExT8jirmKnFHxg1J2vSEvOLJxmKa6yHHbISWEnIw80FQk3ev1xg351SgGlq5zoNrJQUKfYbEgtM2awE5V+6hLzDVJ0Zc99vKt3D6J8/UjCIyGtb8n6lGAQiUsCthahN6aYHBLuUlTIDeTgmVDUkZrHgtDknorv9FypKS3FoUCxUnJibYuO9Eqa6y+kWh02RZwUTBEaZketTlGvSIEnXDZ5iEQ5zJuYI1sHVKqjlMIKBR3g6SpXWDFGJQZOFKTiEajLDIcVevxJDpUPc05mEX4xxqqwlXg8yZiQmCr9hLqKF5YZDYuMSVvLtGmlcIsilWRg06aNFYuKvCA/XYiNme8KOC2XPzlOZR3DsIyWeYgti6x3UDQm1CX5neU+qlNPRwkhVpKg0musJaQ4PKCrxKw4ri6b/NnGIE85KHBXWDHJKBLJ/pOwzJbSASbH7mGycpCpvJw07VxSu5Lv2Tb3LUrz5fUruC7TxilJj+ZMmllzXLxOCJZryhlNeadGC0izAJxF4IiVxy5t7pJK92pKuzR+WZPvg6knIHwmJHcHHL7X5q5nDvK6vTdzy/BWWq1auv1D9BS62Vzez52VI7SFFlc6Sd7uecwNFdaoY9AtQRyLkzDfghMSGNT31El4TT2cY0PziBQjsDiuScV8YpbPiUmL+n1x0keSbIjXcF3rXD7fvISfZmdze0sD6x34XWEbj47/gd2jd7Oj2MOkeAf6k8TFPkhX+XiY54Cy6dMha9ws3ytU+EDPIJMli4mCze5BeOxAgdvHBvhxWGDlSYoP13fwqtaFtMU8qk5AGZ9xOaUsc2rjDUUWJE5clNWa7gqcJKj+KKyWiDK5Kc2Z+1rTzTgzdNka+OxRgUlvKQTaZNiObop86sbFNqhDYUtMXRZa5kdjsC8eeVUQS5nhXoGAYEhjUjbW20nGrSRMdbNz6DC9ecsg2cLJEy70XNthjpUxKLEc381YTCqLQTlvxIxdREK2mD6JT51wRSXKrw4aKow+7TO5ObrGRyfe/+z/vxmpG+nbz003/5Tux+5m3UVvZHYq5KvfuZnrrv8U1vA23n/11Vz2+qsYVs2sXraIO777NX7zp8e4v3uU8049laEtf+SK176ar/7kD7TNW8KSeW2mCPHLI3zhn/+Jh57ZSZ4YqWQNhd6H2bH7Kc6/8BJal53OuuWd/PHW7/LeD13Pbf/+M/rLKVRujH+99n1KzXJHAAAgAElEQVT0PH4PTvtxnLx2EdLtLQ3v4TOf+mf+/PCDnH3uBcQaOth0z62cfPaF1C86ieNWzOXen3+f937oo/z29l+y7XDIyhWL+OVXPscv7v0zX//Y2XzvT1VOWjWXH336OvqTXSyf28qtn38fv+nWnLJmkYS28IuvfJBfbbc4Y8187v7Bx/n358rMSRX4x2uu5id3/Iaf3PYnTjh5PUGun590W1z3sg4+/rFP0nXi+TQEA7z97f/AnNWn447v4D3XXM0HP/1l/FQbx61eysGn7uX6T3yW23/579y1qZ+1a9aQTUQd854nH+LGr32fh373XR4eTNKmRnnvmy7lk1+5hVTLAlYt7eCBX93MZ268iVt/8C0O+42sXNTAJz74EWoXrmdWQ8CHr3kPdXOX4fjD/GmLT3zXo9z98MM83D3CaRtOo71B2jGwd8vD9FfrqSke4nOfvZGHHvgt5130CtqWnsraFQK5SPGreOah33HDV37En+74GRdeejXLTj6f5V1tPHXPL3nbW67ixh/8B21dy1k6r4U/mP34sRe2/TGL+PmXP8O9jzzOt3/0UwK3ndZal0zG5stf/AmnXHw2xT3P8oar3suK0y6guSYW8S9nerT/s/Pgf+XbUeEJ3775QX79+8dob19AoZokJ55ZlidcfzZVR2lVtglcl+k36ZCpUBsyrVxKb6vmeFeLomEgZlC6qphwTsvtTQB4gzLZm8a9eDoRQkv4uwggJCw8DX5PVCDpfqjIZ43gxaNIMTGUs5KKwh5N2C3WClApwGHJbByDxZ3KGIcKGd9JKcoykB7B+MVJETdLK2rbNTurIYfKir1+wMlNNnMyCltQvCL0leAQFhO6yCP5I7xndhNrlzo8N6hNXmVCaofZUUvZwF7S2RgArxGcLhlKNUo4RnlNtlkYzQVu/3OVsDluFLvzbdivQ5JuSFeqhkcn+4xFiAgUquUBM2ha0gYt9VBvp6m6jSwQ37y4xSEbckqT9hUDYRGtNMc5cYQfNmEUolEh8UjlCGOCDGmLil8lqaIoqZSS9qwySKS0hgR1HEPTo7WZpmxZrLdSOHaa7rCE449zuLyPLq+FFjvNdsul0W2gyW1mWC4BoRCxY7RaAU/kj/DrQLPeVVxcmyYIxDQZpGMuYhh5Gq/ALvEErEB9WtGZUTQIIVwrniuDWGoIf0eOtWZtm9ZV3AkZCS1WWxaDTgP9Vopk4Vkqwq8TZEtQO+leBFIOykObGDFB7cQB39yjarGAKZOJdTCoRakasMpOGQuXgiFqhvQRciAsMyuYJFfYztszS+l06klItJyCrfktRtXsCzrnNuBZcXKVQ0yV9hNLdBnBhqB0OixieZFRqxR6IpCwvBazEQxqJ75z1WEKYs8SVgkN70/GRSEuFbCcBpMZK2iIGL0Fxd3GH0xJe1li7GKdRoErPLp2J0vBTjEpaRXBJP0mkcTCUx6Hq4P0lXtx8puo8fsZKwS8fvYGbl50Nld0NLAik0WoCSmxGlJRoRlbpbFEZVHWBiESWoHkMpPT5mZneBS29IC7Pzoft4xoxgagtFOROwhhTkx/bZ7cU+LDG7fzM72VCbeF1tQSJhJdFHGND6PY5Xwo0c75yTQ1gShFNYtF5WlBnYPhGbbXKjrEl7KomNusaKtRZLKC7ipqGqE5q6jvVHSXzB43NiydDVDXqE22c0c7ZMuKRck07fX1rEw08tamJnKlNlarkLV6mHtKBxnL72CrP2yQvJQ/bhC8bFimSVBfy6E37GdtPGRLPuR7lSlqEhXOimV4/ZJa1mRTVPZpQjFeboQpSbPYpRiZgIY5ilrJpl2qmByGwUOwp6SNYKlOjJ3lHKiH5Y2KihO1Tyuj0JJWxFMhr+gt8OaMy+wOi/SsyEIp0wROPhpTBTl16mF0b+QnmK5TTByGfAVSWUVzUjFUwFhQSQ71pJXikN/PaaqJxV4jnqUpiW1UIuC3Q2X8MEevk6YGTbMYPAslISzSaM6xBEl/jBER5ri1RpAkNkfNyREGJgZpndfAyacviQCh6bPwv/v0VyN1MwsI/ApL2o/nS5+7jg9+/EY27TpgbrXLhVG+/I5jGV10Abt3bmPsn6/lE7c+yOmveKX56pc+9g/UFHpYcdrLeeMnvsON157FpWet5TkJOJQuUqyOV172GvP6ja+/kmM6arjryT+w4bKr+PlNn+adr72B/tFDbNu2k3d+9Et84rqr+cg1r6bgNfDm66+Bl32UN738ZKMUkpnosMJnv/x1kp3rue+2m7n+vW8ltuwc7vj+F3jflZ9k4Kh5ffTdb+Az113Ftr2DDD75WR6aiPGZ7zzJn2/9Fx7bfZA5S+t47bf/SH60lyuu/wbLFnYZwaH0vto7O/nna77P4PggN731BhbNambswDZqT76SL3/+c5Ru+y63PfQsFe1h+wFBpciXv3ETJV9uKMf53i0/ZLi3h89etQZvzRX86kc38sG3XMY9Dz7Ffbe+h+88neL9H/oH5nh5xkSSOP2olnJ84UufxZ9/gcnpW7J6Peuu+BS33Pgern7lGdz78KN8+NVvJbn0ZP7hI9cxNXyI0UKeh75/ExPimqiLfPmH3yOXKxsrg2omw4WvOt/M/ZPXvYsV8+qeRwL7uu/nGxt7KE5N8tX/83VOuPQafvGtz/CO172HwyUfNY2uFaaOcNPXvsAlb38/3/rkhbz2K3exf+uTHHf+a3jNh7/Bl959FpedtZpndnXT3bOHd/zjjcxs+609R5jsvpMPf/SXvP7Nb2fZrATb9vss6Ork+z/4Art29rFz833cvrGJuW2CjRtsdXpr/P08CWhTqAgqHRpidtkfpV6HjIVlfiG2C2hu0wF/COChIOT2oMImw021GA98rkxaLGxN4aRDcnvF3kQiwRSVOIyVNWFndIE3PDpp2QiQOX0WG6abVIYimpjAtDQlikfyX6fpSagJTbhF4x2EmMQQ1YL4YY6NajqzIHYLI+IrNQkHjsBXd0HaigxA5X5CqEr3TWgTaVXQ2hjhirO7IHuibBOne8k1FTFuty84k0VXrYdVo8lmFFO7Iy8+sXsQg1F9BPShyPRYWvCSZStIHYfEU0386+CCE2s4clzIg6WqsSDZV40ycwVJbnJsLq9Zhl8aI1kdJhRCvLTdlIcXn8eIP0q2tNeQlXu0Mhyh4x3Fm+psro5l+BNVfuTnmG1p1tk2i3A4wWlk1GkweaAPB5p/LQ1xa3XY2LM8GZQ4HFbYQcDvdMgDwn/D4sN2jAUotoZlnLDMUu3zZqeeg7HZjKo4Dxe6+U1Ypt5KM1EZYlt5gPbkcsL4fEOFPIJFox4lG4zw/tF9fHNoNzVx8bqy6GiC2wSR1bDeU8wJ4Lc+bBvQhgu5ukWx1hdjXoWoY5OW0c6YtAfxGnxdscSYIPuholalSKVW4NReYI5PSUqP/OCkmxFx6URpKsiXIIxSFGlB7KSRVD3EyNRTCGoncWk7wgoHdZXDwRRD/gRJHRqz335/hPVigWI3MRX67NQ2CUEo8QjlR4Riy6rwy/1Yhd0mCD2sjppiTrnN5sJmlilFmexP4cyJebLYsWifcnWYXHUYO/QJqiPGX084gKb4c+rQooQt7Y+KROlFT3dOCKumWFRWAsepx7fTPFcd5XC5l3jlsClGDgUT1Jd76St2s6+4m0xhK7FKL4M6zfKO8/jgstNZVutRXx8SxnxznkkvpmaDqN5BVMuSwsKkMq3CiT2aia2aJ/+gue0Hmj9+V7PlHs2OXk1PjzatPkldEOJSeyvUz3PY3DfFVw7ez++TB9Hp9cxKLGLErqPo1HJyfA7Xpleyym0gqTyGKrI9NWOB+EvCLDHVFUVnLSTi0FsCy9HUe1GxUuyFlgbxxIuWPTkAsUmMAKecgMkcuD4sXa7MOLBXitPWgKLtM7sjZG4szsmpZl7dsIF3Nl7M7U1ncHPTcayzbMYmNjMy9gwHJh9lR3kPD1eHOFQ+gF/Yzxt77+K+Ui+vTCW5MNnI+pYMs+KKXI9vkGhBzIT2UclppgqaVBK2TWh6ExCrgZrWKHWjviJcU2lrwia7YjKZxRBbDt2GeYra5VC1Nfm8i6ia7hiLOnPVA0RegGLIngN7BGLNmHzZcVcKWTgiEW7ScZinDLIXxGBeBobFoUBpznMyCAR7SAWMhtE2FO9KVQ0Z1wUO2xk8iaSzU+z2J1itLOY4cgPXwKS0Yu20UeYLlcB0wcTbSQZBKuStieevt//Tq9nfLJSQC8dOp54r3v5Bhh5fzKWXfxtOvZJq7hB33QU3fvrlLFzcyTU//ghn3NHD1658mVnH5cuWovv/ZF5vfepPxvn99LNfyfiRYZgtgLrF3C7REMHixZ0Em7YC53HxhjUcdIQ09HssK8HiY5bx+9t/RtYVckIHzc1NJDvb6Zg7jzmtQgYSY0shAJtZcfZ557NCiRkNXHjRuVg7BJj+NY6TZPGKpdG8HCmW1qFC0Ysdy7WvuIhVK4/hfZJdnMhy2rlXwrXn8aOTQo5ZezmnrFtq7malrFhz0rksabuJW75/C3vWnM7x647B7ssT9N3O92/uNi7oMc82LdaMI1FQNitXtJqBy04kqFlwMlMj/QwkOrD3PcfdD9Ry6RlnUwpCznnbT3j/yJf5xKf/hY7FxxObRumiXyb37Uv58Puvwup53LzVs/UpigdDLjztZRQCj2/edyff/OZ3uOHBKie94i044iO1dBExgTJVnPny+2yBkUNqUgk6u+aY+SxbugzXVlQDbZ69eB2LapNmm9ZzARdtWEOvK2yfm83gHK2P/B/CWZ/jlDWriI2cycnbRjjUJ41F2PzkAxwMpjjzwjcwMZhjycoV3HbbT6iTEYV1hmtWzgd89fabOG/DGnY92MsTj+5j1oKV3PSBi/jxr27FevQnfOfH3zWkfFnnmWLyheX/fbyKDj1x2GzAVnUkBQ3QIQ3BFEmvmRgO4s21V4LllW14PEfw+alf4sZsjNmWYiKvyWRkoJrmzYldQL1CjYCII5SAK9LllwFNBt51yvDpREzhtkV37vmSZl9asyYpEWNRm1ayIgu7NLld4MtAKlFNFWjSMKtJGfsTmXYy0PSGcIIjBrdQzsKh3shUdUrEBAmLUVtzsecYsG08qenbASMe1HiQFQ88K8758aRpm+Z7tRGFhOJZ5YHKmWx5s8NUn8QFKSzJjRRlgrSqJOc2AH9vSEPG4YvHxDj3lkk+1dVg7CZ2hJbJxp2Y9Dm/pp2D4QJuHt1Ok51mXAdm2WL46QYF+ssD3OvUMxU0kwodTm9X6Hpo3w+nkOaeUp4HgjKvcBOU7CiL9x1OrSF/n4fFWfFWtuqQknLIhzl+XCozoBKsw+M4yzOxVtKsKVsuZ2hNjXIZ1SHLlMeV2mJvahmHCt3ky33E43NpVK4Jkhdj3AXC+cscx65SD5XCFF7pAC3xOXw7P8a3/b180m5ig1/LmzKa3ZMh/1GG16RgVqD4+ZSPHrCpsRQSM5YUFFI8xjSiPWF9GtqLDh8IEzwInI7mHMtms3IZSq82ir7i+N0ob5axKzHELqPyk1NZxkQ5wCQKQXr5cbSqoCt9EC4zLePQERNVi6qycHWVohRHljjHTjBbzFatFIP4PBFUONZrY0O8g43lPlJ2LYGfIiztI/SPYHvLDcKGpEjIQ+SKYc4UZ7JOllMftawMclhGWS6u3UYoxWdVxpfAfC52LGZaf8IU95adJijujH6DLZeFLJa0wGQ5YpFiZ4hJ0ShonpDaygcIgwKjQmB16jkmPofR6hFGhNPnLef6xmPpEtl3bUhR+Ghlyb+NUHArrijO1mx/WpNpgWQfdOegb0qzTDzdilAr8R81sHa2Mj+xKHmqoaZagpRto+tD7lIHuWbLTsbqOzgpvpCDoc9Auc8kYLzKzjJXeYiFjB3kWBsT3hiMlxSbp7ThfjakMSKaTXlYKGKSSTh+iSJ0o5Z9URAxG5pmRbGECSuy8sjMhtED4Ig5chX6+6AgBWCXFN8Qn1AkpDCUA0uFPBrAlQ2wOF+LEHxbY/PoayjSq/bz+NBh7hmR7V4QyIsWt8kcD6Olnfw+X8cfiw4XTeZZNZVgeNKi2dX07IbZwq2twkGhb5Sj4/l04QcXIB4Xb0bNgXJAVdvmZnY4dFgYU+am0/ME4dMk5H7flUQP+F0mxU+LOR4fczhWO+Ymtyhj0zw4vBUaEuBKxrLciBaj5R0Z09R2KA4HgoZrEmWYJdYugebOUM4Fn+22zcYpuLgFpHkwZIfUxsscLmSwVJlaudn0mg1NQaQ+wumcg2ajlSIh2vxyf1SFmnggKepMhK55/t/4728u6sxCD95NrL6Nd37qNr7w23OhfxIn0cD65fDbjY+ydr7Hnd/7Aq+58DZCQ3yAZ7du4UQhwwCnXXAZyzM5Ah5g9myB1aNHuShBMLBl2wHqx6XHPMu0HPL5Kaxl7QzseIILX/c27nxiN0v1Dm741+9TDXyKxQq99z7Gnv5LmN8uDHIZGAy7gkK+RNERsg4Uyho1lYN5bfRvf2FeS8Lt3PCVH5gxrFJyccSnYTqWJjdVpHXBOj76qhbe/b6PcMPNd9EimXgy8Egbpn0R775yDe/+0Id40+duYU6dwzuazmD8Mz/k2289l/03fIayL/y3kCNjvnl+buthtj67BSexh4meh0nWfRF7Ty/LXnUmV1+4lERo01Fbx46n72fD697LdfM8OpesY9EJJ/OBV59hfksg2nBONMqfqjDTpTQ6/SLOX5LCsWppSSq2PrKLaz/1NbyBjay74NWsWfUkyZHdbN68mU7HM0JJKX4DM3AElEtyRw7Pbn6OYzvWI8WoPEK/zFRFuGw+ecPukbtSSS/da3gHZqLp6UwwqPA5inm2j1RpbIxQtTMufCVLU+PE67fDxH7OfvlV3Pn4bpboF7Z9sZBgjth8CxezKnzITSg3zlmvfCvXnnwpuBey6+ZjzOcvFn3MrMH//WfZnkoytGRk0XkqYswa5JmjQ3NBq1oxJq0kzyAXY9cERtdqn81hibpwhNZyOz07YXZrpEo11iNFiDdGA7PQMEyvVg7xw5GBr/Tbgk0aJa1ZUb2WNNVDIId923yFEouxdtBynRYfPYkvEnz92Eic8txObdoMiSrEOqE8F3Y/aUSrpjgQk9F2UYjamsDXxDzF47kS5x2r6FRxcuMaO6fIxqHsaPy8cAQ1v1M2l1GLl8DkNEpAvbdc4Zc0k/2abK0yMTvjezVNSxRahBMVjRpVhsgsLWW5WIoHwfHxBNedH/DI5gqXZj0SZW0sHprjcpsZcm58PjfrvQwFU2TshMn9DKtjxrIinViChNxX4kW2JebTkQ/oHlEc60FbqLjMlaI74BE/RFgxkjYgyQez5PotCQjI34r1wum0a0ywveQ7ToQVfhGUmaUDcspmVMXoNG1a1wTLi3r1GOUxpmO4ySWMlXsZN8iTS7OTJSZtmbBoSNTKaaS59kwmi3sZyW+hNbWCYZXk0/kdvNer5xXpLhN1tdmv0FGA+Y7NxSmH/cWAg1W4vM5mbijiCM1DBc0jaOZN2TRozRplkVWah4KoIFruNvOAP4qfWI4T5Ahyj4LTEBVxYrMg3De3FcubHbU2JavKoF1ltHjJSYvTrjGCiEY55cKCSQ1xJMGhmjc9//k1K0laGj9QzCGkoh2WpJaxMb8dr9LHlPDfJK5LHlJQmUpfZJGFiBsnz7KtbBk/pjl2061YXRk0JsaC5mmZxlwlQkM+D2VctuLG607Mi6VgM+if+OzJMgX9k/avhDiFOSMayfk5Ywgr+6MrvZKcnWWyOsyWyUeh2k9b3cW8JbOOeVWH4qjGExuSIYjVQmU2HNodMnSv4q7xgNSYxXE9iqUtiv0TmtoqJOYoKuOatdKWbTW3vlgS7ZaFttmKoNdm23CRnz/xNJ/ffJB5805izGumUp2k1XJpi8/nSaEGaB/XklgzTVF5ptAYrkJcKAS1ir6iZlR8iGOKBkHrPEUsoxkpTHNl7Uh5mhuHung0vohnXLpDMdarKe6GoENitBROd0jdbEVMwNsRiDdBrmpipw2fU4ov8XrLKd9Edy2rjdFUtOiIL2MjXXxs3hFqwjxf3XWIAck780Iqbpr+3Fbmp9dwaznLv/QXiFk2V8bjdAaaZ3drw/NrcCycULNA4sJ2gL0YijF4ekxzf0yzBotn8nkumg/NXsoAAmLubGe0ucmtpqNxxA5szgyT/HBPwCXKYWmdpj4hoglFNaXZfr9m+RplbFYODYFgQkECykc0dTUg1Bepu/rkpjcGb/HhwVgzv65OsjxRZqTgGFPlMUcxUHHp8wfRlkOXU8/+sGS6CgusuOF37lceC1WFfcYOSYSBvrm5EHRZjkXbFS5+dDr8T///q4u6meXZrseaNZdSrmjmrD6Ln3/lw7z2XzeTyTbz0Vvv41WvehPn35Ii0fVOvnvVWaQbFDe8/3W88dx389i2f+Pen3+dN135RmZnRjnn6huY1SS8LSnAFLWzF/C+S1fy8Rv/lXeevZ43vqbFfOLGE6xvy9I4Zymffder+OrnPsX8GrGq3mCQoq5jz4HtZ/L5H5/Kt/7xcsMpEbRu6cKFeJ6FpTzSiVnGONRN17Bhfk00r2un55WtwvKLsLVPtmMhnhytWDSdsRBP2JrEueT17+JffvVezt6wLhqHxEjJqMc8TnnZFfClX/GaM4/Dth0u++ENfOymn/KPfU+w1cUoo2wnzonzPZINHXzjn67h8nMv4q3XnE/92pdTN2sBn731P3jZGz7AAz+tJbngXK5c0IF1KM4FF7yRq686gUve8D4uPmn18/vbcWNceFGLcdtPd6zmwV//iH/4x2v5WSLH/As/wNuXLqHnTz/nyje8hfULsrz3s99mw+pjKH7gel75+kt4/OrLWdK6AcexsOK1HFdjE69v50sfvIK3X/YeVmy9i5OWiwMTxFINdNTFcTyPM88WLwpwvTirV17yoqLO9dKcJ74KssWSWTrcSeauWc9vf/RFPvq2t5Co7ubi9/2AVScezyeveRlfv+HTzBNy0PPbfhHx6ULSjcdZtfwcY/OxeO16LhIV4Mdex/xWGZDFlmDmiDSL+7//X3QIUx2FgrjXSoR7VXqMeeRil1SOUQd2hUWmxMzUUkwRZ1Nos92yjWfWBU6SiYpL90SII6jubGgV37YhTbk2Ms0UHyVDUBI0S3aPdJcknUGAUyH3tCnsfiiEmn1Paha3KvN5KCCI8Op6NImDULdUbkxA4osyByLz4FQd1NbBg/2aR0I4ZzqabKwopGRtvNw8UeppzcmOx/mSAWtpMgKs1MFQXtOaF/GF4mBBU82XWLQgRe38SHUmVgLG4cRTDPdpvIwmvjAKMK8Z18Qlz7JNUT2icb1p7uAo+K5cuOBdrSnesHmKGjEUxTM8Nrei2FMJeJA0H2o/ke8duoPxqkVCihSx5RD02IrRkF7DSH4zzwkvzZ/LqY7FaAWWiB2D1qQsh3JY4l5/giVus4kHWqhDBDgUq4sTgR2WNhyqRRVlUiMaVYz9+GwTNaz26a2OcUcwRZtTY0QThyyPWuO1FtKtXFJuMwPVw8YWIq1gwkqwzmujLcyTsNM0OzVstpJGwDxU3E1S7G69Jr4+6fP1kWe4Lj2XM9xGdmLxdCVgdQhz4xLTpXncD43SNVVVzBYDaOAxv8wpMYcOZbO3HHA/IUFYQuyFRRmcsHMcTizFqgwQlPehRHIskUbTClijj5diSFAzaaXriqG1hJXD5oIUK/UzGpZRbgOOXJx0QDGQG2hNq6qlLwzJqZC8tozKuyszj2MmEmypTppM2qoUZNIqDyZQlmRUBUZwYnh8gspJIWOnCE3qRdIgayLukDizsNxrlmOyNMUEUNZZELhgKrpgiuhCDBhlnLAkN0RutmqNN582ORkxSpUBipUBltgZxr0WhlMrqMUiUR1laPIR5lcPML/2PHpTK+jVil9K9ySjSVdsSjnN+Lhm214RJJRZ5cQ43rPZndT0i4XkmBbAmY6kMgbXy2Yrxg5LoQXZWYIMaSoVi0qtxdP7DvKug0/SM5VizoINuKHivKDIGU6cQSvNb7XmcnzyYUi/3ECIh1zosz8s0+nGmZrUyGV0liedF0WlpAlqpKWqyE0pRCjQswNa66IbKsns9TIKNy5cEajYmslxqG8GKW4SCzRzllpMSXvfdKtAWpGBdAmErzkFK1dAe53NUHfI5mrIPmuELbkCjbE4Ny6oY0HtQgMbX3XSYv44MMZXb36Op2Ssjw3TGTzAqrr1POG2GIPv6ysh6wmoURaTfsg6JVxS2+QhjwvfeExRycOAJ2kZkvjgs9Gp8OF4iroY5KUX269JifdmQ2Q4vKI9ugnuCjz27QzY7fnMqdpI+EY20KY7IbYuZKGpS5Ea0IwOiL2UIpOE2AQcKkRZzLPEiD2tyI1YDOhZnNN2hFevqfLUPRZn1lnUpiSRw+U8K8XmcNwozxNiZxKW6ZPiTWuqUrhZMWqcRsZLPdgmQselKkoLNAO7ITcCGRm2pl0n/rsXNaVnJI5/5RwErRFVqOuKxFzWJzRKR9eLGYFCUC6SL1ZIZbLmAJPZ6iCgUvFx49E0+akpAm2Rla1nfpIByczroFolCDSWY5mWpee5RiFa9QNiMc9cKCenyiST0scJsWwXx7aolkUNa+N5UZ0qyFilWsVxxK5SXvvIOgoxWF7/53lpbAlaDwOU7Zh5ilI2FrPZ/MhdXLLhEuKv/Qyb/u3j5g5UDveZskJagWb+rvu8J9+UHD2uS9y1DSlXsub8MCTmScZilal8mbj8BkOidHAshSCVpUpANpt9vmqvlIoUilVSmTSuI55NkQ1MtB/k4uc+vx7FfJ5qEJLJZl54b2qSiq/I1Ag0LGNfQD5fivaFIURLdFCALya4nosOfCpVuZi4Rj0nO8X3q4RaJPYKUQd7nmf2ibx2PfHWjx5B4JsizHMd5LUcJ/J75fHSfW1i7UEAACAASURBVB5Uy0xNlUmY/fift30gy8QmnDrEd/7PF3nf9V/nrsd7OP/4+czYnkwv9u/iSUybRdn93CP9rNrwJRqXpqnkFjJZcwbnxdrQYYm9WrPCTpoQdsGn2xxNRzzggVzIjYUDnB3PcFasnXQYsM5WPDChuXAOzJut2BnAmnkKu1YjuYYCbJi4rcNRHqOpHcQXRfhoQucb13Q/qJl3iiItWavSspUtNQqBeMKOCyiiGXhaMzgBrUsUbUshPAxffCpElxWvqIVsAoZ8mJzCoAtinfLcBKxeqzhpjmJ8hzYWPaoFClV4bIcm6yueUFU+vusw37iyhWtXxwkGtQAokgqOmgXlQU3/fZq2ZYqEUEsyorcRJ93o4uG0RPm10l6ubNdGsRtLKu49WOWcX4e8tSPGKdIS1bBdrk065ISEw69Gn+Du0YdJp5ZQkXNdKBvp1SYG6rnSXrr8MWoTXZxl13GyeIam4N5JTZ9W+LrKvrDAPuUZD7vzrSQ1SrFTFHACkIqxr6NZEipj6yKF2SGteDaILFBiUjz5o9ziT7DcSpAPy3QpxYSdYR82C2Wg90fZGpZMm9IJCtRZKWbZCRNJFugSQ5IZGRaM0jQMC1TDEgti7YzpCkMqwWk64Ip0J8OqjsnAZ5Yt3mCQDzR/VPAGMYaWwtuC3wcVphyHSyybu8oBA8pi1J+gIyxR4zbwLX8UL6zg+KOURu9El7ujws60W12UN9e0ZbU/ZIolKcBMO9ZKo9x6gUNQoggUdM/O4CqPYlhgVXIRr8qcRD+aAwRGmXuF5TDLs3hg5EEeHryDmFtH1eS6JlDxBUa8YNS41SHCUrdBCUUwIaiaJXEjVkT/EFsIuehJYWd8foyPj+DOgjqLJlbOjSgeRZd7TREivmByUgjyaNIqgjzF6hDzlMXa1DEcdGrZI/w8cW8u7mR86imDPnbUvYx31JxgRD+tJt4MPiHRI6HPPDfGvtDnTCvk1V6MDilIlC1DPrHQoj7QZJyQoqOYG4N+W5t0iaYSZNoUKW0zcDDPrwa3cpO0iGs6aXOzlNx6MlaCVXaWBSpGE5qbgimOs+IUdZV7ZayW3xnm+KRls86poz0bMjoVda1FuNE2V5mWbl1dBHZ69eAnobIfhIIXb4G6BkV5SCNgqXMCTB4Gtw/8YpSw4DXA8CPgZiHREEV7iap9qKjpG7TxWjSVVIkf7q/ijk1ywnKXdalaltZ4WL7GkkNHjM+bbKNgONCnORyOsbl7L2/b+CRMjNGw4BSD2nVKyoLWDOgqj6FYpas8i+ZVeIZqMRoqTrc0hyxRwCfYXDzEMXUJXl9TQ6ZZG1/NCSnIZmG6BYf74OB+zfKTFSlHsfOIzxd2FnllNUUQj1T6ksRSn426FpMlSAsKmYFBMSV2oKEJJo7AZFmTk/a5CwenFPcVSjzT/CgPveZ4/nBfkmBvyPGLNJ/uK5DJJ7izvItDbrO5SVurbAYNLCWtcIshXeWJ8gCJYjeBnY4U6JP30tY0Rm93O49v/AeOP0loBdOm7P/NK9tfjdTNzN+S+J+oKxe9pSxiYokvxZkW5+kE2ZgwtaO/ReGnbJuYNPLNe5BKp81rOTtNcTRTFcj3XRc7ajNH0whmJt8Xp1atsdwYtVKev+ThTq/DzNvirRSTpGzzsE1xZl6a9fX+y3nZxn4y+pYUfvJwnQQf+/7POPP88wwJ+aWVtHC7Zqad+Swt2PpLHtNbAGW7ZLIzP3Jmu2hiiRTTmy4iUorSLp4w/2RWMwWdvH7xfoh89xIp6dlHj5lpE+nsi9+zbFIZQbuOfhhXLVmCKWgl5/Poh+PMrKsQgaMPZZ/MvJ6ZVlDK6Y8NYhm9jjzont/n0x6Bthuj5iX78ehtbzuu2RMClbfMWsmd9z3G6WuFBSgJK3+zvmdmFf8/f56UhGjy1NJFjydeIjaT2mdXdcyYUfYoh4u1z/3EWBtUOCHw2OUXaIqneF+8kZbakI/3h1S8kHlZm/FxRX8S5tZA7nGNvQzSWW1c0eVYcwcjvk5qIdg1wp3TlIag6kN7s8JrVuYCb1q2IfiidMtrnKTClnDtRYqJjZouMRgehqd3a+yy4jQXnLii4kbz6oxL4QQ5S3OMViwQBHFUG+s7VQvP7dBGmSk3B0WlmZKqwq1wSiU0nBg9odEJhU6AVYraybPOUxSe1tinKXPchxNQPQzxhRFXUAsb44jGFqaDFCodcFq7x3dKeX74yAQX1mZNO1hudluVoqmsOCOxiLu97Yb4XrGzOHaW0bBEq6AasQ72SiVc2M1gbA6dVhtzipoTHEVLIB0il2PtGkphnm8HRY6zkqSVokYuOMDuEHIlzQpRyRXggYrPuGmxOlQIGNEhHY5wjFzTTowRcr8/adCf8cohNiuHOjtpeFyV0KM9NsuE1z9SGaLRKElDowjNWCmmvBg5kxWb5LA/ymR1hKzbyIOElKd6WBCbazzfTrYd9ujAWJmcY4n4RtOG4iRJObBjfKgSkLBCVtqWyRd+2I6xLZhkVVjgUreReyuHKIvjfeZ4KtXDpj0pfnCCZkWhvHK+ywAtvTCxQ5L8qMOGZ6fcWabYEvNiqQ4csUQJcpxVcxLPKsXusIqrK9TokKdVklzZJeZ2SfRAZEEiV1FBBkSMITm0Mm9Da4nQO1meslIRchgWBWeM+ANGbCGt13HTXhd0UTlyIxxDC1/BiCPsSHjhtZl5SFEnBWFYPkhoxQ23VSeWsh+LZyuDJmc2lCDQSi9xbxanZI5nZc0yHvQDasIyT4VlZisbMd4+S8WowWG/JU5kmjsqAduDHG1umi4/oC4o4VoJstUYHdKyF+FQQtFS0cTSNrnBgG2TB/hRaTf3xQNq08eQwKbDa2W1XcOkcthJwDGWJCyFdFgxZimbb6KJK4sP2BZeWMP9QZ5F+MTLNhMxzeEidJagQTiw4xEH1nSd+6SdqEjWRwbn4z2Ql5QSFwp5iD0KEwF0HSPmxBFzxPdNB56pvZCWzrwIFkQBKrYm/SNs3F/GLla4Zk0tZx7fTGPFY1i8dzxtPN7EzQjPotgXIpePOTUwp7WeVVYdDZXFPBnby4+efYonct3k6o/Dis3lsBgc6yrjYYlZVpKUZZtr16gd8vXqOBOivlY2/ZbN2+0sSVebFAih2MU7oLwXJloj7diRPHQOimhE016xudxJGr5erALbhjXJFCwqK3NDW9MJDWEkLBmIQVpsniRqb0j4clAvvpmjkQCs2VMcOVQiVhNy3jk2X/halZp9IftLk4w5MfKO3CzW0o/iZAUbw7Lhnzoq+i1yLomIS44buWkR5NMxDvJbqZh2ixAvp9s+/82r1d9c1EmxII8Z+wZ5PfNetA5RgWGKuaNaZDPFjrx19PRHTRKd4GaG0+NINMMX/lfSMnph+TPzNBNMbwdBzWSgkOXLtEev5wszijabIdsftQL/edqo6Fx6/FksPT76tqBE/3m6FwqumdkdvZ5HLzcaIF+YPvpMtln0u+Rvmf/MMo5+/8XzOfqviBs1M+1f+v5L34v2QzSPmWW9sG4v7Iqjl/KXXs8sU74789tfPN1fWjeZ4oXj5MXTv/CXzNutaee1b7z6+Tend/Pzf/+9vZAWvkkYjK0EZwnXu3WsdGxu0XUm0snVilkxxdvroFJVVGrgsd1whetwyoku1QnNjdriATug5wgELgzv1ayoUbTNAek6VSTvc0SbUHBRnLrzoJSDgkBKKSgPwDMHNYvnKlIiSOg312Njb5LbpY3qVUy+JJB6yAGrU1EchO5ezY/HNe1KUfHFhUEzOaUIxRBVxnqxCrIjwndTC4wLZUkI4Mugs6hQPZr6GthXEnK1RXtbjAbhp0reo9iYTGhUnSIchdAT7g4ETeDv0njHKuz5YI1Pc/+iLjthVnJpFfqwRhw3BDl4w8sT9O4v8s6hEh+Kx7nM1uwNFONOwLxYCorrcSf/ZH60pVwjYBAG7xQWK9xGHCvBs6W9vEf7fDgxl/6wyqAEbqOZoxxm2xGC/n1d4SPaY3g623SDUnTbil2iLEZsI8SPKzBO9HMkccAJ2eRXqdUBC5VFXDhgTo1pvTfFOijpgB3GS6+EV+5hX6UXx22j2a0jXx0xBZ0M9kMSXm/FcELHeHa1eLNY57Wwu7hHmJo8Zqd5LL/VFI/Z9LGc4GQ5QsjsUBvRhvAZk1rEL5pPpSw2h1V+U1SstR2aVJyNdpb9/gRnWDHqRbRhp4jHuwjSxxFMPYY2hHDhsU2YDFgzLmjx1xIebTROaUtMiqXwEzQtDRLVJg7SQZ4H/Ekmg6LxZuxRHissmzY0+xGEsM4oCClsMd8TXpzEghn+m0DMBCgRuxlXbRG9SVu3ig7EOCY0nFH5TwuaKMWkFIGC5jktxhjYOMrKlKb1KsW1FHcFYyNcCSaxJNYruZSilaDfH2N/cT9upR9VPUSbsnGyp3Ni5ljmOR45f4LB0KNbK5bZKXJYZGQ7K9sgqlJMFnWZe/BZ4KQ5QdpmvuLrgc8KO+SUWMRxfLJQZXnJZn3Mpq9a4NtDm/hDYRvEFrDArjVI5OleEx0qzjGeRlTla30pIOGIb/EosFIpbrBcpB25SFIeLItmUuwIQmZbNmtq4KEpCbKHnY9o0ill1Oz79mp2+bB0WBszY69VMX8OTIhYyY6yUGMFaMlEMYWe5A2L4nMXRtkeb9VM9lp4rQE7ihV+fniAfcrifQ11rMy00uUpxD8ubAypFdRPcBSx4fHBloSMekXQp1GdYrcUCEmVk2pSnNWykovCBTw+1M0HDzwN1e3Ut64iacdw7DqOSP4syvgZpi2XrOWxO8wjN0c/rF3BznzI3mrIGcomllW0dinUlGZop2TXQrIG8g3QKoh/AdbUWfy2O6CrYjGZh2Re7KU07a6irgB9k5p4VdGyAIrDkB8C4RKXZOwS9NKDiYJmwrMZ3+wx1AvzOxSXrIqx7dkScemQSGqyciiGJWzlMohNTvs8G0yySSVZbMdpsTwGBVhyIqpA1GuVMsyOju3/hQva31zUvVAASGESXfyPfk/efNHf0yt59AXfAHNHv/HSH2ImeOmb0d9Hz9vM4vmVmP58GsWRguDoaY+e28xnf+nzmdlNzy2qqAVRNNWERJ5No0QvnnB6WS+UHH9p3kevw4s/f3GU2n893UuLwaOnjIrBmXdmVu/Fy4mG5Jn3ZnbBzPaY+e7M+zN//1fPL/3efzWdvD+zzBem+a+KwKOmiHawKXjN75HA+Bc+/jt7Fe17w+FCk1MuG9BG4p4PFZe6CaNQnGeJgtQjXdI42sEONSsLCS5clSG7DKplRcOxiuXFGIcPa/58t2blcmiUqJvFinRj5ImVHVc4Zagc1Ngrom0pIkZzzV0Ka5+GREtkUSJcOj2iCZsVgZDERHBchiNZ6DmkOVXEFIfg34elxaGZF4dsxmL3IW2ETguS0R2rWDZUylCzCPJCRh6CyZJm4jmYmoJRB1OwbspX+FIp5DUNSbyUhX8oQvTExy4QwL1BsiWlBQt+AhISrL1Do8aVMSa2xP1CEjSGJfxbYZUibk/FCClCEimbs+Zk+Wz3EOW2JiTjMwg1vQHMLtssi7WxPb6AVPkgQTBBQcQTyjMh8ONaE3cyvCoxj6fKA9xR3Msit5nFdgxfaw7okB0azjcHWsiPw4D5OuAk5YrhP0u1oiZyWmHClDUO85UmicWBACbFv01rlqgSKSvJn6Wo1T6+5JBacTrjsxgMxUctFfHCwgJTpSkCg0RFlA5leQROrYmzEsVtdzDOrPhcTnBb+GNhB/HqIImwTM5r5btTz0JsDg2xDgrKpl4H1Cubp6TroTXrfMVCbfOA5J/qgGZCrnIy/Ngf4/bSQVJuA3VuA8XqKCLWsLy56IrIIOumHaITKKchUpmKO6sYAQuiL5w1Mfy161BBztALxHPP9tp5pnKIjNNAzGszWbmD0whLLiwwW+44RHkq57Yoa2WEFcTNivi7QlMw57oUi8I7EjRDkDcpNKXYllxX+Uwmku9bLkpaWZJ2YFA+MTAWtG7K/B4jwqgOUrUTNGRPYUl8Hrkgz478NlSlD6+0x8SQBV4rc9LHMek2sRWbfGWS36ss59oes3SZg2GZY1XMmB4/Lbw2QhqkQ4PFm/HIWA6B1pzpWaggJgJZ1ko7L68pVDyD+N2c20J+/FH2yO8RFN+K0xGfw2qnzngfzic0sVJyfGUcaBRKABK9pwgdzVIpdCqau/yADSKgAO4J8qypZnFGFLWCrofQ1oIROFQdTZd40xWElaGNV2V2Wol7ZAg6RAkrNkVp6DmsaU0qg+CJ6KleKUb7NXsq8Nhgheq+InQFXFrbTksyRqLqkK4PUJ42RuFum6I8LK0DsOshLv58IxptKUp54WCL2bliaotmeCykuRKytibBvHAV53TM487cU/zT/k2MjifIZlpJxmrZb9XxrPSDLZf+4l5OtVxWJpfToFIs8Hx6Khbf6As4wbVpjmu6YtDUpHhkBDrqlUlhi2hSmqa6iGP4wB6fl2Vc8nnNaFqxt6oRSn37LMU24fnaiqQTKcgDUZmHkKkK7UjTamnulhuOeAM/313m+kWw7AzF/fsDKpOaSxIOd4qwRznUEPCLIM9yK8nptsNhHTIUFPEtz+QLB6WD0XFtrmLRjdLMDdP/9ML2N5sP73z2Ie58ZD/HLp0bcbTk3qoywU9v/hl2pp3mhoy5rsxcfP/ihV8ppg738LNbf0tj13Jqpluz/28/5sXLSUcn/3QFEhUxIbu2bSewEqRTLzannVmPmefJkSPs2jNAS0sDu557iN89so+V079JpjEPk8IgF0z5NcqQ8w92b0Og3Yba9AtV7fNt5OhXP//9o4oZ856Z33/+lTNRau5MlNpLiJJBeZRvff17ePVzaGvMvAiBnPk9M3Od+Xt6s7xoWplmZr88vy2z7bTIPgv/chE8M7+Xzn/m/cmhfn714++xN59k0dzW5+cvO+d/R6EabX/ZBzPrPrMuf0/PM9vjYO8Rbv7hLprbl3HIT3CulyJh25RCzZOhBIWDozQNriLbAo+O+HzeK/Llk2OktFwQjOgb1Qw17Yqudcq0U8VOINmh0F5kmGmJo484UUickAAaIs0XVEu4byEGxYtbYB0TRYpZWUkQgWSLMr5PrlgudUCzpWhuVkzsgw4Xmn3FaY2KZYsVHSsVNWPQkIF0WiH0x6ILazsUYQxcIQbairEBIWNLh0YSWjT3l302hUVWOnGOUXFcUcwGkYLN7VTYIu6QvNtmGNoDWTEd7Y74PDKoqulkCfHa0+ManZabHokOUziSyz6uSeyzONVW/KYg6tGY8YF6QHJhUwo7mOCxMIGMRLnqEeLiE2UnESVag1LktbS2kiScWh4u95EPxDQ0SxbLtFNlF0iRl7RcxLRjUpfZEZZoVI65kItwoknSP0Jph8FwGLLEdfAtm0LokLc8hpRikWWT9fNGRLEtyNOkS8Zjqy8skrViZJXLmPC4KgcJKn0Id01XeglK3SYbMhQJc2k/YWkvB4o7mPBHmNJl/OoQ5UofccsjpjWPTj3Fbh0waadpMEH3FQqS7ECVjaFiLTE6LIf7giLNyiVmeVhWymSnSjyWcupR/hh+IPlIwvkdjbi+npzPUftV6RJhpY9k7Vl4sQ4qhZ0YM1+BewwaJjt0ECvWRdxtplIZIO40MCxh8mHJWLgcLPWyNfcEnuHwispV0DbRFctNnxO1raVQk/UQxM4SMok4U1vmtQTMSUyXiVsSkYX41lkpY3Crq4PR9jPbUCwjgqggDEYNj47UarJOLU5hK1tyT2AXd5qoMLFRyabXEosvoMdcgOOIXlHj0Wm5dAdTjBGyUwvPNckCEbAqm+3KYr0SBE0407YRt4xJVq9YFYXSiodcEdbaHvVWgQfGn+TRqecYsxyOTR3LGcklZOMdRkDjUTX8PmnxDwqfXMMKx6LPh1Et1kcSKygqYmVEFCtsG9FFPYNlENAaHRDXLnOTYjysmAyVoSyITcdwSZEINV5MMWuxYvQIDB+Bqbw2EX2hD4+PamP2a1rEGZtqDPYMVvnVyDB/DhXrj3c49dg4Fy1JMtTtmjaub4eks0rqLYPOlYXnOxp5ykm+s/jqiFWSuHcIOCvCp+FuTW8PzBURg8RxpTSVIKRreYJTO7q4TM3jpEaXvUMD7Nq9kyGVY5mbZ0dlnMvjTbw2sYgFVhLXC2iPi0E0zA0t8XXmz3tCeodgV1ExWQ3oFPNgsdsRb08H/F7jlMSuqZAFjmVasRK9J2PlgA9tkm6jlfGw6zmiGbVBeHci3JUI5E0V2BMoEo7Lpv4hNqwpcOrSFtJ5KE4FfH73FHYqaRDwA2GR9U4Nq3TAYV1grlNHPTZbRLMjaLTx7xTkW/gmvdSkPcZHpnjzW86gs0MQ5798Lf5rr3d/NVIXFU3Qv+cR3vxvHq+/9DSCShnhslkUuOqat/LLezazYmErocSAGZJ83JDHJQpLeHUiUpBczHg8Rn6wl6vf9mY2XXw5QmOtVIR0a+M6EWcraqPKXW5AxRfCvfvCcu5+1ixHPiuLAGNGpFEaZMmKFfz6rme49PzVpggrlYrG1yg2LaCQHSSihac3/pIzX/44Wv+Egb2PcPXPYrzxstOpVitGBDKzASvlksl+izhzAb/97AoeW3U7t3zgFcYwMtouUcFRFRhDPJQEhph+SBRWiIWIB+QhMVe2bSORXr5QEFyX8ohEqa3mnqd7OXfNbFMsH71cmb6uJo1j2hNmBDTfD7Cmt9c0SmSKQbkDlu0SCUukGBLxhwy8lorivYQDObPPfnHXJlYsaH2+AJN9JcWYbb0gypA7CIllsx1v+rdFbWkZjDfe+U1ef+3nuf+Zbhl+TaCexFvFY5GAQsQUsv7SVpLYNnktBZ9sB2GkGnHx9Lb6/8OTLaMEWRwryZCVNK2aCQ2/rGpeJxYaAXTEIBvDIG1h1eKsRBI3ZkWuDtNiAklNF1QrVaMIT4XtD2na7w5pXqbwFkYFmvDkHDEQ3gWh8NvqxE8FhGMyekSTXKeQ7EQDiHgQPqlxWiKPLLVALEogsU/iibQh2rc3KMaVJpuWQg0yMWhuUPgS6iCdtRI0ikK2Q67fyrRC6+o1h8c01n6IpTVZD9ZXE+T8KmfpwCRU2PO14edJt0ysHEzG5YggcwZkMlmX8TZtrAMESawV52IBYyZg6qB4z2pqxRZihyY3FrWXpb14bmeW7Tum+EMxzynxNC+LWbSlFBPSW3Ky1LhryE5u5EB+M0k7wx+xOdlOUKPkTJI2szj1ruBIcSetuWc5lFlNq2lwBkzokBVY5q56tik4bDYqn7nyG8ouQyF0KtiNJqEUfyyNMB6MMRFohvBN4XJvWORAsYeyxGHpMk9KRiYOhbBIUYoR2TH+KIHYawjXz/BsxERXOGejkSeceK8ZRapPv3o28lkTpaigOpMbTTFUG+YZLe7kz3baZMo+Lblxusocy+WA18mDiS5OsDwyYZXbsVmlhL/VSJhazK5yH0qXScVmUzExWxrttqGLu1CqA63k4iNFnvDdfML4HHRiMUw9E7U1hRcnvm4m403O4EhbKgbAk+W9VFScjNdMbXWYYrHbXIjLboPxHA3yT6As+a4ItCT8V0iUUsTJfOQ4lTQLie2IowMRSMg6BKZ4k3gw8avTvo+uShEnqJ9txBv/D3vvAWZJVW99/3ZVndhxerp7uqcnZ4ZhAkMWCZJREERFxYgXQb0YXxUzYkC5r95rjiigGBAMIFFEwhAGGGYYZoDJuaencz6pqvb3rF2nca7hu3Lv9zzv6/3ueWBOn1BVu+rs2nvt//+/1nIRjziFVYdzo2YXZuC3dMUSjh534smRwKRkWzJTKUTjhHGZxtxC9vs17Av72SlJmzhyLhkzka1fjh4bol9L/eZVimRZw5DwCxHyF3L0tAgHvKb5AQtylk3lTn7R9RD3F7fTmFtMNreQ+kwH9X4t7bbCUcT8RjJEccRdsmczAa8nR08xZoYx5LRQIGRtmGJyZDi9AUdAeXYYDvdxGnr3xmXeTExWPSVlmeQZavMwMmic1qW8T6lLJDp0D7dOhckjhuEx68Bos+dRUweb98bcGY/weCEk1xZzklfHO07M0LI0kCEyY9ayddw6cePZGkemGHe8KDZYWZ35MCwHnB3W1dHmlibWJgqyDu+x9D4Lte2Ql7qMgq61HmY+jG+PnYjwYe05ZnUv4oh5s9l8fMmVbo5tj5hXn2JBttalvhfXxvQVDP0FS4OY8kOWKUU4Lee5hd2tRcsNlTL/vN9wQTFNzbBHi5QDAkm9eEwPDKvHQ45KpxwBZLzbMjgGj2+1dFQtAevyhultMDZk2Sa9ulwi6XQgsm6JIza15/Uk/n05OHKFx5InyjwUFpnvZxgIR3kiGmWBl2JnGLImHGSpl2GeSfG8Ccko+qx735XqCDskmCeYiMT8FyfCP6GPv3NH2ZpWgq0P8rH3XEw6m+Xm+9Zj/TRHayrLp9i98UFefdZZXHTR63ndOz5BZ/8ozz5xN++6/EN8+LKLyeWy3PrIZjzVvtBMLgi598avcMZFn+CAekT1sXbV73j3P3+Q913yRrKZdn71h/UuSa/j1NXlKQ118ql3vIZsNsuF77qKvft7eeDnP3Rbv/sjn+WR1eu4++bv8Za3Xsz5553HD2+5z3kF+hJ23bGe71xxOfATvviNGyn7rSzve4KPvu9S0ukMt/zxGcLxAb539Yc4+/zXcvbZF3Dz3Q+zZtVdXH4D3PiBt3DtbQ8nK0j9EFGB2274OodMyrLkhJfzmwfWu3Y8+Jsfc/KhGTJLTuaG3zzoBuI7fnANH/nElbztVcfQetrbWfPcTn7+o5+67/+vK6/msSef5effuJKXvfwCzj/vAq7/zSoia+jcu48oLHHvjV/jQx/9FG+74CU0vPQi1mzd7wbApNYP9m97mndf+Eqy2Rou/di/0jdu2b1+Ne+57H18/EPvctfrjFdeRgAAIABJREFU+zevIiLgvKNOom/DI7z38stZt110yD7e95YLufPRRKxZgLBn1/N89O2v48I3vYELX/Nm7lsrg1FFLWHbmkd438VfhKalDIyMsXfrM7zzdWcye+4Cvv7TuxnYv5V3nPca1m7rpWfLGk49/vVs7y6wbd3vOf+1H6a3oJhIEvB0f/wD/+OckiwMrtKMlCIshywxHqtDGK1YjvcMFWvQle0vaHUH+0fhrgMjvO/4iEmtHpVe+bda4m2W8jbwGlVDbpG1zeITDCMP4qJ2jvSQh0ofhM86IhymOVkpFwZg8EmL5EPEWhMwco98kr2yOy3yXPWacMeQCvJIkLRnLITaEbG0LS0zoPdJGOmCrk7YLxeJQVgyxTj2bJi2RAUVcsOiuYZFCw1LliST3DOh5aYoRVtbhvxkS6rJ4B1iCKYZvBaTLJmXGDIdhvYphtRSyF/gMfdsQzZnMQvBWw7+kdCwMgGX5QFH6nUSDLklUMlJ6iTmVQvztLdGfDYs8EhhnF/vj9gpp+5olClBg3NPUEqvMv4M9bbIqrhAX1xy1l6aoM80HsvyC510SNfYc/y6tJfnlS7FYy++8zWd6ddQ7+Uoh2X+ZfgAj1W6uWv4ER4d/D2P99/K3X2/4ns9v+Kmnt9y98CtPNX3ax478APuHLibZys9bAv7GYgrdMYRe6Jxil6egpXmo6IcMzFBq0txGr/JaaoJxJhUCyY7B5OZBalWvKDVuSC4QiiNOSbltsGrZdif5Arzg7jMmnAIP+wjJcsuG1Mz/gx39N7Cp7uuZ1XvLxnq/QkPdH2LXx34MYP9txMMP8xI32/pGbiTuLAl0XGrslldnVvVAkxXX2K95XIX9cEkcpNOcZp1CclB6dGyA1+KvE3oww2We4i9FAUbs2v8eeLCRuTH6sCXzlHgU7VyjskqXVKlVIuJVp208eRqUdnn2LBxeSe2vNtFMl0d38RYIcQgxwulXp3wvHajvyuOhOdWNGLMlg/ghQOIfmWDyY7lq4hfOLyK0shDRKOrGR+6h1W9t7KjMogkKca9LPMkQuylXIS1URjFpJ2V2EwV5wtKWpyIu37Lw41lSRpenUlRyxirhx/jnbtv5JnYUlt/Epn642jLzWUA4wBms2rGTIpLUikO8TN8I6jlLV6eIevxIDG/lmCtDdkunaRoiIKJOBDF5CI4W1FBYIVS7pHh6doRdtRAnez3fHh+1LoFmi5FrZxiJEoe4ggC4Qhkc7qvDA2+T1uuxBOlEb7V28vz0yLeeWHAZ2ZP4ozFeZoKPpWdsSNbVeQ+UWNowjAsI+Q9ltG9UBm3RPWJBt74lqQMUtE5BUsVa/UnwY71ltppMENSTHrTGApa7UpmcBxKQ4ZCk8VmQhpqA049vJ5Xvrye5mMbuLOSprtQZlsYs61k2CMBdQzr96h2Ulp1hoEQdsWWYzz4eCbLsTbF/Xssvx6L+FZnyNpOCw2WExb6yBVmy2jEsLrbdKhrgtm1hmdLMCyCWQP0DFtKitJlk4hrX9E6V4m0mA2TVRdcVtd39h22Nk3QlGNaueSExQ/z65lmsuySLJrfyAkyQ9A1szGzvAyxdBIPAnVhQYGAgP77tZIVrHCrmoke/qKf/+5I3cSedcBw/W8480frecnSuZx3ypt4bt+tTJ1ZR6VUZtP6dRz3xvdx/jFTmH/I0VzwxnOZE4zynW/8bx5cs5bZ7SVeefUtbL3qBI5b0MsPv/wprvnSatZuuYlpTblqNMdjbOwA3/7WV/jj2o2cccLhvPzUk1m/YzXTFkwnLoxx05ev4HMDC9m0eRM/uXIhF3yykV9ecTbz+Dgf+th7WNAa0XLMpXzh2ls5pr3Co5t3MS65j8CjrmMO57/nk9z0zo1c9Nqz2PP071j3wE/56pfXMnvSCK++5rds+/I5bOqs8KWvfIVN93yL15z5SbbuvZ5r3ryYR6ddxrkn/EkzbsMjd3HuW97D75/cSHnDXXzhO9fSkT6PE89/Mzfe+yRfNVtZecqJLFi7nULvk1yzqp2n/vVazEtX8NOHLuTyc8+HT/wLH3vvpXTUlPnpxkG++o2vsemer3PR+ddw+uA3+c6Vn+KQ486msG81X364kae++j28U1dw/T2vZ+W8Vya1fuEIV71qOUOv+DqbN/0LH1+4mC/OX8JbF/p8Ux68q9bwi9ntXPiaz3L24LXY7m7aFi5iy23v5t71F9FSivj6jbdw2TXXTfzcdO7eyPiUl/Dlj53N9y6dzwd/fCJPrJjjyphnLDyED37y1fzzjmUsm5rmvfMXM+3q67j9A62sPO5M2u66j6bKr3lw3bs4Mr+b+x/7JVu2fJTiE79kXfsKWmq0QvmvhZpfaOj/wT8morXlCnzzji6pltFVLjMnqDjm9swgptGz7FM6whoKGkwsHFCgxssyRSE5aSe1G5eWVCQrUN1bLxilTaWX1GpofLthyyY4QooGPRbPT9iutsY4D0OBvazq0cZxkgZu1tElVpHOuCWQBoeYbPUQyzZHfpQZeH675ZBmQ0kK+GmYe4TH0AHLwLhl0gJDvleTgGH3sKVGRtiKIkpiLK/soHVF0ZOOMQwMWLY8A/U65mhI3SIP5uaIgxhvwDpvWtFl3YJUEUWJe4ogLKcjnU+roeaoZI5Gl1Fjv6KPWv3X4Fi5WQmMtpiE9ZayLKw1fPVAPen7e7i+L8c3sz5zvDruK3vuVLOqDas5zDk1lIv7yHgputNt/NqWmaPVs1WKskQu3U63tWTGN7K/9gi68CiGXayIBtiG4eHSbjrHn2MwLrBTERUpLLsGJgArr5SmyRF6KRRBF3lAXhth1O9+a8ktub6uaJhNRL4lPByJ7iv9N9eJRIPQAD/x0L41RCtqXnmBHWqrrFA3Azj3WUskLba46MCcDMXlthAEWYqpNiecnLElxpT8kXCwooThANvCPucU4UUDRNEYzldV7XSRt7Bqs6WJRvZcCUHMEjFZE1N2AQXnzCCRRLVRC7SJ+9n5XuFJCiWeShyO4RU2ESnyppy/S5vKpWEqtvA8xuQTKzJFMNSxhDr0+Hdzm/atTqObR2Qk6ekpQ1FOUrcq99As6wUugufIcqq90zbaj1/jvhcJAKpuz4WCJXosMoVIH5awsMXJslDaTcXPU87MYnJ+Pt3paWSqxxMXd76Rg4dxUdozfEMHHiOxIrY+G2MYGN/Otb2/Y1skY9glnFO7gs6gwdmhPR2HnOgF1GqSd6LMhmYxr0WvMinO8ixKVhVsip+HIc9Zj6k2dBG1e+Mhbh/Lc6ZErr3E31e2a1ODNKvHYIpAfNEwPoKz4JICcZPFEQN2D1oG+mHWTM+J6g6M6l6v8OjuAjuiPlbOrOVzHZNoqUnRVjAUwpiebuu8pO1WKJUsa8esI2YsS3sundo9ZGkSwJGWaR5K0qiUBEq7/Gat+18s+5GtlubAOKcZFSVWZGUYWIKFxsmHeAeS8kmx8QeGJAwMAxtDlA5uGbB8KW3ozMQ8WCiRjjL0iTFatg7UdpQ0ploaA2lzwmqJMdfBce0+7TstezzYYCOurUTsXOezeL7hzGl5fvGc4Q29sKTZMKsdtu+zbC9CV2g5W5KfFagYmK/IqxEIF9FE0VmhZPD3GOxmsI0QhRE2nyXuKdJiU+y2sfN+PcQENBpLo8k48H+fLbJH5D+9Dhrd4kf3/WBZV7CGb/zuXk79wOGkMyIH/Xn/nxgT/uPnFw3qovI408/9Ai87/DBCVWryCfZ29lPfMUNVEHTMWcRt19/KdTsSSY90StJoMRx7pbOPyva/jIWP9brJrrwZrvnSV6h9xYeY0SED54MeUYUFr76Gk5Yvhlka7D7Pvs4BmmbPY7w0zNqH1nLtxz7OgvkLOP/t1/HZU35P3TdezyJg3uw5NM+ezi03fJObf/NDVhU8znzDm5PQvjRg0rXMnjcTMhWmtzaxV9S84z/LCSuXkxs8hUl37iTf0MQhMxu58UfXMrZ/J/4xrTS3TmXatFZmz1/ClIa802qTtMfunU/BKZ/i1JWLYeVizriozCN3/BI4k1eespIaFnE6sHvfXkqmjk+/4wJWrFjOCW+fy4Z0numzZegJC+YtYHrzEPM6avnZ9dcyvHcHh57W4qIhh7bPcJpvo5WYT1726mT7txzOk26ySK5bZayb76yHP/7wXOYvmMFl13+Eyx58ltfNmMeM5nOdvVdiufaAqxMc3/ksDa0zeP2Hv8uR19xA/DLLa674AYdMVb2gWMQek1qm0xg/xg+vvQ55bs7JyrdOfc6Sqp3EjFnTODJawNT6FNsCqNu1gbv+WMc5x7+MXG0D5/zz9znp/d/jinMaecUrzuPOW6/jkWv+wFfuf49L1/5XNXkO6jH/5/50k4/+iRjLrEPOp+VUE61ehm6JZ+ZgQYtPvMsS+5a6FJR8eLgQkWo1HJJNYTUPagWr0V1gRyVBmy3egurNXQd1xxsObIgZezIhQSirIgFiacRpPlWAigGlQcGfaRAwUg2eatREXAz3WIL5RsEUQmm07bXkUobZoXGr6O0HYMpMQ10OR+qY9zLDiPDgTks5DY0nGBy5VP6QIl/UwGgPDAWQet46rSwNhDM9wxKxuVRb42vpmejjWTm6iG0ga4myJdpg8Q81SBLFVaCXLfEUgxm2xM8D08DIYGQPmD3qdIZYkT6dnxQ29sjSyNJSE3DV7CaG9/XxXDbPdM9nqokJiXltSpI+s7kz1ULt6Dq83AJ6TIpQHo2CXV6KpZUDPBxZGj2fsdIehsaeJg6aSZV38VB5Lw/pXpBUiVfn6th8v1kFfgQ2phI0YIMmKhK91YlKjE/gJPScyK9+GNWeKaXqRmoBEkWmdN8q2qZImJilbtmvH1C6kyJmCSSpE4hxqguo81eqUkrPBawnMKUfV+wT1ZdFxEqPipEqyQQBxbDPpRjLQTNlbevXYU0NMTWO4aqfR+QCqeA70UNtq+PqPX1fbJpQUbjIzTFqgyk8z/M9P3GuNX52DnFxK8Z51qm9IjZI8FkoXTIlg0TjsosqE4t84dqlc9Jno45Za5VqVm2R/nfH1mymSU6N05OelZJVu/SGjqOxSSAveUykZfW5USRl4n7UNReAVbGpNlX7Zcel8xK4S0ag5NpqmWoLxCbCSltHYo6VA2wd34DNzGQ4v5ALamYzzdTT6FtGwohDBA0lIm09ulRXVxlg7fij/HhgA6n8fKYKEAbNPOBlWGwCp8t3qqvB07rLOpHdja4eW4aXhiESoNKlBZ+FU3zYG3k8anxmeXlKxnNadQIqa1VGFI3yoC0zy6/j1DjDL/sHGQnqOCkXkI4tg73Q51sGBTSNYbLv80RXhUpU4e5ykcE0nLMw4HXNM5zsRzRJ8nGWoYrFlzCvfjJhfF8LKhgsWZSuHGgG3Ybp6cbpXoYCmDMSb9Xa1uo9mzWYXZaNfZbyVljYAeUiFLudoo0bFuxjlvRCIzdF57Uay/FNlq0laD7cY9eWRJR3foth0bjPika4bTgkUzJOokyBsoHQssIadktn1cARKWnpwUB3kq1YYA1LUoGT+rknhJs2hMxX3W4+ZnoWpqR8RkYs8zsM9RXLaAa2D0G9Mh2erj+0py0LA0NPJeYP6qNBlsFyRDEdkZ7kY3thxmgNDakKj4ajtHo17JfXqw1ptZHz+5ZLzVg8qqUEGVnV6X53ZCAtKtQ30xRq9yfaTeqRE9242sdfzNOLBnVekGbPrR9jw863MbT2ESVDaWupZ7RfRYW7uezUV3DK937HZWcv5PNXXe0AicR2aUm06cbHx9iutFtc4Ungtvsf5cbTjuX9/3YY1330TS+QL/xUls03X8HTO95IZdPDzqF8clMNPVs3kAnyzD9sOt+7/UHOO3IWD91xLbzmXNeRlYxc/9xGFjf2s33PMB+75qt0PvoLznjTeZx4yhhL2xPB47ExhU3uYuf+91OW58nUxJ5KIrmHtcJjD/yGSz70GTbv6+H5O77L9358l6tjK46PcO+Ta9hzzlF0NCV6by1t8+EPb2HjrssYXHs7l1//FNdcKrx/F6uf2cVctnMP8PEpbewc66c+k6zGx4a3UYn+ZM+1fdtGdj/wIO/95OfZvK+X5+/4Nt+/5DFXGziwf7cT3ZXIcqNk/aWGNvyUq12c+MGDXBOvAm5/+DFWzE5z27e/xMlv+C2plM/uXs2ciX2X85myiSzDeGh4yTEv45AHL+WjD8Idj252fUzi0L4J+dUXjuaq0ufo/dY/ce2+z/PzQujGWdcPZedVLLB7aAyTqXVG1guPPIm3nbGITKlCe/M05k4zsPUSvvjc5az74od45/KX8AQns+LQeUmzXdhm4gz+MZ8nroUmG99FBzwik+JQk3arvXFR5rvKjPtivqbYM2adl6Edj1k+x6dWWajqatXMSeYtZaGMwE+t5kLjJEDqWgzLjvHY9bBlvpwjZkjI12B3yWbRUu5RrZ3St1BzVuIz6ea/lCUUpbM1KWxmwLqatZbJhs2b3GzHwA6cztUxkwzjT1n8vDQJoW4wIWis3W6ZpraVwFfUbapx/q2pPGS7oaxi6BLMi5Oanw0ZS7gLOFQipIo4JphEgNUVAksXT/6TymEJ35QTSqnLLNYY4lbLwB2WSWcY52AVKWpZk0zaIoWUdluMoncVKPfFzFiR4X+nmnnVAyO01eQ4KdVEwRp2SIhZSDKVIVe7jJ7Rp6jRajk7jc7xZzlFE6XkJUZWO6HiSIBLqcbyfkJJD+SWkElNoRQNOwKWwKDiRAovltzCJ8BWep1emnFAQ7VcOiepKVcwKfkKqNBf+g6S5hhNCAACKL4WSAJyUaK35mrGJNGh6ijpt4ndGTg8YkOBHqUZNeILlIwmwETAxUX/NMVpOE903wQGdZ1tRQxVjVMecl+QxIoDnxL21gSl1JBTei6A0pSqTVO0S+coQOnlkvcc6PSxpW688j5i6cdp5tExRPbQuXsZbKzaONW2CTQl9W8iMTjA5UgYImNkcOczcQ7V2qIkhDtREaQWVP92hZh6LQmVZLGZgD5XRJyAQZ25NDSkVSdihQPRejNIJFrcJOm7KKkRMPTr3O/ger8ijZFqr/UbCTQLSCfn7xsfv7yP50u7WVWYyrT8CuY2TmeHyTkf3QEf7imU6Rp/hluH7ne5xbaaRYzmFtKZaqTdZJhkAldjVTQeU/FZLCKNjUnHFca8LH7VAUSOms+qvwLzlS1Ma/Fhyfh5Ho6LfCiQx0gWEYLmGFgr72BSNIvIgWV3kOO5VJllsWEs9qivN+waM4ymYLRU5sfDg6yJLYdG47xyXjNLpqSYMjPlvGxHn7YMDOPG8M4N0NAINTIT6YW86svkj1oxnDw5YOks2L4zkQUpCW+XYOh5qF8AGdn2qSsULT1d8PgWy5F1EMpaU+uNhoS9r8i/BIfLnZZA9XXZJPXZudfS0poIaO8bhwXTkvpAOw+OyqdY1h9zf5/lEwdi3pg2dGHZFBvHBJaOpgqfFljIp6TPaSkXkt5+pDysNexlPZ4p+3SWuvlBP7x8bDJ1OePSrwJyIqmI8NItkqvWJhbnCLKmYukTnjMBx6ZS/Nv+Cmd0FjhzcR3+JM8Ji4sx/mBxHX56BrV+PZNsycV3816WzlhRupgWL0/FrzBWKjqZHfXV5JHCF6Oj+vjTnDLxzt///Ke9/J3bBGmBswV89oOXcNOvfsc3brqfQzomUaltoKmlnQ9cew0f/uq/0vXUfLdHMZY8P8dp0xIwlcnWctochV3THL3ySJYdfgyL/vgr5h//Ki4950SOW5JErZKQ/iy+9ukP8cMf38jXfnIvK+a1EaYnk2+cxOuv+Db3nftKTj3lx4Q1C3j8R2+iJj+Jyz9/OW9967tYdNv3qPP3ccFFl7K0zeOqb97Egin55KbVzTB3GUtZx5e/cTPnHJXnfKWt5KSUq6W+qcShy47h1ccs58qPf5xoYBdHnbuQkIClZ76f9We+kR8fu4KPXnSK2+aIl7yC7171z5x+9IkU4pivXvdTTj1zBTd/+7O85eSTSUVlPvPdWzjuiHlsvyVLrdyJgdopZ9CYMuQa27n68vN5zQc+z/fe+wbedcHpfOwjV+AN7eC01x+G0npzTjvVuWVk66eSrwot17aczRQJ6Gicl35eehL/suYe/unid3Lmj1J409/Bz994GuUtqzj+eOXdxELKcuzxC12EYNaJx7q0SnryfN7z3rN4522trFyimVsLZA13Ace97sccdsXX+dgVe9m1dSYrz5CkgMbbpNvlals5pNaQbWjj+gdv4e3v/Sj33lBD3dxTeN2URhrrclzxuuV0LzqKxYuWcvKFS1l86FuZ1ayw1P+Fdl/u7P/z/7hggitZloBoomulQfvAuHwSk8zR8nY4MAzpCI5skkhvAnpUXyJ/1qL02MrWWWYJiKH5ept1UiCtxxkOPBczcgAa5QfbCV4fjG8Dc5SEQg09vRaJbGaUuoytG5iV4giktagAR8lQqYf1vZZHIjgtDZN8g5/TIOgyp4T7oa4D/AJ0dkJ/Oxw32xCrNk9pYqVf+yHXCOkQcocZMtst+3ZAXxrawgypmZoYJWFfjUBqtJFBd5J9dIBO4NUV3oxZjKiDxUQ+KGgx1CyA4mpLzaEGTwBQdTrSq9U5B8a5UHiexTQaotGY2R1pvnZYDRev7iOe3MRsP80MDCW/hnvDERamGjk3V8u1PT+jJt1BWN7NtTKET7WQSXdQ0swS1SYuCULD1sEIYkXDgmYHiBRNUmTJkw2ZK9QXg623CqzExlREzsdLTXWRKn0/icaVUM2colU20j0rpJt2K3YXFZOdkLMJEfpQhEn3hwrnVY/jYVICJKqol9G9pn0hYd2DVWCmcIeLNlWjeYJuLmWrgmwpThtseQyrwig93E2cADGxX10UzMmCiLCmc0hYpwnzNJVE7rSNonr6vtokJqoDmqprkwZNkVisWBEb9LlrX0DKL5BKCxyOJS4U8rsNR53LUCIkrnPWuUxMZXoWYEsWrw7IuUbrdbIIce11UU59L43xs9VUasod37VfGRgBx+o2SSQ17yKusiVTKtm5UEhnT+DVRTxDvJTC3wKrQ+68K5UuAjweGBuFwk4Ybueo2qW05xfy4Hg3jwnMFbaTrzuCdM0ykbaRX2B9UMdUCQiL1RoXWeFlWeil6AmHnLjwQ5JAIeYBalBrjvcNz0iWx8iazrCmZB07XUS7znic30kqxcRsEsmHFAu9HDUYtluZUEk/z+PHpQLPlXdwTG4Bfq9Hd1zgwXiElzZGvHxyDaeV8ixITaLWGlIpS2U8cgFgOcRNn2kY32fJZKXaYUgpYTNqKXgGv9WSH4SpOcv4iKFhSnLvqxsPbobtJYn8GpomWxqak0Xqvh7LWUcY0kqJ6hLXJcOP9C27Cpapyw1eCQpaE2yXuJ0bCl1QdmCrpbDR0tBqqGuEoRQM11oaxgznTDcsrLX8cFfEsLEcgXGRsaet5bfAp41xqeBckEjBdBblRZ24KE7zjfOlzvr1/LYywJAt0jAOryqkmSdNmKLsEGMeL8S0hb7zl9YdqD7UjXEi5ONaJAWT3fXROGkbLeOeZaaXYWUwmTWVbhakmp1f70kSATeGbi/DoXGJzYTUOv3KOqwWfm7hov0rGl7t266v/+f/edE2YVEYOsKBDcVwhTrRbBTQLok1KlZIzPDgMKlM3hVra1AS61I6UmJ6OiupSKFRj7AiI9uMY0CK7anvBrLaMnDvzVdz2g0B4zd/0FmD1dQkIKBcKuMFiTWYjOZHRotka+vITDBOxfwshKRzGTdOjA0PO40oWZJNDBkTl6si5VbJSGhhaf/UPmnSiG0r2y553OaUL49liZXYnJULJUwQuAhYMrhozzFjQ6MuIlCTn3CygPERyZ76jtzh7nWxfH2fwPcRM1aDXyolbaWQUjkmm0tTLo5TLMXuuEqP+EGArrscG2wcOjs0sYTd9kYM2ASbq0u47vFnVm3uN9P1d5ZriX1XICuvUBHTiNuv+z7nX/YePvPDO/nU2850KRk3EVUv1NjIsJKtZHOBG1+1n4mH7Ly0wtRvq8dfszqLxACWhY7aXC5h/RSp/260V8fgrnD2Kz7PH/5wDLOWHcvppobNNuYlPhyX8vnJuOXkrOW8Fo+NfXBJapiHX13L5F0+dlGi4WoVgNkD3hwwmjs1I4hgqPryZocj6HrIsvMhy8pTpEKsuc8iZWN5wNpO3YuQXWnwpLehSUXuD/dZMi8xxBqPQo/H9o3zuV/GfHpGDS3dsWO/KWomUFdSKZDk7IpipVty9YbeNBxxLNj2KshU1nCjdZk/zZ0qH+rutAyOGG6tVPhMf8iGN+c5dI7DSQ4LuJxNyeXunSG6wxXqNprThXOU01H1chWzWPno3mNJ1UB6WcKks1qXiYz5fJK6jQQCFcSS2Ljq8kLLuhH4yvpxtozDuUGavcWdfH/kKdqKz9Hu+aytDOFXerHpDmfnE5o0oaJYGskE0JyorRj7+T8J7jpigiJdQpa+A2oyindpU0XFnLn8YBIh8urwUs2JDp3Gv1QrsasvU5iz3jFdbWV/NQUZJNG6au2cyAgCjbbS44CGk/pQBEmfhwPEAh6KJCnKpuiZq8+rpn0FBBUO0fd1Nk7PTaBOgEvEgWTScBEpfS6Q6Aq29VkVQLn3lOJ0VyO5zd2f2lbfkU6O0r6KEKoNSXG3+6KLOmifLpbpnF9aG316h1Qmp4iEUrpJZDKoydDclKZnMHZlLMnxdFgd7OCHtpt4T8dXZ1HdnCKYAnRKYWfdzeNAsYuWijmrTilAp4e2EcgUMFakT6LQjYlXrPPaKyVFXU70PeNS1i4l7DqoXMHESOonsGUC1S3aMoMOeNfSYccZCFpINZzs0u3ywdX5l+Myi9Nt9LqSUI+yjTjGpGjzsk7iZQuWWbZCp/FZJp09K6u6MtutvEkCzjE+fSZieSrglrFu9tmYIYFPYmbhU/EyLPcCx67/WVTiFRILu4+YAAAgAElEQVQmtjH3xEXCygE6wiE6/DyNNnQWbJ87vp7lNkWxP6a7z9LYDCllCEag9lDoO5DUw9WMJWnUfhEGRH6Q7lw9jNUbHnsuZtpzhmAWTF0M/Y9BJQ+Tc4nbQl+/pacEszKGYguUxyzLZxtGRDgYh/p5yc8rqUKVm0SKiHVDTQc056F7Lewfsyyea7h3OGJsH5zd5uNV4zwSFujbkUj8CQ1s2Aa3F0vsrESs9PPcFw5yp0nzviDHWb6rREHkrzqlWX0YKiRRtz0h9BrDTdEIi70s+6xhi6lwUY0lLqVZHvscSEN/HHFM5Dkh5GdDSyOGDRj2MMaqnU9w/xsP4cTDptLlh7zhB2WOK6gkazdXl4dYkJ1OxqQ53Us5UPd4bNgU9rLahk5LcrTURVxYTzy2kdqsx3BXiZedGnDn7dc4G85qx/1PPb3oSJ0AhqtmSAWuDCZZBSnHPQFkPOobk1TfwS36S/soSB/kR5XW8kCDqgt1GOYsPoVfv8eQS3uQ1mCV6J6lD7Ik84IMDY3Jcd127qb0yeSTegsNQzX1E3ZdGgIS6RF3HBe1SqJc/76didWVmnGwbVeS2ki+KcD4wkODkGubR03VGiw5h2SAytclaV33HekMVb1QtX0gRUb3SOy5xEjSI53NO6HI5FXyrz+xnRRcq48/bZ+8kYy9f2nV9sJvpiH5IJs3L6UaGKhpbuUnN/+OM0/XukKPiUE0eSUf37/1mLDz0uc674OvmV5rte6n0kmf0Tm/YN32t/b4j/x+MmnKBqnWyMQ8xsRlxsmyObbOYi4Xws4DsK1sOWlyDXktRtSdDiSlQ7r0TolBfuXK3GnNpLlMQR4VQfVZ6toMtbNdaQdW9XSTTZJL74dhRfuGktSInWwxAnadlvRihaGT2pWxGH7+9H5u315mTjSP8wOD1jcNnkEBVN29u7RixTBtqnGerrIJC7uF2i2mopU7rmZHUUKBrnw7ZMaNypDIBxXO9gao35fDKk2rctlqKZTLXYpj4NLK1RiKInf1xqWE3byfN9huS7jVosRA1AZ7N1naD5cGXVJvYmWJJjKFzlc6tdMsnpwnVvscenjMuXaEN/xgFWs6umBsJxnj0WXSdBndWwsJFW0KB1FKzLVCQEo3vQNXSc2WZDncik9FiA60KQYTJGDPYZxqytPL47nolE+sSJireUslTFaNDyJRpFqII3H2PDylQF0qctxprhkVDroU4HhibO9ASy7RURPoUl3bhMSHbg/VoTnwJkcFDeG6KJK+VyRN7wlUCVxVGaGuSk9v6UdQtE9RgURfzo2ILlU5Adr07EaSajRv4rU21ftxom3jCj+T3SVRPzcCuE1t7NPU4FFXY9m1XQUxU1lxZIPzlNa30imftU8M0TW2m5mzmhkppOkfFIjW/qtjTxWcuevvxtjqD6/IpdornFY9B5e7c7+bLoXSzgK7WrAr36fzFAgVCBQg1c0mtwxFPMdd9DWxFxOuy7vFgYs6uhtRDN1Ejkk1lJEdJwpHKJb3kKl0OULMvqCJlEkRi6ji1xO7VHwDngnYFw3T5GUdKWKsvJ9BrBOnfkxzjw3pMMY5f2wzHjZoZK3S9TYmawyOqmY8bqvUkar04oXD1Emw2lr2aDsvx+0OXAfUhyP81JacHmKt0vTytw97yfh1NNcfwaCp4XurixyfjZlWhrbpvksZ9g3HIk9THjdsfRbmqVZVtoBDIA6bbU/q4IID0Lkt5rnt1mnCLVB97pBxdlqqod3VbakNDNNbDYvT8IfNlmIfLPMNI3uTNK78o42cbcYhPwOkuynNPLFlW+oTd5r9oaU/BYON0LPPY34tTjfS74KGGhjTuBFCMAZ7hxJv2YuCDFsyFa4dH3DivgpwrIorNEsuy6WyDUdIfijwKGctz0SSb5KXs3UR3etszNtSGTpCn6uHRjhE3SVIc27eY+uo4aHQstdYthPzUlLU2QKrdIEzDRgJEQ7qdo+ZaSp0eDH7aYB4H2PhCFm/jgN49BqPjXEZK23MaMwtzNwiT3p1iny7AV6rW91r//XHiwZ1E4ecAC5JiH3i3eQ5+SxRuf+zT5xZrbbR/xOWW/p7AgAkf8OcxUcxRxORbkCNtX8mPqvX/+6z6j7cEOQ+S4aHg9tZHS5ck9zfyY7d64nvaaWoz9w4os/d30n7XBvlanDQdtUvuG0m9pG0zX1SBal/2f7k04l/k5a9sFuBxOpxJ77x9z67Yx+0/cFt+fN9JOeT4rQLLnzho6QNB1+pCaCdnMMLX/wrf+hYB1+D/7dj/5XN/5u8pfqcyGUUe4xHxUs7TbTno5jTfJ8pGFRN8VRplKObDflCPVLmVY2a2K6uQLajGpmrZsCsUp0CMRLizRvy8vesh+Fd1qVB4j6DxhLNs8OboeYcMM0Q7wUzVZO7JjGT5BIzHt3bxvnqnXs4YeoCKl6FJ0spgrShJbLIkrjmEMOMFsPoA4kEwjOdliWLPVJytpFguoJFnZZaRRKnQGUHlHOGsgrII0OvD9OiMTfoO8k0dWYXpKmKRNZVnwXmVPMi0KoAjopiFIxQJ+yz+BqdOuRfq3o9S89aS/vihDGnKKatsfgSGx73nPXYoCnzTGU7P71hHbeH66htKJAeLVFOT6fgDOeVRBMEMolhvNM/G0404lwER4vBqq2UG0MUjtRkn3O1WsZvSARuqbjom3tfoMtFrRJApQiQgIWb3JXSi8fwMrNdvZZIVRqVXBTJy7lxxKUEBf5VZ+aVsbF+7EwiceKCXj1JZM+WnUiwgJmLHKlhDrTpwlXBm0uFJhmPJP0pQd/ARfRcvZ9L6aouLUnRJuBIUcAk8lgdMZO5xe1L6VUBwSoYVD2b6vzctZkYI5Ix0t286mchtLX6dO0dph/Dpe86l4vfeiYLF8ygrDoSgbp0wKYt+/jh9X/ku994BthJ27QGurpDjCwRXnhowTzxWs9uZE5Svq7NopRUF+rRKLEtJGOztheYU4hJ7XfATp1J16x6Lrr++i00sco3VilX9Q4RTOIRjE5EHrXqK369i6bGpUTgWPurpBKhdT8eIyxuJu7pwa8/1vWVWDWH0iLEp8/zCeMSRa+OfSl1WmmyJW2SXZvrO2qvjaipSl0oOltT6aclNZmSSVNOt1EIh9y4UZ/toMbLMBgOMRfjnC265Z9X6adPfTEtOwmpkh/G9tI+to9upCE7m0eiBr471kern2XZHsPizkk8Uw6ZV5si+xx4ZY1PBl+1rRE0RTCyFZpqjBMu3jUQ8ZQKr2o9clnYuck6hv3UtCFTNoShZfCAYWNk2ePDCVlDn7LVSrPK3WFh4is7GLl1IT27YbRPl966FHLXmOWBQZjuwxNPWeoqsF8E6V7rUqCjNdA1CCMlWDoZmjJGQwTbIsvRmRR9qVouEWvFlGgOctxbrnC/LXOeX8uaQsxxpQrNfsqR6jUUKW83TYLO5SG6vCYO9wKGghp2W/hiWOSLvQXeTj2LlFXDcJQJuDYKWWHSHEvEo+V+4kqdi/V4mYi98RgmqmXYRdV9Gry0qzoZJmI2HgtMwCb1NztMRcz4oJGyyDju+1oc/n/3+E+Dur86YVdRyV/9LIEpToxYzddXJyy3HLioArEEzCRAIvnOn4HD6jGSS5B8NrH9xHHdPqrX6K+9N3H5XhgQqqDxhferEaakSUkkTvuZ2NdEiuDg42jbFz6f2NHfeO+gj//dn9VL4AbNiSHz333h733xZwD4b2020V5dvz+/hgdvM/G9g9/7W3+/mO/+rX38Y78vO6MUO1TsHFeYaVJM9yyD8nFUbYlEQb2Yb3XDozN9PKmqp6Gs6NrspN7MlTMqAqfMlur2q4FhzU3WWFdsLM/UyvPQICHivqQyQ/VrzkVCAE7BCQeUBMCgsFcgzGC3xdz9iM9p+cP46KH1tAx42KJl4yjsjKFZKQqxuaYYckda+jYLoBnMJku5CfyjTBJUyglcJTpRKmcSs3Zwp6UnZcjYmExUwJtVBZsqWpbVl6hfijwqutZjMS3VtKtyvcWJtKvB7rcuxaqiaoE8OwnaphnGH4aBRy31ZxmiyJLe6WOnW3b4g9y8bgNrV6/j59vkj5aB7CFk6lsohNL4Ejkh76JbkYtkKeojANHqUnCu0F+TvFKfE4K3E+OMwwHDTg5DEShHHY6GiATWBDgkgpue4vZv0jWYqMFFgVzqVi4HYsWW92DS04iVUtVNbqRJ0+2AoYsiOYLBmGOrqrbLlYo5q6thCPscqzW5J8RKTYCRAztKKYrkoH2KUepIOhrM0q69iduCLLREeFAnEvjTd8cTdqcT+9UUV2WrOCCnKKVGn+QaJcBRgEqjnRsoq3+rRToXjY9J64RVprUH7N21n0sufTlvf/vLWbZ0HllJNPzZ46gjFrD0sAVc/NZOfvCDn/P979zBtJlT2NuliJ0OMdGOaiTVbV8FeS66qjlC4FQ2YxOetMrmaAr2XImKnhNAp3OvpnGdfo7qFJLaQMm7OF9Z+b+mp+KlO1CEVvs2SqvLoktC0Eqfy+FDdYtKobsxU5OxFiVl4mgfce9NePklmFS7A+llk6ZoLV5qCkEw2dXNqb+EivwaQ52XFSWHwMuSMQEjccmJMpeCeuZ6OepsyB4j8WG5VqgjSsg+xqv00mAy+NGw8yw+PdNBNr+QtaRZYMs0piYzhZjRdCt/rPSyp7SPaUEdNZl6NpmUK6t4hiJPETqS75l+intMzL/1x7zSBCyU6oHIBtKfG4/w8ykeKY6yITK8JK5hasEwabrPUKelZwjaJ0F36PGsLNGK8NJ2GC1Ytz7rk7zSc5b6vYY+ec2qzHGfZUgLupSIFQn5amcooAbiMa7utSzNGfJFUCWTBG4khD4ra+jHUpMzpCKoiMyQgX4Ppnop7uuoY+dokZ/1FWmvNVzoZx2jeEUAm4xlS2zJ2GQRsEZEHhNyR1SkM+7iNel25hmPJmIWkCVlA9ZRYqhS5lC/jlGjxblI+0p/F3lUKhC6FbXGKcbsGBpjWTZGxJq52QV0R6N0ez5LyDLdJL/fVhvREY+zTyQdVxKhcgxp1mlg1M7+8j75s9vm73r5ovcyMflP7F2v3eMg0DPx3r+f4A3ODuu5bUyds4jJ9Rl2bXoWGqcwc8pkF7UTyNNw4rY3icDtn5Km1b/cgJN8x323CpwqQ3u58Ze/54RzXsOcKUmNwsHHn9iv3ps4BzceiWDgeexa/xS/vncNF1/+T9RrEnJDlrZyB6A03M/m3V0sWLzY+eq5t5NPX9hf9eU/3JOuycHX6h/uBP6vabD6aBoa2qjYmDXWIhKqtB1bawIKg9YZRD9uKyyv95hSykNGESlDZlDApsoOexb8Bcnc7yJwSqEqwCA2rCalGkOqHR7+VczJhxlqpoBIh5UCZCtQ05hkn2Q15iRStNputY5Bq/1EOwMumjGJeRWLZAzsgKGhbDkhZXhm1DK50zoJg8lzEqFku8/SssjgzzKJtEijoTBmsQO4IuhoDo6Bm1HmIUwkBabODMhqrJIlWI2FZgODVvoPyfzquP3JwK75VjjDsWRVIydErPlSkgnKJqrOUEKm8y02Zyhs80h5lvUbOnmkbz3vvPEOXOFWbSOzph/HeGo6Q6k2V+OULu52vqrKC+ue9oJ2l1IT8JHkhtgXXnp6UvOm30raaw4sJABGRf8CGC6lp6iXNNZcBKhCXN6tgSjhDKlGyyWuNZ5kEyClhF8wydXbxWKIaqUusOYpVhi4iJs8S1WvJTJDXDmQyHy4UTBJAZt0BybsJ5bPqur+BM4cqUKWJNUfWlElzTCaqPS/i0ZpCpJUyphLyToyRKTa2AoEDYkbQ2kP2DFsLK03kTMUtdMYLFAk8JJy6cAEDAr8SQtOx9EEpOsj9KV2ynXCugjd3l2DvPnis/jC595Bc3NSejIxH/z5bSoT9KNWTmXOrEsoVSw3XPtH2qY30dUrh42J0V1AVMeqgrtquxKwmYxb1Rmo2nblz5J5JCGICJ9KbkalIPpM+/WwslYQkcRF7AT+AuLCc87JQyQK3XCxOl6lB1vZ6347B+hdIYnAo2r0pFqXx0hYrlrrGBe34EVjeLVH4aWn4Lu6x4rzpq3o5tB2XtotAiquL4gBHRGr5lGgzXh45SK9cYkDfq2b6j3pC/p1eNEIo8Xd+Lbi3D9mZecxN9XsRJJrjMfx8Tj9YYX9tuI079LGcFq6DaWInwgH2BtXODqo5ZxUik22TJ0NaDAezxLTiMc/+z4dGBZLDcmq7NUwt+yxekgacJaNpshvxw0/KgQsGjU0ELDEGuanDXtNmVKUZWmtYbf06yqOY+w8XwXcVnVbp9U2KwPzpsJUyROpPlbErGFF6AxBKmbX5DLNAwEr5wbOk3ZgG7S3wWRVS6RhWp1HPCKCBxxxmCFSCamBVmUy0z5Hz6xjmQn55qoyp+QNUSohUZyZz5ApwN6qR/TMVMC6KOBtQQ1Plgd4Nh5jn8mzx1ZYKB8ZY2kjw5MGVsVjTLXDZEyODcZSUfFyZwV/edUdZ7+KkCczSYtpW2KbrTApLpDxa+k3Ps9bUV0MqtrtsyGBIxjpblBqX6UESQQ/iUT/+V3y4l+/KFDngI7SplHF2VBlc6o5qAIfAe5SqUommHgvsZdydVypILHDWrGCe9Z1ctqyFq5adCiLb7mHD77qNBe1k3VWRXoz1UJ8WUyJUBBHoasF015LpaJjcHrV40r6MIwNcXGQt11yMXevOTMBdboWcexuPvmtlSrRC/t1ba5aaaUzSY1aKp2iYdLEajZhboVl5cEDZ4215cmHWHrKeewaLTOjJvUX12ACKL74n+B/tvjvdAWcFKeN6cBypIGNJiIb+owMhuxW//UCGkd8XnVozMwjrCsYjjcnEgASH5ZkiAI5TjBY2aBheShWefW6/zUXiUSqtORC6F5vmbXAuICPnNfKzVCvIIuMEZXaVJ1mo/wXE1C2a6Nlfp9l+QpLHBkCBRbr4BAJfeahrhu2jEB5jyXsNc7/MHuIIaPCZxlRCicOQlZBKsmY+FDxDCNhYhg+J2u4T76SfhkFkSRv4AYr1cMIlOp1DRilWoVzNO8L0KnNetb3dL4CdCJA6LVIJDoP61OZZ9l4Zxe33fcQXxpYBYN7SbecRvPcGUjDs8caxtIzWSHf0nicp6ICEgaOsjOIS7ugtBdyh+ClW7BxPXFxJyYz3UVI5DCQCNjmsFLxV+RKxekadKV35mqzEiBjnATGJBeVi8v7MeXuZMJOtToSRbLWVZpNRf26bgJEClPIbH7Eyexru+S1IkJ6KIWr2iyRMUIX3VP9XVwSFzKxyxJl2M+JRaPOIaSviaTshExdmlAARgSBak2Z1aypcVCg0AGwRFpF7ZHcirXSmHCFQdXolejJuSrKVrO1ClCdVuSwlWvmBOgUmFS75E3amKRczzrnCP7li5c6QCdbQLdQP2iOSLZP/tU1UglO8+Q6/uXqd9DTPcydtz1FU1sjAwrPOCkT7X+iNjABqkQT9OkqI1hAz2UnBNpUm1SN0Ll2ataIknSqXqstOif9r2viqQYvAVs6FblhOCCvKx4Xkro9gWd3U+r4uuSaHxI1B2f15tLaVdAZF4hLct2J8WpWJgDfXT/pFdZiAkWDE41B7d8LVE+ZIpQAsusmISYaY1DgTn0l7MeTS0dpF0rrBjXL8TIdpE2avUEDS72AVglLW8PNeC6FOw/LTAdsLXtsRN6kmePXMxz2sq+8n2K6nXG5HRjP+TVrqFD5rkhdw9JW81JO2mOqMcm6Kg2lME1rZOlI1TEeV9g6ViLrBXxDdRVbEou3y/JTuLNsaYxTHBGkmJ83qNIi7Rleugj6eiHVCeVR2OXhhIXbDjF4WkethidHLLVpn7fMChyAUza1dWZyGw7uhtx0CESakEBn3tI42zgCZnGnxiDo3wPZ42MOnxvwcd/y+10xqV2Go5os/cPyz4UZPjTFkM/7TKrALDOZqXE9jaWAw4lZR5pdQYkHCxL5DGgP6mk0KdZLNqi0g7agiUxa4cVh0hq0aqFQhKHR2NWQznVReMt028uBsJ/eoJ6ZKq/Ack84QJ1VrFUWkOIJaDDXvvTD6156UXAs6Yt/5V//yiuvvPKvvP8Xb02Aln2bn+CqT17Jz391Ez+/40mOOPoYahnm65/5X7z0zPN49JkuVh5zFP74fq68/FK+/bOb+PWvf0/LzIU8cNNPue3+VYwVx9n47Drue/SP/OTG1Rz78hMx/Vu59A2v5qqvXU9t+3wOm9/BDz9/JQ88/iSfueZrLFp+FKt+8W+c9dq38OSmXpYfeRRN6RLf/cJHOPrk9xObQR56wnLJO9/AzCm1Dmxue/JhPvPZL/HAvb/irHNeS/viYzni0Ll0bVvHhy97B+e94WK6KvUce+xRhAN7ue+JHaxcOYfvXP1Z7rp/FZ849VT+WJzMEXPq+OHX3s/DT+8n9GuZ0uTx7as/z/W/+Bk337OWFSsPp6k+9w8fsfuLH/1/3nhRV0B+tjf+9BF27JpB7eQp+FGFlV7KhfW1Suv0DG+oM3SqBOYww0nz/ETORGoMmv+nGkLZWA5ap0GnSJgDeZqnNPImeCIJ+UcwvgeCQWiabygesIzsxrk9SNHdpTbrjJMzMTsTyzERHZ67NUY2xFqJTy4ZglaYNB28Ahw4YAnGZWdj2GdgTDIsLbDkVM9FE6NQbhjG1dQUngUrj1alYcU6LUGuzbC6E3oHK+SnjXDiylaHPaiTBDxYHVSkBp2PxrKccdE+jWkOo+xMMI7XLjsxpY1FNNE5+fSMWP749B4+cNODfPK+G3i4bw2Taltobz9XqsiMpKdRMRmK8r8MGqjzVd+SVrzFeauSmoLJzqIc9uJp4k63OHEsT5E5TaB+Lb58TVVoL9FegasqiFONk6Jz0mDThO/SmhLyc0X0wm2KximnXAU5OllhB4EGR7AQmJOgrYrzB6oAS3IaIlBMwa9diZ+Zk0hpaMj3c67mz/i5xC0B6dvJ8UE+BFUdNVtwETY3IUiKRClDt38h4QmigNKT6lhauCYMWZ2H5zfiZWbIfDMBPK7gXuFS5fvFrBVoEkipgjrJuCgC6KajifcEtDQ9GTwPp7U4NlLmqs+9lWOOnO/GwgTQ/e1bSPjK85LMSW1Nxslc3fLLVdTVpyiKtOsKTKtATHWGLm1VrYkTInQRSn2eLMLdOR4MINWp1NleiIQk4FrnlrD7RZyoTqQTgsuKQjopFLlTiFkrdnIp8el1sYoJgKl2JClqI0Dmsh0ZF4VTO8SiFghQZFMsaQkgKw3uZFOkpae+oUhhOOj+juX5F48SV/od6POcIPMIaXn/Sjsxv1i/YFL/6dWSMj6DNqIQDTNqy+yU5RhwhJdxtXVD0Qh78RgwHkPEzDM+S/0GYlvm0bjMCi/vguF7sM7DOCOWrIsklSl6qj2zbI0jWmOPzbFYtRXa/CwHMExRvNeUqUiqKRpiuqJX6Uk8agOOMPIVj7jFhjwaw/ZKTIMYpKMwq9XQdqghowVkHWQUjCkncifPDcTsi+EVjQFNAnKhSI6QaTRu0SkylsoxCtshHoAxrXOnG8ch2rTaYlsNs5cZt26s+JaW5oCZU3xH+Np5IKaoiJ1EmCNoSUNtiHPyyKcNG0zMrmKZk4IUkzG8rDFgaZyhEAUU4yJb4yJTpFmJobu0k5pwhGFzgCMLrRwxZwq7+yK+st7y+ro0uRjHkN1b6abBlqlPTXJSKHOMT0M4zIZ4nJqgibDSS6xB1MuQ9kuURi2z53i88aLTEn/0v33b/Ief/N3QcKL+bNVvvsxXvrufRx7/CqvuvJ3+wW6evfMHvP8LfWx87lnu/tEVfOume7n4uAYqU4/nK29/Bd99x3w+ecuJXPfas+DTX+SCC9/GKYc1s+Xjn+KC73/RWXXMnXM8n/vRrXy4Y4SjTz+J2Y8/z9iOe/jgdRnuvu9q1tz6Nd55xRbWPPU7fn7l4XzgKzP50AlZLvvU13j46efoX3M7VL7uhHonzro8PsjXv/stbrnvSR49/QSOfcXpLF34PHd/dgUH5n71BSutL81fwptmR3zmE+/m4nefy0Of+RLTvvAjvv3E71lx5GlcdOZTnPvq9/GlG97MZW99Hc/c/DE+//XdPPL4v/LY3b+js6+fuVM1sIvF6u78iSb8z/P/z66AVZSjOMYoEV1ehlnGYzqww8iuBh4dtWwph7xsJHA1Zlp0KlLl2J+aczQHHZOkWyUqbDo0vyb6bS4Fq1Ss5tc8zFsM3aM2yd5p/laETYruGhQV6ai1yfycsc4btrQFCt0ytoZelXIEiVaeK7kS60xuEHNhcR/8oCcRQD2mwTg8ogVophFEPg8lBKx6NzlViDGXM5geS+mAfGMT4c6mnJwCEnFkxxpR9k44QTPPhNKEIlgKrrQYQlkSNSUSCroMwg9BS0BkLL95sosf/eF+bn3i8cQ7bXIH7S1vZCwzz9n2mFjpOhcSJMhM5aWpZmZ5AdfHFcKgkXyUIhMXMek2ptYexZ7ClkTuQimtdFsykQps2QqeojKaaF19mvCApC9qXJTFOokT1WPpP4GrYrJNajJ+0IgtdxG7aF/BAbZE96zkQILYrn7N4W62EonBHUesR6X3ohEHGhU9lPCwxHoVYfMEuiRhpIiOQKmfd2nYuCxNCIVYBRSUDlVKVyhZRf36YQSIlJJV2Fe6barhixHjRGAmLu2GUPutgjeZ25ssVvU9EgoWODTSq0uyGAloEuNW6V0BGYE5Aaak7i2XC+juHOKEM17GGaeu/E/f8WecsowTTlnCg394nprJNYwXdawE1CXCwrJfk5SKpq0kQpgAO52LULRAz8EMYG2vPL6+rxtN2wiMqf3qZQJoau7EZ9WUsvqSjqv3BRj1HW3nXABURFWt2XOrE0ULk4jsxJ70WyQR+1KSus3Ndb97IqmilHFi+6aoq35rpdgFpuPKEEbXX/Vb+k6qjXRamnmeA1CSyYlKnfh+HW5NgJ4AACAASURBVBUisl6OfXERJYcJR5kR1DtrOt0PvSbFIBEzSCGYdZctcQ4By4JJPBIXeZaQBhtzjEgXNuRf4wpn2JiCl5FFKgOSWFG0MZY/jjxnh9lv6rnIeNTJNst6bInGmS8ZMi/NCWJ2Wo/WtIj6GeaXoTOKCY3hW6WIthGP5/OJr+rMfYa2tKUwbkkNegx0e9w0EtEqi7Q+KIjkVUgyCJ68YstKzVrEeKfXOoe+vAP9kJlmaFBwWZUDOUupB1L1hnLOMrnO8qolHg+nYk7bNMaXcjmmeT4VYxlOw7YRaFbNX2jx0xXaM1kn6bR3wLDE1BBlamiT4EA0zhdLfc73WHW0/VGBbNDJZaubeMkJh7pShsXWc9ULkyKYVwpozcxgbbmLVDjKAj/PmrhMlxYXwSRHnHF1rqLqqA+5dMbfDcX+w/vr797TRJXZ8Rdcwcc7v8PVn76SSbOXciqGp2//Nz5xw50sXnQIi6++mcJ4he49G6i3Dzh7qW3jMLu+/gU7rHkLFjOlo0Y6pCw+bBH1qdDNU9s2PEFxV8wrTr+AweFehgo5vnHbFzn95GO57s7PwbF13PG731GcdglLTA9r1m/lpMu/y3FLF8E0zRj/y2mmTZy1braOWW/i/JNXYpjNie2HsuuZ1Ty4+1A+/mFZac3ishs+wqUPbOTC6fNYMOdMAqWXpzdzwVlnsXx5nTNrrmlsYWazlFRnMnfODKZc+GE+svebfOkzV9HQsYCXig6kYe6/ofbaxLX8n+e/9wqUpQBM0QTOM3A3SQpDq7yXWMu6KCLKRRw+OUiYntWMv5if8Vbj6kh0JGXNXE2cypcybuaBvkRzVlptJrKEjdDbD5OHLdnZUNcOqTkGK+mTeuWSDJ5EBDuMA1LDOyz1DQmjLVeAxlaIpxlKfdYJIWfqjUunKqV6oaQOAuiQgkWvdZjB1hmcO1WvJad9VjN1UY/VyE9uAKZk4WdRyKc91Z7J9kRpQ4Vkkkikmz9VTydtPU2bunUUmdTtq1ts1OJHHnHRsLm7k/t2reGd373XgYjZ09Ic4Gim1Z1Cd9DEqGqH4hLlcIi6TAdFk6E1aGSOCeiKQ1aagCCo54DKL0zWpZpmp1rYocL0cIi43EPFDGJs5KJXE/Vnfu0yYtXiaSVdrUuTREUS4anWZblCft/VRsXlLgeAHOhx0TKlV5WeSyUkipwuVAovM5NIdWwCg+EgkQ2dZ2tKMgeVfrzM9GSQD1qxFQHERHxY0T3jNaDjuMlA+XIBCUlyCKipLkz+qQ6bKM07kliROZFizwE/1e4Zuc87wVPj6vcEWhIG7//D3ntASZLdZb6/eyMifWV539VdbafNeC+NpJGXRsJIIEArDUiwoAfsvrP45cGDhQV2l7diFw7moZVAgEBCLHLIITuakTQzGm+7p3umva0uX+kzI+K+893Ikga91b6BeWcPB7rO6aqurMyMyDD3fvf//0xWzfPWHnKI1ZcqlgLCUsF4RppUs31A5Muqfc6fhdFBS31plZ//yVsZGS78vTsWqnKpEzQyUuHnf+YN3PWFn2NsaIDTC2r5CoBl1TKPxTzC6tuaeCCv86JrXQBbylVxCgW8VBrO0m98RSxDb9lzPb9O29TpVVWzX2H1VTtdiPqDQNxmJm8mRMlWKfqb8UDPX8H+usgqlh7seSAoUNh/LFkiXv5odp4kzLEFko5K52rr6zRueB5fxQZe6Wpz0wS2SMdGnkcXCNwlG3Q7a54npwVA0nwcE47hwkEKApDaTxwL4QAHPcA0xDjqnfMcj0YZsRGzJs+DSZ2yzfEKW2Aax5+nLdbjNaajUf6VLdJ2PU72K35aO+4n9OvNLyVtluMV9tvI23M8KnsV53hBbkJJYrzdGOZxdC3EBZjJ41udX7xouVyxXYllomJ5aMnxkcOOtwzBM7mYo3Jtcpayq3EmSankSxRkatyGWpitLwS47BB0xABYyFJkeu3M+Li+6Fg852mxTJeVVGHoym+24AjLhtgXXR2vvCbk02mOu47LD8WSWsNJHJ/C8ZYArnAhv5OWOdQzFI3ztN8bQofSMlZ7sMuWuS2X405jCbsLXEgbFEyBdu4o/9dfrnHT4CCXjcGE/DTFKqmtsy9XJrEhj+pe8eE+Aauen+tQVVaUCn+NamD0150WH9nC1L/geXx7zqAuI5imHH3kAeZf9N382E/O8eL5fUzsu5Irb34HP/BHH+fHX3s5n37vr/Cls/PsW/1ZfrX7ayz93g/znlO/wd+2e3TbWhnCo48/xhWTlyvPmXsfeJTrK5m74HW3vp7b9pcx7uPMb5nl3sUaYwqyVLdmdBh6o7zl7T/C0Xs+xVk7y1Y7xk/+4o9z6Ge+k43HFSXm7xP/U99sEHH2xPu484lfZUvzIHeef5Jf2rGfqypP8jdfvZfrdxX4+B8oSuujRFHEkWPn/SDSjgqk6im4pr/wZLDbiTUTneDxJ5+mceg+dr3kTfzvPz3HS+b3MXrlTZRbczz8xCrf872v9RPj13fi0n/+WR0Br7gLCuwjYAvWD4qHnEQ/PdouZtoU+Fquw7hyk703m8s4ZIHBTmcec5zF56H6uClV3ATs9KU2pooJAoB9RdjEqJRmEC7hPaHGblXL1iHzflWO9XwZiDaPwcUnHTPjhvV1r09CEaZFqWhLIHraUCELsRYYG8wZPttzTKYwqeQocdzEyxfvXlYKqqwtwMA24/k3tTVQ+l/XOH7NaUIazFqswgGqImYFjqxjdbGfMqF5Tbso3pyiVYfEA7ScOr/BX9x/F7/wqS/4eIqtc7O0xLfN3USutI9iMMBk2qZiiyyakAPRCAtJg3WbZ5/Ns9dCN7XeAmKEgJwpcIevynQJbIW9yQbHSejlpxloH/PgSvmeRFPEaS1rjdmCb1F6wKakBPFiIokhMoGDDUd8K1XAUHwYH3sVjRBWb/JtwyC/1atdxYVyEkLIKsXkvKeZJj8XVClHIwwYy8XuIjZexXVkhbGRqXDVDlRFTifDE/pbBKUDmdozXslMkF2nb50iANlvn/oLRRWAEKcK3dcfD3ASi/RkkVLKBCFqB3quncCdjM37E4vUs1YtV2XY6gIUzhHw074IyvQvKr27UZVMoGiEsdFsrN6k6mQvfG7fN18zNqZKo7JpE6xR61x8RgE0Vd10MQlI91uqm21lASuJDzyPUK1q8QhVGlYpQveMbhr9qm/6kmCln3Orx/pt9s2qoD6nB7t9yxtb2I4t7vWA2x9XExDHS0RebNPqC25kHaN2qtqobQ+ytL8C5FIB+cd9VUZy7pgo6BLr+Pr2dcpaKnsTQzVXIjAhI0mNti37MSMXVhizedZjJZkMk3ZOZB57cYnUn0sBcflHRpxM276S1w2H2BkUWErqNJKUaSyHdN6SlNFwiDN+sRlyR3wBesu8vbiDaxVPl/ZQHkjdG/I6thjDbFDkQjjm25BPpx0OhCMsSpXrYl5gQq3nKATG02NlK3uxm5n8vjDnGw80yhKKwegC3DptqArn9EK2p3AHCR/syNy5TWwH+KuVCq/ZUmbAWioR5HRJJn5Y8L6YPkWn5mh2/G5z6D7H/DYIthhs11HZbnwKhj/fPYOrOJI5eO1QgaGFHh+7q8d3dXOsGJV5lOHrmMkHjDdX+Wgc8O25KmWTspLCwURJIPJ1TxkxhrfnZ/mbcJjTrcMEuj7zBQ5Fhvct3Ee+OMMVi/NMm5gp22GrFnX5Ab7cbbHFRh7gDZDj5qTJvXSJ4qXMk1KXo18AaPG0OdBnV+k/9PtzB3X+xrCMTo3x67e9iS+/6Y1sec3t3P5tr2BH5SZ+7Cs/wItf+gpq0TY+/eFfgBPb+JMf+XV+4f84y/FnxtizPyU/NM07f/YtvOMN/4arn/wob/qd3+HF3/MvuOWOe7jncx/hR9/6Vv5ia8Cu1/8k22bHKY7OUpJBLnDbD/xHfv7gT/DDP/g2njmyyHs+9iFu3XMDv/Qjn2D/vtfx5tum4YrbkBBp80su83JL+Iv/+iu854//jJ9755/zkhdez853foYfeNuP8to/y/ejtF5N95mvcsMLL8eagLmrrvXiCK28Xj2z1fMlprft4I3z8B9+8/d4x1tfwJ//5I/ypRtvYey1P8g7Xv8izt79h7z152p8l0BdNuz562pzXy79/OdwBDTh5ZA0ayVtMMWgb73mcZxzlvtIGNAycyjAnbMw359s2gazp9/xER1LrGXd4xIOSMEq+o13bXAY9XJ1fQkIrcB5CQg6jqkZSzER+b3fVdJcNZzx3dmA0zmlPcC+7Zk/1Op5qAxDYcN5Pp1uFM2diuRZaUGvBzfExhuRKoXCKe81zEyH1a7NrePj61LN43VDvudYWHKstiwHcBwsrOOCOa9odRI+CNcoMUITizCA5uWqIXlS7tfqSgR0z8d8+s5n+LUP38GDpx9haHyewe0TnG41KFVeyJbyNZxxHRRAWAwqyMp3OO6x7hJytsxlRvpTKcwso2oTuZSm+ERBkTuSBlcpVBtHGE0QtU+xOxrgisEXcLh9mod8kkSBguuRdEWWF7BRezbwk6iX53ZbvtWmyknSPpaR88MRX2ELipf1QUNAqiqgqj/i6uW24pKVLNgeQ758gGI4woYHpXmmXcpZX9VVWousRvpiCpHvvf2KKB2D3kdNLVKBvs3Wrzh32oYXXPjECwGifovR8wD1u5S7qlhpHFWFT9UstYd1gWSVRoE5D4QEijaBkuedKRKwX8ny1QThoQxAZSCrx8hAwKnjC/zUz76BKy7f6a/NbFvZdfpcv2++5orLd/FTP/tt/Jf//BEmZidZlJjmWRVC3xL24FKVQl1I/Sqa35A+oyqpArg6DkIEm0BOwNAjOy/A8CDWlFAChDiSWf+uf/MIIKkNrwqhTKmTFd+CF+9RPTY9S7dnTU1Wn2ureLi8r8CodZ3GG1ippElIxKnsnsR1BaiXsaUpknCYdqAqW56ia1FzOd5WTjntOnxRIcxRFXodKZMgPydnbWbyJR8yHxTmqOfG6DaPEBZ24pRG4ZW3HVznnM8m1TZzvWVWTci6fPgkPNciQlg8LFEyBVaCCp2kyWX5WXrdi/xJ8wi90k5//hsmZUvOsqdrmTGWryUrnEhqTOan2Y9l1Ob1yfiKS9lqDEdxnI4Nu9V8CODiOkzIcrJgaASOaWHJ2HFd1fiOgDoBTyeO+cjwTCvmRlvlvnyZrbbIl5dDPlTvMlqIeFXRcPUozIWWWeFWg1ewKsJMnisXVg1RzzHhDDllq+nEyLxO8WDi+Sq7eochXYO4l3JzNcRNO37tyQ4vCyKuzxs+10u4vhcw6+CptOPHh6KFMwnsD9UkSXkiyR7fHxWZSwMO5bZRzm+l3TnBo4FlT3KKI6sH+cDIbfxAbpIwGOKMC9lFib3pBR8NpJxoGU4/k9Qo+nV8Fgvm70N/jerazLDOc71nvtXz/l4xYVnhGZJum5rUL5UKBZ0hDSVJwsZGg3yl/PXHGrWahlAfLyVfqXxeNgIx3W5CVFDklqPT6hDk8oSBbBIa9JKUgeqA/5jdTgefWNBva2obtVqNqFih2Pc+Ug5irdYiX5KZpOwhIgnWvU3JI5/5S6557YeoN9+P0QqkXPr6YUu+OUoriZFaSxW7Xq+XbVcr/m7PK2/DwCI1bJIa8oXoG8egXKGggV+vV57es1Hltzrqlx7/J3kEut02r/uOP+YLD9wKW6e4mQF2YbjSKN7UePPOk501vnPI8r+9fITcZZpwM27apgrUCwlUxZM4StUtzR6am2TtoWKJijISZirCp2C48LCj9YBjRNQbA4URI8cKzLzatgb3hPPz1bkvOS4+7dg+b7yBpwSEw1sgLx7bCpxdxduUDI/6WiPdmqORGsaGYHg/tM9AbkwRqMYvKHtdR+1BqFxjCDec/JY584DsVgIePVPnoZ0n+JU3HPDKNq/k1efQoLvhfBKEF4EMOIKnLOw2PHBug3d/+Q7+22c+ACNXMledYbm7QdtUCSpXUsjNcpXJM2dznCDwWZdqOSmPURVR1fpXpYQ3InKH3iNL47zwcc05zhLzobjFJAmjQZUHeheYCIa4MSjTIaXdW2SpfZy7kzbFtEas8+KTJERul7hh2Ve6NLPI3y4o7fNgztctpZCVD5gztOJlku4FgrRDT1y7QMH3iW+deiVssk4ZS0vqe/mbGUsBkdRDKoXtdAVWlReruKjWsYwDp2376pz4fSOZOlP8Pm+TooV+I1OwCogprcKbACtXThGIAnaqGGUKRX/jqV3nFZgau7WuV91XHEEJJVQRU2yY2o+qZvWBky5GvV8ksNO3Zkm7TI4YLpw+w6/8+9v5d7/0Q//z+1qbeQ7z1q/+2h/zK7/8PqbmtrCgkom+xG0L+pVHAU+/P/4kZZUOL3QQf1G+QMJ2shqR2qZDUNrvW56qhBIIrPcvRqlRdXz0Xh5E9/NjfUVQYDprobvOMS9EiYZfB9Gon5yLQZm2sldNxLpAZG/Vq2kFqIKgSEsV0M5Zz+WMvZBFGcGLpPl5Rke+nW2yvjDyUUuhvci/Hr+M1WCI8ytneOlgi33DEZ89eYajzZOMzyQ8uVThCS3kwhzkxpnSwiKtUcvJ2lZq9zF/rhKTY2tQZjXtIvh/izGspx2e7pymkcp/T6e66/mVasXngxJXRGMM4HjApXyHxOa5YWZzIRfaXY5i+aP6I1RdzHDlWq4xITOk5I0Rvc3vv0gK341C7eGJFF7vT5Phorwkh2F/AosNUICUAnTGZUy8Bg8Hjr9qdRkwOb7sYn4xDDjh4K4kpaPLD7gllS1JzBWTES+MHVsPGCaqgbdweuaYI2rByDCU9xudGpyPITS4FYerGBLrY1qJH9D5hG7B8JVWl/96pMdtF0oEZcex1PFxGYa7LjeGgzSQKljuBcarh6Ug3mtyan74Mebj8Rrr9Ue5qXWYF5ffzjvbd1BIT9LJ7+b1pV3cGY7yjqjAsEv4Pxsnua0w48et+1IlivQ4KTX7yoe1/PKLhLJdY+N8j5e/svu/PiZM94oGjCBXYMjLev0DfueUZzo4nMVJbZbSy5sRWboWfU1QBzYkX9wsEEoJoxsse99iuew51dnvihHT1ZF9+fcMgq9HkOl3fZkgoqog3md99f/E1M4r+MiHJsgXI0I/o2TpFHpqkC9SzavM0f9MQRYPpt9zm5Fcz/q/3jPM5fwQqDvj2cdA+yLblucwXvntXfr2T/cIyM9sMCxTNHk6acqVNvCB03Iqv0jMLHkuH7eEyymdI4682gWqWmkE02pUt4Nuj8hkLdglVd+0RJUxsctEFQJt4qwvOybH4FwJNg6D+NaVq/rCC3XKxIVbc9RSOLnk2LXFeDeAhnEMSn1WMPTWoJ2DhkxEK4bmRchth55MhGdgOID2RSjsNt7pois3hkaW+ZrbkoG0dCRz6RidN5zfcBwXCEV2R7q5NNc4b8Xiwajy5IfBTjrM0wGLw23uvesQP/UnH+YZWTHs/A7W0yqnWycYCEcZG7yVZjDAq2zeK/LOiy/nYl+R24blgaTGlTbyLSOBpRmTQ+Ecn3HO82tn1N7SpKuKGM5PQi/HcVM4yt/2LjKRthmNRjgbjbEkQ9fOCZLmBYxMhZMTXhygiVpiiaByvQcBqQ6auGf9ipFvU7bXaMmawnXJhaOMRyOs9JZpSRShRaZEDckGnaBKp3lITVxCtdLiNlFxnsH8NjZUMdJ21JYT6FI7RkrVYAepgIjsStTeC0f9/oh/lc3SDheIp3M6C6H346LafLqoJAYQMNO5SPrgRZWstufN+ROkz+FHL/0UA13AbrM92W99+halwJyMnPOSrfqXZggqG4uzjXzr767Xv5af00C5+Z76DHqBuIFqv+qzZJ55/rPrM/aBqLeD8aIEtca2emFMJl5ZzKqdJdmJqCq5jJE/oICcF5NonnGeW+nb5Kps6rgpcstXQEf869LuaazN000atJMmUThIU55/qrzEq/7cF5I1OvEG22xIXNjGWVXNBPplM9N4kG73NAPtI/zI6Gv4ahyz3DlNoTDK79XbkJ5ge6h4uFEWaiFj+Qn2DV3BzhtKPPF4k6cvHGYiWObTtbM8LqO29hKEJ6CwjUq8xkB+1CfYrHQbuMI2ZJcrzpvU2dv6ubF7gwEG4kWqnZOcjC/ysaTAA70VdmhODAr8RdzgQLtINTfDPcTsCMrstyUOKgnHdb0x8dMYjjvHDxnDinMcdbBhHKORYbHneDI25JzzfDr5ZkaaO3uwJmNzb7NoWDVdHmk55m3EwbTLm03EVZGl2kuZwzKcN7SMI58YjhNSO2/4ZNDgw/cZfmaowA65Qp4PGNwP1XlDd9GRRNnQaVvOXyGRFqY16FyEyasMvaOOwDheuyvHxEDAbz3QZnwlx0AOtroCJzuLnLJF9tgSJ13s2THq3JUIKZLSM5YPxQ2qpKzHltxIxPe/xNJ8oMgfHHXMVKp8orsE3Ys8U9rDy6NhrrcRF7sXGS1s5TQdz5oLkppvvX+dz+oVcBos9e/5f22iq+f8Tr5cqJWgXtE3rd3clQxoZYaQ+vPXgZcf3f0L/HYEkDYf0nP0nvr37Odv/t2/wG8q28o3b2NzO1kZM3t2to8wtesAb9h1oL/Nze3037H/GTa33X/0W/74+v7og+uXbzoG3/KFl/7wz+gI6BrtkYqsbYxfzY46w4ocPSQk7Bnel8b8cKXAhkRuOw2FvkjRC6BkYSLA5tc5QkMZkPIqIpn26j10n8jjTdeh0iZKIFKzcJ9fi6j7JMGBWh4b0K6BOjByYB/ZCqIbmTNqV+BX0VEqTzrYNgkdLeArMDJvKF6QwSrUT0F0pbdV8yDSz4UqTBUMYcfhC0G9jNR8cc1xvmc4vwIVJWJITCFy/qYfrnar6LA9Q6cZcO7cCv/h1Ef4o7/5MkPbDjAb7OACY9h0jWmtmAdfxB7FHrmYpjGeKijbhZtN4LlGqkV9XzjAnWmHCy7lt0OpDq2PAnqJOjEYHsOx6By7w4i32Jj393J8xfV4gY24Lihzt8mzL+2y5AftkHJQ9RYR3m5AlZrcVor5LSRhViGTWs1G0yTto8Sd096qQoOxQF8i7pWp0kk7rEpk4MPVL2KVUtFbout63DRwI+WR13CPxj3FjiU1uskaC62ndHIRUJPliKpH3jBYLVy1eEPliq58vR0o42RvueJ99LSodR7I+HapKnsemKiXH4AHiGopTvhWoq9M6XHfku2DPg/81NJUFIkqYX3vNlV1vHde1v/yYgQBFD8BWS+qVQl2o9am3Y4pFCRq+Mb4/uyb3+Q3Z4pnP5r9f/M1eg+9l6Znedhl29F4K7DZf50Hq/17QGNxONRP57BeSepTPHoXEQhTmTpTMFdw8v5TtTGp+pB71z3jPeVkGO49+yToUWva+9KJI5dl7vqUj3CIpPGQ9xDUebHxoje1VhxY3D6GyhmT+W08U9jlq6+yvroqHGO4t8ipeMlXYNNkL8PBICcWP8bB0gjfVX0x54M9Ph90b9piI21RC8o8nLb44PJFlqIC1GL4dIOXFYe4fOg6ikbX7Qb/prLuxVC/cexJ2qtPcd4+Q/20yvdFprdMeRsfReOdikZIZMlh80wFQ2ykHfYU5tlZ2eVP8W+kHerxMn8cr7HWfJqxpM6TGjhwRKUrGS3to+6lqBUKJuRkUmfWFrnWBNznHKPA60zgBe1fih2TgUGMkXoPFhU7ncCeBC6bNaytOe8/+EA34T7TZdAGbCXgkbRJFATkneVKdew07lj4SOI8aLxJnnp5JRJW/Pj13xsx9RCuKcUcOGJ51bhldNhQ0eWSOJJy3wv6GShr3aNgFq0FlKI2DsmK49qRgJ94heHGd6/zorES1wdFvpa26HVOky/u9EOtxo9d/kp06O4cco432gLv1yDqGrR7juF5x9hKCR7YysJYAxG+cr2L3NsyXBNcxdXRCPdLLUtIQwbj8RqmK/lcZrej+yn7kt5g8wLvP/QP/PGcfer+zvt/E5jb/FsGkDZ/0439DYD3jUf/7g3vAVj/j9/q+X/3tXrPZz+SbefvPpKNBQKA2WDxP9iPb7Fv/6/3+eYHNrf9LY7BNz/90u//fI6A96n7wFc5fHKOcHiQLYQMOcvOIiRlw2frdc7Fa3x/scT07oiy7EckIpDoQDyRgs+Cxk7KXFM2JlnVzSnDVcsvRUeKmyYvO80i8k5VRFfR8NgXHZOXqbpmcGWDqRjfqrWDxq9URxNDKjqUeHYxVMcNuRFDrOJNHXIK0W5lsTyay5T7WNhqfKaqIEO4RYsumXxD1IZ6qrzHDFCibSUgg1DbNjSWutQuX+OlV45na099DhVDBiDoBbRPJPzuhx/h2//qAzx8+DA7t99EPbqRDTtI0j7ubUKuH34NQ5L/u55vedybdtlmAq63gR/6hk3INSbwfJ5C2mGvzVMyEUrrlHRhyhh+I22yhuEBUl5mQjbiFg+7LtebHBdNSCwhS/soC71FbzOw0TpEp/EYI9EI1aFXcXnlBoZskXPeo8qS9JYIVB3T8CIwJaKjELOsUWR74onxdZ8A0YnXaPUWKJIQ2SKtaJyJ3AQzpb2cDSose6uRkIIAhDcildeZ+HFnfKs1jeUPU0eK1MyWREhfYo0xf0yz9qnyfFd9BVAEfk0K3lNts20s0YWR5UbfukQDpyZoPzyKU5e9xqdQ+KMqqw1dkO0sfUHgSe1KZZZ6EKgILk08UsKqHevo9GDLTIm//eQhvvON1zM7I0PjzLfNb+g5ftt8zcOPHOGH3v67bJ0f5eJq3/DYK161OW1bM7cKd3m1Wzzv0Jb2+qxc32aW8a8XfwgUj3nLGnEjTTSYnbO0hq3eQq+0n7Swg1CCFilMFQHWO+9FGF4F7BfuQgKpb8MqC9ZX+eTnI885VfHSuufO9aJJtlWuYl9+KwPhCGMS7qDg+DpbwhI9U2DRGB/w3ku7FF2bL7cO88rSAXJ2kHETUzARiSmxNRdw9f5c7gAAIABJREFUVRjyqmCIKTpsCcr0cgUeNnBfu8mdrTUeSeoMmkGejgtU7Q7eMfgCBu023nhtiwPzG9zx0IM0ch1vXlxJ6gxI1GBytGzEE3T5SneZv+muE8TrdAXibIlmUGVfcQdjhe2Mh6P0JAbpnuN85wQL3XMUjPVJFue0CPHpKQmzxjBtIgaIvc2JLpfTRnQH5brDFVXDngnDYxfhkTZUtGg0lr9MNng4sfyrfMnHrT2MZZqAGyrGpzXpLB8WZSUxvC7KlKqyUVHFTybDVyuQwBlGEsNkZHjXWfGEU+qTCfmGxZ6BQs742yrYbrBaOC9AftL4GDMtoMXo2pLkefVgl4+dXWeHLXuvytWkxmM28lZUHSyPk3K9sWzBeErHWWN5UgudxgluKq+wLbia//KkY6la4rbcHIPJGqd7izTCYR5wXcqkPCmVr83TMxFLqsq1n/at+KzS3yMXxHTqAdt3xNz+1lf/r/Ope4735j+qp30zyPxHtXOXduaf6BHICOizWL7kHNdbh00Mi2uGmbTJm0yLeVekEjqMjOdnDEYZqfJ8Fbl3pE/4VXfrnDhq/basfqpAIoTlvevwsV/qAir4flnGnNsM5W0ZGHTHnFfLhidT6ueyCt3cVTIwN8QnHK6UVQI14OWuNvQec17HIMNhT3pQe1aM3q1QOw7B5X5+Vw8ZN4p3TxdwTBeUTJFRurpDypg17CnAokqDmhNlzK5OpcCjsTy90ObP/vpOfv2+P8XOXMZo4TWcNsOe82o6J5gISpwfeCFNTbRoAM9TE8k4jXmzCfy+nZPtkMP7c92C5T75dYkMrpA2r8lwvkj4DgE3E3Bn2uVgamlo/Z0so3yGFZHIZS3QPoGVfQiW7y7tpTjyOg6Go7zUBjzhLFfkpknbRznUeIKCeEwi39uqT3VQdU5mxT7AXYR57/umBWTo/wX5WbYWd3sVbkPec351HiALk2tsjq4mcgziJBUFYLttbzoswJQ2n/TRUkZ2JsmYTxFQ+9a4sjeztV5JmeDkMePNgRuZIjbp4NK1LCNWCljPpVPlWBdUpgjNxAQ6/+LRSRyh/r9asurp6/fMBkWrZxOOZisHL+DYyN5XsFJAVJzAVCbKao/WWFrKlLLPXqg/15t88zVLS1JGyGh5uF8FFK7qAzkBS23XmzJn1UnvqScg7Hlwqi4WfdvVr3ZEQPVm0eLOVUiVJpKfZzA3y0CyQS4aJc1vpVm5gUbnBM32SVzjIRKBO0XBeeVthJOC2VcyHa57Fpffgs3Pe09AtdAEjQ+7lF7aYh5LJah4498LJuR4GnO9zfEiCQskoJDZtKrPzYP89oWP8eOzt3NdqUgj79jZcKzGsJaE5J3jOjPOJ1zCiyy8KTCcsQkd41i2dR5tx+TXUuaqZY4HIUP5nbx2fA/NeJ3p3WcZLS/z4+fuRrx0f4MOTFCKZhn2ljaOVlzjnqDCPTq/7jxz4SA3FXfSDMp0o2GahXmWVOHtLlCsP0C3e47EJRRyM+RbT/vXng7K3CrD7yDHdxdKNFI41U2UYsjuMnxxI+GWhgRLjlrL8JAxPJKsM6h8Wd3nacpj4kISsC+Cw+r6yzcuze7l62UY3IVyaLg6b3iy5dipQcE4btQaVwk1eXi7qoJLjs980PGVvGO4DC+dhuqMYUKL2ZbjYhumlx2x9EcyZc87GkcTduUH+d09CW9/4hyXlSeoaAHjelxuK7Sc45NqI+NYVuWSzMi5bPM0cgOkzYTl0wmHlFudG2UoGmcwHOCB3iLDNkSxcF/oHEVG0p+u3kLF5D3/V4rls/L381VxXT26Lzcrds/1jvnWz9M7Xvq6dAQuHYH/346AJseUdRfzMmUraiBINCA4znVTXj1dZHoia42qOtarO7pnoaRiSA2M2AK1LHnBIxiJ18RDkRJ2UhU459sJpgLxUfwqNIyF8wyhxgWxl1XNE4g651hbhlwMQzMaCw3xgvPRPeK41BecT5dIG45A49KUjwX1bd3qWbBSuEYGm3d0HnOU5g3RGMQlCNSble+t5nEjPouCyeVVpfaHJWjnM3rWiLg6eo7l/kdO8Ipf/yi188eZnN1PM9xPLdzm25NB+xgd12W2ehuviQYJXcLTzhG61Ece7Q0KnjenqDUJ3PZmIRVMYmgmbfbZAtcb44HelQIjapeYvPSLXIPjc0kLafb22IDF7jnWRVqPl3mBVtL57eweuIHX5MaQtGshabGetLlAyLjNEYcjUoTQURWxdYR8Ybs3ifW5oQJE3gBYulqHs4rx2ok1Bb/yX5BTv80zrCB1VUZUcDIBJ9MOJeN8q0kgz2eJhgMZ9y5e9/YlGf9rzQOzpKkA3Aa2MI8x8qJrY6NxpOgQz07VJeNBiLTWOv8Ccn3zYGv7tirKpRWPWHYlzaxq59MkZBS9kfneeWsTVeGUoqFqZCac8AdUYNX/R9w2fVapYxW7K/Re5T/91qe46cb9jAyXPJVmE6j9f91amxSclZU6/+k/f8SXpJfW+iIifRqpdMOMb+iri9qoT2OQeZk6NZ2scqcVjkrXtpS1Y/1fBVBb3uTXRhO4wm5Go1HmohE1YjlLStd1icNxugpNlmil9jXvl5E4xYTJEmjKv5+OkZI7fLvdp0Ks0LNl5grzpMbQaB+nJWscFzNnUs4T0rI5zsYbPq1hKL+Fwc5pzuR3MJG2eLJ+Dw9t7OZA9HJPJ/Ct5h6UQsfVA7AkOmPP+AzluZ5lCcPgqGG2NMTnlg12p2N2yXKolfB0zvGXD4snWGbr8AGKts5rJ67kBQNNFtuP88nThzm+/pmM3F7YB8UdPj0hVVvf5n1e7Kd6SxRsiWFbZM4WaJAwW9zNw/EqRosfWbiYyHNTJW4YTxqc6J3kTG6Cp5I8Lzc5iqZE10bUugEPJD3OCVSZlJuCkM/1Nvhgb43vK8yy6lLuShRNlvi26VghZDRQigXsVJRqomQbmClDtY3n36k73rBSqTq2hob5Iqy2lEyhIq7lDeMBi4njo7Fj4QScOpjwbdfH7K/kmZWCX8+NIW45lrR4HYDKQMoLB4f51XqOt55pcqVyCk2er6rFCtxqIs4Q+GMxIvshZcVr4RMnSIWbznXhfMBkOMjHkzqvC6tcVdjBo62nGShexkBxD621zxO1jtIt7aWebNCTrZDRjNC/t/wdu9kC9Jt9Xt8ugbrndfguvfjSEXj2EdB0KgUdnKfHteSpSO01nIXSd3sdXKlEMGKwUeaM3lqBuCuQlPHiAgE7gSDFTU4b3LrL7Et0z7dcZm2iatmUXxxihg3hmqMdOtYuwNBOg5vWvGZwzzg2ljMbvMk9WrFCqFVwJ8uQHxhTgIHh4inHSACdlawV20xhYwPGTkDlZginoH48sxNoLIN4eMIGiihNxA3Up55ybF00NCL4RDPBrrdI245w3NI4Bl/97BFe8+H3Qq/N1i1bOG+2e86Ykf1HvOyVhAzcykgwoCAMXyUZIUbFSgGIqnHc4Ys2KXuN3OIMJef4kouZtBHLikSSjYLxxUVPEL/PT64KQy8RJnV6vVVONY5D4yDV4iRXVa7itYWd3OS6/Hba9Q75oQkJTORx8aF4hcdiRyWukcuNs7e6hbXiPi7U76O58VVCWyQKhykVd9E1oefaRTrI8SqteIFevERP8WPRlJ/oBQru7l30k6OEGMtuxQMLbdPztoTevahCFcB1TFjJUh08UJAH2ro3Lc3yXKX0VIq52rICLvWsiqZKolq7qjSpV67UAtf0ik8lZmgf5ATglTm22q/WZfw/Wag4IXV9Fq+C0Z8b/exXVcvEA9K/ggdVWYZdQLubMDFT4q7P3ctnPvc1/sX3vuzZN8Vz/v9nPncfd33xMSZmhlheF7+vP9GFBS8gEVjKfPO0L/p75PlvUqmq2iMQKxGHqmHe/Fm+cT53VebZOzK1clLjZG+Bw85Rci2m05h69wxjOK+aLhd2eJCv3GBV2zd6F2g0HgHxFFXJVMVPptSKbVPVKzebZe0mDS66hIV4nVLnFLvyM0S2zCuNZckWPSe05VJfpYvSJuuF7Qwla/zR4qeZK+xloDDLFVGPPdNygYCBimEtdrxx0FKagMVFGD4lBamjsG55Xc6RV85xwTE/ZBm9CItDRdbSlOPNBidMQhyq6lahF76SK6Zv5ueHFvny0llONb7Ghc7jHOkl2N4Co6X9/vo9Il5GaDx1QVSLrfKycwn3BkOUZCpr83Rch73ROGO2yHraoFHcTcXFXJYk/K4SYnpn2GNCdoSDbA2K9EzKaRNxrLvK3b0LvLmw3d9buoefSrperW5szOlGgKKlVxKYi2BLBBs2S7BZ68LWAuweM9x3IaUpS6UAaj0ZHRvMKIxOwNH7HNVZ+InUcPQcjEzA5w86PpxL2DvW5TvnCuzaEjAwCdGjKQp/UpHt4qOOFw1V+OmRJr+1UGfWrPFYfs4vwA7YAKnoJZYwztEmZdI46vIwVLFbDsfhANfYAg+YgGPOsRqNMyCFefc8aVDGFnbSqz9IPqz6lny3d8H7ZfqVt+euasBXG0Y/n//XJVD3/I/hpXe4dAT6R0CEbnGsQg6YkKE0UaYAyzUpMR27KyHh2oAvoqRrjnTEUJ01XkvQ6Dqfc6h508mUV90ldXKV8HQEP3BJRCFih0+bUBZs6kgfTAkqhsul7hJ3re8Rq3ZnaxROPO3Ye3U2OboN59WvAwEUBwyFUU8Lwup1BlYXYHQH1M9CvB1KYwZ5zLmmISo4TN1RHlX+q8PJ2msNNp5SzFcm3O02HFuKlh0jcLCWIBugxeNtfvGv7+fdH/skhWlD2cxyhhmi/LQXAMS9ZaKkxgurL2A0P0vkEiSQ3Q4+r1Id6ruBOee4ylhOE3vH920mn/GWMIw5+BRddpFjdxDyBI67lFdmDPe7hJG0ydnWSVh/jHeMDjA0/nLu7G5hLRzylTxN6FGyzLvTHjdHY8Qu4VDnvAdl1+RmKBW381UFcXvAZ+iGI+wMpJBLaYXjtMJhD86qJFRzkwQip2s1nnZodc54/7IgN+vblGUBDe/377w3XLZi72TZrYonE6Dy6lK1N/V4PQN3Sp4Ul8lniCp+ajFTwkoRK+CVkYczkCaQ6D3n9NObk2Wed7F6/D2cOGpeDKFSayOr2vkrWFy4PKa401f00t5CdmHootQF4luzmnhUb1R+aogzadZ+T/V4xPve9wVe8bLrmBiveouo/1n+q3ZNYoggsFxc3OB9f/FF/76xqjty3t4Edb6lrPKz3Pe1HbW3I4jGMYrlM1Ku6nxnwM63SpW9KoGKj4ITW16FySN00xbVeJn5oMyJeJ2TNs81xcsY1oRrcj7KqZI0KRhDvbdGmt9CmJshrd2DaykDQTeYxCAx1pYI4mWfwiAhS8GWMfGaTxt+LG4wl6+wlDRYSmNsWGY5XmNZ+xmNkUu75CRXbx+iHT/Ev5yeJm+VL+oL3ASiOISGwanM3ihfMkxVYXURihMwWDQ8uYKPGjwQwtyM4qwcQWp5sCyD7AKDKyl/vdzmynzIVJDnYHcnY6UdXDl9DV9ePsee9Uc41Fvj6NIHGYgmGMhtI8lP0wkGqGHpRkM84xxRvELPdSiEs4yFVZ5JNhi1EaVwmIfSLgMiEUTDXO46TORnMRL5xGs8k7Y4SsJJ8Q+B+dw0Myagqva+S5mxIR9IO/yUrncH9QiGhuCpVbihhO8oLKzIBSOr0J1cdyy1DFcHsEu+cm1olmC+DeUWzFYNNnRsrMLEMMwULbdERU63U/7kBNy9kHDVyS7ff6NhxuWJjzoq8+DKjnAp4ad2lKgby7uOnmE6P+Ovsa4x3kB+EMN9pARpl5qq1xLUpBY6dYgDyvkhlrBcawxT4SDHwyGeqN1L0D3vGTNpNEancRDj2gR6RPeoeCt+oPcmpP17rT+VPI8f/zChxPPY4KWXXjoC/1SPgBdKvP9ujh8eZGpiiqqtcsI5diSGtRTeWV/mR0arXFGJCOay6lznpIx9oSBPOUXijJnMU1XLLR8MK2UlGC3klDalVoMSJ5S3Kp7dcMYrSc9BYRgCCS40H8aG2hHHhXth91UGW4N0ymDHDQuLUOq/jzh8WrFqfJKHlLpxpUGYlLmeOhFtQ6Tsd4kyNAZZQzRlCGqZCEneqDl145Ra1TV02or5gi90ykwPx/zfH/8wf/CJrzA3P0wnrhDnD1DIb6XnYh9x1Owc58XlA1xV3OM5O0PW+oBxKXpFjt6hxDP5FmPQ1CBPsJuCwPN2HkkTWsbylEvZaSPWpXw1KRdcl88lHdSIPNI5xunlL/JdhQ6/PHE9Lx24jsnCJHe2Y7pJjZaL2TARg0GJRRLO9RY41znNiKwq8jPcKoAWr3Gke5GL8RqTLuaNuWluKF/NQnEPO6IhXuMS2vESZ7oXffWi5ZRcUfKWF1XldaYdmt0z2O7FLLTAGwlmCQ5exODbg8sesCjfM2u71j1AM2qx+q8UK35YWM0sRfSYQIT/W+zjvDy3TQ/ItkOOq944WFYZ/VarV8Hq+hDXTApXZcDVsharf6M4MzmWtYomLQ/mtAGBQF1UUT+cV2IKRVOpYqZyRUKra5maHuSR+49xcfkiL77lSiqVzNZm09XA7+qzvqk9a63xXLyf+fl38aEPfo2pLUMsrqlNnH0y/3QP2Kz/3J6vqOpkKKKmgGX/ea6TiTzUNkvkNruc/T2o+paxEyANSpRzk5TzcxQKO2kUt1POTRMFZZ5wMcfiGmvJBmH3HJHNUestknTPUgnHPDfBFPdliuX2M6SkDIeD5NMOamFKlZ3q+cpytQWKtuDbpZEJmcYwbgzjNs8VYZXLo2EGwiGOuB7jrs5n64e4uXQlV4xVaUYpwaRRvTRLeOnC+kFIi4aB3TA8CTnRHwZhNA+5GkTqBBT1kQ0TJcNE12K2wnV74aW5lJcXAhaxdNYSbnApjU7IfDhOUNmLze/k9bkt3NU7R7d1kDE/IHWxVtXekq9MKXFCi4swKJMPB6kZ49+vaOC2oMSoLfNyE1OP6zxudU4idkYjbLUFnukuUEs7TNiiF4h8snuBdtrhrM2z5FK/MNtucuzEcE0ORqMsJuyCuIWp73v4yt2JJnyx4RhJDVuHYWzaUCoZOomjGhtWLzi6KQwlhmUtaCMIliGfc2wvGa4IQrYkAU+vJJy5N+VjqymlgmGkaMmNGZqhY7BumYxLvHf9PLlwgOmwyn4MJ7wfZsxpEjre2iWm01lhb2GV7aOTfPxci1sLU6wg/mNAAcsXgNdHI9S7F1jxAi5hNkPaOf4N2gNZ+kkmlGixfUeO29/6yuctlLgE6p41yFz676Uj8HyOgEDd+z9wH8efnqI3Ocl+U2YvsNtm1bhVTvL2iWEGCnnCbZnjeTCh6p6hs+4IxYMbNV+v0vllu/4sGo+qaaJIxX2LNCEdjRNS1p6BXl8RGwwZv3o3g4bzjzna52DuSosKV8GMUfcIezxb7Qdq0dacN/IU7823U7vQUTzPLuPbrJ6zpSJI1VBf7VO1hD9KApvZhKp5deOJzAUjiqBeCPn0hSX+7Sf+iq+dOsjkxBwXe2XSvFpgW0jDMrmwykbvHKPBAHsr13nBwIIRV874jylMu+DTzmQSYEhUSVRrEsd8EPCVJOG9SkswhmGMhMFcZuBUEvNptRfbx1hpPsbremv84NQN3D51A3vzIzRqDmsdO0zE43GPzycrTMu7ywQ80b1A3kZs2BwT0TABlmdUiWkd5Vob8D2FbbyoOMGu3BBlcWtwHFAgeljlDhOwOxpnGymrvWWfS5s3lqFwkHZQ9jyuuuLCYvGy6j57VebBNj+b+cXZAOvB3qZ9hwQImljlTydCZeq5YdlJz9o0ivvywEYn1bdudD42X69qmn7viyHUFtaJEmDz9iB6XFU6ce8yLzpv5SHAJFDkAaGsG/Q+Qu0CkWr1qh/ue+IZKPQRYYr0glrDMjtb5Ct3PMTaxgYTk6OMjQ15Q3aBr2/+1+7EPPjwEd75W3/Je9/zKbZsG+PcggCqNqYy9bN/avsRxlfqVDnUPmSfwXtzbFZPwqpvtZrcln7LdauP1lK8lniLE9G4r4QeS9tyA6KRdsR08O3UoLeEzllb58wHlAmwthnITXizaymxVa10uRmqaYvVuEmzu0CU1hhUKzYa9Fm+qraWwkFulpDG5L1BccPm2cB6sv1J12XEFliOxkjjFTrNCzRp8pLZ/QRKZ5mG3ACIKeBCaK9l4RkK0PBia330XpaHGu025GRBtNswpMD7cxmHVarkgVXD1npEfsBw3TTsiyyrlS7rJsG1LFOx45W5EmOlWQZLe2kW9nCi06FZf5xO56CvMOdVeTQRqeLOggFSRYl5bp1vvvN02uYGG7ItKDJsC9woNawEEfEqn2g8Saw4vtI+hoMyB/w9BQe7Z1jpXeR8+wTzut+DQSIbMOAMS3V8esRxB1fJXUcU4xjOOPgiKaUAxpVp3TOUujCawLK0IEOwTaKzUFxaQ1o3rBnDthHDQAEGBgzdNcdMLqRajZSWxj0ty+dP9shXYsJ2SDdRLmuBZlLkvs4CiYmQvOkp12XctdliCxyXZ6XypjsnCNoL1N11PGnGuDossc9YT/vokrADy65ogEba41T7GfJaBqjC2z2fCaB0z/rbs+5byZ26QF2J29/68kugTvfjpa9LR+AfwxHILE0e5fjFy6iMDLKfgCkb+arSWup4onuet109TtzJEzZTIlmP5A2iPwkM+SZXL7M28bQl8et042uSE6hTYUTWYwJ3ojWdAqOgABnaS5Spx7cZQmXEKhT7uGNOAgpZlSUa7A0sOtoSWAxlma0CiaYKy09AdXvm+n7+PJRlIGodiv50RUMo9/hjUNYkotbwLiE5L3rEadLRa9fx0T0fbhzjXYv/jWq8Qbm8nVWXwwbVjNMk5aKLafSWuDLtcMvA9VwV5Rj0hHUUqMHeguVC6vhkKo5cFvUnoZvMRE77j2446+A6Y9mQESoJ7bSLhMPLrsnJ+n18n1viR/KX811D17G7NMYdGymVVkJiDWdj2ec5TpqiJ4nLhFWk76tyk5yxBboi0HdOs9w5yfVBmWsKc0zl5xgyEdMW9mlSlMWCNVxMBRAdFZvjHkIOhINcFg5LJEwYL7Gi9mu85uPKUgGJwjafFuHzS2OBO1VBVvuihNi3Yb2Dh2+pyih3IwNWsgnpg6osFkqVNLVwVzPenQc1Qv/yb5CCtR96L86OwIgAoxIUVG1z7SzzVcDNV+yUqJD3mbCeeKnH1SJ2qkvoohOgUmNQ3ACJIvqAKtvRfstX+bGWjY0eU1MD3HXHYd7z7s+ztLrMxMQQlXKJWq1Js9khjlMee+IY//E3P8CPveMPeejBk0xtGeXcQuIn5QzQ6cJ/9r+sjeytW7xqUJ+jvy9B0VecJOwwuRkC8QG96jVLiBBwDZIGnaDMug1JkxYuKNN0XSYlwFFKSVqjrdvG5BgKB2h3zhAK4CU1rggqdE3AqoQExR0M5edYcQG/Of4Cbqns4vPNszTaq/Ryw4znJr0CVbSLaQ/k4CGBPNnseHPswIuadAxvCCrcm8bMuwXuP/M0L5zdy3XjI9hWStzOIvoGFJUym1XWY9EbZUWns9MA2djpPrcViEJDdwPCA8bzXWd3Gc/n1Qrp0VXHSFfZqLB9KGTfFRFxO2X7oOV+urRbCduWSrxiaIQ3zOznRdXLabhhTsjip3uSclJjJFn3Bsttb4C95oFxYiPWesuE3Yt83hSxJu8TE8rJms8zXg2HuVrKqs4p2iagGxTYCCqE4YjPLNZw1HM9OvLnMxElJ4WsI7aGA0pacYYNA3/eSzhk4GFneLU1nku41oZeC4bFT85rvFIEmmEjhMHBzBFqRJesOhJaD1fg9DocjGUvAq8qW8Zj41XL//2844Flea4kDKWGFhH12OJ6JzkRjjFiS0yIt2pCTrg2UdLGuCbrrTMcDq/hmpxsaQLenTpeqfEq2eAoSt6wzAQl7lULVup4gWJZ5/jFyGYl3JELUjr1Ltt3RNz+1lc8b1B3iVP3jwENXNqHfyJHQF5uNWits5gMsyDOjYhDwGMm5lBawnVCxmc0OBuSQeOTFqw6bMuyG8m6XBqoNQ97uoWqZBJJaYqT/6sAnuZZFW9mwEkIcdixXoYpreoPQu56WbxDSy3RGhRmM4d3pICtG85XHHNDjoI4OZOGjXOOumLDzoNt4lfDoVS45/pGx8puHTEMzWSGw57zJ27fmJIhMt/IvOwChgPurD/NL3/xTxkd2qBT3EvNZu0vgYrMY60LvRXoXmD/4C1eSHAw1lIbrgnkpeU8z+oqDJFxvrssVelnXUqOwFdVDqepb8kqXVPqV9WSltImf9o8Cu0TvGfkRq4dmidHjnyacKjR89WDmjWcSZ23EBnRpOvafCrtoP83FPvkNPldwDWfhPwWfrh0OTfnhjzQXFdlzjhP5JajviiN1xqYx/A14Borj7yAdZeSDyq8JazyYDLqW2y0nmF39yxPJE02erK9z+wyTGmcQrxB1DtHL16XJtVfACL9p70LGYctkCJ20AMutfc86PIh7oofU8sxxXUX+6pWEcekBI0wqaLB1MKNM0NetR+l3lTVTc/xhsmVzMMukFpUvDP5wvUtTsSZ8xU9tVrjLBHBX3i+vJBZfGil4B/7xu1rAsuF5YSRyQEGSvCu3/8o7/r9+7nmhkG68qrQBBsFPHy/nLbPsW3HFLUGXFgUoNN7a9cyW6AMvvgrPwOSqgr6myIDrVoNZQbMAz4ezWfnGUvSOY0tbMsAa7xGmGx4q44rMHRMjiPWsScoMeFiVk3EDTbHDdEU+wycIOZeF9CJpnyrvJCf4UzS9pW64WjKZ7ZGIrU6w2ihyAtzezDRLtrRWf5q6RGOLT/CUHHa8w3vcBNeti6vwljN6qDK4bTHlcZ6gLeWNBnOb6Hb2QWVM7z71CPsY44pxU1OwMAU2A6+ApcUHenx7NC0w4xnJhCnxV1BCWSKDM6V8NyZAAAgAElEQVSCaPw9rIJsYa/xRrs37bSYtkPWhxL4StX+Unm9XQeXRTma5+Au6zhzMGavVKXhFnb0Zri9eg1fbT3In9ePsr78oDyOGFMWbjBErXeRODdNJZrgazZH1HySE1qdehW1oiN2cGN+i/+c4wLTat92FnhKYM1GBPlpWkmdK3PT7LJ5XNrjM71lHrAlHm6WuM4mfEcYEFrLLIE3OX5pBDsUFR3IHgkk6DrXcax3YK4KS13H+ppybg1SEPsedmgoJVA77VjrwQ4LWwPDRldVf3iBiZjNazEY8BPHG3xbwTHiWlydr9AKpjjaXWQwP+WNz+XRl8OSJCs+x7fnuvRYo+EqlF3EDxvFcKecMXnuJubFSpixRd5U2Mpft49SFCVAPQlVOkW50ODsF2q66PuLp+wOeF7fL4G653X4Lr340hH45iOgwPpVptjJ32K5nZhdKuM7S604xNJRy1jREe4UvyIDcT6XXJQlGQ4riUFzpZCK3Cd0r2v0UXVMS3Q9Js64ihRaiUrRGsCYYsUGIaeepXqVLRhtQXsbtA47ossy8KVOnrhzoSp1gw43AvV1P177gVJtm+nLM5NiBg1BXZ8nqxYsHHEMLkJF2fUan8T1DQRBLO2Thj984n5++vTHmR01LNoJEifTCFUWJ8nltyHb4EbcYLK3TDM3iYsmuBr54+Wo50T0hjng/g6MWNjTtyYR1nlbaBBt/3xsPL9Ob6zV8GK8xpn2cVbbR/iXdpzb519Lzo6iSKpzYY+oZ9iSM4xaOCzTZAfiAvVUETApRWPZHQxxPK3R7RymS0S+uIdqOM6FQpkDA45y5HhQqRkiYKu7GcFCHp5qwguHDcfrCZ9qd/m3JXl1JTzcizkU1zmu0A8T8ury5X6Vr+rkW9MGK7bAw+qLm5SFaIKWekYkDMvzSy1RiSDU5vPecbKp0KQnW5IlHz7vJwX14WSMq6Pq7T3Eo1OlTi3JNKvw6dgHIl7qeSrVqrwr7mZmSpxV9dLM305yZrWWVP/zlbCgHz+S82ID8fw8oNRMKYDoAajak5uTkQ6MVh8xxio5wLFed0xtmWZptc3D92+mUOh+cYTliLHhac4sJN5z0LfyfVlaV4xuBpWkRfbUqkbCiL4xslrOgQQjw5hoMqtSSlwiQKFqYzSBSO2NeJ3YG7626eTn2F3czQ02j6QOez2PMvC5zLp2Ozh228irHE9L3egsp6Jh2mkTq9zgZJ2St5JR2bqDd+MLy7xr7Sw/OjJFwRV5w8gevm1oKx84c5zf33gIancz7ZrY/A4GpX4FnkrqXuH5mD6iLmCXMh2UfBJJUDzC58/ew13cxNW5aeZtQjRoWFiDyDrGhwzhtCOaMRSVEHbISYCLHfHuK4SjjpKMxyUhHQUzDL11KEo8oKqW3udq8f7kK2morziiU1DKifoBb5iH+JqQQ82UtTM9LpQNuY0R3lx+Fa8cfgnluZfyZ2e/xMfXH4A4x3BlG8b0vJqzJCDdPUsxbRKYIkHpCl/NWo9r3GukrB/y1c6yPPc6S9zdWSWU67qAbtLkVDDA5TbHS/MD/FAh4YkaPETMe7tNhvx1FlCwFW70QpLUMwjCjl8H+2rdnKrmUgEHhid6jlrgmBswrJxxTG3NmhyubthZdZzW+lHHoAozSvtpOIZTx3Qa8O8qAz579k/by2xvnmUyv5Vi5whnZHmTm/YK+kitXXFevOK8w6Duw5wVO4432ID70ph3mxyv9uOaYYWYXm6WalDxKmsroZQQu6J5VL1LVEXXwL6Z8PLN88nf//dLnLq//zG79IpLR+B/eAQyocRXOX5sFDMxxgEzwPfm8p4kLfn+V2PHm6MS26+2GFW5VFg55rxQQpw1b1muSly/i6Z5WPO63ND1fE82W8Kv2D3IU09Hc76A3R6jGAVvIhxtNygacuGTMHG1oShCcd35aC+t7Ct1iErGg5Rkw3gD/lYdBorKfM0EhsG4EikMRvGXNuP5RSqyKX1ia5ZW4doOGxm6Txvef+4h/vXH3+8Vf4s5ATpN9ik2GsUULqMcjZI3IQ3lZ6ZNXl19Ia8MCsw7GNb8XYSTHXWHDaNSkJmMU6cJVK0s+YA9mcJ263zE1mOknI2XuGP1s9wQr/DrUy/n6vH9lFyZkSjhVMd5Tk07hlacxRU95Tl6KY+lMVcFCa/IlVjuxXyx/TTzSY2b89t4W2mnX10/mNQ9ubsaR9TirJIple2ABG8prCdwLoV8gj+/mkvPpTFrLuCcc+RMxHaX8GhSZ3dQ9GKUI8CB3DSXBwO8ODfKtnCIo8kGLxHfJm1yPpA3vyV2MTkbIJPS0Ju3WFIZFaYNAg+cet53zYmj5y8S4azzHrj5SluqtIN+S1XgqF+900Xr0nUvxPD5rYoyk4LU8+zUplXLtecBlMdXqiqoqqi4NF/BE4hT5UwgUX2t7Bx7IOnRu1IepEaVRYtAGNTbhjAKKZRDokJAVAzIleTRZVivC7AJwKkypzfoT2w2U7dmrVZxN+U9l8u2qZg2ASxbxJiCF3YEufFMQGJyBFIHq4KpdVFQZktQIc5NoszTopTSOC43qqk61kgYx/KYt74xqI6p6Dm9dtoEPNW94CuoB8Iqr4lGOJE0aZuIMCjRiVeZi1eYjST6CUmaCVNJyPbSJC/L7WA0rPLl+ilq3ePev1HntSbRhM0xZgIPJMtWHpZq+0YMpzWazcPsGYi4aXg/QduRlg2FnWDFl21Dbqwv8FWBZ9ioi04kaz4VXkWxEOdVnfNJHX4JnRymZfARhAq3X3CeGys7v7yq+haaG1Cch+QkBBuKC7RUQsMVWw3VvY5Hp1MKjZCJdJI5cyXXFvayf7jOXf8Pe28CZNl1n/f9zrn33bd3v96XmemefTAz2AECIEiAAkiQEERRpGwtoUxJJLWUwiqpKrbjOFV27MRxqmJFLpYTKY7DxaIlaqO5aKG4ghsIEMQ+WGfD9Gy9r29/dzmp79yeSHExKZNWWUrVvKqenu5+/fq9884953++/7esPkd3fYHUrlIeLNLTwSJsoHi0JBHXTKkjA2+PckkRaDbiZWe4b6jEj1WHeX0QsmZSFgm45IRkpoSmSCWznMByowk9z/YjyTankqY/HNk0YS0NuKliGVUhFTsfj6gIRoG4F5bySOK1FF5rw1iET5BQRR1aw2KCtz2al6NAIc+iPTIjb0vH1ZZhNDC86iwHjDKlO6SFMeJkmV77BWrlgz5tRHQJzRCTbREPLrOnfi8nollaZHqLUADYEI4lEiSfedqr5iM/XzZ2vo0V3UFz118/yk/sEoVan4scOBTwd957nVP3PTfX69+8PgJ/HSOQphm/+7tf5fXzBwin9zJvIsaRaS78WdblPYUCbw5LlGKXq0lH8hguIWxCUNLXHXbe+FO2Wq2yCvFInfa6vbnXnacRbTuMSMFC6tRNU9v05lwNazPjlXHyvus8BjrRlufyRS9IjbcY616UX524e3l7tVyFosSCI9Br40UV1T0SUeTtnVQH0x1HWDPeTy+4ovawIewYXNnykRef4Zc/9u+Y2zPJqhn1ylb5zxGOEFSO+ySJgcx+szb9LOYd5cMcLO7hOM63ehczuDLQ03FUjOHmID+8nvHefZl3cz/rDB2Xcs5vAgnfaj3Hpa3H+Xv1W/n5qXs5EjYYGlWKRUqvZ7k8gKND0JNAxMH5DJ7E0c2k7IuomYjPbK0zWRnwhniItxQPUIuGyNTOFLk9KLHev8IrqQqZCvUQZgqwrecZwJyFlVRxaaDgjeNYfj+N+Z10wK8UShy1wtCK3GxLyP3tarxIU9FhKuis1RbGd03AG8IGDxT3cLA4zQmlE9iQelBlKev6segodcIXeSVil5Hoa5HWVXypcJN/moQNapWqnSP4xqtU1cu/FuuVVyrybHNZ06safUUgDlo44Qsrb0asKkEfusnfTfFiKiLVUpOC1m/U+vkuUUlIqghN+jvhGEHpkG+Dmt1Wu9A9NcezzBIPEpLE+Wi5OFGLXXXctWLuGjqnAk8fhTymzD9/gVr593J/H1Uw+fekTs45goNdhe/Aq5l7LmEQlD3B/7ZCgweCOsru2NrN626SIf8xxc7pwKCubyXHlD0KLNucLddDqB3RJHUMczbwrfRtWXtkffo2otRf5HB0gPXQUlax3zckwyl3jxS5NZnlJ+cOc9yM8pnLL9NKz3grm+mgQhbUabuYNG0TkJL5Fpxh0HmNXrHM7fWb6KYFwh4U+7ApW486lBu54XhzNUfavf1hLVfCBoeUG6xILOO9J7VeBDP+hfloQFkS9pQMI5HBKIT71ZY03i6kNGooHTbEiaG15TxyN6yzgDG85W7L0XEJSlIWKwH90gT39W/lvmM3+dzT7sWzLBbajJmi56mKdyZEOHA9+q7HotTBWZeldIshB+8oRqylO3yu2+K9xXE+VBplTxbzeNrhKhn1LOCcka7a8rxHfSPeE1S4yUZ82MSsFGJvHFytOioE/jX0EzgvZ5EQhopQHMCLfRiuwvFhw8YOXE7hYgLNAcyW4LL4iTqTpLAZG4ZSIfjOc3QDl3Aq3WY8mqVcmMIOrrDTfs4nkbigTCKKhqgQ8RJx+TjlcNS36i9lmecAzpByyjkO2wK3BoYns4zTyTbF7oteSJMbgOtEn19TUeDotyMOHgr4mfc+cJ1Tl1/h1/+9PgJ/fSNwjfWjZ6DOp9paMp3tZoa2S3mZkPG0ywEHzXCIZD2hU80XkkCRX7IZSyCYEyljd1/dBS7k4+GFER1FMzmS847C0K5CVgbFXWivQlXWBlrExxxO3h+bMDS1C4DIGuAN8pyD7KKjJBCn5AjUhxSR6BJkY84XdEXZf0lKqhawOnxl6KmQURi7/tZlqJWhOCwRRcC/+7OX+ZU//H3mp+ssZzXEc/OeYV6puNu6S5t529BHaXXJivsYdikDbYQGzuO40xmed46ZyLDPwucGcNhCKbU8hUN+dW1FJCXrnN14nB+vVfg7h99NoTfC/mrGkEu4sJbnQ64ljkXR0za1WcO6VRvUcZuQOlviG1mbb3Y2+dCtw/xs2KDbK/BkM+XoZspjLuWic8hmQW2Srw8WmTEhRwoNtpLUR7LVi/mYyOh4QUiiOPwO7jVFapJE2JQvJSmHnF5LrqAbKU6xGnd5AMvj6YBVDHJle79VuZFRIKIXFJgPhzloLHcWZ/nEYIUHCmMkaYdvxutMhk1GbcBm2vZtr642hcFyzn1TweY96CQZDLwYxVubqCDyrT6JMDSZ1NLs7YogannL1lueDMQ0z6EgIXHRjH/PPI9PWbBK1VCL0wsy9PsDjBS1+uxSbFnR5zL2UGdKj9vCFJq4nlqWIojqB/6fv/DU889Nz09X0O7z9P531/6vok+CCF8B5s/NVrw/n4/piqbzFA29Ji/0qPpkkYaNPJb41EAZdRXqxrLlYj7vYg6YgHeaAgojk1GsrHJEgu/48tOxkfW56AY8j/Uk/xfSLjeYiFCHClvkJBVe0njFHQqFMnNCjELLIc1Vi48DPFvMaCWOuxtlbmsc4cH5WT514VV+4+LTvNpdYrh2iPHiXjZsSD9pMZENaBVmmK7fwnPLj/N09Do3D52gF8cMepYlRQouGIq1nPBfLEjl6nAHZUCudSFXyXq/vj155sdA0HFoKFxxdOuGYgPKN+V2STow+to9gnCPQfCk6sr6HVC9CKNdGKgDcBWWP+08D+9IwXLgTny8Vms5Ijl7nP8qOsSj++/nYy99jVP1Jxkqb1DJYrbcmKezmbRJaAUFHqaYZbTNNn9/6WI+hwK4nJRYCOocK47wa0mdR9MOC+kWC1mBA0GdvopvE5HYyIuk/s+oyMF6xrfWevzqcpO7wmGOZhl3FCKmAp+nQjt2XjU/FMCgAy/FjoaFK53c/enmAnQSmLZQtYZI2dlt58VmqTFemPWo54qGXMg63G7LHBh5G19b/T3arafI5CUjPlyi600HvpRO1uFEUKNedJxMDZ+Kez5R5jCWyDna6jwEVW9x4zmguh5EYdC1JAW/n/8dUiPuTX6N/OU9RVfO93O73n79fkbr+n2vj8D3GIHdrYo0Dfj0J9c5c26VsYlZzhJ5Ho8Wj8fimDffFNDslDk05hi9xfpwamWvipahZLCyMmBFPdIJUiiZuOQq3GQppnystgSMu5EJa2pg5V5yIkkXDhuCgkGJD51VGFxyPh6sp1CBMbwFgNq4att0JJCoGAIZm4pYfMZ5kUZ1HtZ2oJ1C40Qe3S6YsSB1qzaAMVA7s3cuo3w05KPPrfALv/sx5mYty26cWOdrtcnUYpCHmNz9ZbrrLUkCCoNL3Fq5gZfDYR+i3nfwWJZ5FdqNM4Y3TVhaG3A6VQEndC31uaktGTiTsdB6mYXOU/wvc4d4uHIPs0GZ1UHCrBbqwCBz901ZPTg4pC4Hhq+nMb+ZpdxuIjaymH+V9bjRdPnVIOJdM6NMjFimVGCm8Ni21I+GJ1zKmoEDQYV5iQhc37eTrbMevbgghCP1sbaeAinUaW8DThYNy72QrzvH26V4NFbRt94zuu1CXrVF3mYtodqc3oZFxUaGAsYSDCtC4iRWSFsskrCadTlRmGAzqFKQF1pQZa50kO1oiiQcJ5XCU5W6p/FbwvJRhhtvZah2G0R7SWW8W1SrUoWQwuplbFz3yFsw/FaisXeisHu9uUHtLkxxHzac8MKMQuPtBOVjmGDEtzu9J5686TwfT0id2rU6geRwhyns8YidF2J4xFCoYRN2jWf/osWqfWv3ivFI3W771VuV5D8z3tNE3xfXUIWd/sa1m343xhTGfHRXEM34aLZaUKdiK5Rt6CPZOkLCrFByw1fIeKNQEyNFIkwZy1guGWHb8zYdL7qYL6VN5tyAQ2GD+23IlutzoX+Je8KGV7GrCD8cVLxadiNtcsj0+eHaHr7aCdk/BG+YMOxR0oouy9hQxtGJMkaiiHtv2MO7jhzgxrblc0vPsd27SF3pEF6dXvdh7wMpQXun2Vu7gRtLe2kmGc0A9odSwjqifYbBOt6+pNCA/hq0Lhkikf516EpMnrMsm76zGr5cCR8LbBXXtpsXfB7h13oi0YQAI6H0BoTIGxlERsbX4Y3DBjOZc/J8/vNZfIE0PAdHj2SMTATM7Jnk8NjNjA9O8NjFAS33AhOV0HPIulni0XnZtpiszSAYphQWGQ6Uouo4Y0K+ZWtsxjsMXMp4UPUqctnKvBIvcSZeYjhs8LAtUjOZ99ZMOoY31SJudWVGreNFunwkTpCWa9w439ouF3Iz9IspnEvgYBmmC7nHXxAaLohbm+ZFuMKik4Hhy/080UcJNioPu1nXK3wVi+azp9WuH1z215vrncW6DqkbcGz4hxiPpjhGxvEwoOEsHWd51PU5JBGGMfSwrMmvrv0sgbirkit7orKuoz7F0NJvtzix/zA//TN3eZeCa3vKtVn//Xy+XtR9P6N1/b7XR+D/ZQR02goDw0ic8jufO8vMvhNsZgXuD0qIqv67meGXZiOqTcuojEQP4a1AJHpQjqtZgHC39epddmUntms34tWv8vqQhYl6uRJPpmqd5kDG+VehfrP1RVwg6lUIW5d9R9ejd+rSFmoGs+Jww7I8MASbzrdQk9XctuTyMzB8wniF2OZ5x+zErpnFai7G0EbQfF7eeo4aBb67vs27v/wJ9tkVNgqHGfgYnb4n5gsRknGpyir50qnQi5NN5tWWqdzAmHhJiL8iM0/LydAybuDzrfwU3bXwioMzaY81G/qW2lfaz/BQtsi/mP4hHqgeIs0cLZdxrm+YKxt6RYWG52alT4qLmKnlnfEN4G8Zy6eTHqY04OfSAr8QDXOoXkesnVoJipOGemB45pLjilU7LuBll3hC/bCsGLIeH0sHbMsL3gVMGeNrbok57xTCURTnD9pdWTQYThgrMbPPoVWNrtJEheIEqc8CVgzZhLHcGAQ0AsNvpTEd53gkCHjdxVRMRGZKnHMDVk1O5D/qMp5JW2wZy3ayTay0h1hpoCmxMdQrN1IrH6UTDOWyCClkoymvwswGV32b1hRmSCXEqJykNPqIR77S3rncz0339ZuNg8phglCniTAv4iRKcLE31/UFV5CbHsuaJr9lWB8Mn5NA9X3Xv4xTqLF3zta9tE3pQ7umCjWRR4VH5IVcjhIKcdPM0M/0fRV28unZFUtoQ/TqVwkyBl5NnXPtdDlEjNsSr+HyMXQp77RF7w0mRKRoAmoYjqhYw3HWpew3gW9pXxSn0WUct2WflBDYkGkcH8sSX7RPhCNsax64jL1SSBs4mw3Ym67wUH2WblLksMmYKMLwkKFiDeXYUM8ME0cM5SGoZY6JdpXbJmZ5z8wBRgZdvrx+lk6ySkkoIzEDo+tkmcXOK7ypfhv3TlaopI5Sapi+3XhT79YO9DehKFBHCPqYhmeXKiG0X6Mq+sKsvNx8jK63IJKvW6B86V0ao94KmZ7rOvJvRQKDV9XKNahVa9UZMFCegJJatZMmt0sUFUSPu2NQ5O7EIcehewPedvMkb91/jOH1Mb7y+ll2dhYoD41SiWr0dTJMpBhd8srPnoytlYFanOGt3pQ54lPA82mTilq1WcsfolpZxxsAB04pDmpxFygGsDnIKBrD0VLAIdkhJQEt2+fzg5jLmaWUGY/OalS1dm5ZGGrAoAnn2nAlg9kATvUzFnopoy7gYoYXschoeNMlrAxWGPHob0QrbVO0RTY7r2CFromDmirecJuodIjhcMwLPUQv2DbwLXUNMMxj/fy5bAp0k3WW2097BM8PuL92UozrM1UP2N5a4cP/6L3ccOcej9zl5Xd+GXy//14v6r7fEbt+/+sj8D1GwGlxNIa1/iof+cjT1KdHmckaTAY1RGFXG+e9Zcdst0hj1FESH60LwWGpDcEp2UGLroQRUrpq7VAbVArZg7vt0NYuyKHPWtS1J3ahOoCiTtUiTVuJL+RRBxUZkN5giCqGrtA4Zb3q70bQ08awz/iUCGW+No6pIDK4FRg9aChL5jkKVjYo6tidcwy2HbWhkAsXt/knv/1xXnUXiSrH6JT0BBMf2eQ5TkLpwjEsmf95KDPXZIOHSoepB8P0TYqUcCcJeU/JEgfwlSZciaFj4btZwu0ly2yhyCea5znb/Ca/VDzEr0ze49GztJj6rNqRkkH2Ut0BTKWGMFQuraGZwr/PEs+X2ucsp4XOleCXCxWOjhbZPw+lQUZQN9Sm8nzJQgyHjOF8z/GxJGU03fL+ctNBmbvCOgoCO+cyjgVFztKn4Cw3WUNf75eDL7XhRee4wxpujnKR8muZDJEdd5iAZtrkTSbhaFRhMcsQ+ihByKw1HM0Mb48se8OAL8Q9nkx3vLC5k7XZE9S4oha+S6mkmx7RcN0zOQJmDOIqzgy9kcnyMdo2pCdFoQn8nEq7p8kGl3HdV3zBJsuURtalPva3fKGcbn7BI23iwKlYynoXPAI2UjlKOljyjyMkIeuLqS+0TO+rLGp2W6ZePKEYkiJZcR4p+4xiw/pXcf1LefqEWrpeXKGL5lpBpzavCrZrrVd9P2+5+sxaFXK+eFNfW1WEeHtq7eZFn/6eV/foe0rVMBZ5ALZMxHvCKjPGeO7h/Tai60TGl/Ah9D50EkpcxHHFWB5zMd9wGQdsyKwp8MqukfWdxnIGeDpr8x4TciKocBrHXSbyBth6HS2lJGQxJ8MZelnA3RrCwLDUg9G5PGheliRDY/hrqrMsumPmCfpCpB6cO8jDQwfY6iU8032ObmuLieKwN37e6l0mG34T+8s1LjUzRiowVDCkaxBWob2Gp1hEmu9C7oXU6fQwLEslMFo39LUOhhp1tVeVOLGiAk7oW76eqJhTG1Y1iu5bOCTqBDgdNI/kHXH/+7q6r+KTZIoHxcvNY14SrUNNg2k6ejsZJw6WeMs9h3jfPTdz695RPvWtZ+lur3NwbIiBqfuoxFTF3WARfY7Sjs9bftVG7A/qzAc1z5s907+MnuJEab+/pp7pL/JEvMZbA3m6lVjMClSwxHFGEhtuLAbcV4y4bW/E0iDlf+s6j6QJ/U4JUDrFqSbMxrnWTODyZibXAHneBf7/YqJUK7kf4CuZ4xuux6SteCPjdR3+CuNem7YdXyUMhr1ptHUxO0GdhaxNvTDBdFDiaef407TDbbbItDXMFixfSPokradYjzc8Wq7TunwuxVU1JDSGS2ytb/Gh/+bt7Ns7ybW95HtsM/9R37pe1P1HDdP1O10fgf/vERC/QkXdwsUlPvbR12nsvZk0LnAoqPsu6msu4xcmQsLtgNFZqO61pMrPnpTTOUh96nlrigXUBqHTsE7VKt56eaiAu+K8gajToqqAVK24ajNuQSC+uyLDOobBKxCtgXLNXVURYY6uvhYnZ2+OKnVfgYr8Q6TDSqBzCSpSxOnE3obanPExtr7I3IL0nKM0HrJW3uEffvHf8ofuNWaGbmOjMIfJxKvabZGJZ+V91Oreh6kc7WErbXOPy3i4epATRctCmhsK32oCpAVZzeCQzVVjn8sGZDbkhEv5bPMllnunePTYnfzU/sOkmxnLacaqxqMnTg68EsNJA5OThtUOJM7wuXhAaDJvUvrxZJt/PlPi5/dUqXQM4wdgpJir4Z5adIzX5WhvWFr0YJZ/Ps8NEraIaZiQYVtkzoTsCUI203XKzrJoShwwwpMMWymseJPkXAV7kyhK7lq0mZzw5REI541jygQcDgoe1RPa86dpSjPT71nP7ao4OOHZdSmPpS2vGBR/Zzvr+IilOF7Baie11VxVnGwQlA8wVJxn2lhey3qUfEtU/Lkdb0mSyWnaDSiENdJ4mVtGHmGsNM9C7yyR3qviHmxprzcy9oKI4gxjpAziTc9pUgHnvBhDAouaN0DW16qvTFgnS5sUKieYqN5Bx7XJWk/jegu7BV3O78snquaHKg4VdDq1uL+IO9MJxhd+El4IjVOhqDaVUDpVHtf88HRBaMrqMYr+ojCVG7ypcugGdD859xwAACAASURBVNI2FRMQuZTLxjBuLANdl7bgEZPPp22WXMbnsx6XXY8dCjwSRJ5f9edZjEZh1hhf0H0sbTHfv8z9xVkez3RfOGIKLBpH34SeOypO482M0K6F3HvMeASvue1orRg298GRk4bsSi4ejpW8Zo1vdUbVjPII1NtV7qvMcdPQLKVoi+/sXKUg77PkCrfQ4I1Th2laWNyEziaMzRuqLWh2HK0mtNYNZVEjZG1SgkEES69AY8qQxs4LnQJB4NLQdB390xAdMd66yCvpmxDIcHFUIiu1XWX46KN0lcuX2yhJg7OZq9NNwWF8ESnEz/ioQmVYK/EmbFp/ALStjEq7zs17D/HBN99Gq5zy1aeeY9DaZKTaILURqbfmqaD3rmhLBP0LdLIevazF4uAqlcK4z4HeE1S5KygyFza4MSjxtd5F/vVgGwkRTqddVm1E0QacSp2/7qedZa4XcKN8Hhnw+X6bj5Jywobsl5pcZwIV7gaOFvHcXR0A1fjQzyScGHKyfgo9atoxAcsuo+lij9TqwDFoP5t7NMr024SURh6mFo7wUn+BCeOoh7JnKXhj67ZzLJqIFwerXN35BgUdfjSBPUondXDfF3nDVRV1y/z8Bx5kbt+U5+ppL/lBb9eLuh905K7/3vUR+EsjcK2ou3R5lY9+5Hkmpg9zPq1yMKh6ef95+jxsSzi5jMtuYI8hECH5lXyPK2nPU4KEFlntWVpgBV7I3kQnbe1zMwYnqZ4W0tj4xVR7WyIpv5C60CD7p4tnHbE8qMZBGe7FBEr7DOGJ3VgGOV5c9HGWvrAMNqCs+LAyrCvAYAlqMvhUW1GUra4jK1sWtjr8dy//Pp948jnmxm9hKRjzeZAepQsVQCllovWxTEE07U+i/WiS2azF7cV5H5uDzfjjBH7EBtyrvmSWK1O/nvVYdgktW/Tu/7+z+af8iIMPjD3IXfVpjElYb0Kj7F8+Cz1YDWDYwRuOGOIBfGMt4/HdfX9ikLJpYz5+S5m3HSxRVGtEp3MVhKt4tFJFtIxLOxqLODdmla1BwwV8MglIg4i+M751dzy0BNbwL+Mmd9iIuz2BH87qFRu40aq1B7JN0VupxCYpb/Wz7xjHF7OEo77tbHkmcz7z8k3GeFRpECi9A5omY8EZ1pVqENTYYwu8kGxS8Zw7h1FLU9FTWY9BvESpchNB6QDzOC5kXXrJNkYEzV2hRBYvk/XOej+3NG1SLu2nX7+b9azr297e203xVlmPrLeAlZo067E5uEoiXzpfGApT3M1d9bmzshIRehb5DVIb9C31e30hvt58EtM9nduaBDqd7LZc9Xm3VetTMTwqp8eU9YkKPM0DiSny3NCcb6TiTbFgKvL0OGrt6o4qDAPPp8vSFaJgiCiaoRsMccwWeSnrcT7Z8ma/T1DguNqm6bZPDlDAvBSWY7bELWrrG8PdLubWIGLKKYwl8DFhvylBSX+RkeIs503Ad43hv/BZrvLUc2yScj7ZRI3uk8EYbzpq2S9e5zZ02oYzy47RCPbKIqiTF1ziroUSIe0X+Gi4/JKQoozhURgvjTDZP8C+rEp7cJGrvTO4wgiO27gjDDg04Tg3gG8uQFEc+z5URg21ML9eZYXUWVeBhjfa1VD1ldk8ZfJLUuO2AQp3YCy3QhHy7vUr+/TeQLYCmQ6NEl6IKiljPWlnlKEmUFUHTZ1EpI0R6idh1pG8w2BLxoursg2HOu47axmlMkyU67zx6Anuf8MRVpIrvLRymuks8R51HRcRKtdY4ptwBM3Pjoq9/mUCE1C1VeKsRd31CWyNRlDlpXCcB22R3886PD+4wrezPq+lPVSsn0/VMhVX1XIwCLizUOSdhyrcHDvO9TKecW0WXMBs4NhbLfh8a0WSSU+igl1iCp1LRZvoYvlW1kFxX5ore0zErHNcMJZI117nZY8aZ65LoX4n1eIs/bTLc/EqFwernm6yYKtIZX0laRMm67R7siwXah17GxO/Sjjx9iKGKxFb61f44Acful7U5avB9X+vj8Bf/whcK+ouX17jox/5JiNTB9g0cxyxRfbagI3M8MjekH2C/FWryWpkOG+nBCVDXxYhEjXU8xaIx/ollNCCrAXVk5jzr03Z5C3aak50lkDCKkNLrZWC89YFnQUYEjInTteowW069aB850qOGAPZJcwb7z6v1svVMzKENYjPV5rNY8vCIUUPOTIhW1HAv/zWl/hXX/kCB/feycVwzhdfsh4OZHQlDWe87DdbtfPEsRJ6l/VOsxXt5acrMzSN9UijzEIbqguAcw7OGngtS/kajmq6ycurX+Af1iZ518SDZK5IwUdjiGMHQrOkOG0oa1SCiswxPDB8Z8vxxXTAmrFcGhhumDD8ykSZm6Yj2tu5uk1h6DuL+PSM4RFo7sCXLziOVg2VGiQ2Lydq1tJeH7A/EtpTQJweEZWE2D1oy3wYtVR1olcmreO/zaR6S70w4jXrmAmMd73fkim0g6MFw28neQqB/OvUGev6oba+5S3AdCGDOBPHMON/dwmdrEtoChyVWtrFlIWSBXXvBxbFS4xWT3j+XJp1fJu0m+6QOrnU5z16oWlp89t5ASTPr2SFmaE30QxqbHTPE8jLTcWRilMJGoSYSPac9bG7/7cyLhafToa/2oxUsEeTvthT4kUeVyY/r2lW403S5rfy4ktRcCosd8fIVw2+qNNJQRubIs92c2tV1HmUN2/0+fg0XxZrduh+kW+v+srCF4BC7vIoNGWSDmMpFee9B9xbognGgiE2wmGKEpz0LvBcNuBs1qEQ1D3nzprQ5/LeZAoMYflslvjkEbVcpVJckpox2eJ4YcKrLpddn58Mqx4UP4ZDuR+JEJrBFS5T5s31ce4MMm/gTcXQE2pUhuM6QEXGI/DxWl4fF6rQleNMkucrjx0xXlm6eSljfdty49Ak3zbTLKRN1uJFHs9GyOIKXVf0FkT7ZKnTh1ShJGoXVo03NR5XTNhualokutdmDrL15LpxNufZBXuFggqqyteMpGBww0Lzjb+zWwXT27VTkg+13m694yO7EYVKs1H9rUOTavFd+kegYkiVkTPejklFYa2ZdwTizYziZcfxiXF+ePZWDrgav3fqFF2zzVwRuv0NkqyVK7DTJoEv8OpkNmLVFlgeXOXVrjI+Up4zgeeo1cMhH8WnbN1G2mIh7dKV6CXr8DVnecbJfgaelVI9NtycRpwIAhJCLqYDXsoGFJMOSz1DyQlRgxec83QIvdT1LPHXZMvFNDGcsFUm5JkpzrQ/lEO7/Txqvcp45ZbhBygGdVxYJ4ym2UqbXNx+jJ2gxOHdlq24eL3BEnaXR5pL0GQSrq5GgeGKZWt9jQ988K3Xi7q//q38+jO4PgL5CHguizG8fmmLj3/0MSanD7Lu5phS646MZn+bn56r0V1zVMYN8+WcuBvKb05tQ3nPaQ9tGFyam4T6Yk4edjn/3J+S/SousELAxpLzbaCs5bw/lX5fxODBOajuQPF2w/ai2rpgbxAxLv95VoLedxU1ZLw4Qy3bvoyF1/JkHS38wzcYsjpoX4/GA554/jXe/6nfY2bfAdYY9/u0+E4qNEyh4SX+nrxuDFZO/yLvm4BW7xy/VruZe6Mxz5cRL0raC3V3pFiTjYmKtNgWuJqssLnzPH936nbuKN7K0T0WE2esdeU/ZRhpGM7sboqzBcNTqdobKWHf8f5BlySLuJ2ED+0t8K5DBUbGDDt9RyRzUrnqq2W16agMG0phnpcr0rSMSIcGuXq2UISNdcfoUIFmP2TKOJ7Nerzo4KCJOGoDZq3ln6Y9hKVEpsgbgpTPlGIu9gNGsT6GSJudcmKltnu0n3E+7TPvifuBT5pQdoMc3zQOstX9YhZzXiastszFeJ0mQgYjX3iIW9ccrNBMt4hdj2r1RoaiKTKXEtuibzkOyEhtGauCyQSkahMlm3nyQrJNobiXbvVWP5nkbyfULlLIvd+aFN01SRyorTtCUBgniCaxQY1MvB/PaZMEOzc9zpL1HK2TD51HV8u49vMibe3my8p4WHFIQusspIr1UjVxTQChAkOKH1X2corTBPf93N3/X2vT6v7iauaQruwgzG7BmOfSik+5RC8coRrNcs5GHmHcGlyl4/pMBmXeXRjhhqDGsku5ZELmjWUk6/Eq2qgt49Z6H7pNRdHJLidp8oHitEdovp71qWEZM6H/nGI8MruI47nBMj8ZlnioPMrhmrwNDeUQH3k2WzWM7zesLzrqU+LKGtIrUDkJUQPSizCuc5A4kdvQuqK8UHWgUyoye85Szq98h0eOHOFTboMvPpWwMlHhpjDinpKhM3Cc38znbb2Qe831HEQd6C/jDyiqH7ItqMznqP+m2sCZ8mIN/XPO+9VFMiwXIicbQhVmDUMgY94Z48MO/NsiLcwZl1M7dBqRK45U8/o9KWhb8hw0uUpfh0+hfOq6rxvfqo1lcHwxYziIuHHPfh7efxOB2ebR0y8z0oiop012euJsOm+EbcNxbDjsD6PFUEKLSTYzWfi0me5f4VmX8oApMG0CwsI4xyW+8lxHQxpWmbdlPpnFnHOOR7sxYdr3HpVLWcx7gpBxU+If9LtsxAN2SGk52dEErDv4lIQwgaNnnEcAF8m80jkk82vWOR0mdA33L5P2XqIYTlAafgudoMIoAfuCovccjI3yhbcoJ2uUwzqvd16i4H9Xb4pYr0LrBv791msdKvXZWr/MB3/h7ezbN329/Xq9qLg+An8TRuBaUdfZivi3v/U6W0NVhgpzTNsSRZcw1lvn3cdGudSE1UtQGjaMzoCdNqR9R3IKIjnB7zOYsV0fOiF32he15KkNqkVVnPEiuKs5OVrB3gr8Dm+UUZZaKYb1F5xH4goyDd5UVJBobvnCq9VSlgvaZ/WcJdJItqCxN5f4S+0WyFa/LCd/GQ4HPHVhiXt+48PeAy8MjtATOd2nDYzlQfSqFrMWVukEHtkp+ZNsGi8xMIb3VG+jq+gl+dAZw1Wcz2K94DK+7Rx9a3lq+yXu3n6Zjx26j584MEswyNjoORbasFdwYwI7XXhq1/KsJJqVgwvOcdo4HsoK3NdwvP/WkNvGvAbQR4KtbCo7E7rrkAxDrWqo1WFtHYaGYb5qGGxCmBomVGGN4PlKExOGUsuxpJQ0pzBvmc9myKLgzYWAg87w5+K3JX3uq1d5z56QM9vwnXTAm62hmVjqGSxlEsmkzAQpB8MyX01jImEPJkTyEr20r8p9XsWGsb59eLux7AmrvlWoDrzGyQyu0MuavLFykqoKuqzrvdacAupJ6GuT8yH2ZSSQSAeXfOvSKqEi2SQdup+Z8g1eTJBGE4wUpjx/a3NwmUb5IDthg302YqCcUwkRtPlo8mVdVMQJvXNZH5kU633OVMAp6kht2IGEEWrzNnJUTQid79nt8uHUww9UkFXy6voa+uYnt4q7xHvdSVyjCe/Szu7jaLLviik80qdiUBeEHkuCDTW6s9yAuXSAQlBmS/54UiXKBieaZdQYbghKyJBY996D5UBQ8xwp5Vro/uIsflfj7xIeCEeYspY/itc8FeCRUM1vxzmbcZWUr8Q7NEhZ713mb5dGeNfRBsUog5Zslp0v6kYmctRdSk2Vo2qLqossCxLxaIUK6zoO1QrtQieCehV6JUM9DbiYbPLc4mP8m3/wDv7eQ8cYlC7wuTMtzii/tRBQMSG3jsJWE/58EfYKEQ6Mj1sTUqd2og4n5RsNaQT9M2CXoTQFmZ6LFgDPL8tROLfmsHtMjr6p/hZtU+tET4WWhBji1YlDl689PsJw2+VSc2U/D0QByQtFvaUyQfYiqyklXhjvhRcUnS8C99SGuesWMREP8SennqTV2mHv8LhHbvvi4uqJaQ6oKlU6hfwuTehznxeCiJ4NaScbXIjXiEzIeVvhXWGFmaDMTNZnKOv5aLF5W+BVa3g9S/gTl/rr6tV0hyeM5d1BmaGgzMeTFi+mbbq24JXQh6TEL4dcsjGf6m6yL12nETa45ByrLvHvuwr7QdaEzsskQZ2NoXu8lc5bjfWKfqnT19XGD4dY7V9la+fbuahHSSc6WWo+67WJkyeEOhhhrLLJxvooP/f+h5ifG87FLp5y8IPtbNc5dT/YuF3/resj8P8YgXwxgvFGkUH/HF/+4z6TM5NUKfF4Fx482OJH947RXnMUUpg+ZqlOGOIN5/lzyZ588a0cNzi1RHRCVhEn0EKbgAjL4rxvO+xe2ZmYnOOihbYO4eHc407RXdFFH49JR15VvRxtE4KnFow45+ZCjlrptG0qkPZB+ddJAIvi4ulv71OmvWXtdcfP/pPPcrnwKjPVo2wEYwSFSU9y1yatGLC8p5NXXj6+KSh6RKXbu8ikYurLh5gLQsZUClo4nfV9CsNZE7BNn6dXv8Pfn7nAP77h7VQVFt6OaW0bUXvYSWE7zjcq+efJz8/H0RrLKRfzbCoVYpkjRcsvzmnDsyjoPC3mgejxImxLIRzAyGFo7+B5P9r8xCNcvuToCoRKYXjYKNaS1jqcXsPbjRwJDa8riUP1nvdptZQc7Cfwe9snk0VODByjrSqTAqUyx7ddzLyV8bT1oeN6+7xHnDHcmjnaLuNM2mbUFHiM1He17rBKIcWjmUeCCksywZUxrjzzkjXa/atMVW+iEU1zhMTrZBZtgXmhWGnXO/j3hJzGG6TtU7jBVYKwTpxucqg0xxtH3saLyQ5jXnSQ0pHxbX+BG4Mq/eJsLhKwJRomYCHdIZJliIidNvJCCd/yDOt5O9SjFed3C7AdXHzFF1m+L6eNym9cKtaMDy73bVRx9wT7yrhXW7ckl4FGRjf5dXUx0b48Z1aWOEIzNOi+TSvCgr7Q7wl21mPr5+Lc6fnl/U3ZkhifhBETF8a4Q6pr5ZFivM/fnIl4wiU8lW6zkXV8QddLW8i25u8Whhi2dU8R+NN4m5dczE+rpYbxB4f7AsuskR9jQDtZo5m2+dDIUU7OW9wQbC7C1auw9w3Gc1g1BNHcLj1CIl75Nm/lkV69FHYEaqq2K+FR44GuZwNXDHxiUGF18Qq/+M593DE3x9sOzfLIvojlC6/z4ac2SCYr3Fyt+sxSzTmxJkxiWNp2bMQm17sIVBPyrvSXMEcIN86BuZSvF6nQNfFodUAsG+y1CDIpYTW0Whs0IccNxkcG4s3fxK1zWpcaBiPjYh0q9aHXNmbwwoleXiBKJKJDpdA7WTD5dUy+fdMF3nRylr/95tspmTpffOEUvZGa72q0glHq0SRZskMvXmLcFijbMi0plMNR7x0XqE2uoo2MtbRF2L/AFQcPFIbZH8BTg026aYcfsxVi+swFdaKgzEumQMGlLGB5EvFZQ8om4On+FTqerxpwOTWspBFX5TGZdpmUMlw2OckGXflVZjE2axO3n2M8HOZHh9/CA7akVY4lHBvIgmaLLSWOhA1fAGbtZ/NDiK5VpWToukpWfaXfqBiunD/NT/3cA/zszzzgFbh+nmu6/4A3HSSu366PwPUR+E8dgWsXoRZJsffZJsk2WLczbIQJtEqk4rgV8+zQquJ4OhDsqtFcyxFIOqkrcsd5PzonkocWRK32ow5zGLLTuUWBIsT82rAigURe/CEf2RSKM8a3ZsMaKM3mwkswcxs0lDOrx1L7VSToLlT0eFrE5w1B4hhJoDACypxODHz66a/xePpV5qcPcikt+cxMsa89qiI6uq8SpfyVuqPsuWeeHZUNKBOzLC5Lss5IcZ4gTnghs3zHKcQiZiwIeHrlCX54vMWPTP8oR4arXNlIeHbFMB0Z5iuOSZkyDxv+ZMdxJYEJC884xx/H68y7jFtsnYcnDG/Zm6N5GlOJDnrLOV9fNKLyIPeQi7ahqszHAQzVYacFw+OGutpFUhUmEA2gmhi6mSMI4I0jhoeX4H/N+txmK2x6/o3KEMfJoMx7Swd5fLBCFrdp2QpzNmIps3woHvCQCThiIm8qLK7Vewvqx1oOETDrYmpZxowt+D1RhWoLS9eW+SOXKMgD+Xk9LeHDYIkfrxxjRQkKLuGJtOl92Y7rsdNt5Ds7byq8nrZoimQfL/u4LiWcxmmb24cfQNiqptJMocGyNDlpy3OHTpYOUneOWRuJMcejGMpC6QIZFm95dC6wRRJboWaL9NNtetqkdL2kTVz/4l8UXt4PQ31EFWPKmdJur4pGsI9iDGR5onmiCas2cT5xnQqzVKrMgs+8MpGMi6XmkemieEd6QP3FXE1rgmFfLArp8Bbc4RhZ/0IeeSbD6ywmUQSYkiSw3rtOBs+KUFvHcEu85hW+m4UpusUxbjERbSwvk7DgUoaDEj9hhr1iVnYobzbWe8SJ63iCjMWgzLircWgyIoodp5dyj0S7B6YPwOBC/nIDVVsWYmWuKtFtDII2rC/lUXzBS7ATwE7b+cOK7iOvO2UFv8x0ftKq4BHjN90yyolb7+TBby7ys39wjs+8UuJ9Nx7g4UKZ9WbGZM954cT5Vccb9hvqo7DyqqMxZIjG8mtaxZi4vAW1YWVevgPBSZN74Sr4Q/RKCS4mcz87p7gXdQjEr9NBT0DrSn4Q1NvveXhqt6rQUz71miO5AuleRXgZFFWoIBK1cf00UHt3oINs5tfCm05O88/2P8JtR8f557/5OK8Gp9g/32ErtqS2QSGo0k57tOMt0qDMnuIQl7MuC4Tsjyb8+ylw/QmX0cv6bPQusb8wxr6gwtWkyZODJZ4yATfYYW8fdAuWqq3QNiFVF7PoYm43Rd5R2s961ueTg6tMdF7xKG9fFBISXu4vEqq4N44JW6WVtOgONCtSomCYKRNRNtZb3mh1lDvnoksIB2ruSknsfBqLSzdwsvyR/6Ow3yzxnZCiAndJufFYwITOx7vLff6/H+xf1eLXb9dH4PoI/FWNgJC1TGfwNUazASf0uEmMLfc8x2XDGLo3GAqlHB3a3ce82izRovharlTzxZ0WU+2Rl5QWD05XqzpS0uCLoPw63lDYn6i7OQG7q0SIBecjhUKFBzgYG4HqUO5Bh9omiiVT2HWSu8gXKmDncsWbDpDukBLJAv7gqWV+6TOfYu/0BJdEgEvWvAGtBBEmrBJEs35zzQarebiskBTtvfJhUq6lUJagxpEs5nRifNLCpsuYsyU2TcZnzn+Nn5/t8Vu3PsIIFS5cjNnuGt9FrkZQrRuWM3g9dl5NOmsdX3KOc2mT6f5VFrKMmzPLHaK6KPB8GqbuNpSn88xMqYtHJe7s5Ehf80W48qojmHIUwlxJOLYPpvZDrQHnm47SiHhwvgvLTAZb2/LGsvxYVOGLmfMFwncyeNZJhes4Ziw/VpzhPBkXszYvZAmbBh7yCtmQ2MWsZC2mjfiDajdnDBvDg8EQQ7bAd7I+57KYksQjGB5LO9Sd2rEBZRVF6Q4/Wj7KXPkAbwQmMbyuTFQfCp/SF5LnMi4IAehfJfUbjpI9KnQGiwxVbmEh2sPH07ZH357O+h7d2B6scENxD/uDEj9iixRc5n3bJAopBTVsNsDFa14k0g0a7CmMM24iuoNl/zNbmM7zLyWIEALnqzy1lhRbpZJ0tzWqVpN6j2Leq0BT1qyQNnGK9LUv7MS8V5vTetK8CUax0cH8kKD7KM/We+JpbmV53Jke36eXlEVE9Y9pgiLWln0QfEljHa9zRrYRSvdQYgfwgC3zQPUEJ6o3Mx2OUidvd34yHbCJ9ekf06ZI3xiecRnVwFEIjPea3PBjHXOut8QHRsaZGzdc3oRS3zBeMpy8yfiiSIcrJU1lC47uZUdvm5w7dzp3olZbvqSDhurdBuy903iUeOiwoVuAyURjt8K2lDsCMUcljsgYwfC+t+/htX90lN98T8YnXjzFz2xcZi1K+XpqaFu4p2FIU3h9AzZ7sLkJrz0FG6dhY9F5dHp8BJ+gIr6b78R2YeMFR1PE0Q3nkTg/3OqGC6jaLehUxGUSV+vtVbEncqzeuiEo7DO4SSi+w/jEC61P0a14TzwJQ/QhSNoJoJV35kFDsplSXs1430138/lffz8f/JGbuHD+PGx9lXr/FZTBumNLvDEcYZ/T63mdvnJXk3WOuwE/ZIu0fO7wCIeiCbYLYzzqlPQS8FJQ4ykb6a3g1e5rDAbLvsh3tug9Gg+b0HvhfS1rUsYwpeu1dIDV8lFflGXtF6H5BN3OK37+RUmLtLfgD0x5vnJIUJyj6hKKaYu7si6THv3tUkg7ZGkuAMkGcoxXO0VUhqa/ZmTCLEGS0zrp8fmIgSIcdXb5K7j9Z22/Oq9yyp+1P3v5HeCv4FVcf4jrI/A3ZQQMPPrN1/jGo5eYmpynamY4Y1JuH2TcHw75dod8y0bqUBCPRdf7BrSkFhiF2hHj2znaI70YYhejcCrmVhxp2ZCWINBCuejY3oR41lCZzaN9rBZbuUrMGs+b02m5vwO1vYbggMElxiva+k1Y+Xqe/1qeyX3y+q9CWnUMHwlYebnNP/03v8u58AphYZSeTpxqt0Yzu0VbB6kjXdrzpHcrFElkP22iIt8nG5SiadJoyvOYbiiMMO5SLpiAy67PC2uP8rPHxvmn8/dR78kPLiPuW7Y6ufWINpUNebn1MsSeGpY61Efw9Pli/zIHwwZvqo7zgemQ0cwQh9AR8qBxHTMoHq2iY7MAmhjKFSl4hcQZJu4xSIy6dgHKY16kyPYqLEm12oWlTed54HoPQgvrMcwpIio0rDq4xceWwaed85uGCvcTNuKFtMmfZk1mrXIeYw7ZiIvJFn8Wr/Gh4hhDznJSnnhCKlMkN2HSWJqoWE18NFHDWI6oneMSXhss86ulWR4pTjIrJbGxPI+hviseEDKwpDijtElvcJVu/xI2XvS5rDIgrmZtpkbewYtBzQsy2kGdozJUTYUgl7g9mqIlB32PLuTIRZPEJ1bIOLoXDhMHQ8wWhum7mPXW86T9iwSlOVyyQdZ5yZPb83ar2u9qveacqBylExq3a1PiVa45auFtSryf3m6rVgWpUDzPCVSrNc2tTNSeiTfCtAAAIABJREFU1UTYVd7qzfTo3i43yT9OUMpFFzJbTDYp1u/xaEhmIp/UoIxa4XmRLTFiIkZ3c2G/krV9coCAJBUPRRMxbwregHpHcx3DjcYis+KvZkKkHUeVjOI6XOku8j/Uj7B/KGDlrLzbYP5BK7Gr35izCsQ7OVgZV0AmwdrvNAflFRl3oTYGaQMaKooasH7BMVQxNLfh8e6AU/0XWFx2vPP2o1TrEpnk/NdsLWNiuMwd903xY/M12mcu8j89fQk7WfOGeO22hFGO7aphcQ16Orxl0OjC6jYEfUNZvnJqAkip04NMKFsgzp1QfIO1DqP8aMHcuo9ojPtzc3Nf0GmtVRf9Wu2ut1438e+kkFVtr+erQ6kml65Dddp11lWGfX+XjmkNRobJM46RQ2XuHTnJvrF9/PHaM7TOXmJiWPF98tUs0DQGJTMEQY1ivMpO7wJrWcy6LdO1ETeYAnfZohcY1YMqbw0q3mtux0bsTTZ5qfsqQbrN1bTDUFBlR6iaKXghxHeTTVZMRJGYqgkZLR1ABAA3uIwNymRpzyvM20GVZLCC67+OjZfZqt/DC4RsxGu8mDb50mDV2wGl8TpO5uXxGq53PkesRZ7ULJDad/f/8n1Uok1ra5X7HzjOg2+5LT8U7w7nD/rpP1tRl1s+6ALPrQD0+ZoNxA/65K//3vUR+Js2AioEvvPtDb7ylfNUJmZomRkCitxo29w31vDWAy+ecow1jOdPeF8odRJehOIcFMR7E/ChBVDODeKkzOUnZuW2inOeXXKegOzdTregOGoIZTIqTyy1TxQJFkChkavYykegVMmLPt8T6ENxyLCewKQ4dep2DRlZUHk/vGA74DfPPs//8dLH2Deyh5VMWY3aWMq+lXDNO8yrIIMaLihS9x5sYrAo9F5q2E1sNEFaGGctXmU8qHnCuow8P9f6Fu8r7eUfH7+NSma4uJQhTlGr57jag5kIuhk80YKbnFzZ4WoKT6Yxn0163srjv2+McJcJmSkY5o5CKEFZDdovgZWH1kyORsgnd3gfvoiT0nD4Hojl+7kO6i4N3aBcTOiUYJ9sFjbhUcWViTenjU4dtACvjrtFMVCp4dk0RWJihW6s+Nx0R2wUY1Rkjwn5QrLGSrLBIBjy0Wj7gjqXXcjx0DCK8YbDojSNWNkVGp8k8Uw24Mn+EqWwxuW0w6n+Ff7raJzZgpiInrJE7OCKbDVcnok7nqwyZAwFWUEk6/SVTZk2fRszy5rsrd7MgaG7UfRR3UTcFtbpxessJ01+uTSPM4rMMrSzHueSJn0SmlmXN4cN0sIYk1mbw2GdYQIupS2iZJ1E6tOgTrrzTT9JtTH5m2+vagbp2QpeVk9/d7f3h3fNnxAV/7Z40BeFnqen3V97glfiqpgr5Tu+CrhwNP++lDyeC6giUScaVQkSBckmRrkBmu91j6KI49ktTDFlC9xUnJPRDqtOHz2P1Cke7JQbcNTFHA7KXmEsALxFxoaI94oBI5MgnAkfOZa3fqc07ibja5uv8Ev7pvmJ2ybZeCaledJw7G3GJ5kIjAzm84IpGjUEMwbZFdk5Q1GiCBVK4sgLsNwDRZkGj0NXRdWcQTnQNV2naxmf671KcfAyP//j91CtRzizmwYRGbIh0TMcs/Nl7j85w/2HLZ899Qq//zxM7auy3gqY6Thubpjc6kTKdh0iyobJ/SBaho8AbkDvEl4dHh402JZQ6rxVbA4a3xXw/DoF5vrx3gVW1TmYy/3szIjQuN02rbh216bAVl7D+OpROdaaCioQd2ODhYZ7nqGALK1pOxm1muHu2yZ4eN8RHnt5i9cXnmOiZlklIg6HvZLUxevUlfMbTXJW3otZn1La5CwSavQpBxVWjOV+A7fZkI7U6NEs3eIeLrkBo/EyC+1TLPcvsWqKDGzkI/+KWKRqVhyYOJibQq9N5Ivx1KfLrPuaVBVp1l/w61upejNJ9QQXgjpXbZkgKFPWwUQG6l7wo+JuM8969bQDVcchRjEeUoeL0lDOaG1t8fCDd3HfW3xf5z95S/u+i7r/sBD7D7/+Xs/o2n2WF17hM3/073nimdeojU8z3qhdL+y+14Bd/97/L0fg2jzfW6vz6d9+moXiHrqlWW7KlGKQcGswRHPNcbkEN5y0VJTuIJ67TrQxFKzBDuctTJ1mfY65fqbTr1odUtONgfzjpFxVKyRegfIxVQgSSjhfoGSLznP1tIDGL+j+8vg0pOo/SQk7kwsoZIGQbebtxvp+Q7kMzZWQJ5bP86svfoQgq9OU35yUhnIu9QpEtYa0nxc958oT2sMRf3DP8QglVKwjH7NANhou5WD3ZcrFfawR8LWdb3FLMMkvT9zMdD9jOXa+Mycytzh8In7NTRguxrCSgISsVbn8Z/A/O8etpYDfGKvy8F7Ly5uOgyOGqtpTmyARYjd2nrQedQxb55WMkRdxes4DPfUJnYxzU+ahIaiMwfZp6HSc39TUAvnDZsbxSIgYvvgeGTe0pJDNYK6oaC/4ShJz3IpDY7zn3FUfE6dCL+KACdnvYs74kKKQOeu4o2j4F4OUSed8goQQvz0W+gF8MskBjVEp5NyAhe4r/ERhghuKM8he8JQQCRVfSrBwmd+0hm3AJgHLadub6/bi1VyRKsNgFUbJFoP6G1gJxzygMhsOUUPjvcJ/WZvBScEsTzYyZMpbiNc47zpMh+OkQYWzWZsfLoxQsFWeStcZKPYrbZHZKlnrKVx8MZ8TOj34uK5dZavftdVX1Aamcskzi3JRhIq3YJigdNjHjbme+pFKfCgR1O8Fb5wsiEdws369iBG5U1/Ln0MVgr8odtE/bZSeeJ7PSbV343iJg7Vb6HqRR4k3F4Z9uPwF4KGw5BM3tqRGD+psiQrhhd4FXjfyQdxhXWR4+SkS+ESKmrGUJB7B8YXeApPxZT5UPc64i9g+4Dh6hyESQCHFqJAqJS4EQsTzlmPnrCMJIZAVoF7TkHwgja9Le2plivqwkftVdrdVUMKqs3xyaZF9jQu875H7qY5EuKLzKQ6eVqhDm9IpOo7alOHo8WEeOjnNSGODX3/0HF/xato6UdcyXHPsaRj6qlL1t/X3pHq3eANeIdH9VWh3oCp1rJJV1CpVVqwKMdW0Wot0bepD34tzrqBH7TZyFM5U5HGXZ636Ol28vI4OTA5nTc7T2/XSs7I+kWbmEgi2NlN5HrX+lMzJ9443+KljN7J/OOIPnvoaZbtKLSzRs0M+kSJOt70yW7nRjbBOwUb00iargyUqpAzZCue8DY3zhtINFcSmwFQ4ysHiHK8FFWrxMq7zYo5u61Sb9f3jzNuKrz1XRR/R4loYx4pfGq95I2+hd8Z1yeIVJkfe7q+BmpSxigI0RWaCKuW0I69nn5ShQVO0nj+UeHQ691jUolcqRGwOeuzrNPiNf/ZzNPZV/0rqoe+rqLu2aYnX0O31sEEBa3PETW++0Le/fNP91XK11jLYWuCnZo7wJTfF4PRv84u/9iy/9GvvZqgU/JW8kL/8d6////oI/HWNgK6BrJjwB19+nOULVWZHZnjBBdyf7fDwyTE6HShLaVmAWKjSdh6+LfsRTzzWYijOnE69+lBloZsAjUaO3PkCT5uGxA4iOE+a/5tLLgqSVhR/8jewWYReB+qjxpvrqmgJ1UtcdgzOQ7cH1QlI9HfWDcuLKT/z6B+x0nueeuU4vWAEG+ZWJdq8rbel13VuvVloTMZoOMpNxVnaPmtTnnnKLrIosaBISj9tUQvKfLb5HPeVhvn1o7fR7iZsBlAXIpnBWhvUQSoVYGknR8EKTrWsYTVL+HraodEL+B/HS9w9CSva4wM4cqNhaJ9hILDgCkQKJE8MFRV6QS6a2L4II8dlGWPYeRom9uKJ72phC+yJYwgHhuohQ5wZepcN5RLUK3L7N5xrwsUWTBSgFBj2lw1j1vDh3oBpG/Kq1rfIMZVZjiuCKIyYDod5Ie3x3e55DoRlimGRXmr414qfUuSULVB38N0k5QukHFZEFY6XO+e5PRxiqDrP6cxRxPEqMYcVUeWy/4u994CW9LzLPH/vFyrXvXVz6Jy7pW5lWcGWZFs2OAA2GHBkMMk2M8MOYdhlGbKBnVm8LGGPzWpgMGYWs8YYBjC2kTFYkq1gW2qp263O4fYNfXOq/IV3z/N+dRnvWYbV2Dpz7Dld53TXrbpVdb/6wvs+7///BKq+z24Snku77PQrXJMXYHeOtHUmyzkTd8c28GQkXL6Nspcn8PKoCXkyWuIer8St+UGmU0sXj6s2RgrQjuezXc7+XgHVJ8Q52vRCUnl+daZBXL3cNqc0VZXOVehcdU4hxWrm9YJEt0CcKzX3kiQEMh3wK+IVDjjJpSpuVioeMfPVVisezopByUq2eHBt28Bl1qqioTaulXhClTpXmxPwE3jMZ61aPatrrzvnbCg2vYKLnBrwiqx7BWcqfIwcjyY59vqWc2mT2bRLRVWTeJUobVP1ihwyAS/1c0x4hqesKnaeA+5TRFzrTPPTO/ZweHiMkXzKgVd4tGYhfQzSoiHQuSxLkPGe4lOVO2k7Wi7P3uWqylpIwoLlR6E4hssrDsS/q6tLadioW55qBVxpLpPY53nna19OOcw5g19nHpxpTFxV3rSMPLFJjWV4IsdLbx/m1YdLJCen+N0vrvCRUpGxYpHhCDoty2DFuOzYriIJhbmlet+b0TP692U+ll3hbHVyywazlC363KHV7tYYpEOp76QhoMftdYtOrchUStbiUuVtYXlRS1SprGQKfqVUqDonkCg+sG5q6zqgWstEXtYa0pWUyraA216ym30jR/jIp+dorj7B6EANGZYIwreCCjLzaSZN2u7cKjiVq+xpvKTOWrLBZ9Mul2yLW/0yXRkJG+PU8TeHgwzl93AlHOZwsgadaZbblyjGay6R5TwJeYH77hxefgK1RzTueeGoO/+d2jptkavcSjM34a7liirwXkhHcXXi/ckUvHMx43rKDkim4I6cqEFHKFlt+S7RUsTu28u8+8e+mWJR1IP/L45yT/5X/KdT5AXdtgDd4tRpfvs3foslIfnSDn7+F3+KtWf+hj8+0eJn3vXGzB3crdCyjXORMgLk58/yKWD6T36Xgbmn+NDeu7gw9wEm+wbdYuD/DQdf0CZdf9H1PfB1uQdkxul58s9qkahf6lVJ2i1S2Q+UDcfnUrdSP7wt46mkU5bgkKEzZ8nPyTPNzVsO7LhlozpcqgAkFqsST964Fb7mI41wjoSswVQD9VyK7bm9K/OmoLbkes/XTlmzE1kskEBfvm7pnssqdStXLKn1eXT1C1xqPs7k0F4WvBrGLzjFrFSXapFJ+CAfe+V+amTy4zVaySbTMqHV6tZNtZl5rLhM/V4fsVfm0aU/48HaA7w5dwvjowlT61DZZtg3hkvAcHxsmQyHwpvW8YCu+sp3TXifKjhxiU/sCbmpbJlpwXCQqVTzmiCWLNVJw8wVS6kMK6o+xlAdgs4sDB+DdksFFMvknYalq5ZqAVYWoDENI6Us4SFXgsExtangQgf2yMF/A74ssUQI5y20WpaSOEGpx/f7eYevD1nLTVWPR5YzMDrmW9ajlNeI4J2r89H2Ij9hyihNQhYKT9mYCWdo6vGk8bhPiksT8ni0xL4QfmLgoAvg/ES96xImtgsA2ZQvkHJPKkDSpWQKrNuIkaRO07UpK8Tti65ykHaXsaVjjMic1ctzxfhcSBrcbNvcGY7wuECstH3xKjPRNS6o5BIMsNe2uFnB60rOVO5l0uRM67yrUORKN2AJSdtS50ik0CNUubKMJN/ivmmlob6aRvNe8oNmd7WHnWJ6FIkgXKs0beAVb0RehlJSuyqNUfV3w9mU6HPki2fSFiZUG1b+XRJviF+g0pdQu1xzW1htg7vO8vi5CS41vkQ5HGGHl3eg9xApdZtyMR/zM6MFppbz7Es2iJIWn1dkm1digNjFjO3wCjRMwv4QHujAZ22XvxP47Mxzk5fyur4J9twLlbYhOW8pNA3hTdC5lFWcvUn15HvVdYc5DZ6U6sphlq2fKlQlqO7NIus2r8HAXmhcylTzAj6lesyXWeOYwkildFJ1TEp4Ve2FDWQ7ot2vipcsijYNaZS6DOP77xnmpkMDvOGRa3znwxf4pS8OcvbQNg6FIa9dTdmjTOf+DESKsuA1M36ffC/THYb8oawlrIWlqB8uxEPdbt2me7w+Jdhou7R20++0TWod6OeOLpLsdToFnCBalUFx61SqFDAU2DucFV7dKaMFrUpbep++W2iIZxKCg4Z3vHwf9w+/g19+5E/4vU98ij37lln2DpIoLVk0j2jFnQNeusFMUGNXYTdxvMlsZ45qEDEfdXgsbrAjP+n4dhtKhrBu1/Gq3CRhOEKzq4HgOZajBZaTOn77Ml2TYBKRSdQmLWcKVol0snKiEyqtti6R5nfTDYdRo//peN2pzCOvhCdQpzHTz6rvDt84cCcNuiLylKiiAbzlVPbpi6WSyHZr74D9/9zpywidz85McfOr3sa3PHCMf/NDb+fpM/Pcd+Aor+vTBurWg2c2YWZqimY3JsyXGb/xZWxuNPDa8/zv/+5/hN3fy+FtGsqzdZf74fp/1/fAN/QeyM59awMuLI1m/UARaMRHKhaoPw8LTaUV4KJ0xG3RSl6dJVWMNMCKaW3lCyVgJ/6JBkwN6FetS0Vwfk96rDlTg7wKF1qaaSWtFmLZYC9ZzGRmMFyczVxRIrVuNY5I56C/pwG5CZpU8kMw7Plc6S7zrkt/Jdkpq0bRSCmeRmAv7wLiJYzwxF1yIEPEtE28aNFZeQynTSb8CheTpmvlmfKNjOR3kESLXNt4nLf3jfHdpZsZHLKOoH24YvjUacvwtMAFnKlbbt4Gg6FhsAOXW/CoTfmy7eKnBT66p8DL98DmMoylOLXYbN0y7htitZZWLH0HIYygT5w6iS03YL0NwTVoqoKSwuA+y8JFS+zEJWCXIanC5ia0z1oO9xv6h+DEguVI2zCTWnYpVaMM0xsgjzHFY45VxEeUR5nlY3XLJ2fgAQ9mlPLQxTneL5qUHeEYr1dMmNpDNuaNwQBraYdf68xwUzhOAY960uSKXeFktMnPl/dSSiztjuVeL+SJtOm4QJtpxBopnzIBq0mLSd93/mqhqkzN5wmkSPYrzi+rzy9RK9/ElHiO2p60xXh3iQtBhS8777g6n042WJe6Mxznu4OSU3g+Gxe5TMKpeJPdtsPZtEtoI0JV0fx+0uZp0ubxTI3iKnEZH861WaVQdTUUrTJ6J6VCSh0XKwN8RvxAZ2zc0nIAG/S5iCWdvm3bcZ6HVh5lndmMO6fAY1XsVOpRi7Z8O/H6Z7K/o89WBVB/QQQxAT3JnEVh6F4l173GZuV2LqcxuxDRPceNJse9xuMGY0mCftb9Kuejdeq2w5rJ8VS8QTMMWLaG8x3lCqfMqCKarDPVXOHn7rydfbvyFLRPY2hegpJMdQuQ2+OKkViBmrJxnDXnzaziiwDLlpOLNjUHhe3GGfaWtNuqhoFBXRfi3sHRtoGLAexL3QLOFXgEJnSNq9KlYUblqq0pV+CpZJzdiC2l1Pp93vTmbZy5ZZCPP3SRH3/qJJWx7YS1QbY95vGyWkpl0NA5klXrhC0ktC1sWqwq/+q+xeAPQXsandTkdxsk1uK4qnFgt8advqyD4PgX4s5pcSklmDoGAnKq7CmDdk3IrscT1imiqV9evPpMfQ+1fCUGE84QAJan5XGLuSVh9y39/Oret0Nxkt/72MNUty9TSNdY9yfd0Cf1dUSXnXiO4rFkI24v7nZpK88ldZ7szPBkfQXykzwQ9HOzXyJWeodN8L0CE7kRSv5LXcV2tTvFWnvKZfx2tOAQKVdm1+J8SuTg7AEiUn+Qvvw2+rrXONGZ5vn8doYUASjhkpGdy6xrETvxT9zFCh2rjeKMzDP6gHVfGs5c7b5oylddD8LOL+i21Vkdm9zFwqnP8f6HfpdZWyRME4rD27jtxn0OrcuhXreovcQv/Kt3cPDgQd79Y/+esFhkc/4Ub7lznLP938b88YcYqfgIoW599gvakOsvur4Hvq73gKVUKvDON90KG10CITVr6eZzTvl2YK/hjgGPim+oz2TXueYsB9IEPKQyk+hPXHOBsF7Mj1v1blOodm/FLhAYgNxE5FJvxOwX9SiCSPYnapGovfNKg90FcZiJBVx3rg5WLr43GjqyTVLweA7+78ufg+nzjBdLdAldzqebiN3EmXPVIlXqFCWVuflLjTHsbC8aJmRD6th4jXzaYns4TL+NmK0/7ZSHpngrYRgxHGYqwP5+nA+YgNlMYhkZMI70f3nV7TaSnGE6Xed0YvnNYpGXlqDTgKqSOHZAZQRK2wzdHESrFrPHUO1TJBGMTEJpNxQGM5GDUwQ3ob5h2VyA2g5DvYHz9RrcB2E/7KzBeNHQrGeFEfGgxD1a8aEte4YGHKnCPWNw46hhLDEMDMDuwx5vm/R40whc1P43Hufw+IhNyeu4mxzHvJAfCPp5wO/jUipems9CUmemfdEBxL/uzHK8PcWPFHawyys77zMpJlc9uNEr8oPyeTM+a/EG3WiZMSlbSZntznG2cZJA9gjxpuOryW6hUDzI9tx2Xu2VHHes65W4Iejn1/KTvCEInEHquslzU26SB4MhXhLk+eZSwDfhM24Ubh9x1lrK3VnXynIGxDYiaT7b47AJhknlqulDtHIN+vYfwtmzsptQh8QSKsNkpHDj17IKr01I5WMoe4lw3GVs5uMVbFDDU5aWrFwUQSayuZIguleznFmvjFc65q5+14p1bgri1Qk0Ku2i5fZBMb+H1eYJvi1tcnu8xHNYzsuGthPzxJzlRJq6Ct15a2h7eVb8ihOMrCfrfE5tZpuwmLZ5Otlg1POYaoS899abee0uZ5/N2jLkS1CegI66z7mMQuFVMo6sy011NIhe9UrXrBZoukYFwBSdNW1pXYHiLuOqv/U5WHwGWvJRVoVO+0ArMZW+tcBTjlXWncsqdsoFEyAUbhbnVmOAsqBTQ+pZ18I8eKjIv/zlG/ijXxjngeoVfubCNH8ZNvly4jGzZlmbhmszMDcPs4vQlP2afM8XLOtPW9rPWIIl6xJX4jnrVEFqjzpu4GZW5HH4XY4zMhrWLwTcApOBNYE/0STq1okp3LgmACfwpy6kFqzCOrPZwlXjlav66XsJLIqjuGhIKimj40V+/Z+9lj/5/vewee0ayzOPM55e1bLTnQ++RD/OpPsqg8Yn8gosOd5qiX2FPdxU2s+oV+BEtMRHm+d4f7TiNkNBebd4RbfAGCkeYlf1bnYOvp5iYQ82HHYVOZs2SOX/KF5ptIjRgiLop2oCJrVIiVcptS/Tjlecj2QqhWx3NrP6SeSQqHJqKfP0lMm2ExB5FMUvoc0PvPUuSiWVXnXrAajeo6/m7gW3XzPklfB/vfcN+K/5ID/2HUf5pR/5rFN9xfVVLi932L9r3NkZaEPCwgi/8Yef5H1pQq5QZfXyCSYP3Amvfg+PvOnlzJz/MunuQ4wP6ctmJ8hX8wWuv+f6Hvh62QNanGiRUi4Z3vxt/fzWv52mlDsGzYgidaeo9OQmr45ULWuNutVq5keJvyNTpbnrWma6uokipdguzZH6pxWuBlBV9QbEIcs4LmqpaKCM5zNbL9mksMs44BdOQ/SMpXBTVt2TeLCTMy4LcvigJU0CLqzN8NMnHyZfHWXDhm7yVQ6jq9Ip9MlX4GNPuqcWbNBH2m3heRVS2ybUJJzmSdIW+fxuiibPifoXeWU0x9DYO/iP3Tle3p3hpv59lMKYxVXjOGV1DwYCqKq9uWh5tgGLgeHpqMWTTcOvTOZ4w7BxiRBa/UdtRZdBWZw3CR4WnH0eOR+6G8Z50Gm+a16BUClewhQaXzRPVAwFL4tFGyzDqRkYHIHRArQE+pay/T+nfadKRQClVubGv79i2JBrfggjO2FlJoseq45ZF/F0f8mwc5flD2aNm0wOeJaLacyfpXXebkJ2eQWX/Wqs4UyyQSlpugzY5xsnuClp8rLyUbZ5JS6SsIThgG+4M7R8qpGR+WtekQdzo/xtZ5Z82mYIn7PRCjlPsVpdp45UtJLc7vuLR9jwitxPqo4Zx5MmbwyH2OcFfEIJHCbPvwoq3CxcZmIeiQ3HN2LmZVBsrfOh86MlIqU+5HeTxCv48ZxT/TkEIyAXjmWxR8qDdaBK81EeQ5BhOj2nCVcIxrYxxaP4hV2k0aqrKFaSTVY7C7yueJh+3/DhxkX643Vaak3pvNPJrtWOPt8pBiM3SfrFg26iTVtKb+2VfnR0nTBCDMTY5edK0n2lc5WV8q0cocs7FIFmPJ73LbuyYpNr83+LybGoFBITMeAp4L3Lx9OWi3GbIOJ0Z54Ha328fXyUwoihsWqJvwyXqjA5AlKWy7S3K0HAJhTuyGgQrhK+xzihgJIVwruMS33oPm0dBaK431BfsXQ3rKvSJ4NQVTLMgGHmMwmUInIiheor6uQtZpw314LVY1Xp1UcU8NFNr9NNFUGBprIhSVOCguGtL9/Gqw4N89E/v8Y//9A5PrZ/J+8v1XhwAwbV+s3BQL+sXbIYMYmWPJ3rRddkyP6+OgD6U6J/CLzq709nVUctPt24JLPlpsWq/SxkITP0rbax1rby2RWdTItYtZSFY9oGM2GdkaSq5gKmGtcc7tHhVwtYtit+Sl/V8J1vPsRTx/45L/nNP2f64iPs2ncv8+aowx0bSQMv3qBT2MPVeIN+L+fMhkUDWZWi1Vjnh6nzKZc2+ExrkSUTcjgccrnHT6QN+qwl9vuoF/ZSCCdcTmsaK2i3he3OOUsf7Rg/bTDXOsVMfq/jvTa7M6T1Z/Dz8m9cyDob8nJU9VgtFbdKVzlSg1EOY1L6yoZlGrz5TXdQKYu/mukPekfyq77LYO4Lebs7aXwe+K6f5bMf/QC/9N7f4PmLG5w8dZoTTz3Oh/7isey8yprO7mIuV6s+8AryAAAgAElEQVT099co5n0un5OrKrxhYIF//Y77uO2OV3NxXlBdg+7WGekeXv/v+h74Bt4D2SAr/KUehELX8To8v5qjoxzGpnXtvedXLPOr1qlS3ep0HZZXoXXaZgafGpzV4tBKXP80z0kppo/VgDiaDaQuFUmDqua1HpfFE1Fbc+myJgxLoLaGVr+9qz2ediIzddQIqobmqZTHlp+GmWVGSzXaMs4Sv0mqMLGvZGUiqb7sLMJehqEmdldB8ghMTiw7GmlEzsvR8YucbF/iUHeWlw5/J0WhsKhBeTB2realTYjVlunP5qZRVS0jy4USHJowtGzEx5ImP7utyrsPhwyq0jGYWZaoIycazcxpaC5ZSuNQe8DQmrNsPme5dsbSasGYcl77oDYCUQUGJ2XjgrMTca0jxY81LE9+zjJ/CdYuw+qa5bFpy5MhfNuQ5ybZW28wTLagoaxYTy25TC2r9bcqeOvncVYUrY5ldNPwugK8Jgd75I9nPLbbmE/ZLqdt7Pg24nc1TciNpRvcQaquP0LB+cEpjDylnBrWYkvHt+5vDaXwnFX8UEpR78tv50avxBfal8m1p5zJc9qdwfr95GUOHU5yLreD7zK4WKvltM1POi+3HM9p0sAncEkXmfmtJrtdxnLSWj5tEx5JGlyJVwmDMn2FnQz4JUbU3lI/LprvpUeocihWnm5CAr2QYjeW62TVc8qZ6tUMvErWInURX3k8kck7qojN89rxG3idfGb0+fqcaAmT2+ZsS6wQu+qAspZwubMiUEb45Vswue2uHSbEnpHQ1dKST2LXFbYCU+AzG4/SihdpxutsuLTikMMYdnseyrL/dt9wsycTWuOMlZU9OhoM8Crj8VrPZy6OYDHmfx4aof+qT3shpdIPo7dDowJrSiApGOIV41KfZO3RuWCx53otRbm4+IZAPDtNdQGEtxknFvALUBGq7kBJ5+keqOWsE+wkKi0vnOfp/A5sRRlfPZcYGRwKTOkmkKdVpLDzSq9qp4qdbrrXsCMUJoV7mjKyN88Pv3sXT334AN/jLfGLT83x41cS/vyUYXMa5p7NFpm65MXplRJWVDNlUVcPGHyZEsuSRcBNN3W/9/QM0TW26Om6xWZhC9l2qSqnIBFtj+Z4vU5U3HJmdG7k1XdNJA91IEzGqVPL1rnVGOl93BgosIwqmIvyv0y58+Y9fPG97+KB8Zdy5dyjDMVT2ULEqxDkxl027ETaYp9fZj2ts+FMo2OKNnV2QOo+/FxhF9+aG3Vl1unGl7mw+UUK3TmnZBawD5RyIvV6YS9+8RB+8YjLac0WurLVGcCv3EmpKM7HsPPv9At7SFvnSJonMqW4lFuqQCdrvSg79ba31K9NV4XW6lEA78W8veBKnVGzG8Md3/x2/uM9ryf185Ty/xPtdkShmOfY/b0Jx5Xks03cAms692595RvpdHRRanLJCvZ+kP15KWiv367vgf+e9kDo2Mwl2skc2+IifxFM8j2LKXcNe6xvyNYCxmqGWPFgmmE6MKxqW49fbiX3F8Fa7RrxaDSgSuSgQVGpD+KLL1s2apldwrAKaXKVELDTNaZWhuyQVrP2YksebItQ3pEBQl+t3GVLwfP5+7k53nPprxkeGmM2CfEkxXU8pnLWQnAJEZvO+NWKW6JJJ1KrQat3gx8M0IgW8MMxfGfumacaz3Nf7UGu+DXOJw3uLO/jk9cWGaq2GVwN8fst47lMoTrbsqx3YS4QyEx5PNpgtA1vGQ4Z2mHdXBAoNUr7pgxBQzFglo5AqiYB8RIbkK/BqIyH5eoxmGVqqqVTPwvBuCEvZe1Z67Jb25uwsgiVUXhk3VLs4PzwjoRQVaHEKHbJqJDgWlpzieVYv3HtNpkYj/mGRhdaUi9PgQvcSCy3FgzDLcufyWA4bXNP0M822WGkXRSutd+m7DI5csS0MbRrD/JUMOiUnQkVDshKIwh4rqX6gmxeEi7YFLWXrtiIYTwGwwHG0gYzymjtZGameVugmWxyX/kYrysMMp+mLKSpIjpdQsTJVB0tyzbPc+78Kv72GbiWZt5382nDBdrXk00G8bi/sIen0w4ltSeTTdLGcy6twalY5U2nyoUDbjoI2km5jC8nMYMryWjrdfJmHnJuldFLd1BqhVSLB/JVbq2O0l4v857KbfxO6yyjuTEXiO4V9jvfOZtsuMqeWmAkasdGeCaPX74tW3Q43p4Q9EZmhOz87FRmLRNG83Tqx1nrexkfEE8hqPHWoMR8GnMTWYzYX6cd1o2Paj07bcpZWbCFfVyqLzNSifjY3fvJP59jNkqonTd4izCxC27ck9l6bJ63rK7DkNIXOplYwtl4HDL4mxaj9mNPEOV4ZmqhSmgjLmcFrs1b8quG4Tz4ew2bVxUPJ+7FNR562SsZahWwEl1JHKU4rqlMOGFkYyRAp+JPSQCv16bNaIZZVV8VZ2ekLB/K1Nkl3XlLhf/zt/bzkSdWeedPXcBOjFM9WGFfbJhdNChxZlgLqwi6mzDQsc5qSBQFcfsjTeEXLfkR4zJl3SFWdrIKUvquooxo7FnOxqHkfLaY9CZB0YaOB6xFplrKOglHIFWsmlT/yr2W8aMEJRrn9F30vfT1pMZ1Y5s6Cwm37y/z0Z9/Ax/+aJn/4eFPsmvXXSz6AyR+3nnFnYrXnQBmd1hjQfYhNnKq7j6v4PKVF5JNchJ+eUW2+zUuRXMkSYNS87Sr0mbrX49UK0htw1aSiTgvqjbn91Lx+4lkZyPwrUg9iSKUW6zKnhSu8bLj1f2DiEgUAU1yjl4Q4eu46hA6L6cXb/brrd1fyAf+Z+BV7qtRLRfxg5BypYTv+wQ9gPaVnySJuf5pr/h+QC6XI5fPkQtDwjB0hMKvfP31n6/vgW/0PZCd73Dzzp38+OsPcPnEVcpa+sYdGn5CWdYiyzj/saAA3StZy1Y8GCOvtYpBlTbHn9EKV5OAFuirPf8orVob1g2KaddQG80mZ6sA0ErGz2lp5HDL316LY5vSK7Iw8eRJi1E+URY56NqZD19+1PVzpUzU5Ku3e0F/j9Tedq0EVV1Sr0BeuaTqy7gBMefGOxHWjYjByToFGzOz9hneXtzJXfkhbrQxz6j96Je53F6mE3Twyx6XJShczSxNnpTbvQfPduHN19o8bj3+0301Dh7JxB4FdX2VBiF/rxSamzCslmzFuKqMWl+VI1DZCwMlGB2TS79R15vkElS3wcaqZfFzll0V2HkDlCtwZMDQ7FqeX7coWHu0bOj3DNUIVkXr6kD7gjSfMFiETsmyPgrdkgChdUHsVaVxFKEWGob7DKeallMdeFsx4J4w5U8S5ZAGvNwrMpK2OR9vMNtd4HJnjrH8Tr6ndCP3hoOEXpXppM0vJXX+KqpzOk05i3ExSEqRXE8b5NOIJdtxvCBx63x3rBLHIxOPUdYgk+VbuSnv0zEez+hYEPJsCsIgG6SowijjYwF/4eF+A6dtwlzaYr111nH29hf3OvNUTYAXrGUjXsU0T2bkMQkc3EpDe0XgQ3yh0PHf3MSnapubuXteNSoPq6Xr10hjrUwykCkQ9uDQbdSTMtvClG8ZvIU9ocLUG+SdrNvgF29wFWKbyj6r4oCbQJ5OOi+s4Zducn/bAcxk1VXz9LetIs5EHi3sZ7V9niReZShZ5zbb5VFr2LTGVVGfT+FbvJB7MJRV4fNyVEzA44uX+XQ4xccf2MnL7q0w+BJL14dzS5YrF+DkI5bmokVApbDPuIrw+iY016DYZ9CCyQsUuwWJRE/SOExbEnHqVK3SRD5raT0vg2XDxqnMjUb2JO26tFJ6R8RNN5cy/ce6zThqOYtMgc1+BRpLWptx9Fxag9qWquhtzeg6PHpO44LoE4pnSSxpN6W4y+N7vn2YMw9tY/v9K7zls8v820bM49OWtGFoKM43gdpBw8IzMPVxK62XW0imz2Z+jjJDdgtNKesVJqzTT21TUUm2TgF58/UsC3WyiUNnz1hHJ3EgUNg7D4EWmhq0ZImiBDkBQhkcayASmBUFpWux4s6qAFwwpLmU4aOGH/6RV/GBH30VV678PcPJOYLuNElSdyrztc4V6tGKWwdr2Exsh37Z1ngFPmjyPJR2uNtG1IIy+wr7GS7spJ6XWbwhbl8ibjztfJLkWph0Z7M8YolyRDEw0B/UCK1cA7ShOs4614WyExCFwPWQ1W7VwdACx3dVZ1WUC/mQqakG+b33MzLa49o4rJSdH1/L/1unwD/5Gaq4JYnIrSlpmrif9dim/5VlQ33367fre+C/4z2wdV0Wtnn036OlpvhHWW80aSS05zLi/ew6LG5YQg1oDePIxXEZJNJySEVdLBkOa4xQ66Zt3YJPPydaLSvJugbBVQg06AmkadDUql3v0SpeXTKBQh/y+wzFHWoFZavkdMPCks/cxjU+sPI4/cEgEjvIbkKDTuZgL6dSPdePFwy4iVttWOUabqVJuFKiTbBekT4bO4L6v+i7g725AxSTCClAvxvfmeaeydc41YicYEHZriXfsN7KDIZjY7iiEkbU4cO3DXL30cC1h2WZsDEL6miYeetUeqog9A1AXp0M5Veq4nccp/yVsGFNK3pxn2TqKu+7HVJEKoINZhfg8hNwbt0yl7PcMmj4zoJhZwLLLVhTW1Z5ueLoyWG+ZFjvWvyW4cKXYeWEpT0vzJxFkUmQIruUDbWqCtAX4CqQccVyNOzjg5U859MOx9U2JKLuV1iK11wyxCFNCsQcNp6zhLnf7+PXgzK3eobLts3Hu4u8VybCNiWR6CRZJyeFX3eBlaTpKg1W7VR/0Kk39+W2s6dU5QPdFie6q1xJmyyagG09CtYtxmcYw4jFhZEsWctfif/eveaiOA7KMqW438VqnbCxyz5V1FGrKeqMBm9NTBImiO/WO7Hc6qF3QYtH51r2vUnMqVOV/Vpw9hMSPnjGI1T2panyrYM7uFu8UC+hL6hyNLwB6icI3UmrsDlDULnDgcS0PeWU1/Kry3IzxbGr4LlkDP19HxvPu610B0KAV/s8jThff8p5Kp5wjxscB/T9FMLyRBrxmWSTUwZOxBv8QftZXra3zmM3H+VYGtB5Imb3Do+DrzJsu9UQ7oSn1uDMJcu8AE4DysOQ1/W2CitftE6U5AWG8DCumtU+aelqF4ZkEX0eiLcpy8f+Iozuzwo7yXOWxgV4likEwbuSpIp/pmFECzENLlvecBu6znuxXFsLOO123dQA03M6ZAJ2eqyZXl3y2JC6z0o5+Mo+fuUndvNXv1jmr9sX+dXPz/JwK2Ephfo1y/p562KsJnYaZ9CtxBdRhB3wUmisPlufq1NhxbqKuMT++rsu71XWJQey/FeHbFWBGzSON9dr/DmAp+fsUE84oWqftlefo++jcU+rD7W0hfOEUfU5DUOymTpj9R96xTfxu2/6PqauTDEanaOUrpN0rxGagDPty5ztzpJ3tjYJZ7qzLMabbE82qaYtZm3K00mTUVLuVNyfP0BaOsBY6bBLZmmlXRqN56H5ZZdL7FohAmdJnbl4M/PI00JDO0JVPFXqNIYGsoMS4VlKkIZ7r3U7JoNcfap6tuv81Dt3c+SQiIdqYOrLfe23rP/5T3zO5uYmi4uaXf7x28jICNVq9YUZCL842/yPb8j1Z6/vga+jPaDFW9eV1RdJZXBp8qy3UjplGB0w7Fq3rG3AeAtK1axdGF+D9WcslWGpKNU3y8ZxAbhQVgEaO2SJoAgfedIpKFx4sQs5Dawd5T72+HfaFxoMK7IncOnztLpKToDaIUOwD+SP93dfeE5kPfqCA8zEKV5O7GQRdnVNS7WYd1W7VMaczr8sJFH7y60+NbkXEJE4Z2MWGye5LdfH7up9VFRZU6Wp7blEob3GIwkGudi9zM3+ADvyhk1jWRVVy4OpJOGRzgb/Iizzip1gpy3abG80M0rdlAhEYLUvy9LU1xse7CVKTRhnoppWoaosXA8StYNUaVuG+KR1VgkCv80VS7QG42OGviEopFDaCdeacGbdcksRypFhvm1d8WDFtxwOYKNluftlWQt3+qSl02fokyDRzwBmM8AFt6/5UPPhkRXLmIW7i4b+MM+/i1o8YS01W6c/6OMt+UkeThrkvDx3m5C9dImICAmpej6jNiGxCbujBS7ahGGvyB6/zEzaJFHlKl4hal/ERnMU/RqdaIH7hl7Bm0cGeGW3zTetTjsVzYxf4g0mT80mdI1xn30utgx6qvB6PN1ZJu6oNJNjrHQjk+EgTyd1J7AYtDHNzrTzqduqyKld5VSpzsJExYdM5aPqmNvhOjDO7iR2XlzueWd2pvltiHLSYLZ9njeU9nJ0tEaaT6hEhpnllFeMHmM2PsmX6osUFUVmVIn1CCp3krTPE8UrhDoLW+cx8SZeYRsek5mAQxVDbZMfZX6KLlC9QxL0MdSdYYjUVQGbikwLQn43TbjLJBnfKl5lKllhdXmV33lgJ991+CC1nZ4DDc0Ng6rb/fca9oxY2kpkiY07jxY+icstrd0AQy81pLOWhVOw8QwM7LBy4CC6aGldg9IdGgKMWxDoEhs9BL5s9lStkoFwUwp1w+XU8mRDPcsNCFWi7g1q6pVrobfFjdV1Ln6dFnwCfcIKAnF6vQCmwKCAkXgBupchsLgEqvbptbEh8S3D/R6vf0uFx27fze/88QLv/sgVfvLG3RwueOyrW27fa1x2bXsd2qegtD/zpkvqGf8vnc8+TxVEeWJ6M2C3ZYtMtVuVX60K5FbHwQFCAU59F227rmnFpgjASQWr5yWg0Hu0SNV3UTF4Tw9MSmmr4VTVXx1yY/FX4Z1vvJ3kUI13/+qHKWw7QV+4h420j3zScLSR1ClWq1SCIUpYJq3PWRmne3lX4IxNSNn3KZByq4XTXpFuOMmNpQPOC+9U5yKhFlexyH11x6Ezyj9257uMsbXabrrMa42fRi1WXQpugazyox5n1brsQOm3KpYpVzgb590bXoT//klQpwqdAF2tViOfz/8DcNPzajOJI6ffVyqVXpv1Rdii6x9xfQ98w++B7PqYGLgDeIpYaoakn4V8g6UrFfrClL4ihMpd7cDGU5ZwjyG/x1BoizhsiOasS0BwC0C1AnvVN7nAy1dKpGVF+vhqeWgA7K1uNa5bDY4SUuhfMeO6mh0Gex5kTZDeqjnD49LyCj/8d39Dvm+ShbSAF/Q5Dp1VgHlSd0IIG627SCnX/nLqiy6elydVmytquvdocArU/utcYPfwmzhSDthWSrGJcT60sz6crVs+hyr9UBtKaTY8ZuOMJH6ubvjLVpPXjRb40UMVyhPWtZvVahVfKCfT5QqszENNHL4h6yaK9RL0qX+oduJS5ruXV95tHea7ltJoVrDodqGcwNoZ2PFSQ052CachVvJZEwYEquezSVZhGPVAKQ5ZNWtEnnjAYy0YaVj2VRSnZhgrwbXFjFxeqsHAIJyZl+lwRnFaT+CogtrbUE/gmFekhuGTzWXeVZjkNj0OcgyF8BkRyq3ECjFXHQDr8krP54Sp0VDbO1qgk3app01MZ9ZFIgXiNkq4kLbJqwzpVbi7MsJji4bHWuv8s9IuCl6B0MYoDeFZLOOiY1lDUUm+NuGJ7hydaB3CIe4KhskHA3wpXiewqcvFTdpXnKu+/LkcEnBtVxEbt4yGM8sRZ87qh1jlTbkqm5CF7DiqyDwtM8HwnN1Du3PJofNXVm9g/qylNGbJDRiWllMOFSu8Y9s9fOmZh6gWJ2j4/eRUYTGGgf4H8LqzXNt4AhMvOo2tkhS83Hb80tGMnK5ne359TpEbDroJe7mzTr75PLvKx5iO12mIR5V2uULMteYUNE/zhtI47zxyD3eVBginE7rtlHS3YW3AMijwpGtsxODPWG77VgkjDPPPWBqruGSWcs46UFM+agh13B8VJxPyeShNQiBhg/whn7ekg8YtwuxAdp23l6C8B2ZbcHxVHno6IWp4bZk698CZxkQBHoEfcW+1i/U7bZf+6bEDa73XaAGo1wW9DFdV5vU9XDs2a9s66zQVWiK47YYS7//pHbzirgZv+/XLTKQ5vn1iuzMGv2HIuuQoUd8XpzIqggyUpVpVcMn6jAyTM4qH89w8YWXUmFmXqFongKlt0/Wsv6eKW2+BJjGYM2XW75WYo3FMC1ZFrc33ZMqpcYKvLeGYU8iqSt+zHRI1w/dTfuCb9tFXeBdv/bk/pj35NJNhGXWGQ7+YteSjFRKvxFm/yHOpYdiNtCmpErJsxMPSuqYRN6jrYELWfBeJkVn1qDoXjmWG17ZNqvMsnMDLT2Z0A3lE6tx2tjqS/ap90vO7sUo90RfVl9aB0k07o0i5MpEZi/SefTHu/klQp3arboVCwYE2xX0JzAnU6Xd6XrcoihynTs9fv13fA9/oe0Ac0a/ltrXyesN3TPDbD1U4e63r1KvqesiDszqQ2WrEnYwrpgxSiSH8YaiKgyMOnFbm8rYSVrqWgT651duZHo9GxGXZWU2bTF0mYKMsUy1+a5lfnavsyfhTXpAKEI9h4yzUP51Se9DjT586DotLDEzuYynWBK0x13cpElvKRWcW6wYnsZsV6SfaRcupLp3difM2UEXuPIdqr+SaN8DnWwm3RoZqaql6sNsaPouMd2t8pLPE2c48nXg7a92IVurz+Xidy8kM/0fhEHutlRWUW7rKQNlxdbYbR/IXuTw3aLHit63hBBGyIBC9Twb3JlFLxro2U1diiSjzpqsJ+PXB0H7IjxnaDctmF4YOZ61qVRsq4xD4hsunLc0O9Pk9c3wDuwI4sstw9bhlU7kKE3AkNrQEtjVWL8Ba1zIb4dRziuDSyDlcVSXQcKabOF5bJ/X4zcm93JkG/F09xY+MU7iK8/6YfOciSQ8N7wryLjFjppvi2Yg9wQAr+JxIm+TFpZN9QrxG2p3Gt4nLc91fPMRnkwnm0w7nDXx/UKaCzwQhBRJO4DnByqPxKmm0zKIJ2UxjSskGeePRyG2jIYWtV2TaJs5AOonXszQHv4a1quZlJRQHmFyrvuZano7VrnPEzdi9WdtNar30dr3WCym2z7O8eZrvmTjE/TtHGAst1W2Gp8/Djq5hwqTMdXZA/0E82bXYLn25CdaTNjEeN5eOsNuEPNc8TSeawzSec4sPL5wkF44TRUs9X7s5vHDcTaIpkas6b7bOEJZvcNVPWheYFkm1u8x7+gvcs/9+7h3bzaQOen9CcqFnpeHBhKq7FeiuWEJ5LPbD+rKl7MHYEVh4OqsWrz8Nzzcsh+4TT9RgjkBnIYu8DUZg84oltxc2xMM7bpFfdmKtYmKdKbgfW0a9gMHiPE9+6Tl+8CX3c2RoZ8afFX3DzdQaPHr0CgE0ATxV7DRcyeJIlS8BKD0vAKhDop/1XlX69Zxu6iAIX6hwpMeyDPFSgtDjLQ/2ce/NAe//g1X+19+b4Qu3jPGWhZAHCyljZQh3Gtd5b1+GzVmQL3VhQvmmjjaMNwRcAFO3oLFH4FJmw1K8iloilxvRzaRolxhUNEvR1LTdoou4AUzNgmzMcuILLUx1Kun9WmHpNNSGiygnaoXWDpsWfy3hLQdGufu938u/+dkP8kf8LXsO3MfV7hImuoZXPEySdvDTLlWvQN0mbPp5OlaLzTaHvSLnFQtnu9RsylTQz1kvJGpOu3PR5nbhyXhd9k26JsS1a51155mX344V30Mb5igKvfKbKoTimW4lsNiUQtDm2lTEzoPbect33+0OyVahzD34Gv/7J0HdV4I0TXQCdFugTn93i2d34cIF9/zXuC3X3359D3xd7IFiscjkpCJkNCJ+9bdCOWG9L8JcSLHDCTUbMR0bhn3LSNFz6RJqh4pfoYWqwIGc19PT4EkRqwFbPBhNEhr8QpslQsz2KnWyBlGRRoO1iMoFaIi7PmYoaKBzg322QhcNyo9AtCVz1tBZiZheUT3KJ9WI64sgo4KRHHRVDvQdp8T4JVLJ+7szjuQb5iZdlS71inguKswj3nyWmt/Hmdxu3u0bxwnT8u5KZNkbSHiROu6Y3ysrPLfe4oFJuDIr4+GET7au8YsTZW4fC0iqqVMMyhZPIM6UbMZJFFDVY/Fvep5gqpblB+QRlhUhfF+EfBBlS4WRiwottzAqaotUwmrLdqxz3y/Ls0/ta8XzVPU5hpzMhMfh+AnL9DR8uyLDNF924FU+LA3BYgOnDJyegf4a1BuWxIPtY4bleYjbMIXlSWLunw0pm4RTyQZP1bv80ctLvLrdT3c14TuqMN2Fx5YUYQXbggoLNuJLaYM/iy1Ba57PBsMo10NWz5sIgK07H0C/eg82ukbauYyXG3OGqLGNmGpe5HPxGjeHAzzqVzjqFdnhB6ynAZs2pmUTzsfyuOgQpqv0ixfoFWiHw5xSuoRavjr2NnLAT/5yassKxLvKg7MoUYkkdXy2LPNV7X05YLsTNavoSSbpEIWgbTa5iUNVSpss2xxvHTrKkV0hrSBm/qphftFywy2G8cjyxVMF6H8FdvUvWW9fJfCrHM2Ns0bAitpewQAj1bsYTOs80zjpbF7a0ZzzstP2+lJqx3VS9/e1MFLsO2xG84yufpp9hZ1cSIv8L2P7uKVyhKPeOP0ql5ZTuhMpNRla98PmBUtpAqYuwa4qiBohEF+60dD6pMV/CZRKUKlBqDDfEVh6BJ76e8tNh0Bq89oqLF12X8cBQq5JFAXCm0O7ISlC61zPh03dUYEUGcSlz7P7rpuoHVZUmkRRJgNuaqEKzEkQoX+6yHTtq7Cnxxqu9Jw+R1U6fXFx1ASA9HuBJ4E5YQ8dGn3WsN5gMB2jgBhMJWXXzhK/+K4Sr7tzlR/9gyV+4qTHjx8b5TvmDYe2paymxhVh9d2d0CuX2Z3kEoO5ZGmtQn6/MpgzEKYIMkfi1DZomwTMtK0TSsrpbYvjHSuYOvtOjt0h72opfNQ+lpJ4m8HOWdflV/VPvtbupt/JEUf7qRiz+0A/7/ul7yP83O/zB5/6EnsOHOJKJ8ZESySK7lKnQYDWBsoAACAASURBVFFeqsRFK1S8Epf9InNpm0La5nmvwJT8NuNNwrCDL6qDNl18DpmHi0ZRvsXlwm6K56GKtjh34qYqCkzXh2ufeM4dIAN5cpbXwtjDk4yXDqMDRfpk0Pki314QqNsCc19ZqdN2qFqn301MTPyj6tcXeVuvf9z1PfDfZA8sLS0xOzvLrl0arb+am1v/kk+LfFvnGA91L0OnQLGwncEK5BvZmKZA+50anPRnRCCbhvxg5krvOLMab0tZF0uFhVCtVoE7edhVoX3O4m8YYT0X2yOxlcZR+dNxoOf7pPFDbvNqNc5Cbb8qfD6PP3WO3/7bR9kxvov5VG2GPEZ5h8rcdN5JIvRpgBItL0crlalrQOipIWaJ9H/Szib8tI0dfL0b0JaUFmANs3HiVJbLqaVPZrom4L1Ybs2P8HDjWW5YmXBO/gvRArdjeF1+B6XRlO4gBPnM2sBxchSArpSHAUM3tiTLllhUviFY74G2Ws2QrloH8NYu4tzycyXD3rtlE2OyvFtVMs6JpCcLGCsTeLoxhAegMwMbDUvfDqgFcCCCwTzIRnOkaOh2LVMqsfqGTmzprsFsyTKtWKc6NHPwWT8l3zZZm9daxgQqghyfjyI+1t3gtbkCryyX8aOUOcWI1Qx31WBvzuOpBvz6srJdQ15jUl7mFfiS8lLjJYKkRSvoQ1YyiVjoimkTuGtfcNUvVbD03H21VzGrGb07y7Ox51R+f4thzrYom5C/s4bLXo5qftTlosWts7RksRCMubxVeeAlLsKrQdK65I6rl5sg6Vxx7viu0uASHKTYVJu+lB17AfUtoribsXXua/Gv80etWnnWJfSnda52m/zg+Di35Ca5Mp3QktJ7Aw4pHm4NGk0YCWAiv9P18nLts6wENW5WdJgXUrMen02WuTvoZ2duFwtekWtJw8XT7TY+ZxrHWd74vBNFOLlmMITx8/jhIHFQY6B9jhv77+Vfhvt5XTnPxCjEfkozjhkYNY7O0FywSuLDF4fsiiWZgcZgZnatarqoUwM7e9ZBg4bS0UyZ6VcNN90C00pnmLasLxnK43D4EJx9ShR5w0smoCJAMwyR4owvQ0nXv875umFxKuZ3NiSSKLIuKxOtrRRDphZqDx9/JVZ2AEjNMn2mAJMAntCHU732ftbhEMbWAlE3fU4v4SIDS5k1iz5XmMP6xilkCzXDAw8O8Kc7yvzRH17lZz9ygeot41y80MetcczQZBZppmZdKO2MeIZLmX6mcMw4Lp2zV1q1DvurYufapupKakySBaHiwHomx267VEEUUNP2azsFAOUkLl7eZZmpW+Rt53zr9J21aNWKKNdrz4reGfoknYSJl/Xxa9u/jzT+IH/4t19gz4G7mYpivM5MxhPOT7p7iWk6tks17ZCagFraYSeWhXiFlo3x0w6x7HKEMsWPE5nAy5PkJml6ZfKmiCkVadkuRokTnq6LZs8xQMdNGypVrO4ykVHidsQqX5zdT5z8NwZ1Oge2qnVbwE73urn77EdX0fhaqxruQ6//d30PfB3sAYl/rl696irRX00rVpdISkp1KOTt//pmHnrbFyE3wTkCXq1UGd84ANKn9kUelFEum47BcWhqVVuB0lDWVnVdLLVWNWhL+KD2hQw9p6G819C8YumUDUUN7ssiD1u34naEarVfVN3SirhosMOqABlmVyL+w98fRzLIZlIhdRVJjTrqxxTE6OtN3EXXYlArTHw7z4R00rZrYRnx7mwdOfv71bvpKKw+XuPhYIjxJDO+lbpwzoPJomGy6bEzjkg8w3zU4myjTTks8nHT5HdKE+wu+3RWEtoNQ1VtV4HXHYbOZetsXmTDIkGD1K6Ny5CLMh/U1TkovMTKMQavLisTCJdhdJc6TllepSqgqSasKqhbImxUGMsqm6kqJTKCHciyYmWzcqUOUwZeGsJU2zrT6NGqMl8tQ0W4VBfQNa4KuJ6DMIELS3A6TdhrlKmQss8vsOhZ/jiO2d8/yI/vqhBOeUw3Us6sw75B2JBXWWp4+Rg0E4/3lD3eO9NgNTaujXrIr3BGsV9qd7eWsdGii25Lu/OuKqDJRbmuhxF/q4+myfG+4gSrCfyNgH6yxqe7y0gZUhfvMRig64UEnatOveeXj7EzGKAerzLbPEMhJwf9lCRZo5Tb4YCbjJLNVvvdtVjVnlfvv9BTo6hMlNmaZP17nUOakdUv08WcuvzMqH0W2lf51upb6RvzGS6nPLcIf71q+ZF9hrY4iR+D45Uuh7BMV+9mbuXjLnLu051ZRvOT5E0e0UGXkxaRX2VfMEhB/mEm4MZwCL//5ZyzKWvRohNYOOf+cIBK4QD90Txnoi73hOfYVT1EZSR2+KcTGXK+YekkjBzKDLE1N5d3Zwkj44OgBCd9XV8LLvX5D0NYNkQqZIoXezW7JoNR2C31c2iYysPqimX2FJR3GQL5Ii71qnYqoJ2BzibkkszuJM37rLDKE/NPO0D7pttuRUbb4tL2ptmsRakyjHa5AE+2e7N7XetbFTuNFVrE6LUCczoOwhY6LAJ+whH6UN3rsSp4eo8gS6xWahY1psd7D+f4qZ/Zw9F7F/n5X1jmWqXNr2wfpHIVXjOYiZbSoiHuWFafyehnw9st4THjiv9JlLVmHa7R2lHVN/1dZVCvZ3aYDnSquqhtWLLYNAOFjhPc0xcggYa2dRWMwpe3bsoCk9l6r52rEFWvLsuahJHtffxvP/B9mNwH+dAnnmDPwZdypZXgp7HznxOXrpQbyayCuvMMhSOcNh7H0yYFr+hMupVd7RYtboGSLXpjib5IGPdyrEfLxBrtW8/3CM6eo7DoAGVJKzoo2rXiyqjbmTBQimixye/98oMMa0DRfu/hqq2v9bXcv6BK3X/xD7iLNgN4quJdv13fA9/Ie2CL1/CinMu9laYvTykaHArX+f3OJslSxGsHFQMF6x2oX7J41SxAO85nPDnNnaoQeGoRqkC0BoHmSPFVtBJXOU5XbgoNrXR9KPbm0OqwIRnLTFA1sIuXli5olWwzDynxQPZu8h/mHoeRQVrO5V+cWa1EVQI0GC9zQTbK3RR/JBjA8zQgaWnsO45S4m2SNE5AcR+msJti2mEpqfPGsMT9vqKwlHsKYQrrOYvfNdzfafLFoEwrN8bx1kn6uZMfq1R5RaFKPJaSFoyL9kpj6yqPgfz4FDygsPSKcWrBpqpzZUv/AFx5vqf+VUdwGZZVYRuBg2OGbh02ZZJ6LfPBal3o+SnrcIhgneuZqcq1PsVV87qXLatrcLoFu/MwUIONDizFMFA35Puzik1RLW95v/owNmnYmIbtBcNHW/AldYqSNtvJ88v1JhNpkx9MxhlfMTT6UmfaemMKld3OAoviCrQX4SUFQyNf5XvLRWhHzHkRD3VW6YtW3TGSU77LQe3MZHxG8RvVQkqaXPYrfKB9ueeb1c+YyXO/yZGEw3xzboTHky7GdsmnXWdd0ZFK1CvQ37nKfHeOKNkkMCFx8zmIZS6WIxLAa19yylcZTDtgpxPP73fbYYSMdc64GKQt1JBNJ8bJHWVvsuHwRNE2WGpfYdfwa9mT28HKekwocH/Rsm+LZaC2tgCy77MzTZnIjXG+cit+6yx9fh++X2YqGGQ4GOBcssFuG3HI5OkLR/irZJWb0xY7vDKt8jGurT1CJeinLUd/ui6eTwijr3KEDy4/weHwZtjYwxFil2iQG8oA2tAe4QrDhm9JClAqZBVhI66ahAF9mSeb8pkdqLpinTdd5YBi9GQzZLD7IP88HBnNMNbJy5bKPOyZxKnduzoPD0BFh1mBD2vZgmIlgYeEANdXuOVIiUNHhzNVq6514QLtYgGfrZajAJDGGIEeATed17oXQBOg0/v0WNOy3qvHer2e072e66gn3gNIihzbavPq45yZoYqsKUEZ3vhNY9xxY4vf/OMr/NC/7/CtN/TTt1RmWy3lwLxHupJ9p6An/orEIRw0Tkih000WQ6muRfEZBrLxTibFdsNgVI3TBsmKyH0npYT0Knbafl2jar1OWwdCZXbjnLPFn93W4weKNaD9o9dOGNfFdsBuoMr73vFOCH+fD/3F59m97yhTndDlJSsGpKsxzQvxjc9yvOquA2W6RiQYofRkPWunCtXHC8RJXSlm7PRLTvEvO+H5zhS+eHP5XW6hY4Vc/X6MkiScukPfT19Md20KOR2MDkcPy+In8yl9MQMY/klQ90In6DhW1EXv5Hihb7r+uut74OtsD2ydwwsLC4hX99VU6f7zV8quh0AOpdxAu71B0czRDvcxYUJGBcJ8Q5jPVLBFjRQaX9qwcS0bH5y/XT4b4OyFrOrmRhR9tIzNxdcR0NMgrpZG1eB3LZ5EFhoMpSbT4Kj2izBbaNzjz848BcwzZPaw6eKeRO4NM16dDGIDtWINbRs5r6cNeY+lIV5Q6fGrVDqUKfE6Q/IRcxP9mhvQKsbjAinHgmwYi1I4vZLNIRNhzdlonM6NM1uf4VS8yJ/eMMm+bdCadQlRiOYiqwJ5wTkOoH63KWWqxYpX1zXEPsxfxDn5twLLtjVD/gBUFVu0gvOyc7vIh+Y8lGYhL5K3AG8pM3p1dghTmaBB+0itNvFyogRuL8LRqqFVh7JnOeQb52EnPzq9Vv5i06tQaUC3BepoXlmzVDC804f/RI62OGbtq7wlN8GdvsfqsuWpBVmgWA71wdWFrK3c0vFJLIHEGYnl7oLPX3QC8kGeAxSZ80tU4lU6Ekb0kKhVpU5gSjN5skqam4TOPJvBIE8T8i4TsNMY5ohppXDU80lsmWkd4+JBXlvaj2/yPJx26Us2iONVrMhd8RpxvELgFeiuPozJjfXG9dRVHYzpc+35rDKnXv9ydrq7vp2QhyYuxStp5ZFNlqU0YKM7wysq9/G9AwcoFlO66zA3ban6cOuw+ILw3HHLeFGVMI8DeGwPcixW7+DT3QUG2hcgP0nDysjZZ1s4QojHJRtnAm+vzJNxnZ0+3JGbYKi4l0915ygHw0htG/sDeEEvO681xJ8sPcvb4u0M5DyOTWQFFlWsxClLm5ZBXypx8PeZrN2v+fqAWnxKPuipOtW6l72QKuSp3FQspm1QrrIDK+O4rNg7bzdsnLE8+qjlQMu4tuPKMlS1aNgmKqLFdHyuLq67oHm1qgdqdxGHQpQCAz3UqxN6C6BttVwFNgXu9Hyvy+fAW7b2yhaA+lmVMb1OWEIfqXt9ngNQvYXiFqDbQgQCeXrZpkEqY3n/bN9T4OffdZjXHF7kPR+e4/Wf2MY7DxZ444Th5f3WiUhi+egpM/ai0lYs8t4MdhrCMUPyvCVuqdXcsyu5Ju6f+HY99bo8NLVhVatAjYwPKBCr79CjpDn7Flf8shnIrWUmyM4xRN9rC7TKE7DPI7mcMOJVed+7vh86H+RDn/oCu3fv50qUuCiwSMKJ3PYe1y2g6BT+Ps3ODDm/n7R90S12FVPndp6NCf2y43I+E8nwuk7QOkMqbzrxkZ1fTJZ57PagU3HVXcSiFs0St0XqvTNGarWxbi/37l+cu61D+F/+NB383m2rFbt1r6dVwpyZmWGrupGxB77iTVtvvn5/fQ98A+wBndtSdW/bJsvWr/62BRBvOVbjJ396jF/71TU4eI3Y63BhIyTMWwbzhsV1y8gBQ9w1rnpETf5PsFm39G+t0EXJUAyRQIXIzJqfpJo/CM1noaQxQo7sGncUti3VZCheU+Y+4SmeR5/Vstic4cRpEVQ2yakiQQHjiL0Bxu9z9hmqyiS2y3ZTYT1t0vKqToUZNc/h54Yx1sM2jpNW7mVnOEbTxpxzcv0ul1Of4SSh5RnWrVoRmYuC5pSSMSzETbZ7Ac81Y96/f4ObRnaT1hJ84QOBvwnoNOTUD5HGyMQQ+NYZDCs9yVuxLrA858HzHUvFQvkm4xSxUs5FzSw7V4R0hdyUxM8RrUsTzZIllQO+lHalzOk/FxjnVh+MGy4+Z/GXcMCyI7+6FK7WYVsJQk364qlniU0u0FzKRYnaNjetawfvDLL3iEj9sbTFT1YneW15lLFSTJwYdnYsc8r3zUN30XJlDtZS+IyXcI/nsdQwXMZyKl1j3ZboMznOGZ8j+e0s1E+y0rksbbJrq6paoUlD83Xq93E0v43U5DhlAp6QZ6D7vccqKUEaMZdsclJ+LUGN8WDQBZdLRHFe4eXBII+nkauIVcK7mFebN5rDtXn/H/beBNqSrC7z/e2IOPNw5yHvkHlzzsqseaIGqCqpYiwRGpBJRhVH2kZAbbDRWjT2o1tfayvtiMpDEERFAQEtKaliqipqzszK+ea9mTeHO49nPhGxe307zkXX6kbLJ7zXjRkrT557zomIE7FPxN7f/v+///c5jUUJqakCp+24Qsb0yPk28bPUEQhsqsFdmKUjeaIjlZSDrough3pmH7uyOfp6IuZmDUFOeoyGZgb+9jHLrn7D1jKU531kJlBS4YZAaGE/a2sPsNEBteNejhwtNkyO/ZLVchjCZ9wvOBJ820YUszvoqZ9gNegnpVJlLWGVSKTU7Djt2kEOF69iX2GCwIsI68a1owpeKuJ+RbiorDTeJCKsVKEDaiL163pS1WZBgMy461Igqa3A5YqldQ4C6UBK7tF5sUFxp+GGFoRTCSf29GlLVQBU/AQ5lASGh5tz1KNFd8O+62UT9Ac+8Ubk3BOc4q6AjX5wATE9634WuNOkQN6vqgzSsQlPCytokqdnravCCIEt3Yx6rVFfc4KiPGlt8rmkTsRJc3g8sSJz4E8gVYq/UrwQ7aBkufP7+/n8nm4+/kcL/OJ966xG3ZxfSnFjr2G/JkNNSBcSv2n5yrYXLIPXqh2TiKb4dLE+jxJZJqYUKU/OzcmYKE2rPksyRToHnbuiiqpiXxCvIKFhuEpZnYfUAzarZqVPKY9c7V9KPIuGqD9iwBT5lZ98E5g/5CN/PcnERJYzLQE0OZ0sO4tDLz1My/GENxL8HEu6ZDGRI1FkPFzB91LOh3mlvUgq6CduLtNqnku0EV3DqtPqRLCVVvVyiYadkbpAk55iwMzUCu/6ubu4+uqd7tLcHCuSC/Vf/v8/Dep0EehaUB6os+hv91qdmvGY2D5BOqVu+9JyqQX+z2+Bf1mELjn/zcB1Ji2pDPW8NW7wGlwhAVixLprWuSI0lw0LD0HvzdZ5M7aVMVK6ws1iO/6vGk9Vxbb172frUlm36nDXIZAnpDplpTEK4I2ZxHtRXHVp1YloLLHOtM/RqXk+/Mh5sl09rEWZpLJWla5BdzI46zZXH95exQa9BErDNS9QEnoJulmJ6gQyqZYocX4vNfG98CjaFqs2dN6iz/ZSRK7iFb5kk8DAqwOYszEbGAY1SHhNesIAswzeEARXGaorlsojUNqbREbMKesqVv0thtaUddqymgzXIkW6LHu3QFuSE9PWkbVlu6b+VIOZCPji3oTi3LShPdUJRHbSrTZnWVyGLs+SKkLjonUFLLlB6F6DmRnLcA/s2W1QFq+Qx/nTPt6C0TOWCaWtMpApwZTSSqoujgxn4xafaM1xV5Djbr+fPYWQYtmwLPkTARDx5xbhQhbiNNzQbxhqBtw7a7ksI99Rw/H1CtZkqNomV3kZTmiGH3ST9nY7vS1pBMZxDY+QdiAF/Mu4PLvdcSFfKUFTG/OxuMVGVGdf0MPDzQuEcdWlZwUU7w/XudPLss8LOGpjLih1ZAL25S8jlxpwavxL6RHS+ZDW8qeTvl4kb5dqjZxTg6uQFmBT1bSiE45jp5MUmuhcQ1EVTz9Y7irKQYqetOXEObi3YXlNt3EafecvOuUL9ncnsjTVtmHSRFywHs8KoCczxlfyB1htXaCWHmQSSzmqMOZ7zJCmaQzXGc/hm8/aNqVwmetSvZzP7eah2nGC9AitxqSblESpXgrpYY6G8xxce5xMeYztxqNXUbetoOunPteRChKWkL5hv6VdMqw/DkUVLzYS6zpvHMxEJ/25mlRPyz2hcbyjB9ch+0dnEvDVLd/mQWg+bbm8yyDccXbSMhZ5nGk3ePuxw+wIAk5Tpn+06ICMDTTx6ETd1YUIyAngCMBouNVDYU7JnOh9fa70qxYBPt3L2m4T/OkCFLDbjGjp77RmKgKFnb5F62u/AnftzndrvVZi1+VUzuqW3btSvOffD7L9pmXe+I45vrGth7fHZZpVy4EJKG4D2Y11DSRzzfCIbFogI/kT4XzNA3R8OhZ932BH3kTnpnmBZI0E6GRePNWpntV5Kc0q1KJJoLbRcSrRoH3oMtakrTORc/JPoqBI6iiMGGgpYvcWyP8hH/nUYcbHJLieJkiPOmAnR4fQyyd2dhL5rp/Ehkt4mYmkAKI977ITCtUrUixpbE18VFz2TRFu56ii+Y0QZTopmHByJpKWarlJhBB3EPhkpVyuw1f7fhuXfxTUbUbfJDIsNPkPEaUGiWZD5Tmqfkn9C1NV38YzurSrSy3wv00LJAhp/4G9MiaCVpOg2CAoFMnXoDUAC75l6xZDSqToglWRrANr0lazkuJQFOC0RcBGM1tVxCkFqGoyv6YoAI6r4vxiJQuQgkipS/UXAlSyyhKwU8pDHqvBIvOLZ9k1WmKqFSRcOhf9kfK6RDJTjrPlpweZsRF9AmxeippK8fHxoxqt+il6is9iKejianGxTMCx1CC057iIUofGaY/OtDT5t1JyYCX2eChucauX5VzrLC/qvYaPnatxc3mDIVOktT12HpDlbjg4Z7lm1dB1hdJfyuomel6+JEXOwomLkN9t2L8HVlUxrMpFScEIQIuXp4z3eUvtuGUjwrlbOO/cACctIU1c8XzSsxCpoljBiwKMDCS2Yk+fN3z5jOXHRo2LmlbXLLkBQ18R0ucs3cVknydVMWuMG3s0LlaN5T7pUPkl3pIusatb+lqGxYqkSzraZhnD3n5YzcDGjCWbhm0+PEvHkJiIMBB087dxg1EMJb9Mvr3AukCcZGRSA9j2BXwzQhguMur1cle6h2M25KEo5Mc9w7CX4sUmzbLJsGxDwrhOOVxjzYZkTYqpcI1PSNQ36KLHxqyE69yYGXIuDifjhiuGyQTdSKdO4r6J/pYiKRqoVO2qi0kjqUZfXYyJ/6w0DJN1VOWX/N3ITDCUv4x7oxan0xXK1SK352MnVXPvSVyE9fpBQ1bpQ2m4XYDnb3hO729/4LEjU+Svo8uJ1u6F+imyxWuo0eCguICpQfaIyuCleNQVTlpO4xPaiHp2mxMVFqHdy+10QNWogMIvk85dTm/1KZ6qXOQ+fyuv7A1ZUfSwDl2St+mGBVWmlqB5DlpnLVlxU5U6E4ATvWExua+iGYsZMQ7vNuYtOWmwacIlzTUBQBX8KPIkZRdNOLqhNASpmuELj1jWugxfbZ5RnpPT0yFvf/NOrhgeg3WbROk0QguUKSTpq3pbKffOewJEAkHi1Sn6rPeFqfWeup50UrHtwk7qE/WZHoraadkEVKo6lTeeJoY6Rl2ISv1qewnqqS/RTaKiDQHDKPFeDYo+P3D3ILcN1fnAZxb42T+p8YZr+3ndrM9VWSiJu6rqWB2+tKhlkThvyRYT+RFhGmlzSp7ETiWH7IqmBdY6YMfJoeh7VaCic9f9qmicCp8E4lQ0pkyFeLUq2FiwzojDgdIRA706B7WRIaxGDHSV+JW73wwP/z4fqZ9kKGtYinvx01sSGR+WEuCmkiGFXd31HXei0uqkW/jZvS4yJyeduH4CI40aTdl0f7hJjTioOmiBuqoTKdaP45uYppTQKXLgwO7Oj/Dtf/Lvueeee/5Xu3Uze83ADCwvLyO7sLW1NdbX191Dr+v1Oj09PY5/pH38Q9D3v9rnpfcutcC/phZQcFv3xOhIib/78oM8eqhJpneEvswweRsz0IaFDagrbbdi8ARWdhhn+RovQGGrXKNVlWlcNbwzzNZMXKke9Q2apXYZp6FldiYzad2vLqqnNK04depbhqV6YdhYM/zmZ0/z8LRIZnmaVpWMJVfab1TG5wynM0mnZiPSqR6accVFfqxfJAyXiFsz5FM9lEvXM45hlphztu3I9u36cTJ+Hy/v76Y7ZZ3rz1rbsGJMUjnqpTgSbfCVtTl+snsvRKts5CKG2t2Yddk3QXMIlpuGdhsGB40oYxjRWsYNJ89DPA27roFx+WXmDWqn/E1JxZ5VUUUozJGkVgNZim1SYURyr0KYlQBxwp8KNKhogJXEhIofhpOKxA8/GTJSMgzFhrVzlpEbDT0jIM/ZhZXEWuysbMsE8uowveHGeJ4CPhwu8zNelhu8guOPqZowk4N5aZyl4ApJQQi8ZZPvbFQMF1W1HEAthuXYuADJXFjhuqDkOI311kUWrPxRBbRWCJvn6fGLVGvneFnfPn6ofBW35C3/phRxNkpxlpBzNqmg/XJrhiEb0p0d45rMVprENCQ7E67iNaZo1552vrIjQYm2bTEdyUhL6i0p2uEabS+FFy45i6/EwzKZVbiiGR30JtFLg74aX1V/keKxSoldDn4f+dweqkRsW2mzPzPA2EDsgkoXhS+U6oxgcMC49z4/B3sEZjOWfNkyvAOmFvJc5WV5tPokJa+EbzwaXprnB0X6jeV3bYshcRhl90XMrLUUTEDDtohbF/GzO8nYFuJPeX6WwMuxTIvdzVm2BBNUMj4a/8s7oDIHDUV3OveNRKxFuUopuiwxXaUQNyNBijJ3GWdkL6DUPJzwQjN7k5S+Ikp+l8GrJdqI4Snw6qp4TsSzx1I+X1+q8PbT97E1U2dtdYlXvOAObr9ixCXP3VUsECbOnBahI4V1BLIEFgWw9BMkOejktYCdFgfqkgib+1t9hfajzwUOtR/tW9E9Rek2BY71ufoWcQvEN9j87qwKKRJAp+NweF7fH1t6JtLcdHWRW7vq/PljCzxay/P0bEBtNWZbybh7ZPkQdO9JoqBLUm1ZTPo6AVZHDVmxGEmTiFus45lLBKBV9a6KEvV9m5ea4891TtNFufR76TjVH4g+0evkEy0MsQAAIABJREFUB5Jz0vsCdYqy5gxRFFPyc1y9ZT+P3XuG4+UZynFAQ+DLL+P5BZcu1TVMvO6s7hxxUtHsWDo0CwTlW9zkKhSXTtJCKRGb3V3j9FqsbSQpW22jggsnc1Inl4pYu7jGNTf28L5feD35/N+7dHVO59vytPmT/U8702Ak1ka5XGZ4eNiBN/0tSzA9C8zpff2t5RKg+5+a8NIbl1rAtUAc+zw6t4W0V+ev6xf4/Eabfs+ju2wYFzdOYEOK671SZbdkZ5P0g8jaQkaKPEnaSN2G/tPs1L1QRyX/U4WJaomTgkZTSZyp01UqJlAH6WbZHs2uCn907H7I9tF0Fa6atLmRAhOUEkHZ9ABe0IsY3z4eJVcEIWcDhf1WiBrH6Mlf5irFZDUlN4LJcI1sJK8FnxvtOufXLQ9tQFU8IgOXGfjLOGZNNlnNJf7b7h72ljMMRMN8xp6jVbZkm1ATgHsc+qpw5kuW2aOWZh5WIzjykGXyBAw9OxmTYnlXLloaonKds46AnBo1RF3J+45TWBZXUSkf6L4CsnLVGDdu4I5LhszOZJBtBAa/3+BbQ7MBt3k+1ciwGlpyBUPjAu5xWM4eMfSlDd2RYX0dDq1Dw9O4angganKTX+Z5PXmKhYjzdSn1JyLQIymohxIrhmYIF89aF/kTa2VdUnOdLJiS2TvwqfhF9noZrozrnIw26MlsoegXSCvVE9eInfVcxPO27ePAENy+xefOUoGf7ErxlmyaN/s+B4zP5SZw8iino5j5uMmCLB5TPZRSffR5aQq5XeSyEzzSWuCRtYeI175KHC6xUT3IeKqPO3K7ieI2qdwBvPzVCUkTpWAVGkmidK5EW9E7N4gZNzia3GXI6UEC1U1VWJsUD9o5aqbJJ+dgahX667A1A2VFWFqwOmeZ24hJp2G0T3xTw/Zew/fusmyPJrgp089S6wI23CCNz71xk/tQAWSGYb/AaGbEadI1jQSvA24uXIU8PTM2ds4UUWsGG9WJBEa9EieieSbrFzm74UHWJml2uboM42R1PBU0TBgynevYF5hZFh8SnE2WKjkFdEQrLEF+myF7nXEpwYSHlkTNjMrAJTd3I6heY+0iNCNLacxQ96edzocs9TRLe+nwTrdPJ2Gk+12gTYBMwE0IRs9Kueq2VQRPi6J02lyjudKQAltap9KJvOm1Q9mddVR2K0CniF6f61WSVKfe00vtW4u20+tcB9BpJ9q/PneyI8ZFopXu799ieOlPDvNbv7qFZxXg4eM1vhDGfOFgzMkFS64M1ZlEtq1YhtIOnDuOA2FOpsk4/lz7FLRPKrKmL+58b65T2eoOymAUKVX0U8Fg6afr3HSJ6f5U8FiujLJSlKeszkHnqcip6G5NQ9SO2DbRxQde+hbi5XG89VNk43XicN0VIDmtzlARuxoCaCoIctpzrlhCqdWmuwbj6lNJ/6nX0nzUPaB7QgfkFxOagqfO2An5OeFw/bhPHLdELvecNPO3+/9vCer0RQJqSsHK97VQKDhAVyqX3LNe6319fgnQfbt/lkv7+25ogeS+sPT15fn9e55FS3X/WXi2D1NtOKfKNw8GpKlWTKpeBVakV6mUqSbKSq1ISy3+hvJ7iU+io7cK2KkDWwb5LbrOVxEEdWaqxuukW9260rcLYeN4g56Ta9Cl6bE2lrq6eruUm8D5XpacdqqIHX4yY41bzjZMJRW2vYwfDFCRRY6XZU4dmY3IemmyzicxRTdVFGDMtWFWqU8DD1rLbuNTCRtcaZvc1jPODi9mPF/ic1NdzMytky14bCxa0t0w1gO7+gxHv25pnLacEFVwFL5nFOKVTgBhJSkoEWZ1g17ZFei6StTMiCG8YFk8klh+aSCMTyXbpmSe0WcwLUtFROxxQzhtWX7C0jxn8TPQn06iC+EOw8WMZX3FcnHRkleqdJvhYmgZ6IZbxmGsC74YWr4c13lj2jrh4DDruarmnjSUVSHpSc/LMKVUuiRZ5DWbgcXIOBmWrpRx/G8VSfxVJBFin+1+ikNxmyc1kARd7uddiKo04ipZY1gK21zd28U13oDL+EiYeX4tdiByfMwnYwx7/Syl1FZ+sWcn927t4ocInQZcurXEYvMcs3HEitfDkF929mAFAXjly2rHoH7UReoaUswXcnEp1rzzvzS+vL47JuYCci7UIi+oxB/TpLdilCZ2Uq0+LRPQ5XfxQGuDxxsVgqbPl+oJD7KQNmxRRDUDlYbFz7ZcJOv8InyuFrFxylKuxUxn4HmFa93EQmbp8rSVQ0Y1qnGtxnQT8IZUmp/2c/yol+eVvryMu4my26mGSywHffhBnxtoZe4uF4EVv48PNx5nLGxz5oKia5AW17EKDd0WBaVfLW1xLpWO7FFldnKNin+noptQ95mcYGRk3wX+gnX35MZTlqbwmoKXY5378qSlPQ/FUd26Pk9eXOWnzpxmZ66fmaVZ/vtdtzPRVcKux26i58CKrm/hG2EFBX4EVDRqb6ZY1UcIaAn4aSKnh4CgPtd6nXvfpVq1L32u9bVPAR691j5UiCBwqCid3te2UjXXs75bnZH2K56bcsnu82R7zxpXOBSbmBsuL/Ce/5jj376rxR8fW+L17TqfXYTTNcvSkkXRdEXQg/0JOGsvWZa/Ih4oLMntJQtpgcY+k7jqSCZGYFNe10cgfizGagKrmhIdSj+EZ6xzm1FWA/EcC53iD/HuBIJ1TpogCwiqw9hQWDbkpjd18//80itY0iSbCwTheaLGaWLl3HXfuZmx+sNOrlecUvWVUYWodS6JSOtkdG/oIepKLKqaCiWUTkm7CY4mO9L+7NFvwjof+rWX0Nf3nQuG6af5losGpU1gFwSBA3HZTNY96/UmoLsE6r5lE1764F95C8QSwzSw/zIhrlWGg7YzNpgOLQ/Owqfm4cuTMD+Dm82mKtal6RqqBFXVq2apIv5vNS6dakQcVierdERHsiQt7t1mBy3tJs1aJUkg33V5rIvuEcHHTzzBJCEjqSJNp58kqQrlmTzXGaWDMiU5RtgQT5Wwfp6q6wOyrirWxquku+5wfWQozomXYczLMiGul8jA8lK0TVSZKiPtKfX51jkoMRu3eH9jgVeWcixcDLi4GtGXyfKjqSKHmqvUU4YB0fLUgW/AyH6XuePQRy17JOq6x7iopDBHdtjg7dAgY+iSGkF3kp5xOmKrFlOyKKqi6tV8p40kR9JWBy8uomRTaokWoDhR+r61desiaEqPDQ3AiAfDVUt3w1DsMajgpaD2NHCqIkkTy1rG0EJVvm2mo1X2ZbOuUjfVgNEhQylr8IpQW4djG5YdachHUG2reAaW1i1rdesigipeFlFbY7AA2fP8kM9HFZ6M1p3+mgaNdlAmas3TNmnyATxZyXFkNuGzX7iYAMaWxrNly99YFabAzYHPbalu+molJzT8Er+L2zNb+L7Clbyj70p+MJ9hujFDvXmBhopbjEccrTLU9WwaQYaHJC5NTLt+iKh+DCNR4tSIe3jZ3YlKvvKRMjj15cVWVKmGi9pJhLlpfG5OD7ElO+5Eyn5740lkunS1b9hfNvSUYD2Cc1UN3jEr1JiMLH0x3Jny2Jg37K4aGmlLb2qcn8nvZbV+gmy4xIAswbCciKvUMMxEkI4tN3oB1vqci5u8Mr+T/vaiE2lO5S/rRFPkCKs0nHKmS8yEZyjEPk8uxS5TN7thyXZDYY9x3LjmiQTYOCyTNa4KPT5mHXdO15JLhWoiJiA4aJw2ZG4A0n0QXkg4eQIX4le2y1CfNSwutvm104+4EG0c1KDq8ZzrryR1wCOWkG5Tumwdd5iGTaJOir5JeLcnucdd16oRXPe+QJ0ApEuddsCa+gl9LvQjZwqdgI5XQE6faWkmUUS3re4PrSugqomiwuxa9LwJAvUdugn0kYCu9u/LZ0F6c5I/iSnthje+vpc//6Uyr1xa5r2LVd65HnLfWsxSv6LXBnvY0jhhieZx2pArj1jnziJtSgW7IkkcHU74cJJ9c/w59XOSlBFJV8ej9hZA3Gmc6LjOy1FPVOyiSl7NRVSprGPWPHYtAXUKnkVVD7MW8frLx/mtH3o10+dPMWDP4IWLLgruKASu9FbV5opCq3G0KBK7QqQonav2Ni5Kl2Q8Oj+GOilNmF1Ho6pXnUCTnH4DVrjiwKib6MmR6zux6Cj+0WUT1AnAqSpw86HXm6DuH93BpQ8vtcC/4hZI2F0wHozzvTyL2cppPhYuccg2XTSgpk5JfOZO8KxZkyUWRCL9q+hBna9U2CeSIon4YscLVf23AInMxvMdn1f1O0MGo85PHbcmi4roqbA1J11ZefPkMApBaADWYKzwhPPHjGnayKXoBNAE9my4jpHdhVwG6idIi7Tv9zgSsCbTRQIqqmY16hcUksqx2I7cwJiTH2qHzqLqxF4njT/Nrb0DjEqCKy1+S8RYdoCfPrfOuZMNamseCxpQi3D6LKxNwshthsy0dK8s6d2G/B6DChdmz6rzt5JnIzxkiQ/ajgSCITyT+L/m90JBpOsEtzohVEnFVBSR64XcDohmk2jLlhFDNicdOZiXnJgHF04lkbZQ1bMbhi0DhqIP1w4oZQhzZ+DIqqXd9nj/lj7mUtCIIg6UDXNyCamBZP802OzJGDcWr63D7EXrRG5VQat2WozgaeBpCy8wHldby7G2autUQ6wxVXwzw5Vemj79sHGNWmWBG7dexfU3lMgWY1Z1PAaONCx/MQfXibeF4Y7AZ70Nf74e8oh4nCmPq/0Mr8h28dr8EDvSW6Gwl+cVL2N3Zoy6CmH8buZsmQU7SJ/SrwL99VPOikMSDSbV6y4okx524C652DTQth2YM9aSjhVVbPBcL80wlotxnYmgyOnGFKcbC2zJBFRa1km69OyH7oKlb0wgOcPVKc9Fa68veYyMw0DWcGfJ44gN2ZrbybBcI6Iq83GL0KQ5bjKsitepbJuRMoOirXBXynKrJCvSI1jJTsRCPpHDAxqEa343Xm4bPzv7GOlCkx3dPicPWydqfU58uWKifVjcr+iRofkkNM9a2ues6KBEm5GrcjKBcmDIZdoswZUGf5txVAhNIOKqdSnbzAB0Xefx5ZVp/mSpwv6uEaZOzPG2f3OAnc8bgfkYl64V916RJ2X0dHGoHxAgEy5Q8YKWzYib/t6MuGk9gT+N7JpHKvqmbbSJUrxqAv2t93X8WhTZUmpX22yur220rhaBQe1X2+l9PTRbUyegaJ6ApK7zzn3mJrIBvPzFBT744TF+5M6Irz12js/X2/z3Bz0mF2OXXahWoaKiiQzkVMX+GKwcU4GJdX1YONnp82rQVrhegTKBNZ2fAsP5jsTTejJ5dQUTOieBPbWVVKl0jNpO72siJywmrTx1TgLgqzFvve5m/uDNP8TMhTN0s4QfSotGJ79Z/KBKlE4jOoQYYttCjkrjqz8VNy5MwJ8aKm4lUTw5r6h/tU08E9OShQi3EjtJAtewSft+m///J0Hd5vdtRuNUQKHH5uvNz7/Vs9Co26bz/K3Wu/T+pRb4bmwBozCd+pcDRa77uWGYnKU7VWPGpl0E7tYtxmVU1AHLSkjqJ8p45ZuwcZ91Ufxgf1Lhak90OielSiRlIpmBBkwqXaSOa1OXqmlc5ZzAncZZU/eYO1/jyOE1VLrZiGLHdTKZUae95GafGhBbsxg3q2y72aWNq05NPW5eJGzPUctfSSOuOZBR9fIM25CDcZPz0gZT5+XLyDLgVCvRUNMYInuqs8bjt9qL/ES6xLjJ0FO0DPvQI62uTI47gwxP1assLSWBg0OHLLkiPOuFhmFJgKwnHCZR/ZTVWH/cUpLor8SIpWC/NbEfiqYTLS1/3JAS/avPULsI8ZghK36hxrF1KMfOVhVf5up7QZHO4hacLZPavl6znAqhloLJDaisQSsL63LwmLNkQ0NvyrDkxXyu0uRNEx435TPsqsPu4YCVFMzNWZdOenghEVyVxNbWwCD/7qwHyxVLzZPMDOzSwBMnkU1l2BatZVLgrb3sJGC2eBlXRXw+XKEaVxlSmG89zT23DzGR1+Q6Zv/+pFr6lOzKUpZJa7lTWSYBX8mtSLA5hsE07C5YhoKYxSjmY/U2/y4o8uziDo5J1ia3gw+PvpxfHxri54ZXKMkpyORcFFEHq2vEVfTJ4UIRDKntKmKhnKUWFz1TBE2D6xbaXpYSAaW4xUR6hOHys1lvLVBQdUTW4GucW0rAc6PkUUlnONOwLFdgpm45dM6yvG4ZsYanrKWYKvGSnhuJqk9TiDcomxTXYjiqqCYR86rmjWBbHl5RSvNX7ZgXFi5jq226ItFUdie2Me0G3dgY+l0e/AyHOEc+43FEIGocGkdh9a8tccrgDxnaK5pAWNbOJalzf7txFZ2RqkYVXtW9pkmVKl8XoHG/JT4pZ5dk4hW1JYVi8Zc9Hjm+yHsWj9KT62NNise0eP733OJs8KTfxnwHwEibUvptHXCH/IcVQdP3aK6li0VRqE3MIWC3+dAPkGDY5MJXtE2v9dBnAmna1kX5Op6r+s20f90oer+DY1zq1VXeCvx1AF3SrSVAUesLGGobRSTV54WKyFmGtvv86g/18sD7C3zGn+LB6hofnIRPHEpS2pLqk9izl4LSPliL4czXoH5MIurJuXntBD/FikaLbzeUVPA6EKe+Ruesyasrs0+20Wf2WNKOOg9x99z5FK3LgLhshuYgcrsZhTfeege/eddrmZ+dosusy+G10why2RFoEzhTua28ZVWqq4NSdqKTr5ZdXVz9ZlGE41cI1LloXExPwTAzvczb3zXB1VcOuFvlmWIot/I/479nDOq0z00wp4PR3//UojU2o3mmw717Jtv9U/u99PmlFvg/rQVEfM4qr8c8+6JV+j04vx5hTstOy3JMESQBlo3E1cDvgdwN4ICMCghc+qZTwq/0SBXXGYmzs3vAkJPVmMCBOmvRPBS9O6XOLlFcP7pwjj85eJzdvSXWVZYniymFq6I6pqNRJ6FNpWJdRawT21zDxBWa4RLbgz6+J7ebOCgzhOV24Izx2C5vTxszIUJ6epgFzWQlHtuh34wZg8yuRS7abUdY9VIUtli27jZMlCWtZ7gqPcwDtXO0M7FzdJhbl+E9DI4YF61UZm9WgO+IdV6b/c8x1FRcchEyOUNmWMKrxqn9q4rWgVt12PMQ9IFXTZwlZIOqgXDtPEjipL6s2b4hJwkUDUR5gz+W8MJvz8BgHj5Zjzkn0raK4TY0/BoqDbh/CX5z2XJbj8/d+32CMUuxZZhIKwUIQwXDRK+Rfi0yEphrQn4Q8gOwWk0s4hZXE2szjY86BemzLtmQL9qQPglvRRXnI3swbrNiLYvtZRrRBr7CEOkShbWYpqQ1SoZwHUrnDaYQ8bl2k6GUoSdtOBLDRRdoMWz34LF6RJ/Gp7bs0Cy7Ap8x4/NQ5Qw0Z3hz/gB7vRLPTXfxfeUJeuW/1ph1cifWlRgqIrHuSOTuuomjTmVsEpFwaSsb07IhQWOKr0QVvmIjxrwULS9Df24Hn2suc8pcZGDQpzxsXWXw9FkI5uFl1qcgiQuNx+JketKHM0yE8Crk0xqys3SZc8ZYqD7tqAKie6WNhyQeS9ZHXMYhRb/XLa9JeVwrKkFq0OEdzyvQ0gUqGzEMS+I4DI3yriOPcP+ZllPzmOg2bNsOa7NQXbJE5xL+X/4KKGhyIAkTRZsV8RXKUTmzdBEFasSR32XIXiW6QEIRUEowpSKdtsf9X2tz8+8+wWJd0amA84vT3PzsCW69ToJu0nJLaBYsW6yqRJXuvcVgLjfflO74ZtRMAE2LQNWmFp14W4q2dbCGA1oa5VW9qkmfjlGLgJAWrScgp2ctApWKAAowbu7DSaN0PteTwKAuWtEtRZdQIYjuH/U/Ss3quJSSFbBrQ7bbcturtnD8nu28ctcaf/jUBe5rtfnciseJKUtW2poRhHOw9WrQz7sxA3PiKkr2T7/lgaR/Ex1FE1lPgsrCWMMGo6ibzkffL4ysAqVzCS1DnGPXjmeti955o8ZJqLhjFy6L5TJj8bssr33hi/n+q1/L4uIsZU8gTSenc9IJ6YT1kF5jHftN3rH0VWSJt9qZQSsH3MKockYVtAoLOmWZ5McaHi46GSPtVsHR78TyjEHdJqCbmTzOyanZb0bqvhVIS0C+5egTX+d3PvhBPvrJv2J+re62+1bbfCdO8NI+L7XA/98t4O4d4DWvuJ5gfx5/tU1gmqypGjHrMWsjzp+0klLi6IwlTENJ1jnTUJV9lTpa9RUiXqlTlsjupoeiqjs3LJ4GBHXECqxkSOyw3Cw2SY9kmupUYkIb4MlBAmlNDSNfT/U6qng1ft6JDrv28tIYIUqJDLfPs5rf7+Q1+myLJZPmjFWKocaY+F3G41h7iaYqM43vUmGPWsOjcYyiIdW4xtbYckVfH9Qj4oahVRQXzLLQirk+yHGmGfHQ0iKmGnDLVQnoas9ZF41TP5ppg6xJVYnqjxhknyj70eWTlpVpWQIlYDbKJXJR0XQy3vgThubDNrEO22lclbEif5pEK+299rClup5o4bWXk745lzGMZ6Degq3WUMhAXkT+yLIWWR4LLY+ZkP7Q8oYdaRfhm/GVRrWkZEvWgC07cdFB4cXTkXUFF4pGCE/39xniMTjXTlKvh2XqLj69ha3GJzAeh6Iq/bblwFE6qjmphVS4jIlqxApN9KapHs870VVhPKXpD4Yxv9COeWsQc4W13N+2fFYDigfbJQNnDHdmfRfJOhK6wmqeawI+317lyeoT4G3h+3rLeJmQRxZ8/uuRw5yvnGV84GVspEbI+P2JyKoJsM1pbEsaMq0OL7OjXyfXCZfr0hirUSBiVODfyzqdw4IIU6Ux/mjqNLWgRvewT99O2L4HGr6hv60HjPUbru01FELjNATH9xnuyMFS3dJqefx86RoO2CpTzRmebC87iaAl1TIYy4nIMifQFMAOY7lgAkbSo45KYDyfHfldZK1lODPKrvQWeoN+5tqn+WrzLMNBQLsSsy7JmZR1mo8tEUR1X9Ugt9fQlk/tqST96IRuNydTmsmI6ype/UbCDZPVneZO8bTh1JmYH5w84sj8W4IuolQFFtb5v37gJnq7fBfVUyW88zUVjUKpRgEsYQuBmcsN6L4XRUvgRRM4FznpABqlHPWe1lc/odFdgE/HpwtM6+pvgTL9HXYib/oOiXRrXf2t7RQdFAjUdqqW1frqK5RuFbjR02ZUMN/h62k7Vcq6qF7S6xrfEkfWce32XFHgbf95nN//+RYfeewoRy5U+MUl+NyUpVUDX9mJpyETQnEIvHmonbVUl60kMHFuE7PWTWQqRy12OslYxEeTCKkqXsX5aM5bmrq5vaQK1mU7+5LibAXAXbv2Gxd4U0RPeDSuxRT7DM/t3+dyyZlI91oFG693OHEqs7cuGuf4HLrOFbnT9R/XknSrdOnEvRPHWBHhuOG2zQZtLp6ps/vACK/5/u9xDfOdxED/DFCX/EiPff2veeygphDQbEUdcJekZDeDdzbWoAEzh7/K/mtv5Runz/HLr34JP/6fPuZAvK4NXROXlkst8K+pBbpLWZ4djPC12ce438Yc89OcqEecaQYcTcfOcuiyHYaSB40VUGTJkx+mZqWSU+hOOiLX8SpvobtXtLiScVE5q7TN5qQyneiymTPOgpRHj6unL9NUb+1lnK7czlQ/XekxWlJOb8+5jsm2552iuqQfVOWodFvcniefGUfVlxvhBgGW8zZk2S/ypA1d9kmitX0m4KvteR5sL/I1jHMdEqg7Vo94w/YsO3t8giXLyhLUp6G3y7Cz21IlzW6/n083zrJlMMRUjBNxbYcwq2LdCejbowhJ4qChtIokQJqDcs+A+umErxSrMkBpFxdR00BqMauQPZB4v/qhpZgXfwsaDZBmqAoqVPGoNKgNLYvnkkKDsy04JS5dCJkslLfCXA2OV+Hx2FCpwYsENpctjXOQOprIYFx+h0f3bhB3aiOAsYKhGBmJ31OvwtF1SxjC9XlDj+zCvCSSl/bh9rRBkc09xmd7XGERj5wiqkG3S3UqDS4UFws0tSJCUyRqWpYX4MmjlseJ+GEv4EWpAuMpz/EDf0DaeIoA6lhlIN8yfDqyPEiMIm/TNuZL9UkuZnbyzuIot45Yxi5LMZuZ4s/XH2a0+w5eXr6Fy4hcIUW6cDWp3H7SQTepcIFUOE/GiQl2/INdmirCi+t4ykeaNC81OHmVF/o+LzUB12VHuTfq4j88fZrVyODJEcSDw7F8UGGL9PoWYF0BXmW4lLqswvAuw0ZZ1cch1xS7ePPQHbD2EMPROjJhvwPDiA+jfqKbq3T5UUWOiNkadBN6yvNl8SSULUkYVSqmuqnKFcMr8Wj1ETy/yYmLHq2mZeyWRO9sTREh8SUXoLJiaZxS2tw4gCfua7SgiZfEamU03+FwSp9OwrhbLEHosXoKfuOhx5iO5xj3B7G5fuaePs27XzXKTdeM40K6SpEKdGmComIIyV5ooBRQSwI9SQhcKdnIYoViO4U+LkKliZ22F9jSaWs7gTJtKx6eXgu06TOlcMWP6EQXFel2YFCAT/2KAJ6207MWHZvblzbuRPL0PSL/a30dpwCfXCi0X+1HX6R/4rg1NQmLZUrDm1++i8Of3EW8+yx/dnyZJ2rGATvHp9WhzcK6oofq89qQz4OsDtM3ypUhKfoSuG2oSGgm2be+z9msNnBWbdK8UzRO1exKG0ggWvgkPJe0gRM1lhmKAmoPWbzQY6PV5jOr6zD2TkKvi5STMhHnWOBNJ6QqVwFYdRY6EL3fdhE9yaA4KYNwHS815NwjnCK1UKRr8JDFjSJFWdN8h5dnDOo2j6N/qJ8//p3/xrvf/VO86rU/zNeOqJVifvld7+TQxYRXEbkbCU4c/ipc8/P8/n/9AB+69/f41H/5YzZE8NBJXkJ1m0166fm7vAUUFRPLrLuY52d/8PnQPE23qXLSNnky2uDwTh6OAAAgAElEQVR5WctsHDNbsXQPJ/wS3R4yl68/YmlNKgUkkGadNpM6Z3vRYjWYKEAi/9FzSbpDHbHGVc16ZR7td/ls7K7y64982TmQVyQ/YNt4chyI2zRMyg2+mlUmDpjSGRNSVCis6kr8/WDYWU49IS9z94VVF+nri2t02QbLWMYy44Tq3U2GQlzlSmBcpvJRyP3xee6M+1heg/yQCjaSIoJWCFeNeKRSEX4wyBdjy7lCg3bBoyKiuvxXO+ci0rQG9qpSoSoqKScdelHVqnsSeShviyGQHdMwxOLJdcs2yLp6EPHnnDSM2mUAF50Tz7kwDIF05iLVjhj690Cu1zDZgAtekmGqxHAyhEmlLI1lhZDnHwi4dTygoMGmB87XLPlIFnBJpHV5CuoLEBVgR95yzWWGVAQDGzDXtjx6TNGLZKBZDRO5sIfaEU/amIeiGifjJhkvS1tVxk46JnRgvMsYZlpN3jLYy1V93Ty0HPGpk/BwDJXYd5qAlabl6w3L7aq61gTBwg8MGi4vy3vXugIWAfNa3OaY5EviJldkd3JN1mPxQsBXDk3zB3Of4PaeOwjSWzkf1VwU14U22guuoKYR9NLwsjSiFeqtZUpKSWlUlWdd3HBRuioeN5iAQeM5ypVc7rIYbifNa8t7+f25Cp+6/yz1+YCmBHotjG2F0R3GWXLJF7dmLCVFjXIwVII7+2FPCs6GEQfMVq4zfcy255AczyoRfxFBvzUuBf5ECBMeXAHclMrx3NQAG81znImqrPtl5trzHItD/Mw45fQIj1ZPc9/6GbZmPefTuiCw3gW1eaifT7DKWVEanPi3FeWRpmzBHG40tJeNu8YkcRI2lN63hJOGRgX+dPUJfn2pxnimn5Wo4qrENQN560vvICM5Hk/iu500qiJOAmHSY5M3q15r2HSRsSQS5sTGlbOXZpDAl3zHxAfUxE7gTvuS9Ie20/ZKz26mXwX89NA6ek/7VTZgM5qnYXwT7G2CQmEadxzqmTrnrH1oPenhOdDY0dETqlDkRp9pO7cvg9c0uHopL+bA9SXe++49/Me3pHj0kVne1G7yS8uKakOpPzk84U5NgMSLrZy0rJ6y1DfA27DO13nxVNI/1iSVouio7u1FSEksWh6xkoPqgFJ1W6K36XJ3s9DNlHMvmF3GXSiVoM4XZpb5qaEX8GODb6YRt+h3vGKlWBPpnoSCkExgXHGEZhwCel7WReYcl86lSpSP9xxHOaMbn1Xe+LoXkxFZ2IGfDjj+Dox9/2xQFzfXmB/ax7vf/+v8+ntfx4+//ZdYbVhe/LrXsaWsX1nh/uSAb/3et1H92i8wd/ow73r+W3n9L76B7kwC613W5ztwQpd2eakF/rdrAc1h1LGloOcG5WfaNIMlImv5K3w30A36Kf523bIh66p60s/K2F7itJlx4+x2vD6DVcct8pA6WGkyaWL8+D8gC3d4Lirp95SqGRUJOqIg7yLZGHTEMU24wkJzhijoJtBOFBHKbsXLjOGl+vAz0grxnCE1+T0O6O3DZzToZUXSru1lLrQXORhu0CsD+fay499pKn5WYq8YN8n/TLjO1pRAn++oPgOKMBYtq51iChWGlFOWcpDmHZlt/NnjF/BaSfSrIu2vFi7doipSRS7lSb+gIpGdhqpey7dTUgXtpIp15ZQlkvD7cgLcdLriz1XrSbBhYw4iBSu3Qvsi1NpQWYC2VAsWLNnAEPRGPFzZcOnQ56dAv8NnH2/zNevxX9ox24ptXjBi6O9wl2YvJimcTAzr30jAdCjqjX7PluXYIm4w6xIVLG+cDlclC080LfORCy5yXywJMoMySg9jWMZjS2qQTNwkcpwdCaPWCZTDDduM5nzaLY/iRgJcDqpoRWwx6dPtME6QuDtruF9jjsbaiuXpunVjbL8b8wM+1Jzhb5ozTp35RhVE2BQfXZziNYf/nJPZ53E+3c/55hm+YM9Q6eqjJzpDfe0BxupPcVnzAtnQp8/fwViuh7X2Aj3OB1jRsICWSTnccBjD/dYI5yD2wP40vCjlsV9cttI2fnjuKIcKc2S3BTR7Y2cLt9qGgatgmwZcz+DLPqw/AfXi3U3GOHBaLHh84Lq7oHKep+unuCD5HQxfj0HZUX1fVkR8LDdn4UB5DKrTbDM+12V3uOxjV8d+zWZ3OV/YD6w/SS1fp+j5TM/EbizXPTh9DpqTMKzJQcbJ5RGNii+WRBJjiYS3rIucVg8nUTv/qCFatvzuY4/yE2fXGC33Mm8NA6U+qk8f5Pfe/xy2XdmPbcR/bwmm4TEZIpOonMCWiw4JFwhMdSJuAlsCu+OSOemANv3Quv/V16h/ECoSYFNwSO+Ji6u0rYCWhmh9JlAnzKGHIoT6TIs+06IAvz7T+gKGel8PHaPApAijel/HqXUUuXMhsc7nm5E7QQNJLWk76Wg2Y4YG0/yHN/fwn/4gz95CnQ+dWOcD05YPHVRm2LL18qRyX+LespFbeigpJMuIJ5sy9I0mFewCaSpKMXUBPshtT/CHzsfZJS5CqIjeOUuxrxN007mJBjefBN6Es774iIh8/Yz7Md25y7i+604W6pOUVfzj0qwiCYozrMbX79IJIauiWuKi0RpGvA7do5rGK/0a1+lTVRcNXn73oMsUqDr4O4l/NoOryUE+g//rdXjHa++m7EP5qtt51ba/ZCWKufy665Pz7Fx3wvPZYonH7/0k173g1fzKH3yat73l+5LrtbPOM/i6S6tcaoHvihbYrHTaMbiD11y7l09cPML1Q/vZH8c8WZOQqkE1V7Kr0gRaYEUpH82P2osW22uIJ62zEvO3gd8jAd3ENzEcgQshdOdNMlNfBSMwJ5KWpAbkn1rxXYcrzoef6if0y4xnt3HByxFFFXLquIyPlx7qKKsn6Sk5AuTzB0jbkJu9LEcx7PFyLAVNltqL9KR62ErEXFzlGwqtteZ4JOhmB5YHbYunW1P8TFDCy+YY7Ao5dgYG+w3LkrNYss6daJsHI1mY8gp8ZG2G80SM7/PxSpbMhgYACEvQf7UKAiQhIikSy8IQ9PbB0nHI+5Z0wZBTZK/eIV+LU5NTJR0U7zCIoydljtwqrKozVzqm7ShebhxSV12ZsQQNQ6aQY8FYLksbFlqGWgb+qC39kyqv7h0lvwhrOVgswpGLlu3DxlUua6BTFqpYgI0B40SgxZ8R5y2fS/htot1IQ29/Bg42neUpVWKO25gNG3KLl6XuFzgSrpAKNzBB0fEXFQWV7IxKeKcXDadSlm19hm+sxPzZxiK/VyrxnJ4c0VrMegOeasV8NY6YJ0DFJw/EloeJuJmA4+IBhSuQ282+oMz1QYrD67N8YPkLBFvuYiC7g/XwJPPVMzA9Q1UjeW4I+oe5tXg3ryl2MyoCveeRTjf55cm/4GOLf8do4HE+zrrob2BKTiA4trHLyillrizeRY0XHvyMN8iXeq7iB6cm+fjOMtFGQG7DUj+nyAnUBhRxcfUrVGtQ64JtWw2XT1o+E1pel44ZtGU+uO1a3nbhG2QyY67I5/PEKEMpv2LRra7xDRcblj4/gNwAJmowkhrkkJfG9wt44bqr0ha3rto6y0fPHuI9V93I1TnD7CSU98FoATIBVJXek0D1jsSXuKGJxzpktyQacErLSqstveGxERn+9MRB/t3pOjfmD/CYrTHgeUytneIaL833XX+jG/91fauYx816BL4EmpyMiIGSSQSHdR9Lo05ASp2CC+V3gj4CZopQq6BKYE4PAT6hLIE47VN8v85bbh+bEToBMP0wAlsCfVpXIM2FyTrpVx2Pvk9oQdupMELgT9jFuU10QJQ+176EYaRhJ+Si97RP3Vxa3xVVgDLhcRQ7vvANz+nhL3oiPvrHM7zvbxd4etc2VmfT3BLHXKtJkLiU0wmntds3bByBdlo2clBbMuTSNjn8IxZN2nI9kO5PAL3jm85YV6+gYxEXWZWwbsIrxztNoC0srFk+Pb/B2/buZoci9LHPj4y8gnW7xonVh8mmumiq6kPXkHPh6YBXVX1LzslVxmYx6RE3cXZFFlGVbKrNzLk1hvY/m4EBMWz1s3TaK3n1bf9fP+U/a0mnYz7xNw+462Lh9EH+7mSNoucxMznJal0cO4U5JUYIhx74UwfofvQ9/zfPvXYHR544SKWd8PB0jVxaLrXAv5YW2LyPB3f57HjxMMycZchfZz/wlKL5XsxkK+LQLMiLPL89IdqHEaxNQ7tqE3Ps68AbSSq4ZIKttJSyXgV1pBLelYWrPColtbAK4dPw+UfmOdwI3aDWils0jE8qM0oz1cdtQQ/XBt3U2xr6xSJWxCZxighrh/ByBwi9HPMmxTkbsg3LduO59FrZL9KD4YCXZWtugltkEO+XKbUX+VLc5AGFHNvLpCTyutKh2vQbxzfr7TM8sJpUgW4IZDVjIi9PKezli94CXXs94iVLS5VsGhdWxbWxrqBM9kBBE64awlk4NRQVW0vStSmpqpikatarQH0Kwjwu9ZkaSnTr8r6T7nNORI4kvQbrF5NIn9I7a6sejZTHuqRFIvjbyDKpUe3UMb56TYb9+/LE+y1xGpbOw+7YSB6QlSXrKlzXTie8oP4cFNOwVXIXITROq0oTvBD6u+DqkuGubnhuL1xl4Sgxx8Jlck5zrUVTqVzxeNqrruJUvJ1QViMEXDAxsyH82mzMv603eV+uyMuvLpEvxCyqUrht3Pm9pSQte8tPxy0WCLkRyx9HFdbrpyhIJBXLizIlvtFa4QOrn6dcvJ6J7ACV2mPMH34cps/zofddx8OfewVvftmLYeZ61vNDBEGZcqoLExcZGhzgvc96PaOV53F+YYY+TyRyQ50U48aX+D8azqSpK+u407FkXTy6Yss1mSFOr4/xrD84ypeOV6DfRyKtK0+6PxnohzXxS/VYhWI/XN1luMoaUrIWq0TcVNjDfr+bJ+onWTMeEzbkFJZiDNeUDZWCrrOYu9MFfrVriJOVw8zGVbZ5Odabs85FQ8LPG36RTOzzO/FTzJo1vKpPO7DMTVpKEp/ug6EtiUah0nqS61PhxIJ4dXIJ6VEFp6XrssDpBb7jG4/x1sMxY8XLmIrXGUgNEnkx+dOP8TufuJXBEd8R9D2FEzVgCkgJBGlRBEzvaYbQKb5wg64Gzk2ApfX0tyJpAk2Klgn4aT8CgQJ3AlJ6FhjUov3rofV1Y+lvrSPg5kB6Z93N6KD2tRnh24zoCdCpmlaLSo/1vvavdXVA2pfe1/FrXaEMXWpaT9vpe1xBvHh6hpiY3Zf7vPffj/Fn7+khfPQMv3Rxgd+96PHxSeOcKIa3QSC7wEPiKMLF85bJOUsul0zylMHQ9dGsQ6OVnJIqasVLbJ4B70pD6naDGXAsBjfRdXxN8SBjj9P3L/PpT1VY6EpzMIIr+iKeXejhVb0vccevYiWdTnIyOvdNpOredJQDo47YVUOp9D4Rx+vKRzQrK7zpZWPs26sfUpg82VNny2/7k3/PPffc80z26n5jY5i7MM2nP/Y5pqYP8Ye/9Uf8yPvez3U7evmVd76TkZtewHA5LZEDPBPxlc/+IX96ArbZGd77jnfx2wcr/OQbXkpRrNhL0bpn0uyX1vmuagF1BobV1T4++ckvcb5nlAF/SMM2y8ayG59m3bBV1ZFrsLxq6RlL0nX5cocvllOCqaNTtbXDn5uFLnWcA6qY6/gzyqBenK2q4RNPP8jXD56gp6eLSnY/2/J78b0Mcy7qlnH97enmGXyJDivlKg6gzKvbc5jyLUz4BcpYHrMRO2WmbjxWTMAaMUvifpkUazZm2C9yRhyw5jneICV/Qjaai9zRu5di1bC2BuebsDQLs1UIYri2bOgeNmwpGKKGwfdzfPL8PM/ZVUICtRuLsCrpEc8dDrmxJAVbnUqKG9riqNWgS6Kv80m6RQLOVblJ9CfFCSJLZzS4T+Osm1QIXOyBlGy7JI/gO3MKx68TZW9qBk55sfOCLUceT5iYzzSb/OL1w7xkYphmTRxASIfG2X793VFLf2DoThv68jiJjiNz1nG7iipQ6DWO87MWwcy03CQUqYQjEpqOE0/Yr4eWrPHo93KEUZWH4poL3MhBoiEtFb+EaV3Ay03QVpVmeoTJ7HbOxg1u9zK8sStHwbMsy0pOrkgWzlgJQwc8GhqahOwGTsRNpprnWNJVF9eJ0wOEUY2vzH+cXHGC/lwfk089QWt1ig/89NX8zj138T137mZsoshIXOJDnz5OYTDNtrjMXF25SEsrjCkHGd74wl3Qs8T9Tx1joFigGrXYnR5lVFZd4v77oELJfhLAWc7CjcZye6abWrbNX84+zN3besllymy0YkoFQ1bjoIJgqg1pg1Jv1SnDXNO4YFAk0dpqisDrZr72IEvBIFf5ZeT/GmPZJ2wRQlZizJFPI6rzmbCGF7eY80t0If9iQyXd5yrA816OeusMQS3F1ZlxulLWOYOkVImqOoAGLkUskC4cE+QMhXFDIJcQObrlPB5ZWeXmLz7Fo9U8W7u3MNNaYKvfxZa0x/EnHuZDb72WF999WZJ2VXpSgEc39Sbw2oxqCQDpc0XHdH+Lt6ZUqsCT1tkEaArtdzCWqA2Oi6fXWk/ARttr2dyv/hbY0meKrDnw09m/AJdApg5IYE/HpItJwE59i2ZBCm0527COXo+217HqPVXKqvJUwEXfp3X1/Sqe0N8CfPpufa73VUujSvyUJch47N9R5O6bsixsTPHR+1b4q4GsE8suxxYrDUmdTlkFY4bcuvqOJKipiFs6D11SjRqGZidq6dUNtpDYAvo7vYS6krXYVUM8b/GUjvUN9z56ms+sGZ7OjfKjacP2oqFUsIxG3YRRm4fWHqU7yLsJsSr8HajbTMOqkdRvBn3uZ3SSP+Gq048MbI16xefHf+JurrxcufLv/KLL4hktnn5M4IbbXs6XHnglzUqV2E9RLCSzvV/43Q/hB8nudJ3o13vZj32A5o8mV1uSfjKkUsk632Gw+ozO6dJKl1rg/8sWSCZGcPvtJYa2ZZhbvMCpkX3slNqVNfyGbfO+IMPXHrXctt3QO2oIm5CWtdQcFPeCnQQj8vQek8zG1aFL/1UdsTgr+hJxzDTelqElz9LWetKje32EqQFGvCwzUd1poIk33J8eZtAvsBBtOA9NJyIcrROlBhnwcmRtTNWR3SPWTeDGoAEMj8dtJuIWM34XzzIBy1GVxbjJCJYcbSphwIu39PETz0450jirhsFZOF6xLHjwom2G2hoUPMhpvGjHzoJwcC7PZ6fqvOuFGXJexPxnIH1b4g8q/T0VMsS+ZWbeUpLkyAZsTEB/N0Tizgwb1p+yrnPPFZJUTHvG0j6T/NqanPsriahxW0oFEXSXYPqU0jSGXQNw3XrAI+02T3sRK/UWbxpt8GOXDxDNRE5ORRmYrHTnFmBPFhp1S/E6Q9CAZgMGhw3zraSwY2DZkB2D9SaOtrKatjxegfMWbo1BXLytxuM0MSMm4GhcIyeJmWiNtvHZYbK0szuZaZzG1k6QCtc4GgwwYJuctJabVOjRjDlyIrGaexzLn9ZhbxCx0ZKafZaXBCk2QsvxqEYQrrrfuR23yDbO8EjjGDv7n8XkfIbVk0/ym2/ewV2vupOdl/W6quBYo770EG8tctv3DjD1+Cwnt4zT5/nsS1mK4pBnQnZ3l/n+kVfzG940mdYcpdQ4j8U1+rBOL1Yc/m7jcWVZFmeWawuGsvxxl2Ju9se4r3ued558gPf13cJzJ8YoFyKWTydBj+xOJ0WIo2T3WVaWLZWcYSwwbKRCtjPMD5rn8O6lp5kZkN1dzmGKlVZSSNqTMpRTITU7zAvMSf5GEb1UD7UoYDFuUHbFfYZ2doyuaIHfPnOU7x25gltGSkxOR+z2Dak6aMIgoBktdK6lHolgW7xej42a4e/Oz/GyBw/Cyjgj3QOkGkvsSI+wGliOPfo3vPG1dV78lhc5zKTrTlWhLuijMVND4yawE1CSxpGedWybRQ2uCKKTVtUhCHgJiCkELMSj9Kb2o/SrEPS6JnadAitVuOr79BCgUmRP2o1ChPpMr9WP6BgEvPSs/ei5U03qwKW21XsKVml4V6pVeXXtV3hnc3B36yl1p88772s7LTrmzfNKWbcL0dJU3bXttiK/vO9KnjMwxX/++DQfjHfwkdUCb5VDSAk25qwT8/W6YF76hkqjBtK1TAqkzkxamIahbYZ2bNkiNx4JmD8e4/Ubx9OM1W5iD6x4nM61+eHJdW4s7eAFJuWi4EeXIZ+yNBURViWUGsz5uqpxO52GDthVYLQx6W2OlywehyROFJpPmxbLF6ukB8s8/y6V64hu+MyNG9wG/y/+e8aRus19O3swY0hlMqTT8thLDlL2YZu8oW+u+w9sxRJ7sX92tndzV5eeL7XAd0ELyEtZ1IQ0tdUMX/7CIUZGr6LLBmz3PLpN5FJUNwQeg4PGVWWuqAhOfYqsbQoGK5uinkQ0V8Ka4oaYAUPjIYvt/h/svQm0JNdd3/+5tfTe/ZZ+27w382bVSKORNLJGsmxLluQdWQaMjbEBY2MDxmwBwmqSkAOBsP1ZEnCIA9jYwfwhBgw42MYGI1u2sK19NNLsM+/N+vat9+6qujnfW92OwsmCgyXIOVPndHd11a1bt25V3d/3/pbvz7iJt6Lr5eLhlYwDd58+eYTPnJunXL2BnqmwmrRYj5vMBEPs8rKMGJ+jSR2/u4QXjDiah6R50jnQB5lpzsU1PONTsjFzSYd9JuCgJ2Ntwpzx2WUCl/lg3na4YgzleI0r4Q4eiC/zznLI/soYm6uJ81+RWc4bMdy511AowZEMXD4vIZ0KmSnPcDwT8uipi7xiT5nRXSFhx8paTWcpBV9Kol4YNZRLqTlVvoWFHZAfMi6KVRzKBfkjjslcZxzFiYhkwzzkbjHORLO1mI5bIpPdlB9OAbLi/BDfXAs+3Y05nsT8cXeNJVPjg7NjTG0EZMV+Idnjw8YGrK3DTWPGCXsRQOd7uBRhYWwpKdpUgS0xtGuGhlKKBcalBRPdxqyAYQBRgks/JnaKg8bwtygaukNeqeSiTQ5lJmkQs965QOKPcI/f4x0Td/D7cZE7/YCstQx1DTPDHk+XLQ+uwp05j1oU8atRm+v8NADj93qbGFHXKCdlUMIQMcpJgsxOLh/rcd+LGvzUu+7hLd+4n7HdBWwQO5oX4yt6W9QSPjPlPO9+30lmtheoJ0WHO/bnDBUrDW9CsVdmK7fF/WceopAL2Zm7hl5+n6O3kfuVi/3zIVSwQxtqnTSLSNuHjzFMTIXPbX6O2dYwu0sjhMpoN2Lxn+/RugBhAboNw6XFhIZvmY08ByimCwn7q1VqwRwP1ixfnZvioEk4kxhq6t++Qmwr8Tga1Tnu5RgioRK3WIo3mAlHydqIxaTGZDDEZneRF1YD9gfTtLcSF2ijHMCBS2WmCYjcHixys8zlfBZLbd712ZP8yPuXGRvaA/kMtajj8uyO+T6PrDzGO65d4eff9XrGJwskvZQr0gEwgSGBuj7QcA+Y/gsgKVPDgCJE/6W5E5gaACt3XN/HTgBQXHWDYwWuKmmEvKvDRY6nz7jupyZ+ctMwmlEJq+j4QRtUh0CitgvQyYSrugUu9QIIkEkzJyAY9NskP0Apf5xWsQ/4VI/T0vWP0z5t03WIJ09Rq6I8cRpV4ya3tmkpFgOed0uV25+f42ceuMD88TqbU1mirk+haRgbUW7lVOu/ohRsep6koZdW+ophm3wMfWhv9nGvCIoXQPNbUeVooizspXa89/ELfPLJHtXhcV5kMgxFsNGzmG7IsfYFfqF1jpHyray35x03nVEWEg2weI6HzoRjGLkymIwb31Oqky7FbJd2bZ3veec9vOoVt5ARb5H0krr+Z3H5skHdoC0D8rwBkBuAu8F+9yvU+j9s0M17lq/o75zv6t+rPfD36QE9v4MEy1p/5kfHfyWeWz36GnwyGUMh1+F33vuXHJzZwXQywbpJOGV7LvXR3dkyXt3iZY1LixWvyEQA5lKalUnRd/KXM0XjiG9Fq2DXU187pTSKL4FITM2MwW/AZz75KPc3NqgUd1ELqpTDEWaMT2RjCn6Fy0nTBdFd6pzHN77jqUuiRcZKt7pxPeNlGPOLae5vpS7wS8zis540ORY3WU4aPO0FdLwMM16OrdYpztiYw36DHxrfTXwlx8qKZXkN6tsNu6uQr0M9gJumcYNop2Sc75uE5ZDN8quba0xsNbhlYpRixjJSgovnrTPHjoplP/XRdwFoCydw5q9COZV3ipDVfjmxSwDWz6fk7jIkyOdQF1VKjEtFJiWDJuKba9BO4ETH8mTDYzFu8mHZdHqW3755lFvzRbwwdlF3nTWIBEyKcOksVEQua9K+Fq1CO4JyxdBdtewZNuSNYSuxrIWKkJS2yrAnBxO+MjxAxrM8QsTxxDJufI70Vh11TJB06HYvMBeOUosbZKM1bGaSy71Vvn7sZl6bz/P6kuHVIz41C+etddHSdyUeV2zMUOLxjkyZ7X7IX0YR9bjBgcwYS8aj1DlFNXqUC2eLTKzC7/zgNN//ozdz+y1DzkwmZ39pj91zL0EsYZQkjA8Xaa0s8/5PzHPH9Aw7wyxXIhjtpW4CI55HYaPCBzc+Tztap5Lbw8HC9SxZy5hJ2CYmZPnYKeenUqeNGZbl+2kM63GPPZlRzvlV/ovSfgTLHMwMMVnJ0WkktK+kGlb5YE4EhvkwIUnSzBkbbRhKDLdVx/nC5jznkrwDbiL2vjnwHCfg+R5Mhx5PRTGPK+o3GGVREippOxeC6WCYMRtzTtlUwhxfXDvDPe0dXPvSPK2OddQwk9cZugoYLsnUH1AbtfzR55f5pSPH+f0nQ2b3TnOFmNeRYUcwxDHPo8MCV04c4d3/+jXs3z9CpDR9mkAIMGkZ+LZpXd2j7QJUCm6QuHQasWcAL2EDgSuBMO0XvnBarj4QVB0DjZgKhMbx/QnQLD1sKYlTXOa0uk2zYPmdJtoAACAASURBVIg6RXWpXp1bWjUJbiFwnVv1a9E2fXTjdN70Vqa/epG0CNTJFJua69JtAzOy2tQv5s7TJ0uWj6yrUOfVOyrMJMCXWLZvz/P226ocKFzm1//sDCcnKqx4BQpxQqkuq6fGUqiHCsCBzJJwpnFZKsbETxnhCMdXN1NtnSbCrScs2RFDdpthMZ/w7z9ynFO5Mi/MTrj4gNsM3LHNc/mUf2b+b1n3e4SVuximR1x7kJ6c8dyFiGaogJ/b078ZQslRqqlL2sxUO6ytXOQXf+E7uGbftJMvA4tn2jHPzvf/NajTy/5MQffM9S81tV9mUPZ/WuZLha+uXO2Bf5weqNVqXLp0ic3NTTY2Nv6nnyAInGnwH97CdKZW6g6x+MAFPrzY4IbSPp7n5ZmIuvxN+xzDZoxs4LuUOfMXLNKpaGBV/kiTg6BiXHorNzmUpuc0hDuUpNrgKQ2O/O4UWZ+BtS2PP/jNIzwxHpPL7KKrg2xMy/hMCIj5edpJx1lOvN6yo7YKolV6GG6pvMiRyBrbZYdf5DhdZm3sNBkNLHUv5HzSZKft8dpgxJmxtC3XOc9Wb5VMcTvfOjnC5HSGrZ7lsw24NjRMCQQNp+bUTAWWVw0jIRw8bBgeMVxb8bhpy/BdG+f4jvIE3fOpb103MKytGtpZS7ljCP0+QbP8qiTn5Cy9nFJOlMdxXFZ+DjoiMV2D8i4IQ1h9Gmp1qCpEMpNSpsg0ulVXXlc4kDecjyyf7lzmR43Pa4fGKJUS/MSwuZoSmmZkBqrD8WOWoZJhdBraMn+LOHdReV1hPsYBRxEoxxr/Y7gkvzpRnyhFVZT6hcmn60jiOW3oJ5MmXatE9SGN3goHMlNMZmZcdHEmXndUIddH88zEh7n3hjw37YPOJct4yfDRdsRlCS2T8J+6CduMZdR2+EIMH0valOOak92l7pMsnP0zNtZzfOc7nscv/Ozzefmrd1MczpAkIuyTT3Sq5XXCWzJc8io05ApwXXGUf/ehZcamR8klFfwoFheui9w2TUO7l+N9InfrnGGDHjeFMxQyY8p8hbEe14ZwnUB8bLiyZV2wzcWoxxzGTTb2BUNcn5vgPcvLnN6co7xVYdYvUK2K09C63LgbEtKJYapgyGYM6zUYnbBcuzfvANpvnDrGbHmK603IiDEup6goTsaMYZyYP27PM5bZRla+d0nkzNyalMjXsCXttdZbp8n0uhzI7WJEoHxFuWcNdsOjUzA8dHaVHz56hl+eW+BEa4yx0e3Uei1e7VfoBRVOupSqF3nssY/wnp+9g3vv2e1cDAREnFuWgJg+A+1XqshJQZLcKBx46ocMC/gIEA1Al3wIBPqkMRNfjHzpnBq0r1VTuQGA6oM1b9iQ8aB7FLbmLPkaaSozATCBN7VDs4OBWVQgTOBSYEvnV536qD59tE37VE4gT9o3FwTRr0+DZR+oOfXU4PgBiHUaSB0owNrX8ukY9Y+2SQFpLcOjPjffOMkNe4Y493uneArL8XKRk00YjQ07lJrPGPZuMxSmDI+ftcitopj0mTe1PgwrGadAZXS3RxRZvKbPpxaX+JkvtGGkwtuDIaYxdGNLIfT5fOsCf7hynKnSPoKwyr5wnHq0RbNzEU9Ovl4RL7MN4xWdada9JAptjzdRFomN9U2qO/fzPd/xCsbHSu52PBcY6P8a1MWK0ddz83e0Glf//49anqv98U+zPwZaOT3HV65cYXh4mKGhIYrFIqVSyX0G62EYsra25sr8Q1/KgaI6N+pxcmWev/qzZY7M7KEc53l1LiRMepxMEg7nSkSribNclMpQEX+agMothlCUJuLwki+dAg5cznXjuNoU+Wo6adJqb9PjsUurfP/HP8Xu0QJr4QH8cILQ81j3MrzWLzPr5TiHzz6/zFi0yrneIkFvCRtWyWZn6ZqAjWidnnyUbExbmQL8PL7tMSkuMhNw2vhO+7RTk38v7wIoms2neNfYfrZXJslHCQsrsKcK+7LGscoX1P4MiK2/otyYwxBuMyTnlJLLkm8VWKx55KcbPO8lw7DDOiLioGNZXIfL60ZBa/REBNtJTaGSRXKczvlQv9I3jzbFqSeDdxpEIctjvQP1JZG9GpYkEJpwoWFZMTgB0Ut87m8uMZW0+P7JPUyVoN6zdOuGnOgOuhC1oL4Jy1EqxySI6pctSWRcvbsmDVuKqm2mZbYVDKWCYd3Ch3uwU4ED4kG11mk7NU6skLCPxKUEu2J7xNE6Bwr7HZ3JFbooAs/GGzSjVe6s3MpYlKNUtqy1DUfnLAdC3wVhfG+7wTYvRBDuV7sbPNKepxxt0OhcYaj9UebPfYrX5u/m3b/xet7x1uuZ3VYgycaOvNXklfsX55TuyHA1yHdSJn/XibFlaEeO2W0xv/L+U1R3TnI9IbKed+Wb6CUEYZZr8yGfuPQRprKTfD5pMJKZYEdmhG1Bmq9+e2wckfbTW3Bwu2E1tvzHRpNX+Vlkxp82IQdy22hGBf7NynGWN85SLQwxWi6weNknqSfs3G54YMs6rXYpY9gxYSiNWPYlJQqtDX63UWMsO86KTcgqC4i1ThiPGcvn2lcYCYZJvAx1PDRJGfNyHHS+eMalvCubkM/WT/ENhb1cN5pnpe2zft5yen2LX334FD+wcIozzQz3Du8i9C1FfL4tHCXyMpy2Xer2HMcf/xDv+5k7edu9txIo6bsmGPJ7k5mzD7ZcvwqACd9odqLt0sbrgXZgqW8WlWZMH4EiBS7oGJWTP5wAoENB/TrEFycQJc2ZAFMnBZKhIuf34p7lrnjfinof++ZbATB9dIzaoPPoHAOTrOrTol+V03aV0brOrbapGfKvUxlp+FTPADRqfQA01T6Vc/X013Ws6hO0UNSvrBOKjlXVgeXgrgIvv3eEi2aFP/nzBR6dKjPk55Br20xB2V8svoBlDKsdWJFpPEytGW3lkh4z1J4Aew2sb3p0z/f4qS8scXJkmH0ETHhFbvANw4HHiUbML135DLUwJAqnKQRDrJjQvaN+7bPYoIqX2d6PdhX/lO8y8qQ23YhqOWFj6Qrf+86X8aZvuNN1yz9UdqSd/3/+Htym/3PJfoler8fly5dpibDu6nK1B/4f7wEJU4G2XC7nNM/yGX3mIvCnfVq0Lt/Qf+iic+oFf8NL7+BX/9UHWGit085U6RVCdtoxfn1znW8KIvYYTwGtzpFe5qXyGuQ16ElrsqzUNyAOVyPnX/kpn7BkdhqXScIN9i7npAglLoB3wJkFnJkwsUyalBdt3NF2bLLPyxHmlYBzzpmjKqXDrJmQEokDarO2R8Uvc9R43OLASMw5JZxX9om4wV8nHW4LKmwlLUxQcXnOZoYi4k1DXLYuynNfxZDfY2ivWFaOpwnuw458pmD9nFKAWZfwfmlOAYGW+0pj/Ivfv8TLb+8ydX1IXLUM3WWYfsgydxYeumC5rqrUWqItsTx4CXa04DqZnWPI5eHIWZgqWnaVjKMfCddTIuhlAUCZazvw5LpyuqYkwJ22ZU1ZHTBcl52g2/GJWxGlUUOzkzrJC0hK61NbtRwYhlMb8Oi8nLiNyyQg2o6NRTjfwQUI3DCcZpy42IBTSbptRaZeLG3f0iHHfTnDXBuO2IxLKTabrNH08y7q+AHjUUh6ePEWUXuJ68vwylnjcohvnLCc7cFWBo71Yv5LFLFD2ibbddkpbiLhfPcymaROMXmcuYtbvO0db+Hfvv1OpiblsJak/l2SnE5Yp1LXkcTqQXd/bUoa2xF2MPih5WtfvJv/vHOFP59/iL3bb+cOL+MA6kobbhqGG5V/Lc5TIKJke3xu4yEOTZXYW67yodWIoOuxZwT2TkDQMHhRj52eoWU87vUyTGQMH+tETGfl5xTx3vo8733wCD+wr8qLTZWiP87eVkg+ingyttynCOfIIm2uXBzumz3ITx77ImfaF3hhfpYgiYmNoUjCEfLY7E5O9FYo5K/B2JiwPU/NBJzxiwx7WZKoSyczCfHTfLJ+nImxWzhVW+SDF9d4qH2OS+Eow7nrHMX2x7prvDG7nf1+nkXrcdzEXIyvsHDkI/zmj97JW15+R4rFpKkUwNEQIuAis6SiRQfgSSBIQEn3QUESWtdwNCg7kNZu9tIHRNKSyaVAsVAChNLeiUS404+U1bl0b+Wbp4CWrnF+idUXGXqzBnPCkijZvThnhgSITJqpRi+VQJZAo9qntkirKMuj2tEfh9x+ncOZi/tRr5pBDfbLLKtrUJsH16LfWr9+PWMCd4qM1Xa33m+/nr0g9V8UMExyCdM3FPmVnQd4w45LvOkPjjJ3eQfvmR3j59d8Xhl47BhOfY/3z8KxE5ZHVyy7Vc8JQyawKf/6WQh7hlPZNT7CBYa4gYteyHbPY9pPuBwF3N85w1yyxmjhBlp+nlVFukYb+HGNaOTV0Hwca5uYRB2S9AMmRE7ccVGvkbiYGOWGQ7e7bkxfJHXos78MHpO/95kE6ATspNUYaDv+3gdfLXi1B/6J9YCe4U5HnsrS5Pj8XVAn8BWJCVjjkkwiX8Fl5PohrnuZYeHhUzy4ZxtD6zm2i8A3bPHRrTrX5Cusdi2XtuDunemMdeFRy/hNBl8UCyItnUgJNDkPybqEgYXR/pRX44prc8YxnXt+hUikswJrcYv7MVS9LLPOwTebztBNhsi2HHFrQawEcuaT9sKE+JIu4qQzvhOQc9ZyQD5Itss4IfNJzJDxUWTschKR2eqxbQoeWbFsm/UYOmRYWrVsqZ1bIJlZUHRdFUrKSrYKdfm7iSC0l3Crl+O+3SV++8F1fpRJTMY6ItLM7YbZMRh5BCbuSjM+VJ827LOWbVNpRG1G6YUysG8UzjYVDGGd/CzUYJ/KjMMjZy2z1jhi4CuAEh5mrc9xWnxXcZTz3YDIV8qK1KQloNNupf55xRgekPkngYoHT0fQCOVcDY1YOWINZ5W7FvirVcv2yFAJlK5KlCaQd5HKhvmuYVdoaBhL1TPsSyL+KO6ylTS4VbxpXs7dKy9aI3LUJhnCJM9CzXDLLmhvQXzFEAUJv9+LqRmPKS/DXNKinnS42DlFuTvHhQ2FTU/yH+57O2/4tmsYm1X2g5QzNPVp6it6pHTQYy5tndb1KA0EugSsOAE7lrHtPr/8nmu59fs+ygdWzzFePcD1cUzJGBpJwrbcTl63+x7+ZPFJxrKGumnzsfUnyUS386owx1yU8FgLfm6Hp4QQLDYNSwq+cSwZHhkMI/gc9CO64Rijw5P0kja/duk8v9b8DF+VmeKF/rXsmChizouUOZum1DqREFQTbghD/nr2Gl525hifCoe50Ss5ntzlJKFiAl5vLL8cbZIzPjN+iU0vxxKWKWvZ5RUZ9UU6XWOlUObnzn2Un5tbhbwSBWch/3zHv3eL8Rg1Ofyw6nwhG9byhG1xMZpj4cn7ed/PCdDdidexjl5FZtfUz6GvaRNIEiATMNMifKD+1nAjMKV158zf156pbDoUpfv1fgtQqR5Jcv2X+VTBC/pIa6W6BZZ0I/vn0Sute6sUuHayn1nhGBhlrJjo0yKpPb2+KVbrOoeeAwcS+0BT59SzMgB92q826L+AqAui6GvsVE4aQwE+AVnVM2ibrlXLQLOnddWhY9QHqlORwhXjCIsVkPTSN8/wF4cKvPM3LsAjNX581ywn/RzXrUfkMj7vuNujmoWtU3D2kjg+LZXQuDSFZbU7FDH0BWhXGMn7zBE6s/6a9ajHXd7fPEUmGKWn7DrBEL7JUG8/xmxmmvLQi3n60jw2WsOEUyhdmCYGqYNij3xgWb7c4cV3befVr0hpTDQMD6w06cU+e99fFqiTqUoaOgE6LX9XAD57zbxa89UeeHZ7QODtmR+dbaBR0/bB/69EK1Sf6h4ezvMj33s393/dJ6h7+2kkO9kXeLwizvFbPcPLc5Ybs4b1C5bWLhg5aGg8aIl3KU9pOtDJl98oxVUdQkXFapYuxaLIR7UuZOYkdIukt4zxdmBNQCszzO6k7ciER/0M223Mpsh/MxPUowWyXgHxtIvfbLcxjkOtZhMXnXvKxuyyMSdt7MbtUSTcMrzK97hAyDnxfYQ+Vzo+p1dTOXjjXuOEtz0FE7Ow3pGmCzLXgLcGoYLHdsHZc5YdQ4acAg0uJnzzdVW+88g8X3dthYN7cs4XJn7CUr499Q86eT7ND1uJ4ak2XHeNWBkMlx6xcpNxPGLDHlyoW2fmVS7ZYzVLrQbrPbgUWs5Zy+etZcpYziV1Xpjk2WcD/iLpcihWGgFdv3URj5WWccBzVaSmyhiEoe1Zijbh6KJHLCdvPTsd6/KutmI4EBgmS+JpE82JaBIUHGKcmfmJFiz30oCyFSyPWePSxonsOfSHOJe0uMYLuRBt0JQWwM/yxXaWtbUUMC56UnJY5nsJ55KYu+WLF9dY7i6yM1lj/vQXWOMiP3X36xjbfBHf/PYilW0JcWJTZ3kJVkmB9BFPBY8DAWnqNQcIJGAVUa1i/eT2STfm8KERHvrFe7jt6x7g0eESzwt34JmI3lBC1S8x3dUD2cDEW+S9Iqs24kOdi9yV20WVgGY34pfm4GBiOOFn3TMoouSLsaFqLS9TirzI51XGciyxjgS7Xb6O1UyBj3dW+PjmIrONRUzd52Rvlhd3J7hxvMwhPe9Zy+GxSX5sc5NfkBPl8G18Ku7x/X7AO7KwLEfLoOLAmDJe1L0cfm+VdbvAjfEm3e5ljjeeJEgaBPEyYfEkmaHXYb2ic1UYthF5DEXj45P6+Z00Hiu9C1w6+jF+88cP85aX3OnyFzsyfkWIql8FXPR+6qPJgkCaAJDTxvV/NccUQNJ9EeDRcdovYKV1mSZ1w1SHQJ/AoLZrs4CcQJBQhDSBql/3T/dX5bQ/MZjApgGc0hwqgv42pVUAu2AxSjumtklrt2XTFGYS+apLin9p3yb6GkY9PwJqOrn8FzTmDBa1TeOPAJzOr3FUbdF/XYPM0NouU7La4epJJ6vuWgb9pe26VhU3yl0tl4uE2w6N8Ne/OMyffnyet73rDO/dNw2VEm/zPU6eShjegCA07BgzLiB1WZO7luW2rM+pKw0+vppQVlqK7iqvzu1iPYl5SBGvYiuOVglyexgRPyQ+F1onyHQX2Bp9FXuDcbYqd3Nh5QPOr04N00RZ6cNcRHnZUFta4sf/+fcyOlL8khwZdMuz/ftlgbqBcJN24yqge7ZvzdX6n4seGGjf9Gw/8zM49+CZH5QbbP+H/g5mbuOzirXvMRZvujRdbSVcFyeJafHb5z3+w84c1283zIvUTOSrATRPWEJxNGjgl2NWWQ7PAhIGI6dpzegHs2rZFTXK29D5fJi4SUOJ2F0xyyNJl9cEBT6dtBk3AXu9PI8XbsI3OeKkS8bPugl1y1oUsJezMVaaIJOjk3R4wMZMB2WipMOijVzZx9vzBGGRH9y4xO/Yndw9kaen5OJK9zmdypzCHoMXWraegpFtKbWHzLDXlTw6py3dEMZvhEuP+LwkqfB7Dy3zr0s7yE5Css0QnQR/yrC9Dk8tW5bnLZMCu9sMwQZM32KI162yljH+fJjyPXonLfVFKK7AtTtk6bE8eS7N/FO1hi+02xwoNPjhQyVnur33iEdeidk3LLkKlPYbinIeq8rcaZg6Y7ltLyytgh0y3DphuPRQau2KujA5rIwHNuXB8w29Ii5Dxa6ucq1bntoCmWG1SPYNYzhrY57oLfPSoMqWl6VBhC/+v3idksAHGW7yW1Tzlge3rOOZPUPEz/Q2OByUeTrewk9q7EqOMTf3WV7/4pv53l3fSLA8y/qQR26v7l/fEX0gYCVkB4JXz42EsQSqtqmMHqG+TB5sNzJzxjG3Hp7hb/7VHbzk3/wt0fM7vNrbw945y8xIQmW8DJc9TKFHN2mQSxpcap3lAyF8z8g+3tjxeaoHZ7F81D1/GQdohWOUu3U1gXxkmVCqt66hGFt+uGj5QHYb17S2c8jAH3W3c9Q8wftrj/H++ghBK8OPrIYYX72ZZ6ML+3oX+VRrjBuCYbbbgAvd0EUH056nF63RilvOTzGTNLkUrfP/ewVH9uxnd+PZrvMbrXU3+UZ8bgxGaZseY75hoWv4Y9vjorXc7vsE8ZOcOfYg733Hi3nrvS/Aiy1J02IEkNSX6lf9CnzpVy+h+nYAZgSQ1N/P1GAJ8Lh8qs8ARAJDOkYmVaUBU8CBwNEgFVj/frltup86n7YJlKmcwKEAlmtLn+BYWrU+gIyOK22ZxVP6Z2n9daxIBqXhE2AUCBRyEKP04BlSnVoG4FK/KqNzCwSqjSojwKq2yzwsoKa6VVbbVNegjNo2AIxqlwIwtDRTVxOBTj1/w6OGt3zNLm7bX+fXP3SU9/xBhiM3z/JvvzDCi4KYu8oeOw8Zlk9ZnqzD4gXL06uWY/V5TrQ8ZgojbCY9BOw7XsBcUuPxzQfcBVqTcZdoldO1cRST30fXK/B40qBbPIhXP4Dtnk8pCdR5AnUmwUprR4Xx7cqh8t+VA+7Pc/D1ZQVKCMzJYbxQkNS4ulztgf/3e0BgTebVSqXiJioCcZqw6Ff79NFzrwhZlflK+NSlvZYOTtXRMqvNs9z/4Tn2z+yhRN5RJzQ8zyW234gTRoKAzWWYHjUUJ3H7c+PGabjcwKkxOpOmKKpvpGnG3FhqPM5vLfG+/3qEkbGdNLxpbDBMYDz2CNiZDNu9ED9JWCcmsTFXao9TR5G+U46ceMgEbhy2Xsg2UQUQccAvsiITq+05x3NlJsh4eS4mPfYiHrB1hqMVWqbIV5dnOTyZJZOzjn4lt8/AiGH9fusiKYvbwJNfnay8anQTavMp+HMGgYuwKyzwtjNr3Dfjs0OJICcTzBVDMJSCwZExQ0nBaFdSYZeT75BMh2VDtwbzlyFfhrzAQQcuH7OUthvHS3dpCU54HkGS8DHT4z13D3NNxkf0KBlleKipnLCzYdcOHLeeomzDBuwsaA4vbZKhFBkXDVtqQjdR0IAlKyE4YriyAV3lr+2mqTKnM4Y1US0Au0xKQiyqsCPWcjIWc3TAfr/kCJ9P2JhG9xJR+5TjwhL1x2zSoZ2/idNexqUR+9lE/IYRHStNXoP5lU+wcf4pfuJV9/KTb3gN14dTHHky4dAPQHWvSQMiJED1UZ9L+Gp94N81EMySTbpAgZKB7UhlpVXRonfET9h13TCHkjI/9acP8YXJEvdlh7m27DHqVXnvwuO08yNkTJ6WyTIeDPGIsn0kcHumjDj4e0mPj/XWqfsZ7hQBoDGcjOBwIcUtdghHnCxusf1G2TgSLmO42RhuDQr4wTi1cIRqkOfWKOT99YAH2j4P1Os80llnLVqG+ucI22eYa5/j9+vHONpbo9U+xXr7vDN9h7ZLYnv4foGcP0SY24UJpA7XNZaIugvkgwrZ7A6yNqGZWAIT8AgxJT/msbVjnDvxGf7jt9/D27/xVjFdkHjiiOuDIvXzALQIzOjmC8AMQJcDWH1wM9imbhagkXZNi4CPjlFZ3SPtEzBUponhfmCEwNPgHg0CLXTewb106i53WSnAU1366N5rLjkMpupiT51WTv8d+HJADBfc4gCduOu06Pw6Xv8diEuDMtyDJXOzFv2oXVoEztQng4mD2qW6dT3apkVFdW3qB/WVjlHdKqN69BuIjsVL3/PAMjGc5bYbJhmZ7vAnn3uYx7oe4XCZ3+taihseO7rSTsK1GZ+NXoOfXPock9lhlk1IJRhmmzHs93PEnZN8tH6SauEA3WCEnpelLTLCpEW2dJMjYq/FWxh3g+skzadcLlg3+bc9qprHzF/mB37o5XzTm15JoCi254Cbrt9z7kdd9fdeJNDy+TyNRsNFCQ4iYP/eFVwteLUH/on1gADb312+0lq5v1u//ktG6jy5rOFtL7qT//QrH+S/9lYpBuOMYllLetwZ5viNrY4jtHxDKe+c//N+OoZKA9UWi4HmV5K5E5DMQbMGw5pNa2DUgCgfO43oRpq6NomoLUxAI+nRMyGjJqRAj50m5KhoSMwSe4r7OR+3qPpFAhvj2ZimjdyYW/HyfCqJmLA9NpMWysrTSLouk8WE8XnYthENbU6EbWaEuew6nbDA+iZs3wHdRUs8BNFof3yvpTyetS1QtghlDLCVlHZkeBbKOy1jFwPet63Iu4/PcdOhGyk2Da0li80alw+3WITyqwzhnGXlE9BQRGsGhoet87sW4KsfBWZSt57iuGF1zjpt3EIOmpFlvtHlX74sy7V7A+LYYjOG3KaVZYaZFxjCHixfhMqkywmPuJmv1CwCaE3fUs8ZGluWeinNGpFfSlNKybQq8tvJvEEsFNUSbLVhS+ZiC89z7nqWqnzhbJfj0SZj4QRLGPZiWYw2ORHVyIjLRgjMCnz32GU8TuBzNK6Tk8AxMX73KOdPPMq1eyN+65e/hxfs3ks4l/DIB3oMv9Vj5/PTKFbxgLnnI6XVco+ni3SVcBZo0CKB2gcXRvnKpZVRPlK1QfYvVSHQqijWkZjXfN8s7wXe/u7P8dPPg//Q2c3MaJ6bR+DxTpegUKBnI7aCIcZNyHvay+zpDfHiPSUOBR43nyrxSRM6rr0bjOGWAPb6hvEhWFV0cpQwHBjkDzUR+9wVy+cOHjMxt2UztGyFqcwODnkeO1DwiZj9uyS0uTa5ll/ZeoJL7ae4pATAwRCel3ecuVH7AkHhBoxmAd3LLmdtFI7h5/fiaVJnMnh6ILMbXNn6HHcXDhLK10pAlB4jfpvHH7mf73h1nW/7l2/l5ill4UhcAI7zoVNHie9F4EfadUlc9aUWace0CNhou/peXdwHUG6f/uu+aL+GKq0LJwgUbvXJf3WfBrlitV8sz8oUIaAmTZvqk6OkANGAckR1pLcxrVMnU/msxahdBw12wpIcSa0B3kw/8MLV0wdzKidllNooDZyuQ0BSbVNdWgbaQ2kUta7Nuk6dWyCu3KduGQzDaqsWXZPOpetRWe1XcIXaraU/sRCXos4bNxMmp0J+4u17edXzjEM74QAAIABJREFUhvihXz/Nn3+6zszh7Xz7Rp6f9nO8cTSgEnj8xeXzjljdZGaIlPnERpSy2/jL7jqPbjxC3stQ87IulWLSWyVsnaakgAmX8TpmxmSpK4CseIho834H+ORAoTHdc+rFMWZ2voScmMXd9n5f9Jv+bP/orF/WMj097aJfxef1XAi/L6txVwtf7YEvswcE6jKZjESAGzz0TP9PTa7P4nt5zR2zjExkaW5c4A+Hx3inV2bMy9LC41+EJT7aTbixZGlsQWFctA3Q8yyOx7U/uCUX0jyiGsOtBlAN3sIBG+qQEBJpQELn1J0zGdaSDpueojgN13s5TiUtLnaW+WfDk8xUdvNjy4t0PZ9N65ExWSd/GiZgFuXUjHkkbjCFT8XPs266jn5DDkTSapholRU5/QUxDy9usbVvBzOlBPn5Z6qG9TOWmUMpAXE3grpSL5Vh6wT0bkjNm+2TcPqSZdduaeRiirbKB7/Q4BvHNrjvHmVSj4lkEqoaskoPttXnSp1OyUY7c7BwCaoKwlDy9RL0LsGyOKhGdLjh5JblXNdwudvgrvEub5kedc7hjsMqMvhnYd63jmz2wKyhtgxr53EcgJdrKZnwjpLykhoe3LTMdFMiZVGGSUlRVoaPnihUUsZ7ZUAQzUk5C5NKAL8Gn/fkuZjK+Y4jL4NZ5VUwPueTBnO2S87LEkszIBZ7BRJYEeHGbCQ1ziVNxqJ5LtYX4OKD/PK33sWbXvxypsfzCE1udC0ncoZXv8Kk0Y0bAip9YCC5OQAJEpoSogJ8EspqlKgy6pakJs/BPrAYaFa0QXMFyVtFVI7FfNOPzFItW7725z/Jd1x3N9/Zq1Jwgr5NLKjviQevR9PPMBaO8WMm4d+v9bgt45GEAd/tJeQSQ0Unt/Bo07Lbg2HfsFv+ncqTrihmP6GyqaTrHrNd6GYDTLYif3mmsK75wp0PkWUik+Vw6PPblbt444We85Wq5vfRCUZpSaDnr8ULxmnKZ1KUI0IPin7tXHHviy4y8XJkcrudCfxvG09QqtxNjw6PRufh8Sd4y+vgZ9/xWsYnK9hOQtJQrlG9dnL+Esjqgzk5U+q/Pvoa9KX6W6ZGbde6gI0ma3qPtegeDe6TwOGgjO6ZwI+kuOrSPdOvtstXQmDI9b+29c+n6FItap/utQOQfeAnYKb9rl7r0u1pvz1usY/bNGxbgZ5C0zp2ALAGbdSve476+7XuniVH2Je2xWl9++127Ru0q38dqlPXqnY7E3E/KravSXSmaAFYtX8AdBXAU1YgWYLxLIcPjfFnPzvGf/nUMd7xk0dhahs/uWMHrfWEA17EI/V5R+zZsTGVzASbNubR1jxn2qfJdS4S5a9zJOw6QaLUfLJqBMNESZuuqEuSNp3eClZWnewuksbjrjOC0GNhQZ04zOu/Zrfr5oGbjfvzHH192aBO9A87d+581iICn6Prvnqaqz3geqDX7XFlQbGPGgP/O7Bz//vm1/5O9/OV/ErBo6Uymednfu5FfNe3PcDtEwdZjDOOemEBj7f4niPN/MAy/HAGRwVixyErsJO3eJLzQ2mghNLeGGm8NAMWPYFm60qs2p9GO5LMuMGCF+H5Ur9EXIlr/ImXZcwmEC1Q30y4NlvkliDLo1GTGb+I73m04iYjQdmR0u61AesCgnGdis2y3XicC6t4GvSiNQq9RbZX7qSB4eNJxB+e6PLm4YDJMaUEkqkElMqr0cQBUwXZSYBNHYZGAVqnYPY6WN0ytBSVV7bsa/v85TWT/OLxRW4fH2Jsm6E3DtFxSzdWmi6L2OLzz1PKL4Mdt1x4HOYuWFazcI3QroHPnYKRPCzlLLmGEssnNMuGN1w/TJiFjU1L736Zay1+M8XCI2VDrwa+TNzKUCFg5sGunPwCU0LoYQFu3zAhkmEDx7EM+bDetSwnyoAAO0PDorX0srggi92BfOrkZC1tJzzSXXKz/ilFEBs4Ea074aE8piIjjuIaGdtFypMjUYtMtMpuLnLi+CO8kQm+6+e/kxe+bA+ZDY/oYkzQhIXH4PA7DMO7IFFKKAnlvlxXt+vcwnIp0Oj/KihCRNebqUB1+wcPvkCEFh0j6eESsqfALpOP+Zrv3slDN76S2971ad55ag87hw7D0pN0rdJi+SSdc/QKByhJ2sVLfHA9x3JmEt9o8hCgNHGiixE4GA2h0YPQsyhl+FMJbG5Zmonn+l/cgcuij2lCMWOwXsKiUjfJpOvDndKQxbIi9tieCfnd6h38fvMyn+pcpiQSbs0ExNCtQvrfvyjlT7bJJl5uLzZpOJ5AL6jQy+7gc40jeH6JpLvM0NwJfu2bX85r3nw9YyMBkZfgZYwjCHfBCsqjqtxv8n2b0k3uv4p6N7Xo1AI+OrVAnQCQtOx6ZbVPvwPw5QDXM8CdyupeyGdBEztV2a/WXYaOH4AjnUthxdrmbni/7MDc2e9v1w6dU8BRJl/NA0tgDinVngIhUnYkp2GTlk6BF2qDJpADs+jgnAOQqutTfYPnRdenMmqrtuk6BudXnRoM1E9yMnXbBYhl0tVF9CeqApT6r/2qR/XnrTOv6t4nfszQlMdb33SAGw/P8HsfOsK7f/cKP7d/Xxrp0V2nFO50Kfha4SjT4RgzSZsz7TNY+XQYD5t0SFrHsb0r+JUX05XPtZXJPaQeL2E0XsZ1jK/IENkmYsbKHpfXF/mt930Tu2d1MXrXBjfF/X1Ovr5sUDdolRj2ry5Xe+D/xR4YaJj7niPuEjrtjovy0kv4TLcClR1QnjwbwUGy/krj9vIXHiDDgyx318gHU+yxuOjLMzaLcsG2vIQLnihIYEik/2vWUYK4VFgbkC1DS7QgKxAocEIDngZtZzJztl6xokGSIfHyDCUtSiZw3GJ7JcRIyNHl6SjiB0seP+DleUvNEBjDXFxnH5ataIuul6VpQvZgeMp22WkMXZOnisfZzmXq8SYzcZMX+sN8MXY2O57Y3ORr8+NMZxOnMAgmoKkcqZ+2TE8Yl3IpKKSkvony2bZgJUgDKJzFZVWs8fKaLxKuZvjw0WXedtsUxovJ3WpE4I6v6yz2k3uLn69imHkNlGR2PWr564uWbAyXs6BIWWlTnvBi3r3V5oEXFBi/23OEwtlHISNtYQFMEx5/FA7J/60JWyswvj1VUgQ54/JKXmootZch37E0CpaRqke7CeMd6EaWpRrszoppwnJmU8BOdC0wV0utZFKgaEoRW8u6JFQ4zpRfph7X+XxvlULSJvIrzp8uideJ/WGy8RKj7ae4Ur/EiYVHeOubbuUXvvH1TI6UHepMMgl+27DxRev80Q++1LiE518SgnrihQMkbzSM61l55ke+VQNZNBCcTiOcgqSBQHY5YfsRloP8sDYTc+srZnl4/D7+2bc9xYOnlhidqrEh0mrnjCVC2SVWkw4j2RkeNxmatk7OK/N0ZNlhjIIwuSZI063NBLCtLD86S68NSp+px3rch8U6fBK4QY9327hI7R9OPL478HlBxuOpFkhf4kWGv+wm3BxWOFgcYhF4onWSil+k1lvC760RmDFHRGyVkiSsYmTujhsuF7JCzRU5LsVbsXeR+vpHORge4Hd+6uu5/YWzMG5JGuKw7IMraZlka9civjcBM/HI1fvBTQIy6nelwhIYEtARQBK4E+ARYBlovLQuMKb7IQA12K46pBIW+NKpdE41UBo6abUGYEltEOgZ+LcJBLpAC5tqAweaO9WvOgTGtKgutcWBTOPyS6MI2Iyl93kIqn1NnsrrutUGaRbVRl2b2unqecY1qX7VJ+AmHzz913VrGQA1XahMsqpPbRUIVXvVFp1H16L6B4BX6wKV/cXIHUDvmLVk2pYXHKpwcPsLednUWX7isw9x+rEzVKvTtJIGnWiTUnbaYcZL7TPOLC9SYRt3iKWh6y3i5/djXIVtZ87v9Vbw1blBhaTbwio4Iihj4gahoqrYz+GDh90LpKjn5yIt2ODaB79fVqDE4KD/1e9AWP6v9l/dfrUH/qn1gECcPgoAUjCEUoVtbW196aNtovEZGRmhoKzvX+HZVzqTswyXy7Qai3zkz87BzA6usUXWkibLccSrggyjYcK3tHrca3yu3WbIjIlAE2JRJWxz+dkJGi4VoTOb+dNyavY431jhfZ94gpGxWWo2jxFZsPEoGkUa4rjBXuspwlUs6ke5Lz/OK8f2ksVwvJ3wJBF3yEZnPOdHshm3yfg5PueFTJmAR3urDqgdCMos2y7N7pJzTL+5eD3GeJxxMqTJVwVFRkYC/Go6lpeLStljiLKwIP/ArdRJe1i0LeuwqvDHjKEk84+F/JThwhVdasi3X17ijVN5pshI7sKqdQJVFH+OgkEDuwInMsouIeoQ6DZTEtMLsst5UPUNP9Ta4KdDn9fuzaFME05B4qzUxoE45W+traTpvcaqUFcC+hgaLcvpLcleAVpYMIaH2omTRYWsx6U1y0KieBDDVksaPNjspS5U6zGclU+dBdH1SZm6bC2/Ydtc6C5wjRe4lFWfd478Z5zjvlUqq84ctrdAYGK69VXq4i7ckeHfvfmr+fZvvYeJyQJxPXbpRUzLOB+4j/+pZfZtHtVDfc2bLlCWTWnrJAclePvbviSE9X8A6PTrOsU99Gl5CWJt17H9dVWl/nZAUJrHOGFmusDX3rOTXV6NP37g00wOFWnaDMbdpATfK5B4qZP6hfY8C7bLG8MRpmVCw3BwyDBdMoQCIwk0W3AxTFwidsUFTBrDMSJmMY4bT4rY9bhJ1fNZtR6FyHMWTDU1SGBHSY7yCdNdy1RmhM8nHZpxi4xofnI7U9Nwb8nZo6WJ0zVZ2067Itqg3FsUuzc1r4ddGuIj7/wWbr95jCQXY8t63vpASuBD/SJQpW0CLQMQIlAkv7KBhkzlhOpdI/sjo/pbi4CgAJJe0gFAGtQzuG8618BcKm2W6tV4oMarHpVTOwZASPX1NXDpBercfWCpc6o9AmkqpzaprNZVl9ZVT944X1azmL6vrqCUUtqnY9VGtV3R7qpEgFT0SqpPdQm8qY3qF61rUiBgq2vR8a4v+n2p8+r61C6tuyr7gE/bBXjlP6h69V91CAT2lK0kja5NWvKd9Tjw/CpTmxX+8JN1GtUlOq0VJvw8kVdmuXWKhfrDZJTyK7MdP7sNG626j5/dS5IIvMUucKpMRDPawCQdbLTiNHnK8jJajrkyd4U3f/0r+OY3v8QFgelivjQ5Su/qc/KtW/AVW1IB9RWr7mpFV3vgWe8BTUQU1ZrNZp1GTsTag+AJaebkbqB9+kif5XL+faVbJZyRgxtefAf82u/R7m3weW+UCRO6QAWZo3Z4Hoe9Dh9oGarnAqbqHhPXQD6XTmCTBetMmrUrYOsWgZB0hq9XXH5CRfzMTnq9DUcb0TFZ8kGJfFTjSLzIZpgecM/wjSzU4WQ7wz4fPitgaOVaFbNhPCpBhT1elhfYhLN+mV1JiytRjVow4gIrLhjDjmSLV3oeDxs5vwecbl/kSNuyX03qQe8sNEagUoX2cdg+BufXoHFJ/FPW8e7t2GvchH5rGYZEg7IBB8YSelcKfDNVfuncGv9fKc/opsXKiTuP89nzbzTEx0mpJM6mgkM9cDBvWIngfC3hE0nMhxtdfrCS4503Fwkalo1Pga1CZgRK4rpbh85DlsmGZT2C9cCwFVuKPVjNwP198971Y4YzBq7r+MxYy5VFy1wMUQC5ukUZmB5spfLnsAST3BwNzMsiBxwwhgtJRDFuMONlnRbpQrRJpzNHjgDrlZzGyPaWCBTdKtt6M8dHf/rrecGLqoxUQscNljQTvBEPK3B82jJ3DJbuMszc/t+1cE4mKrm7NESS/QJi6py4b5WTcJQAlWAVmBBI0EECB9o3ABeegJv4y9JtDiAKPAxMu46wO2bsUMCtX7MDfiegkG1iaovORO/5o84vU3J4Sydon2Iis50vZmf4Gyz3epZWy2NbRsFAhkVFR3uwlhg+GfXIeAkbNstuZ7I1DhwLJ9wQFFmxniO/nhbwAy5hnBl3qGtdNPGy0Vvs85r8Pv6qc5Fa0iaMmyQyt0Vr7nqtm8RkyNqeozzpdhdYibbYld/L11bv5j8vHuV46wrP9/Y4LZJRUlv5HwpkCJho7jegMREliUCPM332QZb6zWmm+oDOgZf+oKIZjPYJlAz6XfU6sNP/HQAgaf5k/dM5NbtQPSonUCWwJMCo+6c6BXYc6Om3QR3mTKFpcIRDHzpWz4bOOzhWYG5AGqy6mqBYEzvbP16BWO6aTd+ptQ/w1AfyIVx/BtjVM6I26PoGz5PaLI2czqP6tegZHDx3qlsAUJMxaSCl4dM21aVzqC91vXqOdY26VnWCNI16vnUfNHFajfn0afjwLfdxsbTJ9x35KAv1M5SGGnTbZwm9bD9GQFkhuk6d7hdvdf5zSbTkZkIdf4hE42RcI0m62Gjdmed9X9mI9LKUuPcN1zAkNxO9Yump3fmfy6/+MPNcnvLqua72wD+dHtBEROBNoE25XsvlsgN5Anpa1zbtG9CcPBstlx+IlntfOsZL7quycO4oCS2qXp4TxPx5p0U59vjBbJEp6/HJKHEDhq8k9OcsrfPWJYvXYFtRUtE2dPVmT8jso1FU/nKbzoTkB2XnspKxPTcWjhqfB22Pp5I2+7wuk8rjKW6vErykkKfa63CHcr0aj5KCN0zAibhF0Xi0kjYJIl9NqNiIGfnphWMselnmZYjwffbr3H6Jk/EyW+uGy6ctDdF7yBdNFy2fQAu7yrB/BLwLUMhBeRskShIgn7TTUFsXTYjhmnLC1w6N8P5zMR//4hbJhuzRaXSh6ustw9w5BRJA+3I6sBZ3GraysN60zmn+JQR89w6f757JsrpsOXfaks1A9xwkytBwGVZOpHxci9PKTGGpdGS2NYxUDTOjhns9KEiQbMCOuoIlLEEOsgWYzEA1hnoEi3IH8tJsFX8VRxwjcbxmBxwGsC4CNkxq5GOxSAdsKiozbpCLNomVuaP5NHHzSYztOjDT3hjmxuft4PYXigk/dK4CSSfBCCFtKKuVpfOZhOP3R9z9akNW/pYSgOocCUsJWi0SOhLeEpDarf3apY/KSQOq65M5T/+doE/Lal24x3GWqW49azruGdLETfAF4GW6Yooku48k3E6nt0SrM0++M0dWOYatxTcBl6N1Hm2ddvL989bjsz3LXy/Dp7csD7Ws8zmUBf5+o8BMPXOGDyeWE6KBsZYv2IQTxudYtEkhaXIOy2exdG2Ps1HE6R4sRImzEFZICLyQqHDA8dDF7XPYqJZqL4mwSY9ctE6t+TTtxsMUstO8a9sb+YmJr2HEG4f9k/zrTx5jfiHC0wRC3IXqP33UBwJLWh8s/T52fa11ARH1l8CNgImOkblWH/Wn7qX2abv6Xh9pwvRf92GwTUBR6zpG5eW7J62Y/ssdQb9aFNigeymAKTNwf0hwvwJs2ifQJB9E8cfpPPLB0zWorEClFv0Xx67KC2jN6NO/UJXRR753aotApp4dtVePgK55cK26lgFQdc9cP5BEZQXQtIibT+3Q4p69/nOo/hN+0r7UbS2tX/XoWIHRQr8NjZSWSmDv4Sc6PPzBCtfvKfDy8gyfuumN/NvDd1FvnqfYnSejQai3RNKZJ64/5B5wXynkMoqEvg4vGCWJN2k1j2K7C4i7zgrsxVvkQmWQ6HLXSw/wVS9/fr/RatA/zjLotn+cs18969Ue+EfuAQmfAbCTn+hAKzf41bYBoHOC6llorwSqhNvocIZ//r23wtYXiIJLVImcBuGisZzqWcLIYzIIWO7G/NlKl/oFCORP1oB4zLhJdmbCEA2LiDhtqBnTCCxzUsOBBqUKM9I5xk3i7io1EzKsXK3xFiO2xbz89Tz5NsGhTIbXVQo8ZhNCZU5w9CfSNPkOtElWnDe+oztZiDbSYD0bYQn58SRxpuFr/AwTQYm/iJocu9ihGhjiERgWz17GkN+XWoBqEWzVoLgrpWdRKqNcEyZGUj66rQBkOrW+ZSzr8eZ8iW+5tMr5QuL86eJlix/A1llLdC71h/bGcLlYayuWtU0ohYZmyfJUN+FfvSLPvpcGzNwL5oWGzS6OZ+5UA5a/aFFWrYXLMBkbLkve5gVQLJuKZK2B2OlLeWjHlkc2LH/YsSj5RFd0Kvk04DHx4JgwrYUXGcthLyAmIrQ9Jz9XrOFvFawi8tqgwsneKqXG09S3PkOrt4xnO1jboRxMkJF0lVP2tldj/G3ESl/mNGPSvMkJ37qcnVJQHPES6q9ps/9GmV3TRO7SqjmthUb8gbzRr0DBYJHAlAAd7Ne6/J90TF+754S8BKDM1BLUWnTc4BiVdVY9CeGEg9VdvP3Q85hb7jE9cge3TbyVOwv72XCg6SkKraeJ4waZeJN66ySnegvMRQ0+FCl/reGLFpbkP9dTZLHHq+RiZS1lY/jWwONazzBrlIHCsMcaqsbnb2zMI9bjgrV8NG7xqaTNMWN5KI45kSSOCWSPrywSJYrKnKJ0LMmmQ6lJ5yJe43G2OvO8NLeDndVv4DVDr+Tuwk6mfQ8bR3xLeRfzJxKOHr8ILeWO6wMgXfugT9Qv6hNhHoE+vY8CNwI86leV1a8WHSOVrsv+0gcviuwW+BFA1D0SuBpopATWdLy2CaipbtUlIKXfARhT3U5D2N+ufYP26UFROT07+mj50rHPqGdwTYN7rKKqU21TebVhW6oFtMesVKOwMjAFOzyvYNAUYGqioLbqGH2koZPTrPpJdQ1Msnq+dC3SzKmcFp1/8NH5pYkcPHTqH9WloU59JC2ntM+Z1C1jcyPiv/5Rix+4p0B1m+XcYo89xTKvm3keDB2glruJRrREr7dAvn2W4WgZ45dJuhcdhZBrhFfABGP44bTzo0s6552mTjJhrKLGX+HHf+TrGB0tubH82ZIV/d743/58RX3q/rdnurrzag/8E+0BvYCDzwDA6feZ68/FS6pzjBbLHH3iJA89HOFP7mQzsTxiI7aZvDMnyam+ZDxnIp3HcPOEcTlTZTEIIvCrhvh8Oohmpz3max3e95GjVEfz1LztWAI3YIXGI/Y81k2WyXiLvfIPa51kLH+Dy/UaRuJDC5goWH6zGTEblBgXXUS8Qc1GDBuPnV6RKS9gwmRYjVZdhMZitMZ10TrfXLieJ5IsSzZxvE9Px122mYjn5yuUswkZzeIVWefhEowH8lHbgmYMlfHUz67WgMoolHeDUn0p5dgXWzAea7DO4m1tslbucHulTBBaEvmSXYCxvRDVoLAXFi9BbU7gxuKHhgsx3Pdan90zhqQMuZxxuXRzOwzx9YZLBVh7yFKqCihArWs5smo5UTfUZcVWmqxIuW2hIIaMRNx6hqb1yArYxbDQTuWL/OW+AOwz4lQzrGI4Q0jGhJwEl43j8bjG6aRNPd6i3j5Dt3uefbnd7Mtfy/mwypgJWG08TS8zRrGwj7bZyeTwAm+5ezfFXM6B3DRJvMGs4oI9/uRil/u+KUd5WiAk5c9yWhMJQwnVlDEEM9Cc6L2UsJZskqx0mpi+pkfbB+BkINAVXax+UHkJ/cGx+j9YdEwEuRGfx1ePc/8n2+zffh37w73M5vbzpsJ+DAFHW0cJegvu1COO8qTNFZeKq0iDxJFcNzHsVMpV3/KhTpvQCxg1HtvlC2stR61xk44bQ50y5E8Ty7UikzU+u70sM16GUWMcTUwLy6dtwuOJcMEmm80TVJII25mn3b3sInA7xZv4hpGX83VDz6cZjjvl1WgUo4RRXzUmhViG/8bee0BpcpV33r9b9ea3c+6enEcjjUZZQiiBEiIbMFFgE4zDGmNjG0eM115/tnfBAQew2QNLMGATVmCSBYogJIHCaGY0Ud0TerpnOqe3+01Vdff8b1WJWWzWwseG8XemzrxT6d5bt557+z7/euK+ksfUoQM874YNFAIfW1B8twSQiR6isyRJAlFSUYr2srFzzgSJo4BAiq6nwFq01Lk2tSHvWYFqgTgBPgGptI4kWQl4cWUlpUtBncCT2hGIVJvqSzIeTpWajpsbuxTAaZ+MudoW0FMf1DcZMQqMalNf1AdJywTCUqAlp1HZCyr24TQYhXGJYzd/F5hKKpiqWdWeA5TJfHPATG0qZo/a1cOTewJ2enft9U76pccCfqKDrmkTDbTpfaV+9Qz/dF+NO9/f5E2Xlqjl4MiwZU2vxz8c3c/dK1l2dt3AxaUdZG2d8flHWaGVnkwWWztKI5iPRdJWaFEvr+cJScv7dQHPWEq5WRbnr+Ydv/IiBgekxpXkO+lH3Jsf6v/68/2hbpJI6GcUtf9feLLsmVIGK+rIi0XnqiMme247R4H/KAqkf4jpXs858/g/6rnpczTHewbLvOHVV/KVrz/AWHQJWa+DfHOeb3o5tmZaKETyU1V8JJ+vL0D/qOWlQ8ZJ6wSQ7ElLbtBweiGiNYDWJXHuFpbCKj4VFzTVz/Wz5El1t8j52W6CKMdE7QQTUcBtxmMptBzx4UrP0rFS4MZsyHAYuSTmFZOnQMiyVF3hPP0Kt0HgtDrT9TEyUZMlv5XTWHqMdTzptDhDZHgiV2eiHOA9DqvkKdpraM5K4mPwNkBnDcJ568BJwYNcF2R6YeUU5DIQNOHgAgzlYLOiyXcO8YsHTvCcQhvPXlOm0RaRUbaemsvyRLoe18oQLMO9E5a+5xl2XOIRjWodgkiMqQHZNhjqgKFew8rLob4Mi6fhyIzlYKGBV8sx0DROFVsPPfZEsMvCRs+4OFfHgxpD2RJDCTD6RgQPRpatXpOZyHDQCWwMc7bGarvEXgsjpsAlUZX1wTwzNsIr7WSsOcEFpR2Mmiw91QNMLe/jzW1bWNN6Hb+riMf1BYwYqZZCrd56ngRN0xZ/AZ74asTOF2fo3+EpM1xc1jFmhYNIVGs6l5mVwpakgE1t6ljMUnupAs9coHXsrovxJwYDOldbqqP7OhaAKMbqWcfghRVEaGqMh8tEXpXVJstuUP98AAAgAElEQVQlhfVcn19Ns7yDvdUjzDUmmKiPxPFsglnuK+3ivlwn67yAkDyro05COcAYw6RTq1q+FFi6CRkFLjdZjip+MIZdBOyzHoNehkVruSbjcdxGfNsFzF7hycYyNGfgmDwej1IvdUBpK+uyvRz3O2krX8iFfoHJqOmA/TiWA3i0ZGOt5mgt5MaWDXx07DC79xznmms3YwSERAeNi3i/QJVoKnp0JipBASH9OQqg6VjjJ7CiveqqrPaakyKoaCpw4ABQ0rbGW+0q8LCAjOqKvKnES2U1Dnq+pFcCkjpP60m9qvb1HG3qq9pI9+lxOq66rudoPghAqa5sCAVOpXpVO4n9mvvq7MJ9LNiZ+O/Y9a01cZiQHZ/eUSBXH3VqV6A3Zevq40JSRv2XiFvgTn0QvbTJ9FfvK/s5qYz1bNdmYvunfmtuBwYvAwrfePcXarxyQxsdkvaPR2wd8hnLLPJbkyNsbLuM1X4rV+e6mItga7nMTd4yf31yH2Qy9EYL1JvjhKZITfGLlAJM2giBPGWQaDOcPHaan3vbzZy/QwaO341zmvT4h77TEP7QtigK8Tz/X2SUDuxpfBPg5oCf7J0SxPvDYq4/NGKce9A5CnwfCtxyy8Vcu/oBDi6Ms7ljkEqwxKPA7da6DA5SM5Yalk1icDPwrabl8l5D6zoIFaN2wtI+ZbCnIrYPtvCO67v5k/ueYmBLwJzX4kJLaI28ItNJQIYFk3dgot02kXJWUftX+4YFY+kpe9y0WGSkWWXM8zmmtDpa4MIqvTaky4MjXpGsV6ASNZznpmfy3Au83SGGEN9GvCHbzkeDRQ7Wm9zQn8PM6Yveunh7DcV51Ro/iIsV15zDBa2dkW3bHLR2w1IN8g14QRl6PcOxpiXI5GCmyN8+MsJ6fwf9k0o5BQXZso3GdnuyKxzdbZmrwwW3Gq65QZ6hSkgee7VaxbcQU+iDUHZ0IZS2GErT0N4Ga9f7FO7KEz0Ga6+Uh6Vh756IozRZ9nKM+YbD4TL/O1pkvqF4Z5ZTwTJhLkNfSyeHlpe4y1S5iFZWvBzrogaXrnjsLGSpFDOcaPZTzA2xgqVZPcz5fpkpL8v4ygGmF7/Jszou5Y0DV/KkbKSWVij4bUSh7L4SQJDYhivN2tjJkPt7qtx+Vdm9k63F4EsSJCumLI8bMWHhNTFStSHgIMaqvcCC9rqeB5NP4pPpmpilmLI4hoQWSX3xOLdEi9Gqbd1X6Ampax3AEfPVxbzDL8qsns/k8Wmww+ZoKxS4O9rIy9qu5VkNuTVU+fjCfqrVu5meWM3xOaEUj/8xuBl6MmxvnKaS76KW7SGyy0yaEkVrOWoaTBif9fL8iOYpBoYHQjjqGT47Pg8z1RjZM8tFXXUWjMdPvW4tdx14Mc+e3cxkvoMPhEv01YeZDOaZM90uJZk0+cJGDY13I8/XJy1rc5ZslGG5dT1/9Y09nNe1no5W36mjHS30uppT2tR90UZODQJwkt4JEGkvsCK6ir461paMz9OSJ42NfmpDbQrsaC+JlsZJY6JxU32VUWxB0V9j5MBY0r7aVl1dExA6U6KleuqnU12qTxqzpJ6eJQmcA58CUQo5kkjh1FeFE1F9Sd1S+7ytYGYNVilT1IzsNAQGZe+n99XHhdrVbc1JvZ8ke+59kufqPUQX3RM9tdemMqqnnwCf+qByGiTtdV/qboVFyXp8+/EFRr/q8dbnZWjttMwe9+i9MOKTBw9DNMRWv8TNns/X6lN8OzjBPdtuYaiaZ5c3wv3Vh/m7sT3xwkNELtNCU0ImAToXoqdOdUUvspHX3/5sZ5dr7Y/G4zWhjtuJXD/QloKttNL3nqfXv3evMRCgq86Mcu93DnLNjTfRmpV1jwYoRrcak93fuBPbv42Lt64jqi9x/9e/xt7hU+y88gauu/J8F1oyHr3vfcK583MU+M9NAYVykG9e10CRX/0vz+XFv3EPs5espT3TxXy0QsUrOUlEEMCVHZ5Lb7VnBT61oI9oy4s2GmyPwZ+2bt2UpCrbbaXldMY5vslioqaLzzSYGyJPwIlgmmkZhhuPBa/EOL5TdT7csFyfMRS64YoqfKkpx7McGdPkmJJf2wYlk2dB3q3BkvPc7Ml0c2L5CXpsnSuMz77IIva914ZckcnRspzjS9PzbOroJxyP2BAYen0r+2RyStd1CvJ9BluG2oglW4b5ELLTUJNkbMgwNmOZXYm9Sq8uWN7d1seH56f5k+FTvHvLGqJSwNQEKA2s3n9L1tDMwAdrlr/b7JGZhlD2ZQqwKz6gPGdidOINYl4timFHLOXqBtOmDBAe3VnLdiuPXMOmPst1ecMqBQ2UCVFY5sUny9zUp9AslmMnCwQthnW9Gfaf6sRra+eiss/ivKHRKNG7yWdEDho16xws/msjYKk2Qptsy/Jr2F09wPLkN/jVzbfxyu7tzC9EHAwix4AONw2+F+GLqcvoXfH0ChDthy9/JuCGXy/QN+ARNRO1q9RhlTOkSGKqYt5i4mKIOk8ZbMoUda8Z8015GwoQuswTkoxo02sLTOifrqWARPXVprPZ/K4UMP5Q9ymbLMejBktRyEszPrsGIx6bbXPqrXaa7OrY5ARL13dvYba6QuEauK/9FM3oKGZ2H5/7gmVfNAwN6fWEEvQiQir6iZsr0qMGU+LfaSi3QqXIr700R7lFTjcFLlx3Gbes7cQPs9Q6szxydJ5dxT66eqp8eXKGExRoIUDS5SwZnmXgCMalL1PYlZKXozWUP0DIW9s28bejJ3jr5EmuOLSewmCIkZuuAJC6IdrqWF0V8NKEE91EL50rzpxUm3qNlBOrjsgsOmp81Y7K676uSTWp1xWgkdRP9/XquqfrCpsi7aDs/KRO1ZhoPNWu2tDXnMrpIWpLAEvX0+dKjStQpz6k4E99lWRN80TPEiDVXj+h2LR/6bEcLtosZhfInds+ZTEy0lVftBipbb2z+iyQqPkkOumnd9KmlGoqp/4IMKp/KqdnOs/epM+6rvfUO6kfCoGUAy9jmJ2M+KOveDx/W96p7k+OWQY6fT5+aJQ/G12h3LWT7fogtZavNKb4i7atbFwq0ciFXL1uK5XFtVzb9zzunt3N9OIR7q4+SbFxipozUPXoaDNMj0/xC7/6Di6+eHvS8R/9Lp1Kz6gnZwI4xe4qFItPq0Y1AE/nuktaU/m4jmw7qhzYu4fP/c3v8a6/+TJjyw1as9lYrWoMMxOjPPbNe7jlFT/Bn37maw7U3fWpP+WWn3w3v/OuX+Y5V/08d35nmJsv28iPKqjfMyLSuULnKPBvpUDKVLNw6fO2c8Nv3M29jSfYmrne2XF82ytyVa7oQq5kpPZT5icfbs7ByALsHbZsaTcU+3Dx0WqPWVpuNtitWbhPGosVjNdDaDwXQPNgVGUqWKA928eCX2aTn+GCjM/xwDJm4XzFWBu19OUybMnknOahbCMO2yZZBW6NGmyyIecZn8M2oA3ZlwzweP0kF1jrbM1K5Jg1De4Oazy7pZOPzy/wwmydteUclZqlf0myGciXYwnb0lOW3ABUs5BLbL9OTUNPu9OGUBLIaId5pbtaged2ZNnd3MifjZ/kqkY7m/w2jjUC2gxc2GkYH7d8fgj+vx2G3jAWUGQkOZCETsxWx+IfMoCTNEB72c4NJteayiELp2ehdTcMXGdp6zfsedgw4MdOeisVj3o5opDPUKjDrk1Z5mswO2/JBT4X2gwLgeWpJdhRssxEATPWMqlMGCZkSRKixjgm08njlUfh2H187ra38tyNW8guh4wtwSYTcVi2YVmP+flllioe3e24LBreimW6Yll/s2HndeKUMm9JmKf2KTMUk9SWAg4diylq3glsOACRSF7EKGXPpebESLVXO2Kcakf3VFHqPYGVNP6dGH/6TN1ykS+UIPQkBbvCqG3l933rnBumvYgmOTbmevmwjbjGNOiT00Mmw5aWDk4V4J23dpIPziMqwWt3BJzcO05+e81Ji1xO2loIizJwlL4+NuuxzY14ey1BLk8bney4ss1Jga3U/HpngRRlFTm+TG52gVVtnaxvKfBzS738+vIor8h287WwytV+q0JAskUp22zEP0QBz/Utg54EkZbNtsClHZv43LFDPOuyNfizhkhgRrTUgwRSBHoFwERj9/CE/iKf7oueAknpufrnwFYCqgSURPsUdDlAloxHCqzUro7FzTWHBarUnoC/zl04kwRACjRpbAWWNPcF8gT6VEHPUT+1F40EpvQ8ta1zbZLGpVK+VEqn9lVf5dUXvZM+JhQEXXZtqifnCbUlKaKeq2O9k/qn8o5WybPSd9VzpWYWEEwe4bxi1Z6e6SSDSWgVAVG1I8AnmtUMX7unwbbHM6zbmqWtB4495bFQbPLFCaWp2cjtXolWk+X9jSmebz2u6VhFuT1i+hR8u97EVLOsaenhld03sVy+hoPVY86EoKU6TK0+TLk4zjRdvPZVF5FXwOMfUbBhNy5n/PeMQV0K6OoLp/mbP3svjz51mmzXen7v936H6oF7+OSeFX77rS9NIlvGs1cq01RtGgVL/PV/ezt/9bmHuex5r3fzxo2N+1w2DO/7hgN0utbbplEM+dxPvps//PR9/PorrqN95L185lu7HaiTcWIyVc54lXOH5yjwn58C8d+LZej8Mi9557O5979/Df+S8ykan9PBCu2tZVrDiNGKpS0D41Fsj9ztw1efhEom4updHrmspdoFYU5ZIcStsy7eR+h3Us72uFATfqbANmsZl2gqCthY6mdjS5aOivKRxqmWFL7Aa8BLsx4fblTo90v0myz9UZ19Uc3F7+rP9XOgMUu7NcxEKy4YrIKdbLbWYacLbMiwyRBg2OCXGW4G3NSeo4lhajJu3/SDL6lBwTJ/EMqroNwDzeNQz0BT6bcK0FmAXEusbZFvxmIl4ga/nflihVfP7OcjvedxdU8rh6cjHq9aHhkOecubfK5c79N8wrowKZK+WRm5JRIHl1ZNYK4j8byVhETMRmYz83DBJsOBdZYemV4VoFqC43W4VDHyMobGIjTrhrl6xOSsbP6MNIVMZC3dGTgRWlZ3GayygPiGfxqJyPvi3R4Hmyuw9DDZbCdj9VFoPsidt76VbS1byEUBoQenGh6RvF2DKuvKrTz+RAt/f+8M7/yxNmStEu2Gb9wNu34752wDQ9Eq+VNw+SklvZD6VYxZDDsFAFpGxQF07UxOUBfPjNV7coZwTFmMVioygTkd66LAg9oUrbSJkQu/iJHrfg78EHau2gF8m0DpU7wyJ6KA1V6WlSnLwWbI+X6BVVGTu2oBP9uax1YtufbQgeiZNYahy2JwsS3vs+2mdTEgUN8lsNMYCqgITAqc6LlyBs5C+AC03QR2MHISIVOPXDiOaMLid3jYFoOfh0LTwzQs27NxSJ6nrOUir8DjLuCsz4VIMpzhEpNxjhl7ZbeXM8zUQvpKQ/zlvid53bPHuGr9WshGMS1T6ZskXKKR+ibiCESpvzqXSj2lpUBNCqp0T3UEmDVWaRnd18DqXJueoU3XtYkGAlaigdrQmKZ11Y6uCwilki/Z1zm7tZRuCUAXbQW6BJJ0rE0AUcBJjeia+LbC2uiS07TplrzTEwmtTtUf1dkUp/6yjwukWczWBCzqGXpPvYfmkPqsfqp91dU9LV06T+ene2BSJ/2g0L2EVjJL0Np18mCTz/9mldt2lnnWTsPISsR3liK86En2L2Z5Tl8vWT/HsfoylWCMn2vfQmXZctpzcQKYWDBcUrAcXQ54NNDU99lWPI/+cIHHm/MMFRY4fmSEn/3Fm9i1U9Bfc17E+NFv6d/+v9qTdGy/fscHGW+7nI987GO89uo13PPtJ+javIsXXqPcOtqSF7MhY8ePcuSpIxw9PgaZHv7ysw/xxF138MhXvyVd7NOlBRgvv/G1BLUlfl5xoWSPwYqzI9rRra882Hr9L/HpJ6fi+XyWEM917Nx/5yjw70mBZF3VYvaCV1/k9BVPLR2kzfP5DjW+WqlR1Je/5yJGsN7CfAT9Pmxtg/sOwBMPWmSXVt0Tp7es1+IVUcpdhWyQd0JTKteo4aKqC1zI1mlUuTSrTnPizGa03jfqsLthWQmyeMEMi7bBdYrx5ZcZ8Mvstk0eimp0ZzpY7bcwZPI0bMh1xmOTghXLXM0UmDMZTkVNcibkXUs1HputM7UMw6PyZBWvNy5grxj1wHWGjiGt04aOsoLOwok5WJDqFOhqkebKcDqA5YbhMi/kslw/BAHfqe0nGDIstsDnH7dc9HyPKzcpYbzFKG6fYuSNwYJAnMDOvItGEC9bAjNiJJKA6bpy6bbEYUpOKXG8D8vHYfQbFmGsWmjId8rmz9DaMOTK0LPRcPdJy7ixnGjEplRzy7B7wjo+uSS1ZgGmPTiK4Xi0QoeJmIqmYeYxvjj4FnaWt9C1OiCIDLVI+WUjlw0Ek2dSSXNpZXlhFGEkSYUeO9Lg6HNX2LgTItm4pUBADFWONbIh1JKqJTdlmjrXUi2C6p0Vv07H+jmvwTifvZFRv8rppywBKpu2JelcKnGR04WKijkrQKweqfIWbnl2mfO3hozV29iYKfGFKGI0tLSWLBVTZxuW1/gZB6wq7RH9awx+3dDTbXjkLsvsUpwNRJKQMB8SrITURwKaRwKC2ZBgWOA3JCQiqIawEDIhNfxAAKdDTW3spCU6AtGoVO6S6hg82W01G8wr52+n5eRSlveUV9FmG8679lbjM2WbfCCsu1y8Ckgjp4xjQcQXV0JO2ZBr/CKrShv58L37qI4FePLAFp0F3rSP2VxMN9FQ9NE1gTjRWn+amospQBOQSkG2xkrlBHBSkKM62pJxcmOh8vqlZVQvnQNubHVPDxdwSiR/OnZzIB3f5Lr6cybQEwiUVE/V1Zb6LqCV9k3j7kBVci19rvqjCZGm8FJO3MsVIRrsqTienpGNnf7mVFabo5UWtCT/awra1U91Vupgvbc+OFQ2DWMiyWgKipP5ePf985SGAvoKGex4RGbU51vRND93fD9rWzc521XNl4/NjvAL7UPs3Nbq5qPfMBwL4AHf8lAYuZBEMtFTeJ2qbIZroy78TXVFsVs8fuJ111DQOpIC2/hNfqT/OzI+kx7EKLTB4d1HeP7Nz3VVnvuqt/Lyay6gs6efi3Zsdog9BdLN2jS/+/bb2bplK7/wax+i5oCaVCjpDIyfmoaN0JzxczlnRxNb2uWco8uyi5wpL5aTXNJXcHPrmfT3XJlzFPjPSoE4J61l0wUl/uQ3bqR55BGK3gLUx6mGC87zciK0TsVY9qArMYtRqKvBPPzxuOWOYxYvb/FOw/ayElkH1BvzZLMdNP0ii7bOom1y0uTIRlVKzVMcjCynGnG+0FWenAJiwUyPnuHDC3IDFKIK6wxMRXWysp3KdNBvQ3aaHJ1+gSaWqgvfETOaJQV6NYZnRU3WkCUwOaIowz9WlyhFsFZ60iZUj1uCk9bhBiVRqJyC6ril0KfgvUqwbvEbcWYBqUqjiqUoL8tc7IxXx+P29p385VyVjx+eYXjF5+brDc/b7mPFPCrKAQqFrTE/KJWskyRo3TEbwPSAkWedwvnJTs2XLbSlfiDOGdvTZxgdgcVT0OnF2SCqoWVkxpLri7U+86chV4GdnbC+z1BvgeEZGIhgehTaM7AcWL5Yi9gTGjK2yfHafuanD9IY/g5fveEN3HzxZoo9AdG0YXEmFpi46BChHETyFOWZ4DjqMn4F5vZFfPrbAS9/dR4jOySBOq3qerHIOoDn7OHkuKc/CDFdx8yTvQBYyqTrsUewuy9bp4yiDiRtOUIl9cXEHYKL6e/snxIbRcd0GxqvRIqjep0Wr7uTxtI0LbZJi8kzDOxftHwtgKuyhp1Zw0s6sjyi3MVKeN8LNQ8uUQaMz1jYHzvLhCsGv8+QKRv8Ng9vxeAVPbzAw5OURpES8oZ1m3w6yoaGb/AEdpbAKEOeVO7qf4sAuiE3U0RzfXlZ3shxHOsX+UU+pmIGbjQZrjOyqWtymfG5Fo+NxmOX8Vw2kFUYbu/bxN9+c5YnojGH4u2Cwl0oMHUCQhwASSRfoocAiECfAz6JtFG2cKK1ymqM9NNxKgWV/Z0AUnrdga+kjMZP7Ultm46X9jpXOR2nbabSN3myqp76oBeV2lLl9MHYjG17FRPSqUsF5mT752z1krY0weSNrPKir6SO0swtSxqZSPMEvgQI3EdoDMrMKoNRaplFy+KjlnBBUS0E7mysopVDiaRu6rf6rD4pRIvadX1O2nRzWaDxu04d1lMQdo/jR+t89msNLh9sY+OGiPkFw8OjVSq1PVDeSquiCIQepxdO8OyhFTpLPcxPBXSXDDVrnbOVRFTP8Tyu9hV4x/AQloeCWfY1p+jP1ZkbG+Ntv3wTF12oCMzaNLBnxyZyPqPN/XGTY/WGbh54dI+rc+/fv5//+eWHWKksMXI8dmFLXy1b6OXPPvZVl0vzEx/8ZQpJug6rz0tFafL02TvBxz70YfYfHo/7EAQ0jkC1qc+WPM9/0wb+8DN3cvjQk/z573yaF1600c1zLfDntnMU+P8tBbROiUFl4bYflwFuN/ONxygRMlwf45Go6eygR+rQ2Wq4Wp6vbbj0VMoapfhqKzk4eMyyb8rygmtXs21TgelmFznnamHxbIBvQ7dgKYdpISNdbY05P8J6xpkC7ZYAwYceI5WiZUOh7AzJ/6KxQIcxrDM+z810ukDEw41TTDtbriKDXo6nAuskafIl+KRtKowyl/uecgswZJr8RdUwmreEOXh03PLUAajNQfsyVJ6yVE7K2Blmp6DNGDZuiJO6L4dw/HQcaHRXr6E9K6AEF7v223lOy2b+4OhDjNeWuOU8n2zextIr2feUIDoV0zUnkBNILZnMIjlJyCuxAHYMrLJR5A1Rv3G2fbmtsNhi6e6BkjJfeNDTCataDePHLaM16yRrQeKgcOAY1KehvxuCFlhXivnzocCyw1NwGct9y48x8dQddG9d4v63vZkLuzYRdQfOyHt+Dvo3xzljT4aWgzZ0AaEDT7q7VqypIOHePZ9Y4tI3N1jflSOSQ4OYqFZ1LcRijOlWjwUcbh1Pl0+VczZfCfNMsJqrJzWrjO3lnai20oU9BQm6nwKM9Hkqo2O1nwiAngYTXhs05ui2VVqNxyEb8p3QsJ4M02EsSAprPofnG3xjuuZy9h6uW6eGn6rAww9GzJ1MVK4DBoEDJ93qTMC4AMqk8kHFoMkvwOwKBIldl2zopOJ3zjAqd9TypX2WWrbTdbFd0kC5WBRjO80/7wiYbtZo832u95Suz3BPFDBhG+wLqy6wcQXDKRux1+ahYx2f/NYYtZkIf0bAMk33lQAnASP10XkhJyrYNByHpEuah6nUUx1J6axrAnYC0QI7orGTViW0VlnVFc0F0ATadE3PUt1kLJ6+L0ApsKe5rj6pLf10Xe3oWqC80Qlg0xgLlKnt1PZPx6mKWPNHfdJz1Y7uSVWbqnhFB0n71Ce1pfKqu1NGr5Kgxt76DrBJmqh+qO+p3Z3qacrrusCj+pgGXNaz1KbAp0gkvLcCd/zDPDdWWzi/mHEfnpOBzxONE3yhWeOnS9u5PdvNt8IVHl8+xVtK27go7/HErOVQ1TK84lJJc71vOBrCPfKyV8YcYKg57VLHFY0+SZq87lVXkM8ZZ0un4TlbNpHsGW2x9Axu/vGfZ/K+D/C2t7+d933lEC+96RpOPHYPH7zjgbidGP25z8VyknKptbXk5qMK+Nkcl138MrIZ8ZA53vDmNzF8Sp/Smn+GrhdeRVs+/jx55W/dwUVPfpRt2y9gwy/9Ma+56ar4GWcTBeMenfv/HAX+XSkQh/KxbDm/wB/92rXMHzhJW3aGmpUsLGJFcRsNjK9IPSWbH+PiMDUacIVsh+uwp2752mTA/HFLv3Sc4Qo2rBE1Z926OevlmYvqDGb7nTpVq/LeMHBZdrSG9hnYE1rkPFkqQ2srXGyLzuB9HR5zXokaIZ6S0ftlDkV1Sn4Jrxnx3E5L1GbYpLQKeBzHMuRHvNj3WJtpY5vNccfSHAtZKEj92mKZXYwzQITT0LYmERLMQFnhNZbjKBnNmkWCkKYHAkiLsrPrgHublklCivKCKJYxzWm8uolTWUm9tRQDZTGHaAUaqWRjMQF2dYgEGsQ0kpATzRpkpSpagUIII0ugsH9Ly5a1vqFaMUzNQLaq6AnWCawOVKC6BNUZy9wKrClAPYCFOlSWFO7EY5GQRyqPcbTyKYd+vvKWN3HN1vVwWOpCQ/UodG6JQWdewZY7DfOBx1S2i34F6+troTTmM/zFkCeWAm69riVWTYnJpdKclMmI34nxJczdqUd1LgasTeWSsk4IqGNJbrR3TDgpp53aTxwi3H2di5mLZrlYumkkOUnUt86zUQy84dFs1tiSyzPgFThiFQLb45ht4tll15Q4QDZjeWUpx5/uNy4u4tXbDYUCDK2Fb4/A7KrYS1pSWtf/TNJ15aKdtghbSeLq1J9RHIT64EHr1K9G8QtlZSDZg8jRZznW4VPyi9JqU83BrhbYkIHBTssrV+UYKvt8qxGwYgwXelmuNfDpcIkpa51Ub9jKDMCyIfJh9Ube94lhHh9XZgp5UohWAp4x4HB/RKK51J+SFop2AuA6Vtl0jATEXN1kL5ClTYBGTQnwCPCpLe3TMRLwUbo8/bml9c8Ebqqr8vppTMRiXdsuXkvsPJG2q+fpfvostemAWgIY9Wxp3gTWtAn0qW7ant5NQEsOGeqf6quofgJ7okvR0rbBkNkemyNYpW1RPaEn9U/PUHnNLUmBQzlG6ZlneNvqngOjAlaaf4bDD9b5xbuzDPaVae+KmMj5/PXYCT4VHuOC1isoeQX6TJNvLYzwh9s28vxSF5OnI65d47G2V45L8WuklgZFFNg6ojtcYTxcoie/wLHDT/Izv/Bcdl241r3+2WJLFw9GPBTp8f9zr45Lb9wxuIk/+dtPsLC4SHPlhxIAACAASURBVKmtg1zGI7z6+fz+1Ul1J/uPj1U+3RRsWNvWq27igYdudF+/dG2j0WjgK/CW1pZsgXd99j685Lx/44V8+K7H+Mtqg3JLi0vqrCbPNiK6zp/77xwF/j0pkNjW+Tn4sVdt5tf/uJOVlSmafpaRYIHHst3cZiy7GoZ2Dzpy0Cfj/AysCWNp3aEWw/1P+KxpQouL3rtCZONE1DVvS+LZl+GAl2PJKzEQHsOzkZPlydFBWjAx+mzRcLpqObpkOT+b5+qmYa0NudrLca+zTM5TxmPABi7UyZgNndBkpwsSC59oL/DexQofWpnjjfkuhgjpzJf56+mTrKvkuK7cyuJSRLVgiSqGwS4obTQEc5ZyAPVTsBxaGlWYDqASQm0J8nVoRLoHa3zDQ1GVSljld1su5ZMLwzx4YoKb1vRjqiFWqhupeqeSNUkfhgIiPcmgiYHJ5k6qnjzICaM5YVEGNTHJtV3wqZLl8GnobMKppkVh6U6rL6GlmbMU2qA6BUWpHhO+/uSKHhmwhM8FJsvJYIaPrjxFz/LdTp30xUt/nsvz61k+FVCTjeCSPnxjBjU9Z1mWxK8Ma/KGjnqOBS2AnSVmR0O+dHqSW/9bnvZ1OcJmFNvSiRlrqRXD0zumjFevHS/B8V7HAhgCHWK8ooWWYdVLNSEO4J0BAJM2XDmBN6lrpe5VnxKyOvu9WBPnwJKjbpSh4LeyO9fLuqjCs70u5qxHLlxkxGQpZ2GLpEDi6aHHm/0sD0/DeRsNK/pgycCt5xmGH7Y02yCj0D0DYPfEAXGdell9lwRK46V+K4h1m2H2RIS33WDXJ2BH9JEN1AnIL1boo8zCLPR2QMVJLQ0LK5Kw+rxlk+H2Q1VMkOcFGWWwyPJbmU4+0GzQY0MC4zFlPAaJ+NlCB+8PVnHHkWEu334Z/owFzS3ZJGpuKduEgJlDlJIq6ktJczIBLwIxaSaIFJSdGVIklbrpPXVf9JZDj66n4DoB1O4ZAlhSaUolL5Ck8ikA1LmOHXBKgJeOJVUT0BQd07mitnUs+0yVUT21pT7ovYSCdD9V1epYZQT0dF/H8paV/ZuO9f7qc2jiHLK9STzEqnWOREYx8FRGfVB9lZXd3kCi/tX8FMBUv/RsSduXLH6rYXbc8v6/n+FtHZ2xQ3TWcO9Yk3+oPEFXvkRrtoum5/Hmxad420ALz+4eoDITOK3G4VHr/v6UM3vAg+kojoOscEw9xuNU47TLIFGtKhdMmZ+8/QYKeS+xpVOHz54tHbpn1KMU2JlMho6uLgfoBNz8TAblyPzeTeXTX3pPgC0n3Yc2Y8hms3hC7vEFsrkcvmL9aO6o7WyOtrYU0CWxl+LC/+x/lVdGiu/96fq57RwF/vNRQH8Xlo0XFHj3bz6HxeMz9JoRjtRG2GwjFyX/IJa6JF0N6CpDuyRJQRx4/XzfsKkY8ejBIvXFCyA7SyQbVa9AJlxhuTHBoWiFWSuVa46OqMYaLIqgvxHYQfz3e08t4ugyrBXzzcANmS4ejmosEnGhkXqzSpfJ4dRRTSWjXeGzsxG1BakpLV7N42VenhmTIWc9tkvKR8SqTJH77BhVTw4UxuVpjeYtS1rg65bqKThy3BJmZSBvyLXIk9cwoowRVWitQV8Is8rXauCuIOJN2XZe1tXGq9tXc/P+RR6ZaxAtGKeC9ZJAp76kI8sK9mudpMQZatfAk12TANkMzhGivMvgtRvHi/J5w2CroTEPoW84JgZdhPU5Q58P++tNqoVYMqe4eHJnWFBStkB9zJEn5C8qh/lAZQ9bqruZri7yoYGf4Yadm7EdIWHJUNpkqMmub7Xs961z1ujaCqv75NUcYsI6B21EbzbkPY8Nc/iFk1xyjdTmkbMBdOBGzDZhnpJuOAasaaSfrosx61iB8R2AOwME6r6Wce21ZKa/tL4YrFuXxajjtt0z3cVEwpLwZFfVSUMjTK6Ar6S+1UXWs8J1nudiGcpO7WZiO8oTFZiuw0LDcoly6p6Gk0uWQifMKEh1zdIxDEuKBqzGxehbDVaquBWFEUlUq1IbSnInwVWf4Typ7Y9aJ+mNJmMnEtMvUGgwE6cpZprMytzAg9mmpT2KJazTCxGbezzet8OytznjgkMrxVi/NfyMlyOIqnykuchjYYNjNuS0AizvXM0/fWKM4xNKbOa7OefoKZontHO2b+qceJ7orHkkMKdrAls61xhKVKQ6upaCG7FFteVUlMl93VMZjZvu6ad2BXoE6FRH4FBjqGtqU3pK7fWTil119fx07qTt6ZraEiBTAQX0VftyItKmNgUEZY+ntlI1q4q7Yx2koq/Ets6BwaQ99UO2eEWLOR+MTH8V+mRC9ndnPjv52FA1zSkBRv0E0NVX7ZcMn/vHGe58PMOPteW5NGeZmPC4+/QByBbIFrfTbzKMNWa4tDLBmzpWsbBoOToTx1LWx6FCKPV6cdMi2z02ctLkO5pzPN6coN+fYnbsMG/75Vu4aNe6M4iQHJ4luxg9/QCdiaVkcfw5gaX43PlIPONWzgRZZx6rgTPP1bbO49+/LKFLyy8tLXH06FGOHTv2z366rvvf2/4z7vC5guco8COggAQtEoRlsvBzr1nHrWzgRLhEb/0wB2vH+aYLuQBjNVipKdyHYU0G+rO4lDeKrnCtjbixmOXlbeuhPkqrVmkbYZqzLGZaKWXaGdRKHSxglWrJRtwVKj1fnN90VAs3xtmtrc7BvLG0egWuI2KsuUBJeTkxrCKgyysSennyLh/nNHMZqVANx5sRnV4r52dauDNquJAnl2O5JNvF54Nuvtao0lIwnI5gyoOlMZg8DNOKH+sbcj2GgSIMtkJLBm5aa5xk5TTwSF2OFIYve5ZP9ZS4eU2GyAvY0dINfhvvu+cQ1arUfbHU0QjMSEvVaeIQKmL05fiemL7SOirkiVRyYqzRkpxBYsnb+Vnj+tdegAsKcHBSYVhiw37byNMYt1SEmX3o8yxXFXynXTvYXOZTtaN8s/Io6+w8R8YO8ObLrufWa7bhbwyxOUPLakOpBQYuMLSvNayMQO82KF1kuHsEBpuW9blW6anpVJJX5nndbXkXHDpU2i7HfOWtm6i3BK7ExMVEU1Ampqhy2nyDEejTecJ7nepUZZP4gK6e6mtL+LDmo1Oz6VxbPJ1iwHBGNH1n36T7kuBkDJHC6nhZPmM6OUSTFdtwAX2fMhkebER0hVJ9GjLKK7sCm1rhk49aNxeqFdg9Cvl1ECkRgGgspt9IPH3VJ0lypJpLHCHUT1O0BANQ0VjqvTT2As11zeOI0aU8dc8jKFhKrZbB0GN9p8caSc9yhqALLl3dwhv6WnnjgiXwPeSYJKnwy7xWXlPKMes1mUdxHQ0Xey08UWtwcPhorMKX6l7AWVxWfVTfFK1L12TDqX6JvqKlrulc7yVVqqRucjjQsX6SvLkyidpTE/qM93JgS7RO29L76nmpLZokXwLA6kcqdVPZdG4InGlTu9oEnrSpfLpPQaHeRz/V1X31TY4MTiKX3BPo0r3U5k7t6pfOSfVHYLaUAEJdV6ALYSXRSM+XVF0Sdj1L764u6ifgq30QBwqXo8yRqYCf+p8r3L65lQKWBT/L5xZOckdlH33FjazK9TMUrnCsPsKvbt5FaynPsbHQhUuSxF3ajpM2/jC8P7TcFzXIRDW65FEdzNLSHKdgJKXL8bpXX+OyR8gb+2y0BBO5/g3bP5fA/SAvlwJBPfjM4+93rjLfr/0U+E1NTdHR0UF/fz99fX3/117XdV8A8Huf9294+XNVzlHgh0aBWOpg6Tsvz0+851I41onJjBDURgiDOQ7Zhlv7Ti1bFirQ0mHoyBunPjik6A6RpDwwqHxe9TzZcB4bzmJtnWwUMB8qM3uWrN/KbFSj0xhOEvEFIg4ZWGtxKR2HMvBIFRQDrRVLZ7aNL0Q17rWWoWwn2/HJhEsExqdu8tzQnGODF/MdrdGVCG70PHZjOWWbyCZP8aRenevkj1Yi7lxUjk9DrQ4TWsDFk/LQ2QPhMozOWKfSrLXCmkHI+dDRiTNUvnNpkXevrvG87R7ZjjhI/gUtlvcWOvnYKY/PnpykNhyDGEnlHG8Q6JFdr9SsJyActQ7I+ZIYSG0WWOypmBGGJyzhsGV9L5xUtokqdBdgUIKLvLz2Y3CxrPRtPlSqhnsacF99mc9UjvJHS4+zr3Ga1X6G45PDvPGqbfz+Fdcy0Kp8r3HoEBmMzyjenfozGXv6amiWFiwtNegwPneRoS+quly7Ek9E+SSya2rnJoAl6aOAggCOpCda4XWeSnDEgMVcJV1JNzFftSE7KR0n0jZXX+owMdyEiT69FuuaAIjKKr6gVLmpGjZh+HI00X3l4g2iVnoL7cyHPtdtjLisTSKRJi1entUC/pKUroKeIRjqN1wiJ5i65ZFcRMeFhqlAUmhD1Yf5r1gaU2CGkrRYAgDKR5pKoOTEIcBhjTNLMOMQjILfB1aOFHvhwIEqHznRw0CuyIYW69KKSRJdKkGxG1qaEM5ZZg5YrrUlfnprgy9Xa04anfVgfYfh9s4St+VKfCa0zIUVWvV1sGEL77lrNzO9Vfyij03j+glcaRw0t9U3nQvYaNMfiGgpsKJ7KVfWNZWRPEJ7jWHiZepUjynQ0X1J4wTyUmAoI8V0rKVSlTRLY6ayeo6ONQ/EWPU8AT13LCCV3NeYq40UBAqo6R1UV+pydz2pp/nkQKVx2VhcX6UeTQMVa16p//pNJs9UyJIzNz1LYHeTfMNsbA87lwBH9VX00C+lj6S1mrLLlrsfW+Kicjcv6M2xusVw0lvivXMP0OF3MukVaA1m+N9LT7Kjq5v1hXb8akjPoNTnkI/ggMwolPDLePrLomKkvShRVYyjlSdpy89y/Mgp3vGOVzzt8Xq2YgmR6t+0hWH4IwdIKUgLlDdJ61Sh4PqUhknRfalidV1bs9l0auJUuucunvvvHAW+hwK+nArOpk2Lsg+3vnSI53x8A/eMV2jPnaC9uYavA2uyvbQbj/llS2M45uUXFuPURvNVo7BdeDIO6+yh0lzA8zsIm6fJKa9EPkPTeDQznXQbjx4i7lSubhuyzfPcOv2AjZ0vygrNoThdyO5OhuOG2XAZk2nj/mCBkfoY3X4LM16WKJimpnykIYw5BitNj2WV8fiLqMGVBGzAsM0rcImX5UO1ad5R7mOtggsvxrytaxWMn7AcOwXzwgDWUi4ZHj1kGV4UfzN8nhV+rK/Jy2wr4YwlU1HYE8NCxfL8Pp/T/mbeeN8RKgeL/NRNHeTWhtjIYBQ4uN1QO2bJ1KGwzrisBU7CJMbXY1yIE0n2xFAlkSvkLR1HY2DxSAMXUqVz0vJQ05LPx5qsqbphPyGnbZMPVU+4wHedmQ78cIGx5ZPOVfdX1r2GwZYyp48GdGz3nESt7kFjDJa7rBPMuDh9c7A0Ateuh8UQnjeR4bN2hZ5A0QJ2kYuSXF4ClY75xoxWYTscmBBTFu3P/DnpSQISNMdTBqkyAmeKaad31nX9BBTlDSn86KRoiTG+AIX+TByoE3BLYtjpusCkNtl0ZYwTOm3LdDJhp3mh7/P4FKyqNyHfzjqvwEkbcX0RFIe5ZRU0BbxPwW0dhk8PW7Ij8Ox+w9AGgUOY3Q1+t6VNQWy7jYuVZ1Lg2rRYObyshmja0rbK8PCIpZyDrmqiCZKOza/ziqzlwm6PTDNkVFgmY1x+4SU5vgiDZSBXgu398O7BDG+eHaN1eYiXtOVY1sdMGNEsGH6l4dNuc9wnaWT7Ou59Ypw7907wmh3rQR8Gq5Mx0Hik9JdDi2ziBHrEnvQTgElt7FRWW0pnSbTkZKBxEtfWfY2T3lt1pc4V8JNkzIUVUXy5pH2No8qprsZG5VNpl9rRT/dURhkfNK7qn67pP9XXNfUlbUN90PP1PjpO70t6pfLa1JfUtCqZS66P+mhSH7Sl76J2VUZtCiy2WIxiVkscr0VHAFD9U5mk/7YIXtFwz5cavPt9GX73/AKetBY5wwcnvw3BIh2dz+HGTA931Y/ycs/nxvpGjk4GLOWg2G5pC+VgZeSf5DTGJ6zlQzag1RqGbI1D9VOUjKUqWxD6eMPr0hyvZ6+ASCT9gTYBo/HxcZQm7GzZBNJkm6dNgO5MUKdrAqACd8PDwz9yIHq20OxcP74/BYrFIkNDQ0/Pqe9f8odzRx/QWne7NhV4xzvO5543rKL//L2MLJ8AZXOIatRMnjVNn6sysaF+5Fln69WoGaq1gGxhgF8a2MyfDo/Q09LOfH2CwBq8TAuzal3R8r0Wlxz9Jgx3B0tcZjpZsYYDNmQqMLzQ81DIzS9aeC2WXr+drwdzvCbKcmumjc+hfK/LLkzKXpPhA/Um11mPnZ5x6/HdCuCJx896BYaxLFrLAadIlCRmmodrGdZ7vbS1h4xMWLLyVLSwVIgzE+SbhslZ63KWT3pwf7PJde2G12/vZmleHq2Wcq9xnqpNxYHzLNf7GU73D/C2vSPc8qLtbO0tOgcIebZGxyxFebxJtSYwKDAhJiViK+TJZOxRWf8OeOuh2zNsCposDxr6rMfJ05aZosEPjDMzEk8+aev818acUwEN+lJFy45nkr7mCHZiP//rZ36SzS0dBCsBftlgxKxKhnrT0tqLy1Hqt8PCk1Aogq9QKG2G043Y67cnqrHgyeIxz74nlrlsF3hi1Oq3A1nWgTB3LKAnZumYcyxIcypZMUbdUz29q84TgYtivDl7OV1TRg0xZrUtekgKqIwBrYZIgXsF9nS9EselS6aRa8/RUc+VRlB2zraFSXsS5Xj98tg8+aw4sqFkJZ81DoeEDYvGuFiGRanE64YXdhjeO9Lkry7OUBk1mHboujgGffVWS/FiQ6gQJe1xNhB5XJqhxE5M4Wmq0KN0bhLDKMOFU23C1HHDZ2ZDfjq0FKxh33LTOeNUKznmG5YtV3p4GUtpSFJYSznK8/pd/bz2TsN5LYZlgd8cDOUtL2rx+Uql6OIOXuZleGRgI3fcc4Tbtq6nowpRXblzE9syAS2xTtG5PwEsUoumqlSBLdFbvzMBsiR06rue61jdGY4LKitaa59K5aS+FXeXw4aOJQETKJI0NQWHKQvXvNE8SOeC6qUSthRIpaoyPUOSUYFPSQeTyeXU3U7aZ2MHBgEv9UmSSr2HPGF1rSsZG7WvP24HBuVUobYSQKe5J3Ans4dBi5GH+pGYHmZ1LIF2+V2bMDUb8L/urvCbq1sQBsw3PB5YPM59C/vIt99AR6abejjPfLPCq9fsZIvvcbgS6luAU7KP7bHsn7DcVPAdpv5Ircmb3BeM4UPBDC31E3TmFxg9/BS/8863c8FOiVW/+94/HA7wgz3lBwZ1AnQCdu3t7Q4o/WCP+48pLcBWr9efltJ9L6jTfYlKBwcH/0WHjv+YXp1r9T8rBaanp92Hy7p1qTHsWfAmyTpy1a19vOBF83zpG1sZ7PwOs/VBZgrrWTRZFBziWdZjo1IYBZYjy9Dhx2v7Q8seJ+aEjmZoBouYbA9RVKMRzJNpzmOM8jCHzmHhYuMTSPpmDa/KGAat4RshPCbmjMePE+edVXqndWHOrd0ZL8uOTCtTIUxl+2nYGo2wyVNKKWZ950mrmKN1Gzmbu/WEHCfDE1GNi70spwobeKh+kqu8FsayefavWLqacFunoce3+E1Dz5ZYPbl70vLZZsBbOiNetbmE8a0DQ7nIuBhyy5VYSLF3Gmer9cJsJ5OblvmDLwzznlXn0TbvcXqPZe2tBn8LGEWdFRPReq2csK24gMVeh6E+Zl2WiUjqrCFDZSBk3z7LmkwBxeAT81vvw4NBk/ctV/lOUHECkqzxWLQZGsE03eE4Yyf28NqrXsKPrX0WfiMk6jSUTsX5bJ2x+pRlJmuJAsOAppvSi43gnESkajxV85j1qgTNOVYVN3DVZo+3fPxRXvjSnfT3FYny8n6VeixhlJKQKAyEC/UQz9/U7s4xe0nlHINOJG9Svwq0iXnruuqLKZvYCcHxMXc9drBw0hiBQtHNFbRxYF+BAHnvOkAY35eEdX0mwyukno4iNha6Ccp13uiVWapYJjzDgQAUvPnqKgx1Q6k7BrjZFXkXhzw+7XHxZp/qDBRLEPTG8QRdPMf+2ObR6c8kBRKIUW5g4cZWw9rz4cmvWi7uMWQH5BEb8lefiPjjCxSex5IrGUarIUq1J1DbpaxP+lOJ5GBjnddRYCJeNtjCRy6a4/cPzfFHxT6O16wLyt1jLIomeYGfYXeUIejr4B++/hhvvuI0t1wxIKlCTFMHchKRkACQ5o/isSk2Sr9An6SdqbdsoqIV6NFPnFo2aKqXcm1dF/3VjkB6anuntSKWccTSV6eSlVOQ5kci0ZO9ncrL6UF1Zc+WtpW2p+cIiOlcwFN7ATT1M8U1Opdto+aN3k990jzRT2XUX80JtSPpn/YCrs75Ipk6sudLPzJ0XR+x+niQZ7WmlwDkBkDATqFPZDGi4Jmh4TMfq/LRxzNcs9pjTdHnQGOZN03fC8WdnF/YyHlY/m5lP+8prWdbPk9rMWRVxrCwCGFBYYg8rs7FQcGHJaQ0IZ0m46IAyO2+jZOMHn6C2y46j5//Lxc4z3Th4hTjxn9ZZ9f/6fR4Rr2SxEsSOgE6bQJPZ8OWqlNje4//297v6cmmeZ7NnjXSl7OBbuf68M8poPnS29vL6Oiok/CeLapYLSKa5z19WV73kx186R/H6V57Gadm9jKS6aBLXuIYjkYefmBd6przlE7Lg+46fKsmpruVDYsPcDSYp1DsJxSQa04TySvAK2KMxY9WyGU6eBYZ/ipY4n5T4jKTwXcWb3BRxrBkLXXZ6xm4KdvJ30c11hMgE6+JKEuv38JEbZJroxUqmU4+EdZ5V0uBamg4ulLjy6bJNq/s+tsRBSxHhu1+maO5Xj4bzHPjUr+zERdjGlu2TsigwMb+aciEhi/Xa7x6Y4bXX1UgmrI0EzBWi5THE05OwtSkZaNCX4RwYgw6SkN8dGaW/s+d4F1XbKStXymlILOs8CmJhE72clrSlB5MoERCkXZD9jpoHrNOcrdjc56H9lpW5eK4d8NRncdswGeac6zB0OHlGbcBmWCRnJenEM4zWxmHgQy/ct0ltCx7zO8LsCXjcpWvLuEAaz6vNg2FdoOZtVRXrOOJ2UJstzd63NBVG+Gx3BA7/RZnY7etXeYkEjlK3yoOmjBmcUKpQ8VntcI75powXKlRsyZOzZRKfcTMXfUEWOhYbaisjlMmrbZUVkBHmR4U+03PkSRKdBNjTtuSxEhtCBRmlWlK8rhu2nMFVoUBhxZO01vcLqdFisa6MDrCCvKEzA8YghIs74fDVctQlGN3BS4vQUcDmstQy4GiTGTmLcUthlIrRHL+kBRJfVK4EAEXpX3dApP/CM1vWdjuEbbVqZbnWWmspSeCQFLtet6FVZmoW+dotLDH0jqIA4bZksGTFHEh4vkDLXzyn+Z4cNMClxY6GDARmSgWbgoLXuVZmmaAw5u28+d3PcXlLQN0DhmidSKHpKCxZDZ2EkhAktKWCXgp7IloltLbqU8TQJeCJNFZ46BzSdpUXucCcZJG6qd6KUgSDXRPC4hAoY5l/yYHBNUVMNffj9pQ27qmcUiPda5Nz1M5DZLKChDqXdQHlU2lvpJIqr5sF2XXKdu4gsGk6lzNDwE/AT21rX7qXB8Rak8SO4E53XMx7ZJzgXVJtWfBngbvKDyyXOUrD0U8v6/IFR0wW7f81vgTKGHyhvYdrPHLHKvs4apMH6uya5haDlwGjEOy8axZ2nzZThoXN1Ghib5CwAGFbQNO1E5Ac4zIye/mef07X0nvWlkTn71q12Sk3HCkx//q3gEkja/LCvGvFv+hFUhB3fd9oCZZsn0v8Dt3/l0QfI4W8QqW0iGdM2fLPk4fBi+/cYjff7HPvrFuNnRdTGblMLZxGiXRu8da9oYwrMwMMltSvDQbhye5orCWrPRPUQUjtZfJYLwSkV8m7+WYDCu8u7KXXwyWmLVNrrJNjoQNFA5sHsMchn4jphWvxVsxbCJimzEcVhqvCF7iFenN9rtcsrWwwiqToR2fzzcimpGlxcvQhaFulasTNvktZBQnz4Ys+B38IwojsMg1xiMXWY5HMT8JMHx70fK38yFv7S/wM1ty+F3GxavzF6CgoLOS4DTiPLjnbzf0+tC7GorrLOurHr/dt5H/8dg4X3tiCn8lEwcsVcwyH+ozUN9vXRYLx1AU4FZMbFIh0S3ZHQazEdavM6zeHHF4OeQr0TJvD+t8JIroxqPLKzEeNXiJn2eLDVgJl2mlSWN6jD+86QouWjsEW0LK5xk3LsorG0hqUbfkcob5qlSpFm+HoVyInUZWppRizGN/sERkF3hH207CXJarBsv0r5SpLnkxk5VERAxSDFYMUpI4SUWSzYqZJ8DASeRS1ZvuS4IjpixaSGLnmK7UdIl0J5W+pIxYfRboXNLCmjBxYQHd16bnSG0r7u5L9W4JMg3ae3oIbI5R5RluBFzZCgPZgOutYVDdj+DuKTgxFatXvQFYqcBzCoYLA8OTBy35fsP8IszvhvLlhqgXlpSQdVbevolqUgEW5YAiG7rA0hrCtXKAkRAsD1/f69FcbuX8fkPfxYZ6NuJ+v8Gp2dg5R8mPWtpAUUqU69dl1WgqbZmlbSDDH7x1Ff/dX+QUTfbXjcNK68q49H1yWC0TMljeyuNzs8y3zUNViX4TemocJB0TiBOJUkCWgiftU34lumtMdU2AR2X1RyM6i8ZqQ+OmcqojAKc5q/O0bArOFGIkLaf21A/VSY/dPEzmj66p3pmb5pKe6SSFkvYlEl09TzIeBT5WPc0r/XRNKLdFqeQSyZ4eqTLakrno+q9j9S8tp/dLowC7+ZyofCVJHQRvq2FiKeI3fnWCg16GF+V8Zhoen5x+gt2VB1ndejmtmU4+r5hywRx/umo9O/KhC1kyUbN0y9ZTtsFykIgs+hg8YSxZY7jItaoszgAAIABJREFUy7EQzDO2vIdOO8HpY0d59i3bufV5V7g6T49NfHZW/v+9Q/f/7GQKns4W6UXaWTHhZ7KlzPrc/hyQ+5fmgOZQej2d689kXv3Qymgdlbqo3fCK39qghMgYv59t+U7mqkc4FtWc3degZ1mbiz1frVQNniIFWI5bw2G/n+7mtIuzFce8EFMOXZYKz+QYDKZZY3LssYYRr0gpXCIb1V0+11aM8xLbaEEZDx+2oQN6f9BWokmdk2GTfgVjVSA4v4WvE3/1XuZnCGR3FljW+z4DJse4sgpEDQ4nqcoUb+9hLFf4bexpznLc1sjh8SiWRyycjiI+GDRY217nls2GaM5SfdASKdu2GFQ1lq6tjMYZKiSoWZqC7BJsDgzPa7PcVmzjJWu28vLP7+fJjiXKmz2aIwp6Cop3rkghilHnPAQ7jVO7mnWGYBhMBZZmLSMPWhZPV3hnNMOj1uNKhXfxi9TCZcZtjV2ZTo5YmImWydsa84vLcFE3P37ec1zC+mgRckOxiq2nDvlR50vhPAbLDZdJi8WZONByR8nQ3Rvne39/bZQX5ga5qdTFt2o17g8M9+8p8tV/krgqkciJaWsplNTOMeBkZkrippVedFIZbWL6kqQINEiyJdCgNGPJbcfUdSIGqLbE1FVHWwoGBDAUT01t5hUfLnm+6glsqJ68iGWzX4ILi1kerTWYjOCb2XbGW1YIlHc1gvUenNcWN9XWClWp9KrQn4ehFhjMwf1HYGLB0nuBobNdY/t/2HsPMEuu8lz3XVU7h87d0z3d05NnpJnRKEsjyaAICIHIJph0fC+2zAUbjPHxcQTjc/GxweFgg41tTDgmJyNysJAQkphRGkkzkib3TE/3dE67d66qdZ9v1d5Cxhwbn+vHIHvqeXbX3lWrVq36V3X9X/3h+y3FXmjcbakcslhZohwgaI8zzqD0BP53GE49YB033yPzJXbMF7ii1yczbzl+2lCoJRjMwznb4sSL5GbIFqFji+6NGCQnRgymGHHRSIo7Lu7gzZOHuDsKOBEYV1dYeQ8nMGyzERdkcpw5PMinzhzBXqJMXYPVXB+XvCwsthJSJCu9QAhUuTlozVN7rgSOdE2SveQpIKdFMm9rb1ne1F5WyjYI1Lo9PzqXQJjaq6+2vtQ86ji1a98Xmty2e1XtdV6103dZdnUO1aHVNrn7BcYE4Jda4F7XpqQOxc7JrarzqX+NXafWb41N/emcekNsW/fUpj0muWHbGdrarrYaqyyRGcP3ynVmpju4OeOzyff57sIMfzR9J/0dF9GTGeVosMjVzWne2HkJ9WqSILKcLkEtZ+juiXkrVyzsbUr8lsdsM8bNUY0zjRnSNClklZCU4O1veQU93dmnDHtG+980lvi/8FdgTkHk5XKZfD7v3FP/wiH/LrvblsO2Iv7BdXsQypIVRUR7aVs+2r/Prv9zSsA5ulycRHxvzMzMuPv8J+3lRbMjGhA9+c65qJvf/+0Cv/5bU+y57FyetnyEI7WjfMM7l1EvxbkpmGzC4QUrIwZFE3F+ootndOzhm9N/TS5cJvAyiFPDSBubBInUemaooWqY99omM1GTZNTgnsYsW1J9ZI0qGqjodVycQSwSp/G4LrC8IJvh90qLeIpDkjUwNUg+KLEvatBvVBTbcsDCeXh0eClWowZfDkq8OtHNsrH8sm84GkR8Ho+cX+BXGvO8K72Wjfi8SYTG+HzivCwvTPj4DUt5BZLnQKYSU2bIJde/NdYvywJzShipQHMRhtPQlIWmFrCHHma29vH0U8c49ND5bBGz/fqY98xsbrmAFMAuXSMF0gPevOHwXZaH90eE1qcjn2KbncWLknT4KepByenMjmDZzc+jjVn8qMoI85yce4C/+OkXsHFDD5HSkCehGVgaYsKXe7wrduMmRMXSD80DsbUskRY9iiXX7XOgorIUDS4eXc+gZ3lWKcOxWpnt5Dm6tJ+otC3W0/GtEStGuUgVjK9F36VY9VMKUmuRB8ua1wZf2q42+igmLmti16o0hBTxk2Ls3MkE9tSPFG27PyleHa9F/WmRwvYF4COeua3OF/efYqasIKkcmYrhp7ekueOAahBDNYKdcuuVoHMt1LPQsyCKHlj0Yc285a5DlhdtNg4MivIlkzKs2WBZ3Q/JEUhoDLomgVBVENEcrsZYKHGu4dRCxPGjq1w6kmVFrryarDQBTyv4Drzl++OYr9UjqhtrnMVOMV1G7naVHOswrM5H7Jrr5tcGl3nf0hH6Os+RV5dr0/CIQEIEQ8bw4nOG+W9fepgXXNFge2+KqNs6N24YZwfBNou3xmDkdnWyaslT8m67UbXdxdK1XJ6SqyxyWgSMJG/5sFtdPAHcNC8CitoumbQtfW0Qp7X6EkgTkJLLVmsHslrAqw34ZTXTfrlstdacatH+dikwjVnj0keT2b6X1E79ak601qJzqAsXMtDKkNU1CAy2x6rxubbfB7IKpVCM5OkHqtz25kX+7NoBRiKf7y5XeNvSnQwk04TpjSxFdSqlB7m5eCH9fpG5WpPOlEdn3rpqL7c3QhIJj0oEF+rlD0M3KY7YGlP1CRbrJ+n3phh//BhvuuU57LlcrMha2kJu/fwJXf2rQJ2uQVmBSpZYWlpyyPUn4boE6lKplLOytAGdxqXv+sj6ojanT5/+iYkD/EmQ29kx/GMJ6D7R/SIKnOFh2aJ+Ahe9ECs+KQGvevlWPvHuh9hbLnFDYTellQcYrh3naHYn9zRDzvfgcVWVss47xWYPTnmx38cPSy6OyfiqeJ5yVSaMlyQMSnSFVdYlihRJsGiSBGGJqo34KA2eo0xBk0S6b73xOBJFjDcNzx1MMB8meetqiW3JTgaTa5hqTpO1DY7biHtJ8BIvyUklY+AzZiMu9HMu/qwYRpyxCQZMkgoNHlO8XXOJ+6tH2ZQecTXBfqW3i555jxOpyMXBh0sGWbpyiq1ahfoKVKqQ6zBkUpbsKOS6DNUzlqU5SNUNV/pxWbGwcxv3HD7GlzMn+MXnb3CxcgJ0KhuLKhbIQz1vmTkIY8IkB2J6DJUiahYsw1GS9QzwYNAksCvU/AJRVGc+mONMWCUbLJC0IQtNdZbiujVbkStRZVu99Qaj8cxbPPGsOWUZx26tdEB+GhL5OC6wY8Dj/pmA3xpb4s8vGOKy0TzN05adeZ8T1VU2MIrvL2BkSdKz7km63d25zlrTIneVgpeildKUbpIydkFsrW0CgHK9lvW8bClgtW9bVOSK1dLeqe1ql41fjZ3rVSBBfWq7A4JxG5uwZHqa9GxKs/jwNPeFOXpzfRyv+Vy8GLocAWHMSh2O1uHwIctzTWxhLq6HbBJSFfipjYZPPRzxcDpi1/mGkrjtapaUYifHLfVZS/Lc2AqjqgSqc+vAaR0KA3DfZp/5E5NMzBlevj7PUF+A32eY3uexUzIU5j4ZH5e1hrTmRuBVtBo9Bn+twS9ZEucbskXLm0c3cPreQ9y1vMDTcwMUc02XJa7KKpUwZEfnGj67N8OBfeNsv3EzRqClaPB2Sy6W6v2QPmPxu8Aqe7erNYECYFqknSVnyVLb3NzJSiYBt+bIoeEWiG2DPM2bpkvASK5QfRfY0vy09+neaG9X//qt/rXWcW1AqLi2NkpoAW4HmNWPhiGLmhJTZJVzt0jrvtL9JXepYipbw3VD1vujsza6xvE5dW/ppUGbdA3yf+vHisGq2oTiWz3Qo6syA+/9VEA43EVnw1BJe/x9eT80FpjPncsGeQFW9/LLmQHyyQFmo6aLBj7atKzz4P4FywOhZUfCkjXKug74duRRiurMNKZdElmmeZKlSP+7a3jzW25wL15tHBEL/if7b3u6fuRRKtlAWYFtbjhd7I970VgENGu12hNcde0xaXzarqSOjRs3ns1+bQvm7PqfSECgTssPK3n3Txr/GDdomLIujmxO8473DfD8n3mA9GVX8raOS/he+VE+Vj1MKbsJ5aWqwkR/Us9qw6QJeXVhPZ/LX8wDwSJZL+/0RRxNL6WexzQP44sHgqLjrFNSXiLRwaK4m7wOvhpVKIdlLvGLRFYUJwETQZKHp2AwyjBAiQqiSVH0+CQTmS3MGMMleiOOLL1ekl6NPVih7mW4VXWeoiofTfSy07NsUGktjTUs8YnqYdZHNT7YvZ3RMMlDUxGfDwyvGYQNfbBwFBo5aC5BMgfH9sL8UMS2KpQPQHFAhTIsUY+h0APLj8BSCoZDj99YO8pbZg6SfyDNqwtrya6PqC8ZpsYt/iLk5uHYfssj0xD0wgVb4cKioTJuOVVPkgkMc8Yy4qXIKlHE+JyJauQcObCh204xeXqStz3/WjZm+ommQjxrUOFyI0Pm+jiLVNUTZMVQ8Hc4Hnu4xLVXk9KMLPeIjbk6x81P20YmDVOrcOwYPB5ZrtndyeduLfEzz15k57ndROIOlaZtK1GnLKWsY2vZE9ulvGU50bmlRPUIV0ZEi8xVN4UrMaZ7vA0CWnrWWf30XX1on2Lr2t8FGtSfPmqj36I9qUesjFeojHRT6BqCyRJ7CiN8bWWRG2tFLs8nOS6lq5rFBRgvw6lJmPct20VxomzkJegN4YpLDB88Zvm9Szy8OesIhU0/JHuh+QA0t1kSXbHr3Lkq5f7NGxLWsI6Qv91XZaizQLZumVbA/AqcaBr61xgicZydAFU0Sw47Y7Pj5mtfkwiiJTdHsdFvGWwYfnPdeq6ZO84HG0nekOrkWZmIYzXL/iggDNKwqZ8Pf/0Qz3rOegrVhMum1TuUrKi5qxQ72MoInYHoPou5IK5w4v4x9Y+utpKn3LAKDZCPV7Fqkn0bOAskSZO3wWB7PjQHAnP6d9ZagE1zoraac33Xx81/qz/1qd86p84tICeAJ2uhqquI9FmWRZ1T4E/7ZQluyG0tANdy1+sFQYBP49R4dJ720m7TRh/ar3MKrApPyIUrq2A2roHsYjt1rhD2frPBxAfT3HKTz5ZBn48/forPl06wu3MPhdxO7q48xsv9ATamL3CWf5XwE0YUZl9pxRhvwechG7LFeC6+t0OeiXCVhXCJRPUwQ4Upxg9X+ZsPvIL12xSsq+uSMJ8aS1us/+rR/rgVXxs5ay03WXd3N3PzYils3SDtK2rNRU93D+l0+omYqfbus+uzEnhKSqBllnn2c9dzy2vP8P4PP8y6y65gILuT/6f6GO+rHiWZ2sDL8xmMF5GtWdYlLQ3TzbrkAA9UHyeZ6CSI8phUJ1b1saIKNqqQaM7RlVpLibrLDBu2dacvao0pqlGZUUSwGzEVVugOK3w4vYaPRTmuUhKEl2OidoodXor9YYXBcJ5LM+t50CZYjar8cdjgdxIdLg7tYzZgJNHFSuiTDxYYTw3R76XxrWVFGivVy2a/nwuHfTILkEt75Gsx/16PMiq74nqf/jbDalm0GXGhdnGcKZj+ZNki40epoWs3dGYMpcCSMiojlOGCcBO3fPAEmdcWuOGeThYeDTi8CNkZUEUuuV4vPsc4HaokkLtOW5ohXLAJXrWc4MvjHpGXpho1qUTLhOEqJtFD0i4yXxXKLXJT7x6SfRC13F/WGEeLsHLcYtZBr6w0i7H7t9kNWcXH1aGjA8ZKht96fJ437e6htyvD8oGIxBj8dK7IO0sehQI8fo/H/NwU+N2x0m5x7bkwEylcKUspVn2UEKEYOClYKW8t2i7lLXef2solK8XaVu7aLy2h56i752IePKfj2opaAEAZtYrJUzsp4DaAkMLGxzaKLo7MD5sMG8uFnocJcxwb8XnRCJQOQEe3YtMs3WkckXZfZEiplLCIhGXhKcLTRj1mqhFjd1h27YKGZ2jOuuRtxEFfvtfSdU0r81VgRmNRbJb1yEZVPn6ixG/vGGWoOyLTMNz9gEXYS9VLhFsUo6cMW3/Z0lD8owrObzV673CuWs2NuPLEO9dMRKy3Wd69ZS2veehRrspfwE2ZHMcbIZe6gAPDlmwfX5x7iIOnlrl8XW+c3SmEIYoQJXOovJksirIS51vUHpKn7uWH43tE7z1KqnHATHMquWtOBH5k4ZKs23LXWnPn5kuybwEwXZzmWdsF0jQvOq79XXOvfXJb614QaJP8dJzaymIpI6y4W3pb95COcRUuNJ7W3Gubxq8sUR3n+O5atDk6Xts0frXTWveXrHqaJ43bWfdkEG65YpvG3Xaebzj4UMg7PlrjrZfkuGzQ44HaDD93fC9kB4jS66iJvTuoc8vGC+jwDN3ViFtXGiRNknMxfDWyXGY85olYthGTGKatTy1cZqF+imJjhpQ9zvj0aS7ncp533RVOJE/mUNaV/aQv/8eg7odd2L+31e7J5+vo6CCdSlGvN2gGTedulXUumUiSTqdIt6pKaNxPPu6HXcfZbf95JPBUegN78qxIeeo+ThYNP3vLLt7/4Q/zF7U+XuBt5nWd2xlhnN9YPEy32cw1XpIEHoOeodAV8IWFAVd/tR41sME0UTOD54t3soGX6GQsrFCrT7IcVSn4Be5Wna6oRp9JUQ3medAvkqqeAD/PUqKDWlRnk+dzJGpywstgowAjkKgqFcE8u7xzuTesMq7sMpN19bo3+0WuC1e5zTbpMgmWMRTCCvcbSykKuSm1gf5kJ9/A8GeTVW7pyZNtWK5NG8YrltlmDJSmrXVWNOkKswzbB40L2cnL2qT/9RCyNRzNS1PlBrHOZXzERlT9NMM7krz26/v4UO/l7O4rkOuIXK1X4Z/+DqNLxJRg4pQlSMGFnSLMjdiSyvH0cJzbgyq9yX4Wm9NOeVgiTGOSenOOzcn19Kp0VCuL1CqTUqS9yggVb4KU6hTufGqWakBNYsvA6oLhzytLUIh4xdP7yPcZGn1Q2AFbJxM8tlBjcyhN2EUllInt+wrTle2SsqzHSQpOGUsYCoaUwhVwk79Th7tauEqcUgP3cIyVqxSuhKrNsvS1LT6im1Af2q7jW+BBrk6H6NSvYqbaHhz1gUe6mMGbDFidq+Hn+7gjCljnpzhZD0klPYa3GJbHLckVyG00/H21wYVN2LE1jT8IJXHuqa/Dlv5VuPVYyOC2BB26xjIkA8gIzB+DySBi6FmtxAABE40xCXc5C7TK2nmOkFgkx/NpeIao1JYglYLidkO0JOAPEmtpAbx7LEVFY4h6JGGwpy1myJAYNixNBzyz2M0H+3v52W8eIL3uQjZnfX5vJWCtjdie9jh6cJRP//0ZLv6F3jgcTbIUgJPL2wHollVKRI6uwoQFhZHqnAI69Zg30Wi/MntdWTCJtQUO5SbWnOh+0qc1r25u56yrZewAvaaivU/zJ7lIPs461pKXq3Jh4/MI2Ok+EJBsg0e5WXWczqd7TNY4N8etsWq79msc7h7Rn9Z3rdVWbdSvFiEQvfDot65VLzXarnE6HKmMdEOwGPLBDyzx0ok8l16d4GDUZM++fe4CLs3sohAFfNvO86X153Jht0cyFbGt7rErmeRUBPetCq82+PvQ8jg+VxNxwNa5JyhB9SiZxiTVcI5MSlxNHfzp7TfQv8Fzz1hP1/4UWvy3v/3tb/+3Gq8U5I/jo/HrvLLYJZIJx0WnuChZ5tKZNMlU6okqEz+O8Z0954/nvvhR5P5vde//ePqRXKG/32d29ST33/pFal1lvlOtkfEH2GkN/9A4zZ1hiS/6eao2yZFGyIPkyazeR01v08meVh0okQ9XMMECoZ+nll5Hl5ehw89R9DKMGhUz9wmSPVQTnWQSBTyToGIDFCK0I9HNlcbnKi/FSQMnqo+7OLNTJktXsp8rvBQZk+SQbbrastd6vnMViu6iLvqSRJFK7QSrtQn+a7afl3euIWp6rPHgrnKF1VSNTX6W8op1BoexEhxviHbMkG5CpiumxJCrZW7R0psxLNct++WKFaG+Bw/WDV8Im+wL6+wj5Fi4TI/XSac3zYeDCW7asI49fUnWD1rSZcPMvGWuBCMJgUHDplFDsgnTx+CxFbg1KrHYnCPtpUkn8gRh1VHL9ESnKc2c5n0veBFXXjNK41SIt9bgyH+lHyahXlGCsHEku84FKx15Bmd1OlCO+C4h7320yhufX+XV16/BV6qeqmT0KDPU464jdVIpn2IiZO/JcZ73zC3kO724pJdAmMCVlrZ1ROcVhhD4c9aRFrgTSJO1Rs1bhzyhcAU+tE3tFVslk4WUslPIJnZNyuokpewZRx/ishsVm9pW0AXD8t1w4ruWraNJPj9Z5kDT5/pUD7mE5R/G4doBQ+9aw4nvQKrDOBqTh1YNxvfYgnEGKSVUe0uqXgFDXYZkh+GBRy1bdhgHAoXCpJ9VLu3YYUvXOkNaVp+CwdMY6oY//WaZzYsFXt5bYCALZwwcXIKsb8gJXKj6hEC8vMoNyO82JAYNjWVICOArgUbuVwX1q25w2ZJYYyiULL0LvXxm5iRlW+JF3UOUbIP3NgybTJL5gSYnvjXHc2/sp68zTRRYTCGOOXMUIU6+LRk7gBSDR9NjHDmxGVCJNoGrVpyagJrmpW6wylZSprBAlOZRc6PvrRhBza2VXHQ+7XMgruV6Vwk40dLMtch8dZwAol44tNabjdbarnHVwCrHRVZTCUn9CaBJdppvfde4ZF1Tsk17m/brRULufn3Xoj7bi/pSMobrT+BSN2qcFa4XCMUQf+Qry0x/xfDS8wukc5b3Pno/3zt5mLW9OxnMbGJudT9vzW7hpqFuZst6UVByFnQbz9ENeaHokzyqgYes5R1RiaAxS9nWMVGdRlhiMFti6sQEv/Dma/jZ1z6NhBtr/IxtD/WpsNY0POWXtvKWZU5uYQfmUmm31m9tb7d5yl/s2Qs4K4GWBNrWunQmwTveejPPuWoLY8eXOeBHfLC6yEJ6hGRmM8ejKtMNBajXOVEznO/nGOy8miiYdZUkjAvyCRxvnaKYg/JDpAXikt00jc9mL8G2RDcHvCw9fpFNYYlVG5H083SaJItRmU4bMCkyY2O4KjXA9vxOInHfRWVmw2VXFiyKqgxHNQ6FFT4T1ZnC0GMDPBu48WUSHVyS30E10c2MDdhpIl6Ax/MyRd4zE/GdRRWz95wLdEcB1mVgrG5ZDawrs6WEiCPjls+shHxrMeJAHeZknArh0aqlYiPOMylG/RwDxmcnhk6TYrBwOUR5vrX6EPefanD/ROwiLXiGDZ5R3Xmqis2djt2izV6RLMNNqT484yFCGGk9xTlGjdNkpVDJMbw9g7iBPZGvLsnVFidKhEVIzEBGIGFECjUOAg9UJ3ceDjUivl2xPLNoeNngCBl5xqfBiL9tWfVJI671O1gIK9yQGuB0bYqkzIkCc9L27cxU/Ww94VVZQqN0iEXKUx+5YmW7lFVECqytcLWWEpcSlmLW5aiNO17JRPoiC5nqjLaAnqpRtJV/y/Wmw7Rt+ahl84OGx6ahNxBqMmRsyPXGcHnBY8WDftFMPM0wV7GkjWVdZGiUDA+vtGLxZRkbMDQSUFgj+hODfxROn44tarqWyoSqBMC6CwyVvZZAoE4WscDw2IPLfPSTMLBhgMlaxKksHGlaLqzD+oTqBeNqyEayTqnyRB6CGSjfbV09WNUVbh4Bc6oFcgXKewzpQlwzeHiDz8cu2MEXFsf5zsoCTzcZdtk6iShkNDXMGHmOnDjp5O7mRC5OxcjJDSt5ygImGWteZC0T0JMAxQsnJF5s8Q7mLEYF70WLI9oQZSz1ttymctFq/OpP85KyzgrstrVdr1lRjrTaiUNOiTtb41qrMaBr1YzVedtcgzpWgFEf3U/6aGz6OBD6pHtHcy7Qp7XuI12PrlX3h37rntJ90u5D7dRY59DYxVUnt3IL9JmU4fgdAX/1lgyvHOlkpNvjkycf40/HH2G4fxNRops7Koe4KreVywv9JIKA7IghkcMRpJd7YHIZiklYrHvc4Cd5s5+gyy+ST49SSK3Fy2ygK5VgYmyam9av43d/6VmkVWJOdWzdhWiQT53l39RS9+O87DZo01ogrv1p/9b67HJWAv/RJKD7Wm7YQmcKUxzgs5/Yy5bh9dTNYAyw/DyDyR6iYInHgyVGkwXKXop+azm1/A/4vvgako6E2HhpQi9NRzBPd/FS0ibBmXCVyCRpmBRLXtJlq1rlkxmftJ+l38twwiQZ9lWRIsH9NmJARMV+lon6ONSOsynRzabUGuaNTwOfnEkwaSMeDxZ4rPoYpcYkL/I7uSy7iTohB+ozfLYRss7POB48Yz0GE2neUzlN0Vg2JQrMBdZx5k0aw76a5ZuLkXMpi0KraQ1zkQiOY4uCvI6nsDxoLctY9/Y+anyG/TxTJkEZn+5Eiq8eeoDv1Zpcnh5G5UOTPngd0EjCmXk4HUBQMvgNqIa4IvDHgwXKUY2wOYttnCITrbIU1EgE3bzpWVcyIJ9g0mJqJs5FEMjrNUwpMSIHxbzcfeBFUE/A12YCTpqIZDlgNb/KG17YQSbhYRV3JZASQvq0wa8H7KvPM5Lr4cvTk7zw6hGGe/NxWTCH1WJrRwzGYiuhnoCuwkRb2UqxtrxpTrO2n5HOLSgF23pmOhDXAgstK5/2uNMISEorqx/158BDnEkqha4u73oQ1lTgoWIDfyyJzfmIKvv6pHGnODEXccl5Hp0ZGL8Xjnni8oORLvDmYHQ4tsypL1kZI9XpVc3ZDsPiCUv3RkNukyEh6pxBKGagNAbCj/lRj8aC5a/+bp5n9ubYloRkd8AFG5PMHbXOIz1UhLVJQ8/uOD4tKAtfxHV57QqkBbYEXvS/ptg6JYuv+z64E+71k5a1uTx9c1neOD/mqhjkaXIMw7BJ0ZODvTPjPO/6teSKquQS8yo68Kb+dXECMophk0jlzm6D5DagdvMlsCVB6ElmMN2q1mDiUlrdBgZayRUCgYr9U1NZ+PpEJC1w3rK+FQWeWi5XdSUQJ1Cl65QLWBY9N/8mBnOad8UT9rTmWecXyGsvbZDWBnkSivprv0CoffsFQfeI2mufbiKBVvX/JIJk2xARt2FuOuB33l/l2ekUF/Rm+3pVAAAgAElEQVQl+dLUHK8/+SCD2RGSqbVMNGZ5fbGP1/ZuohKEsRveGhZmLKrvbCPDqbJlbachrMRD0ulUEvejYYl0sMDq6v30Jo6zPD/O29/3IvY8ffNTonJEW/Q/uP7/B+qsdWD+nwNM/5hD7h+bMqWMnvy8aCuoH9z2g4P+537/4Fh+8Pc/d+zZfWcl8FSUgIsUM5adW7oZJOR/ffp+1g52sM4WeIaXJet5PD/bzUWRx2fr00w0Fun0M6yGSzQaExgFRCU6EL1JFC6TaE6RSXQx5Sn5IUPTS7qKD93G5xIZErwscyoL1pgmrcefsTzi5bjCwFbjsU7bvCRH8AjDEsvBAp2ZjexHrPtlBqIaR2pj1GtjkOjkuvxOrk8PEtmQTzrLV8QCTe43Uq0JLkS6yuM4hjONUwyLKThRoBxEdBJSTMAjDcOqZx2RbWgNh4jIG0tgDRt6YEdkON949PsRa0RhgHFGktttg40mZEJGj3wPJ8JZejIRxUYPg+dCua4KB5Ydmw09KcPj05aaQKu17FRpNhtyd+WIK9tVqx2mO51naaLJb7x4kJ++6gpsJYqVqmLLH7NEZUikDcEyeNsN+d5Yp3kFKJ+Ctx8KqARpTj60yC23+FyyqZNwNsIbMDg3bWRZehzKKz77STCm0mIHl2iMzPOMyze44vNPKEu5zwQMnvxAlTKVAm0trjasAwstYNFWtlLA7e9txds6RjQhUv4qN+baKPtRSlsIQopacXoCdAImIdxzlyW4y3BfIuBGk6YcLJI3hqclsuzKGO7J1ziv4DGU9SkPwexh2DUCjTocmYORrKG4xtCsQ0KASsY+AewyrvJGhwC4yI8TUL0PQiXVKAHlhCWzwec7h1Z53aeWec/2NazLKgHCw9Q8Ds5ANgWjeegU2bTiMOuQX2cIfOvqiyZVY3bGvfcQyIXZBb7AkgDNimhpDK6S1JQMpJaRqIs6Td5RrzAijhyTwrd1hvKdfPHBCa7a3Mn2UXHmtECVAI6SciRjyVxWLclNAMwlQgjstOSquZR8Bcrb86iyWs5d2ar760ChgFsr2UW1/AS0XH86vsXd56hKWlZWuT7Vn/oV0NP90Z57udZ1Xu3TOLXWojYaswCnvmut8beOVfyjc/Prdxvoaa3r1TWoT52jfZ42UHXuVlHRgJnz+PO/bLB2yuM1V2QoByWuP7jfVX7oTHYwHkW8LGm4Zc0uojDivGFDkDVUpluxpB7MzhmCBDSVeFKNk2r/LmrytajpalAv104y5J/i1LEj/P6znsfrfvM6fEfP8o+xSuuqnxIrifT/aHHJBs4q5l4ZfmgfmvO26zNexzGZaqzjBbi8Vhye1jaK/sm2lq3/h/Z/duNZCZyVQPxmL5eYeJye8ao9sLaPhdIkh6M5vqFSXBhGI49rkn18uHs9bytmeKQxT0Up/YotoYFtTmObU+75W/YynKoeomiSHJebNKrRJWscHlu9JANRhWf7RWxUY745hwmr7AlX3LFVLJN4lGzAuakhMpmNLIYr3Fk9zFRUdzprf1TjubbKVfkd3JDb4cxPXw0r3E7k2Ph7/Q6SePQFi5SjMl/Hcr+1rEkUeYhufrl0iodqCwSe4e6wwuEgoM/zGK/HJdIGPHhF3uPSDo+bumGzrDthXEGpM/JdWUq9TpZsk3NMghtTWa5OdnIyMUhnaoi/Hn+Qj5QnOTrmU5q0zruVjiBdg435WEkks4aeZER3UnVXG5jmjKuiEQc71aAzSbRqsUstC4nimgYNqd3GxZarJFaoODkFvStuKQWrctEGPo3mCsmdEXsG+mLFvsa4DNnwdOya69oGuaShP4LhxjwbGCazOh1bV/RsleKVlaTNLacHsUuIeJJS1jYBr3YQuJStLCXaLguNNIMUuhZ9d/vjWDLXr5S/2srdKyXtAF3reGE8nR/xg8KyX+bWI2UOiHom61NM1zgeRIw1YTJhuTjKcPqgF3Pc+bo2yK6oHrBhIAX7xyzTd1s8vSy0rD31KehNQk8WpvdaFsfEAQeZIUiI4kSxdr0eU3c2ec/ds/xCcQ1Fz7A5n2C16vHdsYiMD+uS4JeMy4BNJqGag5qSWZRxOmVpzBi8UYPXYUjUY1LrqImznKqmaU3uYdXr3ahXKxjaGfHzqQFeFs7xudphToYrjFuPMSGcWh97jxxzwPEJi5zmSvdAGxj1tBMTWi5IgTLNg2TcmrN4UlpATcdrbhxwalnxWkBMFilHNyJ+PAE31TBrgS63Fm2I4gO1TW0EEHUSJT+0wZvAmuZSfT1RSUL3VutcOkT3x5MAnQCeQwXap3G5cbTunfb9pGN0TQJz6kuL2jvgajCB4bvfWOa+D0T8zEAW21jld4/e5gYzlOpnPGpwQ7TCr/bsIpAV3AeRXyxPW2ex7S0YagXDnb5lsgqLs7pUwxdDy63NclwlJ1wiGxxnoTYDI2le8PuXklR5ON2/rSHFA3tq/f1XW+raYE6ArLEyzb6DJ1k7OOBeFp586ZofQ8TBfXfw6c9+kdOzVUY2jJJO6B/dooySxTMnuPUzn+b2vQcp9Aww0NPB6twEX/r8Z7ntrv1ku9ewprfjCQD45P7Pfj8rgbMS+L4E9P+o/83eviTP2tzBX/zlHRTX5LmPfn5KNV7xaJiIbR0JSqaLG/1eUrbIg6X7XAWJmhe7T/UAVgKEbwMy2e2MyyXrwox8ksbjS8pStSGFcJXlZB/1qMlyc5YiHp2pXuZswEkVtDdJV+LydLBEsnachfo4a03GUQq8KN3DuenNFP0ujhNwyDYYMgkuMSn6jOFrUYUeP0fTBjwSNSl5KfZYj61ZjxcPFRio9fD+xgKvSyV4Qa6DU01PoWns8TxyIq21sWvUxZFZmC3FMd+3BbFLNaO4LGvowKeE5bthwA34PDvj0fR78D2PE+V7uLDRSyLZSSZtqVQNMzVLl2cYScJwwXCwbvmTuk+tfoh6YxKSQ+RMg9WlMa7euplrdp8bK+SSLEBywYpZwziwl1GWp8NsxtUCVQ3Yh/ZGfHKuyr6FkNdc0eQFF3c6UGcGYsDgKVYpZRC9gyg3CmWfd4Ue3pDPHV84wk3PHWBgpMOVkhPlhrO6OMUZ3ydyw8mtJeXp3LByzYlzTNYhgTspMh0noNZW/jpU26XlpOC1Vh/a3wJYzqUn5R0/9FsgT01lPTN8954a37wj4MZRw4XJNA81K3wiSPPidJq5CiRrhvmcoW8WsssQFgxLS5bRNYbeQUN5HoKqYWCLIaVYpyYkNkFmLQTzcLJphKEpbjCE/YYz37ZkN0B60Ocbd5/mHQcDfmt0kA0bLJksfIoGh+Y8rhN/YQ56ug31BSh0Q7oLju4Dz4cOiV+cOKLOEXZfjkFl4xCIiFqWyGjMkhiN69vaR2P6lUIuxXBXJx95+DTn5rPcqyxyk6W3L8Xtc2M8/6J++jszccKEi79sATPNleLpNB+StQQvwCfZCmS5LNTWfGh324ImUKTvmhMtiksLjHP5u0xkudPVt/bLwqi+2q5RuVk1j20rnJvXFqJxQLGFcNRGYxUYE55XlqrmW4v2qa0sjNrZBkSRDDXxtTiQpDbapzHou84lV6/AZKuSReQrvtFw5CC897013rozg5f0+KV77uDjy0lGs12uBF+zXucDoxezrlBksRIwnPYQKO/bGP//nyireoShWoJJDzIiYg9KPGwDmvLDErC6+jDFzHEWT0xzx+/fwsXPUzFF6wxN8YU9Nf9qin7kpW1dUzHwRgiNpZP84Ye+Eb8Q2sgpFXXmLG4qmPvAbey6/FpOzc3x4puv5X9+Ukhb1jvFJszzay/exBs/cQfH7/oIOze/nvHZRT7yu6/ixf/v5xi//wvs3voCjs1W3cPBgckfeaRnG56VwH9GCcRP04tv2Mjrf/kqJg+cATPBRNSkjGEOw6FlS081ZE93htf27+SF3ZezGM67Gq/OJxiuOgtcWD/ObGkvmaDknr/7wlUeDstcZJuO1PgbUY3hRAc9iQIFW+cxP8d3bOh0y4ytcjBc5nBjmnx9jIyqVvhd3Jhey0syG7gx3cm2FFyRiLgcn1kvzwSG70YVVJt9m5dxWYNJxQqGFU5HVc540KcsvyWfZydy3Jwe5l3NBo+EDWeB2ZiAsVZ8eFfCMpmJOFiJOF6Go16sP7anYUvesFmxchEMtKyYj+Chyk39YZIREmzPbmMiuYWP1e7kseQif7ni8T/mQ+6viWrNMtWAw8uqF2kZjFaJxBGna7SieJYWC1y9KgGnYBEaE9aBJwEcl+F6BkK5hJTNmQavD8ZPWCrThj35NMquuHBLL6ripsxFl+0hBaqntciWG9bRu5zbl+Z5XpnNyZBp1jC7qlIVUpytZAn9C7Sf8Lo1lHWpTF4peV/kLrGb1P2nyDzhgEHLnecAXMuSImuKsmkVfC9AJ1eWIyiOrS1GbeWCVYey8in2T/utJThl6dqfZf9Wj4uSGYpE9Ab94GfJp+CcHFw8KJqaiGMly9Ba2DgEmwqG8XHoCGG4y3BoxbIiHjzdA/3GZaNGGUN6i2HHRqjOwJRkuAqdFxmyqx6zR6r8+uQYv9nRy9XrDGtGoZQ1rDSa+J6l0jRUrRE9IymVg+s11Odg3UboHDbYwfhao2WoVRVfKQtSXM6NyRjRpJVkIBlXIVQZq2XoXBNx+bl5/volA3y3OsVgc4a5ynH6c3mW7guZfHw2BkKSuQCObhmZ2mUpFTAS6JKFTtuVifpkMCT3qpb2vAoUqZ2AoECSjtU2DU/blaCg49190aZSeRJXodyyauf6EMBXbJ2AXAuku7lvAUVZ8OLLxrjzts6tc+p4d96W21dAT7GkstqpL51fi9pog8yIOk73js7Rcrv6oaE6FvHBd9fYUy8y0JvibyZO8Jlyja25HlaiKuUg4MMjl9Kf6OCRpSZrMx4rFfHVxq73ypxlPoJjgXWXMmrgSARvJ8nD1rLkpZkoPUB/corpQ3Pc8saLuezl2+PxOfNm/PWp+rd9a/yL428DuvriJH/0W2/i52/5JX7vj/4n/UND+EHAn/z6r/LIpBCw5jfu9uG7Pg8v+0P+4B2/y3c+84f89ms+x2pDsw9zxw/w1/fAtz/0Yf74Ix/H4wvc9umP87dfvZ3bPvFX/I+/fD9wL/sfPuranwV1Tgxn/5yVwP9WAu5ZaS2pArz9TRfxHAV+Tz3Kd1lymaZKMlibgUvWGroLARsSSW7uvdI9cTuFIPwcRgFLnkocRPjNGcLGBKXqMVL1CWab0yw057HBEsOJTg7VJxizTVaxFGpj1IJlvt6cc0kZpi6i4irlzEa8rmvI5rbxmSig5vsUw9Dpr/kGXEWSN/oejxKwy11AwPlGlWItG/w8gS9zSJ2PhXVup85AMuJMFLHR+OQiy2tWl5gN6wz4hm0+HAhldDDc7IkYGA4IDIq1IYQdGeiL4BGBMw8ni8sSHm9ylkzD4cjSZyO6SHBDYQffDgq8dfIhTir5w/PxQ8uJCpRDy0LS8pVKyFK4QH96lGyimzAsYwOZPjzMSg/hsrILwR80eFmIihANycLUAgIl8Masi6VbfNDSEXj01UPom+Tpm6R4W5q1pUgpxBmZLqj9PENvt+VptsDeqAMv1cvnb5ukpCoLbfeplLyO1eNYH4E2KXh9r7VoMZQBq9No0XYdU5Uyjmk1HMjQdilgtWuZZ5QU4cCcTlG1znrmFLuK02tRt3IpT8PorQHP6ZilXvNIpeGaXp+bbJ2wYRnKQL1pKFQanI4ajnpjYAQ6R2D9IMyvwMSi5fxBGB+zVGUQOGIJT1rCeUjK5aoSbElQXk6+GbvZp2cNX1iepVQa5oJUkYdOBS6OcqFimTycYjEpo5hlSx6Ku6H7MoPtjSt9dO0wJARglmIg7ICVeONERDxqMFuMy1pujFuUSIH4Bydj66ffbWiuQGI24oUjw2yK5phqrlJtTlGqVxjefg6f/fxRqvU6fs538ahujiRbyVggy1nTJMdWRqy2K0tWAF8gTYBJc6lMVs2XAJ2OawNCfZc7VccJwKu9rHfaJkusJlLHt611ai8wp3Y6pn2faK1ti63tug/a90obCGrdvqcETPVbx2jRdyVgtNy+Di/psnQPqa3L7I3vF1Wf8ERgbeCjXynTezTixo0+d86f5J1nxljfeyHzYYOlxhRvLpzHBtPJyVLgYma7N8PAOYbjFs4sgh8Z592eVY5HAh61cLuz5KcgLFGvn6QnnGYiGuN6+njb615IpkdWxadmtmtL2k+sNE0/0mJbs/mNz/8lc/3X8qG/eQ/P+6nLOT2zAn6C61/yMoa6BLvbafQRcyfv4/VPjxFwX5+oqL/CsvL2dT919rn117/2Fe742rfc/B99/OvYrTfS19+NCJnesgWmK8r7/v695H6c/XNWAmcl8EMl0HbDDqzP88rf+ClHqnbMnuLDYZnDgXVejmoZTp421Osho7mNjGR3MxHMkRYlCRF+Zgte/nyi5hSRKD+IGBSlfrDCgfoZHmxMMRGusBRW2Nlc4HnGY7U5S6l2gku9DK9LDfLhzo38dmErPdlzHFDsSK1hpjnPXrvMkvgjGpAbhjVFywvSaV5skpRIMIRhwtbIWcslXo43pnq52E9jo1W+0FzhWFX8YnrTD+nyC1ztpXhfsMKD9ZAshv87B4834M4FuNjzeGYSDlk47sOJMsyWrQN/qot7bgaG1sQ6Rl62+7D0+sYlX+zH57LOKxmIliit3sd2YxnD8DkbITfuvqrltK0w5Rco+0WqyUGlhbSiw1WNIkNdSREdkJByq0OYhflDFtMF5cVYj9ZWDafvtNROw5zncbSxyH/P91Ocz0E6clYp9+htJyUctaRUeaIowtyIRVNgvLHCOeu6+dBtZ6iJKLoZ89U565nOrY8UrxaBAudubf1uK2kNXWCw9TtS+S9HJNyqHavm2tcCfqrF6QBDC/y5w9r7M8bFnMktezwV8efU6PF7HP6Qhac3F3DKX+HhAE43YH41YshP8T3rsZy1JFKGxhA8dsyyULaOBHigYPAX4iB4VZZwhi2VLVZN0A5DZweM5OHMqYjlDo8xr8TPfXGCl3avpZDyOVGy3P7diMmH4Jb1Puu8GhvlChediVSSPiVIZw11WZeUPaqsUcV9iwexkzhRRbITJu82JAuGcMLSnFFiEGTXKLzSEnUaKiVLfjbFHwxexKitEJgEJxqziCvlL6ZW2T9fi62emhepTcW8yQWpuRLIcpasVhaqzqm4yzZIUxKEAJ0D/U9SjsJHOl6AdLUF0mTp0yKyRkdT0gJt7bYCbLLAuuNabQXwBOjabRTn17ov3Dbt125V6tB9IBCn8aq9Ps71r+0aQ8sCqf7bY1Pb9tidMGOAJ1ft335zmU+/K+KFF+QZT0zwsv17obiJen2Cherj/HJuO89PdVNNBgzmDZeJJ7CjZSVdtJSrlknVmy3BRRGMFOC2yCI+zLmwwmqwRFQ5RDG7CMfn+NkPXc/QrqJ7iXwivrQlhqfq6kcGdXqO6i48+vAYz77+6e56L3/6dWzthNCD3ZdcRn8+Fce/taSR7ehnbCq23gUudWszOZcfDh0j57Lvq5/mgS//Lz7/1bvo6ryU/q71NJeWaIqLgIjxo5BLuIjbp6p8z477rAT+3SUQR2DBS1+8mQ/9ydUsP/wonj/pqjYsNgyHl3DFyvtGLFvWpXnJmh0IgWQj5YR6rlSYyl1pMcluEun11HPn8ozsOVyQ38F/69zBb6YGuSmzgbWZjZzbcSXX91zL7vx59CW6CEyG7kSGbt9yDXCh8VDlh20EfKVpubMJaxOww4NTIZTqEdd6GR41CRaNR9qkGTE+E0Rswuc5fp7XJIokwoh7G02nA9PG5zWJNM9UFqwxvCNY5nDYIDAeJ2SRTDrjCaVA54dqAFO+pUeVmjzDcgD3rMID5VjfSbeN2gZlIsZshKo2rTVpavnzuLd0Hx9f3sesypdFDT7VXOJkU0VNk1zpdzreu3yim0gazZG+VbFTC2RSEJy0BKp6oO8HoL9LlinIbDakTig7zxIuWwY64RCWWytw3ZYM6REIBcAEmlzgu1y4OAJcPR2VdVk4D67f4PGqVIZKRxcrjxTYe/tqrJA1eVKiem7ro0SJls59wlWqp7/2CSy47EDrYqAEmNqHObAjThiBHlnnFIOnnbpUgQG5XbVdGbHqTx+111I3hKtNvsQkA+kc5xU9ejZa0vkU64JeslgWa7AcqXa7x8RqxPSRCHmRi32GzBZYXoXtmwwLoSUrl+4iLDZwCRiOzLlh8dOgkrLOM9zrcehIg4+NPc41g6M8f22GKzZG7Bk27F2EZtrSlTLsD1LUMpZMBzSXYfKAdQDP9MaZrZ4AbSYGa9FEnLXMNESn4vJX0VTsKjWbjctiTq41RLIWjsliaIkKUCtFrO0c4NnFIWpB3VmAO5M1ONXJV++ccHJ0ZbOkIgXcWmJz6zYI11qWO+2TC1NrgTO3tK1xrTZtYKbSXZovgS0dL0CoeZGAtK390T7F0onnTla89vm1Xd91X6itAKS+a3ninmmZeXQeLWonoOaAXOvea1v8BFrDJ5Fvy1InV7PAp5KY1G0Gvndbjfe8vsyfvrrI0eQiV975NegcYDisMlU5CtmdXFDY7iqL7B4wdGYNi+kYT5ZPwlDGuMueaEIxDUcrcPeKYm1FbxRwVF6G2knWFyqcPDLP+975Yl7xykvjuFUBzf8gi6boR1r0j667atP2QW7/3v3umMceuZ8jcyIPhbEjR1iuir8qToTQ7G8470a++nt/zIFDR/jKJ9/JM3/15aweP8JH/+4zjJ08wa1f/CqvfPPbeP0rbmRp+QH2vPDnueDQ9/jyt27j4N7b+LRiXkZG3bnalsIfabBnG52VwH9mCeilWTWRE/DKl17Jz13ZxfEHHuCgv8o7aw0O+bDahBPTlscmIzZHu0j5Hcwpk9VTUoWqPSgwSm/8DzBrDNPhMovhMiOI6y3vLA/P95Jc5KU4SMTlqQLnYVx9zbfZkL21iKVGk40eXOQQg+VMaohtUZ19UcBtoaU+DXt6jStUv9WDNyQ8LsCwx3j0YZjBoqTGKgk6TJpLkt3cFlX5ZuMM+xXb5/CE4RcS3U7XfLK5yq3VUM5PNhm4OGl4NLB4KcuOPI777ssr1gHMSkoB1YbuJbjeg63iAfaS7AstlxrDZYq7EZddai07e57LnZWDJMsPsstLcdh43BqscLNJuefdbpPi+sx6Ir8LX5YymVr8KuGSZXU/hCqC3qVSthZ/o2AzpLKW8QXL7KQlIaAk62VQ4Wc7AwZO9hKMR64Sgl2MwYPTOQJbF8Qxa14RVkuQmjOsTaUYs/IPDnLPgfud0cTxSet/oK1Y/RjqOzymP9KkWvIxz1kblDm3qp71atNOnhDgc8//FsAUnlWgvECG+mnEVj2Xmao+ta1pXdD/6m0ilauRaxo6apb0Oug7z+eqps+yDejAUJHHOW25IEzyzTRk1kFhHWy/wDDYYxgcgh6N38byWjmjRIlW6TKZWPsgsd24cliZxz2+degI7/lOlZu7Bh0R8+GpmPNMcZTzdYE4j3PweXgxDgOQx7zYcM4hV6M3odhDV44rpklxQFbgRdbRhihOrAOxonhUSbHgUeuqS1RWIOo3FEfjBA/ViO1MhPyXka1cYYucikqUG2XoMRwfn2G6GbpKJeIedIBNoEtylvyc/OMpcoBKoErgTm5YxeI5gNay7Km9gJMsZ5rv9m/1pd9qq7mSe1b7BNpkDVSkgANlLcuZ2rYRgdq1AZv6dvPf2u+yZNvjbFno2uNtgzWX+NACgxq7QIKAn2Ln1J/6b3WrxIjZ8YjPf7HMuy/vI6jBr47thcIG1id6mFi6nesLm7h1zSWuMMVUZCkvwXgJKk2BZ6jXLNWE5ThwQQ6GErC3CXfLOx4FcbJKc5YckxyfOuZ86c94ydV4ifgFwcm7Je6n+qo9hf/idcRhtXD9i9/A6r4P8Ia3/Abvft8H2bZpDSao8+E/fCcnZFJ292N8B1xy/Ut591u2c9452/jvt2/nPW99OY2F47zq1T9NLTfEDVdfyHOedh5X3vxm/uazt3PZhbv4tc99lo+9/rns2nMT73z/57jsPBXAs99Pvf8XR3q2wVkJnJVA2w2bWAu3vOdKtnCKE9VHCUzoXBGPVWFpVSWOIq7b0MXbNlwLKyfpDpewNsRGVUyi37lgc16WTi/nHpiHo4DP1M7wnbDK96ylalLkSDDVVDyK71ym12B5n42YsSqYDbuzBa5O5inZiHRUZ72J+ISFg624bZXjkU7YlYAeTExsrw3A47ZBylrKJuI6P82LvAyPBKt8sz7OV6OAKxIelxnDRV4HqgT5N8EsfbbGIesTBJarErAvsi6p8AYSLNUMf1CxfCEIWZFm8Q1HIssJG/JICF9X/dpIyr7JZcbnSpOimRqEzmv4yOp+vrF8J+sd4A1ZckYry4TV9wTktmNdNmMVBiokthmKu1W+ECIVnRfR8GGcZUmK/OvzlmQX9KRwFsUzNUv/VT5du3xn9XG0ILJOSRaytsgaFuGUXiBgU4F8EXb5HuuCNLt3jfKF28ocO9F0rlSnPDtiF6JT+sLpLUJgp+Rb8VYuJk5WNz22dS59pNCF5JRYkRPwi3+7bEbXRpa5J9FUKABe2kQAMGlcBmOtAXcrgHH9WrLNDKm1lkBuSgsXXdnkC+WAZc8w7MNFXYYXDBo6JmLwpLErY7FQgsMPWJYaMHiJITVkHEiee9zGYxJg6Qbba0mUfe49c4ZbS3Vu2rKbehS6pJiji/Bnp0PEZ3dKpL8JuL4DurIe1Tk40wRvyGBmIDxhXf/RGUv1sCUUR50yX1V7tg6pTXEGsRFVjWSpjNRAySuQ7jXk5M6XysrCwjzMl8X0kOd53Tvk56QaJdk+PMhHb13hyLennZvUio9ODimBMs2BPgI9AnECYJ2tOdEc6LsscXLXao40F1K7kn3bSmHSKuYAACAASURBVCYQJeCk4wXctF/9qa3AnPbHKjruv/1dY9B3ATOdd1p9tFys2q7zqS8XI9dCBEqqUH/tODn90+o8Gk8bRGoc2qbfcvPLMqlSYrU41nRu3vL6v62y4bFOhgeSvPzhe3h4LslocRsnxWWZv4Cr85exJQPbEhb5EEpLIP7k2XHL1GPW5ZXsX4SNvnGZ6l+VwdqDDcbjkF60KocxtUex3iIbl2Df3hexZWvGvfw+1Wq7xk/G//1fTcePtLSVRGFgPX/wFx9npVQmW8i5Mjm+7/Gb7/srvLhYmuOm0wMh2THAr7zrQ7z+be8lkc2TSvrYvhtoNBokkknOeekbKT/7tYQmQbGguw12XvMiHlopo9C7nPLN3X3Wdii5n2f/nJXAWQn8CBJw/7NYLr54LX/yZzdz8y/u4+LLt/LumuG/+AJjhv6CZXOXxxWr50E0CI1JfK+bgBAvNUij/AB99dMkBVjCGnVjGKFBzjaok3OxKHrWn7BNEkS8Mp1hW9PnS1HIfgO9qjqUtuwOO7jDprgjXOHVrHK+38ej9YB1S9Av5Z0ydKRhMYAxJTv4cF3oMUOCRSLOwSNrLZ6f5m2ZUX6ndoqP16foNkPs9nxenVD5rAJDRiCuzHoTcsgW6PNUsN3wuXLkyPavSPlcNWj4zkKDsQYseSn03j5pQ5Ts+It+mlNRxCkb8CbfYyyyjBuPerKLU8lBHl6+nR4vCfkLHehRNmxW0XRCaeLuc9qsgfECgoqlPmXJDBpW7rN0bImVqikbKg/BcndAZiBJ5gycbkb82eEGH3hOP11rxVen5IPY+uSUptyw4rWbhUQvjh0/uxYSGcs18zk218tUCNl3b50zc2NsCbdim5HLdnVDknVHulXKXeBAYE2Kuw0aW5YTp9SlgNuK3m8lAwjACXzILal9WmtRH/qabnHeyUInq4zxqJTKfHbphHObrW1EpLoNnixrc5Z+L82jvuVkGMVYZM6ypdeQnoMDFcuFK8bFpg1eYzn4dah3QbdC0iLo2wOrp6Bet2RGDaGw+arPvcs1bpw/DclR3uN3km6GHF/A1dFthp6Le/MjOBRaOtIeuVrs/k7vNtRmLPk1kq2hfhC8HsvKBBQi6xJvGuIZXAvK22FFJbrAdgu7GEK51yuQWAfhosUrGke14q/AchnGKk0u7+/m1xe7+f0zj9PZfQ5s8njwk6e4rHuQ1JUmTjSRdattxXJybfHLSb6tuWIGUDULzaEsa9onAKVPG7hpu5vjVjKE9glYyVKm7fpo0XadTyhAx+qjbQLozu0K6KVALna1E6DTG4bAZ71FmaL2komsdzqm7cbUQ0GAW+OWW1b7tU2uY2etk6FGeQuGd72/zPn7U1x0boLPz+zj0ZXjbOy7gbnVA25AH+i5lN1Jj5OrIR1Zjy3pmHfykbrlwvUGrwb1Wcv5KeNe1O6xloFcgr3lCmNRlZXGLKZ6lOF8mccPn+aDb/u/uPTiTXEcnUObLXn8B1lpSn7kpQ3sTCJBZ3cnqWSSRMJ3LgiBNBEIP7G0eLNE+JPr6IgBnQiHPZ9kMulEqRfBXLHoAJ3cRfqtdSqbawG6uOLEk3p9ovuzX85K4KwEfgQJtB6yV7xyB695Q4H7936bXLLJH0V1FoylmPSYmg7oDvr5r7v2MN04Ta52GJpTLrZOcWNB9TEIa1jju9iUoUQfmxOdlIg4ZpsctiHftnEpy2XrsSlvuDpleLSxxJdtzdENDEpR6KnvZbkraHJOMmA0ZZgJIOW3YmBq8TP/tLXcHkROd1WN4X1RyP1RwFGrTDbjypz9xv/H3nuAWXKVd96/U+Hm2zn3pJ6smVFOIIJEtAnCXht71wZjG4GNwYvDt2uvw7fL2t5de708zoY12GBjYwM2JgkQBkkIIaEcRtLkmZ7UPd09nbtvqnC+53/qXjw4fA847CpMPU913a5bVbfqnFP1/uv/vu//LW5nnwn4tfWj3J3U2OD7/EgYMuHlKNmQX2jUqKeKEzQucUL6cjHGERdjWG70Qs4Yn9ujWQ4mNfoNrsatj+eIiCu8PHensO4brvU9TiZNqmEfpcJ2VlbudUZiCcvzvTzdxuMuG1PxK3jFyx09EiaQWzEUhgzTcxZf5ZWGDek8rD9mOXXCcmndcOZeS8szHJVF3bvCq19k8BWgL8M7DXahbbxlWGV8VS9W5ZpOWrwNxumF5SN4YZTnfNIFhQlu+5slaqpNqmB/7aep7frKjKwMrcVKHFlxezLmSg5QAXmxgdpFz2LtI1ZFXdcBAp0hpwLtkvgQ2+eOL9df+xwFIJWc8pUWjz5k2ROU2LINcqVMYLmeh3Ld59WpYVKvDzH0Vg2VEIa6YElZni5+zlLeZNh5nWFwCnItGO6HYAmKgxCfwsmR+E8ZFo8k/PJTh8AO8ZvFPq4tx47Qui1qco+J2BsYnufDdWXYHEOyBEtKnFF29FrGeCqPYW0a8nuy5IuRGwzFIUjy4F/X7pOFzH1oz0N6Gs592rqYcsVPugoKQwa/AgvO1kGPqly04Mr+hDddu4PtPSEna2coFyZ459nTTJ2twYwhFWBXu7f7wgGg+TbIExgSKBPvoVqv6gttJzCl5td+AnKaBL4EpOSm1VKztu/MGkM6XmfqgC8dT8cSItC22l+cigSHNQ40a72OL2AmgC/Ktd39GRBss4cq4XZhHJ5YvM55uu0zKRezYrj/s2scf3/M63YGPLD4VX7x6Gk291zN2eV7WE1W+NG+FzGW+FSKKRMVQ1WVRHQaYvUTmF8QaNarVHYid9QtftFw1rS421qmonlm1x9lOLfIwcOH+cG3X8mL3rk3axe1W+f8O+3xLFh+S6BO1ytgp1GUgbB2OQ6tyYIuvqFJ3LYOrGXbZ/v+7bY6VOc4+k7/a9lZpxZ/Frb5N7TRM/kf9VOSJKgU3IWz1v1D4+GZfK3P1HPv3GP9vfDuX/h23nrlKqdnH2TMizhPwsqaZf+SxQsM3zd6A5T6WUhahH4Fazxy4TDH1h5hPppxTJeEiRdtk7PW8mmbkuoOtRHXioDB44uNJtMty0t7Av7fUpW9cUQcGHYPCjUUeL5f4PaoSbO8zus2eQQJfF5sRmQpJdZlIuYwbCeTJdhjDf/eV9audfZpxD0zMtuyJzfMlvwo76md5iv1GUqBz8405fV+mReFg/xM1OB3JSdhDJelxrlpD8SWD07Bo5FlHsvNuR5eHBRoJk2WkzoVa3kNPhOOwLCI2bk1rrlyHamYOEm0lC6lsPIV7lx/lEfTpjtOnNRc7dtWuNn5yPafrTOzHjnJjf6i4sQMsVx/obPhrAeGywo58qlhddXi1S2Fvi7CDb5C40jqxtV5zTTgMjealRCAjPdoVojdW7TU1qGRWL5ra4XNLbhu0zDv/fR+1pdq0PSy+CU9mwUKZNBllGXAOyyPM/aZi9Z2EiK0Xdtr6zJclc0o8Kd9tL8eypJ/6IAGHa8DFNoCyQIAn5sRB1rhHcUcQyuWeAYKZyy5lmVok2HIq7HRwr6ioVvFNXJw7W7D4/stc8esE6rXNYuZnFpTSS5oDkN9GUKF6i1DtF/u6JS/fuIgty5387qujS6rdaXpMWQSPpU0+MVKgSsDV+SDPh8KIZw21onS5hPjNOhSJXU0oNiDS5hw1ylQI/elvE5+Bkq/XoNXruxlS/9O4zTuFINneiHcB+l5q8RlCr2wWsPFsNo+y4TJ8w6zi1bzOGPy5darfO3EEvYxi73TYpV8obbvgDslMchCi/FSG2sp4Ozi6trMmdapT7ReLmH1Q0eqROeuvhIK0mfN2l7/a736U9vrNzpjQt9pnAmIKctWfaypk63aAUJar/3EwF2Y+KDvpSeo42jqMIT6LadRp7KC4AUe9z1U512/XeM1N3bz/gNP8c5D57h6+DpORiuONf+PvS9n0FZZ7kooh4aWMfhVhXxYJpctg3ncvXNmGU6llqMtS8547K/F/OryCqVkGdM8RZ+3yLnVWd6Q38X//Pnvob+vndDpsEz7PJ9Fi28Z1GXXLgDWmdtr/rEG+vp2fwvPOuBOe3aOc2Gb/kPrLvz+4uf/+y2wurrKiRMnOHnyJJOTk98wa52+0zYXp//7LaD7KbGWgdEKt/zyd8Op/WxIjnOnyfOxCBYSw50rMb3Dg/ynDa+EpQW6vJyrBWvDYaQat1g/xJ3xGqPSr8Nwa7zMDxrDGIYXkLINjyaWe+0K97Ys96xZQpPjSS/gw0nCOWvc9mdsxMmwiz89t+BEdC8dg0MNy11NKIaGgSBTylB83HJqGavCv+nNMWQDdhsc2JJT5qtYV0XiTUE3P5cf5QO1ad6yfJr7jE93AK/z4D/k+ug3Po+bmEvHDENlxQElXJaHxyz0Go+tfo688fhy2uJ/N8/xpG3xqG3w1WSZmJRBPLzWeUdbGRNSz41SKl+G2oXF23ikeYYFEhfqFJuQNUps37SNj/zNQxyon4V1j2DREgnMzMJSBPeeVPF3i82n5Itwat1wrtbkB6oBhSjAjlsXD6jkiq+DKBlLgQyxNYpL2ghpr6FrGMaeb+jzDFflDWmln/kT8IW7J92+EusV42bkRnOxVW3GSY9jPf0duGuDCbFyAm+a5O4VYyhXqkChtpULVtvLoKtU1qLFrra3FxgRHhGLKA/lwYRzXz4PewZIVwxLfVC5PEuCkI5evGpp+T6TeJxJoZSDwc1OkpDgKcucYrlUQW4bVMTkboUzh6D+EC5G0USZFImtG9537yluecTyk92b+Im+lIleQ71lnZTNFr/AvUspdd9QLBoX+7YQKTZLGbBLPGEj8iVDeTfkh8FXnovaoUtsk4hl4+LtlO0qllXXJwBkR6EhCdVlS3NRIDwDY/YsJE1D83FI16C3C0YkeCsB47mUq6qjDHn9LEXyH5d4tHjEndvKpMUeaAO2Nqh2fS+wpQbVrKZWPwiIab3+16x+Vd8IBLb7xvVTJ0ZOfabtFSun42gbXYfWd9g2rRfA1/Gkz6ffEFgUOOucj4ul01hp692JsdN+2kbba18dW+v1v3PXto+j318H29D7keGxQzFv/x34vuFhZpon+N2VJ9gyeA2RiiI3jtLT/SLO2Sov7kq5od8jUu6RdB4lGl4w9MkF6+OkbFSSV9EAT1pYJuVLSYTyXNaap6g3nqInbML587ztT1/H0HiV1GblSLMR++z7+y2XCes0gdgYTR1W7eLyQvby2flZbJwm9f309DQ9PT10d3dTLpepVCpu7nyWi31hYYFqVU8a3YxZxZFnwjhRneJn2+TcE8YyvKPIhs2W//0H9zG2YSufjGGfkhk8j2bTJ6iX+UTjEeJ0FS/od8kSiW1R0P+lSzlvGxSNx6yNnHu0xwT0eDlOG3gMyzmbckIF7+OAOLUuxu1k6nNuzXJ1EHIqWeec3qbjGpdGhi1pmWtylseaWR3PiS7DaGI4Eyeojmwp9ZxrtNcJozmPJBMG7rOxW7/bGLr8ErmglzuiGb6a1rnKK7EjCIhIOWpV2izheGOFkZZHSEjDWjYp69LA4TQlsYZX+AUXPvTJtEnTC9jgF8gRcSCa5460RdkvE3o5tgbdeF5Aw+8iTRvk6geYD4do+gUC42PSBiXjsbRwJ2+9/AVsGhwmqaeEoSHphk+ctlQO4zS2ksDwpMoZmYBPP7LILT+dsGt32ZWP8o+DyoM5IynD2bHtCs6X8cw82bSkd3cWWqckiuzxHpVAOrfE9q5pXnrNXkyurRugW1ePbM1ymXY+y2BL0Ffsio6r31FG7AWG3yVXyHLKaou10776V7O20yygI/ynEpChz2MH5/jR/3meb9u6lZv7cuzabgg3tUHVFMw8Dq3AUI48dvQaBgwsrsHUGehehMN9KZdd5pEo06Zg6NluWDkEQyNZXdmoCd6wz8eis7zl82f4d+N7uMH4bBbubBoeNxEfi1Ne7eW4w6Rcq0IdTWUeG4ZC4xJIb0titschu/o8CoOKX2y7nmtZdqsVIyZmS8kfIi2kVSfQJ9ZRNV+r0lvNJFB8ua4FYgrg6QVlyLC6DqvnIegxDC0ZevKSpclBWuC2pUOMDJa4+0TCd/UOs2ksR6ISWQLuak+5tzsgSY+jTn9IS1Db6LzkftU2LllBY6U9UATY5M7VfnoJEEDXpO+1j9ZrG52vgLv6U0BWY0HjRN8rHkH7ZY98JyPijqLf0zEckGu7h3VuAnLaX8DQjan2cdRuOl6SMc8SGD53EH7uXTE3rOk+OsfPzT7Gtsq1zCWLnF57mFf0vIQr8qMM2pTrRjNNy2o3DPTA2ZOGwQLUPJhrJ0NMxbDfWs5jOYHlfhtTqh9nvX6Q8WqeySOP8973fw83v+HKdhNlXsGsUZ59f9X839IURRFTU1PU63o1uTg9F1tAwEygrVAoOKb174IgATh9p+WRI0eyzOW2wXgmtFexWGRsbMxd4zPhfL+pc3RuS+Oe02994w34pxq85V13cvW1z2MtGuGIhW0LEUWGeUHlJr66/GFKfi+xAs5zY6yu3su3ReeYLmxlOm1wk1fgM42zFP0S1+VHWcbnajljvTwHbUTRL9CP4eE0YdFYLg1CnrI+MwS8VuxZboh7oile5FeZ6A64oivh7lkl1mVaU4MGbrfWSa+cj+CGwHK7AuJl4z3YYDxX5uscqZMoGfJC3lGc4K7GFD9fn+JVYT/X+BX2mIRLyfFIrc4PpdP8SNjHa8Mumj7cGiXOPu01nnPRLvo9GNuihc+tScx6XTFaCaXciANsAnsBIalt0gy6CcqXs7z6Ncz8J8n1fwfkRkmdwKks4Th/8chRLtu4h6BkmJ2EvFiz05Ydwx7LSqIIDKupoR/Lp/H5mbKfGcsVSaCYrESYjLKMb2gwErc9YTEDmRvXzIA3YsjVLD3bDBMnDdvOFqhdvYXf//AXueUtM+wcGCZpJs697oy3BovuRTE7QVY5QnF1DtDJwDdspj/X2a4N8ty+zqXWvpFVdKTeFix2gC4DEQ4MrsD8w4roL7IzrrJ9CHRp5/5Sum+WtQJM7DbsmvL51ImU17eBQElxUrvB22woT3mcPWYZHzROk1fiv8M3wMLdUB2F2Pf5s8lzvP3+Oa7fuIdeAvb2KpHDMFk3DsxvMXDKWjZaywdqlv9ehmVVLvAMgwm8MC3z5Tpc17B0nzAkvZa8slsDgz1pYadxwM2ezIAbW0EMod5UwimxlC7UlKA3k6yRa5YFSKYsqhHfVYIjAVxWMWwaUh6SYWA64bXnNnGrv5XTrSdZj0eZmpxn+0oFNsWoDq5VvTspPuxpAyVVrdCxxbxpLMhq6391hc5H/kwBK01i0fS/3liUKatJAFDjSMu5LFvYAXKBLe2vY3WOp99yjHDblavjiuBTZq0+yA3b0SF046GdOOHqzbZfGjrnol10KyjTVSXUihAvWP7wj+Ypnutlfmie3z/9cQa6b4B0ldrKvfT338ze3Abuj1v8dJehqBhGT27WrKSeXK6PzFmHWYMCPNmAIWM4TsJ5azmb1sk1T7G8fpCJ7ojjB47z+//hu/iRN74sq06ma2lj3HbrPOsW3zKoE6ATsBND02FunnWtcvGC/n9bQP3ebOpVT64QP8t2vmAPufviOHaAb3R0lKCdFX3BJk/rj+fPn3cvLps3Kz7q2TPpYZY6JsXw+je8lL9410f44qEHKe58GV9OSryFlJ09Pj9ZvZ6vPvowfnSSqFnB5jY6wHJ07SFeXNjGqglku5gojOFZyxEbcYlNucovcSz1eCBtsZI26Rbb5ZcYT1vuxX7SNgn8MqfTBr6X40NrKW8dW6anOcDVsaFegA+dT3lVr+HGvE8QZTp1GwwsJ3CJajsGMNeE1xvfadh9NrX8W89Q0HmQYzg/RpSsMh+d4wtplZtzw04SpT9X4cbU50NpxJhd4apChe8m4OEk5Uw7XOw6acUlnhNpDtJ1qiaPb2usJzW6vQrb/Son0ogplUnzy6yGA9jKldj1x2gtfo6w+2V4YT8NyU10j/De43fzC82XM1zqwl+JOToJ2+uGJJcF6K82Lb2ekillWXPkgqKLUfJ6s2oT9oyltNO4cmdOWmJdyQwy0GJXsiSycMayJrDbY9m3GPLjSZOfWlG1ngFuP36GrbuHnQSeQ/OKu9LkVP4zZseFQrdXu9/oDHet0ywDKFeojGGnrqsSLFRVwMtYwK+zfEoWMD7zS+v82ocPwe7rHFg0qaXQMgxeCcsnDWkflMezsm13NKBaBj0iVO4rHDB4s7ByxlI7ahnf5uEpqSOF6m7D0hMpURrw6dYcb7/1GGzbystzPUykMct1jyiOuCdOuTQM+W4f3t2KXRm7pcTylbrlMmPo7YERueNPt1hrBrRWPcwVllD1XpfEkKr5DMkDGbAyGxQ/1y7TJYu5BP5WQzwHjUOWQreRLKFzfwogx2eySnvxOpRrEC1al7Qi59bJ05aRXp9fHd3Fq+/RS8N5bluf4ep9G+iW9p5Eos8BqlKhMnJ7lZDSBl0CdOoTDRe5RgXSOp8dW6a+ldxJe3udq0CVk0ppM3IdIKd+1XG11HY6fwE6ba/r1/F0/M4YEJvXTrpyKFvfV6wDazqGG8ICfHLXa1sBPh1XL5Oqm1u1JIvwkd9b4MCBIv7AMr9/7g66u57vvL3nl7/CJb2v4gW5cYaIuSpnGOoz+KvQalgqvYZWhBvriznobcJSnGHcA0mMbxO+ljYot6aJl+9jsBJy/IA0WSZ45VtejuSB3LNPLzPP8ulbcr/K7TYzM/N1l1on9u3ishNf+NxY6p4QaOvq6vo6qBNbp3HQmcTmKaZOLlqxegJ/z4RZAFQs4/z8PL29vX8PsHau75m6VB+pbwr9hr3f0csffOo2Ti0VOdc9xrVJwHg+ZWuhwpZykVtn7mXYC1nz8uS8EnP1J9lW2E4j6HcevOd5eSJ8ztqIOJrnUr9Ewwspm4BXG8O9Cl5OYwpElFSrFcmhWI7Fy056oB4UOLva5Oq0gi9x2kQlwiy/0LC83pXr8lCZzWvLhoOqwiAlCT+r5SowIkmEawPjMMpxrHO97QoMQ5RZ8wp8OV5k1raYlnsWy4DJM+LleF+0yp836rzKJFzrlThhUz6dNtntBYwYn+POJRvRQEr0LQKbsEjEFSZH6IVUgh72eBWqxjDp5R3JkbSmSOuHMIrqD/roMSusnT7K1XvHuHTDJqaOpMychuEuna9lx4jhfAM+3BAx0uL6/CqvfV23C9RXepgEcc3hTGJDBtwqhk1FCMTWaJYrTTF0Ryw11TzdALmGZWYGPprmubRa5JOfepwf/a7dFHtCrAR1BQBk1GWoHYPSZuzkZpMh1jqBhQsNv35H63Vra3/NMuJyA8uwa9u269U4rUKPv759kd/4oxO8YtMlvC0s4kWWgjHkyjA/B729UKgIBHmsnYbKGpxZMeRzkmiBuWPQ6xmOLRo2XAKFclYGSqfi4/OeL03yzpkpvmdwD28s9HGjn6JC8K0k4aBVXQ/DKQEQY1izPiNYmjbiyiDgur3GJb+oHU/GPlenhtYKJKOGwZJxcVq6xoaqSFQhHFRN0HYsnWRNxCqugarmyc2aU11fZSTXM/dz86whHMTVX548Ct0VGO83rMxCug6l1NAVpOSCMg+fX+G0t8yT83Pccs0u+gfy2CB17CvDWQUGq7g9jfW2RqFLktA5XBgv12HZ1K9i2abboEygSv0jjUOVCdP36l9NWq/v2/359f7XjaVjyJ2qNtQ++j3tp3Xqc5cc0T4vxWtqmw441FLHbo8nHU4ubyVE/9GnlvmdD+d5Xm/Cb89/Cb+wFd8vMbr+EMXqC7m8so3vr6RsSeH6nM/uLoOyyFuR2FfLZE0SRLCzajjRhCOxpWQM96UJC/ES881p0sYk1bDJ+alFGJ/jgTveyL59Pe6Z92zTo2v35N9bqPm/6aljtC8ydN90kz0rNxQo0KTx8I/NF174P7bN03H9hdd14TU8mz6r3dWHV185zgMffgPB1P0QH+Jxr8GRZZicTxg1l0PlWqabJ8kly64GrHxR9649yKSkTNImU0mTggm4xssxEvRwwEb4yRpdWA7hUTA+I2mNqbTF5QT0uzJgAZf7VXpd1QfDF5o1Ho4bLFoZMcsPDHi8Lp9yommZS2JmYssTkWXBwMkIehLDnrJxNmpPHq7Lu4ICHMXykLUcSEQstah7eV5WmKDLK/JwssLDNmEQw8slkOx30UvCj9fOcVfrvCtj9jNenvXU8rV4laVkiSaGlo2oBL1UTEjoVfhUuk4rmmUgbTq7uSjLFUtFz8evXAN+kWjps8Sy+AXlzxb48F1PMXUuRfFZm3qhfxgKTlPPlXZ1tWaXz8V03zRL92iW1GCVoTmo5JbMNqoygdliMFvbo9AZUaWLWvwh6L8cvG7jxI6v3xXwdlKqecP8/AyHv3IuY00USS59sI4B1lJgTWBPhroNEp1BloV2Rl9GvVMXtB0/1bkRBAT1EtcBeTL6eeOqaDz5xYeB7dyU72JDwboKVMdnLYuTMNoP3TH4XYZgN7Q8ODsHg1UoNaF2BLoGYHgCNhWyGDa535Tpmq7CX83P8PMnTnNZsIEb/CobFXupxBEf3hfVuCPRuAsZNIbb4oQPJDVWSPjuMCRswdQKVC+H8g7YawyVBLpCaDwIS0pWyCsmD8KqIbfJYOUKz8slnCWnCNSp6aLzyg5tM1xqkx4cEA9KlnCDIUqVGGHp6zX0j+IqVajixHTdupeFxoJhe3EP3abGar3O5+6cc0kEEnpWxq+yR/W7RuKJYu8esljFF8qlKhAnNk4ASm5VnZBYNYE0LQXKBPr0nbZV1qn6R9tr1jbaX/spUUJLB8JsdjNpIzFtDry1gZ0D9oqpbCfVaMx04ui01HiRzp4mbWsNNs4qctiC4UNfqvMjv5Bj3wbLe85/iko4QlduiLXV+xjsfgn/vrybN5Ly1VrCapBkybxiJn1YTCyP1UBhGEq2uGcxdRVpnkrhYJJyIm3xLKiCfQAAIABJREFUtWQVW9tPYFdJczVoTHL/h9/GNVdudM86PfOeK9O3BOo6xlyMy8XpYgtc2AIdgHbhus7nzndP96XOt3OOnbHeuYZn31KZoJZrrtvGe/7ba+HRL/JQcIw/t00O5SyX9oe8b/ilriijH8053bpcOMBU/RC11jSrJuRrNmLaRjxoE5f5Omut02tbsE1mbMImY/jOoOIkP+5NlRtrnccn74Wkcl/qiW3r/NbyIe5oNDmTGgaN5b9UQxqh4S/ShLJnmWoaViJ48TDUQsvjTctwPhOiVX1HxXhfo7g4Up5w1ScCejBEYu+CCtKTO5ascdzGDpBu9HJ8T26YdxY38e54kY81TmHTBld6gQMEyzZlqXmKql9lKOgh9Ivs8vKUTMCDNuWe1jQno1mOJzU2KQu2sIlcfiN+6XKMyZMsfp6lqMbE6A4+c+gR7r7tDEP5gIok8BswUpBwbhajvj0wHD3fZEHyEXJvdtiTAVhXofJaVtFAWZZGLrI2U+QMswy2AvS7jCsqnzaVdOGxqeFzj1PJ3ckHH36cRiPFq6qmb4bDvj6W24Y+06oTHSSDLKDWDrIXBaPf0ySjr1lYTpSpYwtdyKE7bwFRk/c4fmqGP/4zS+nqCa4qwIYJGB43NFuwvmadzAd7IemB6lnoiVOm8yme2LIGhCGoktbaIoy14PTtKeaYoTll+LWPnOM9X1rgXVuuZoheStIGNB7n04QPtRq8LKiyCctH4yU+lqxzOJpjl61zfZhnk/o2hJNPZpmoppUlZ6hkXi02HF2wHNsPnvT95EIUkBOIUJOsQNwWgnbu2TyOdVTFCbmSnXxHC1rH2+3lZaojB5RoLBmW8xDn4ZQ01IyhXoHChpSXdPVxVW6H6+Sp3DGaYvPmLao/qyzbVG5YATOBJPXDmXZmqkCYZiVtqE9kxdUf6j+NEZ231gkU6bMAmo7hwHv7WBpnunEExHQMgVR9L0AoMKex4eLo2vs2TVsep/O7WUxmBv5Vy699HJ2Dzk3xmgVVmfG47dYmP/R2n++9psVD529jLhjH+F0srj5Af3kf23Ib2EzKUmLYEQVcOxayawDWGvD4muUzKU5zuasKG7cZHmooIUf3d8yJtM6RaI6uxqTzGoz0eCwdPsYHfuXHuPa67e6ZkzVM1pfPhb/fkvtVLrb19XUXT5XL5Z5RGY0y0hfnf5k2EFMrV7zcr3/X7aqbRu2sbeR+VdKBHjb6X7Oy4zqfny7LJE1Ikyw7V+c0OzvrXMZyvz5bJ/fiKnbMg4m9A8ytLXPnJ7+Cv3GCJCrTG1u6g34eiNaZWb2fcm6YKLcBL5p14OmFxd3OosjVqlY6bALVoGCzl8MjZSqpEaKYsQJ1GzPghRzGEjk3Zp7TtsWMybHLGA40juN5OQp+H92pdXZpNIQHm4YTnmWrKl+EEHjw8DpcVYaJqqGmOCuJ3MaZLbvR8wit5XfTiN3ELGB5Eo8BL+/YtiOk7Fesn1FmrHEx4zv8MgW/yHtrh9kfLzOSG8Ca0AGGPJYev0wcdLuqNzPKbA266SegJ62xFM0T+0oWCij4FVpp07lfbXwe2ziM8bqp2ykmGl08f8tOHpYxWoSqCo8rg7eZxQr+4XSdF17rc9M1va4/nHH2QKLC3nFlTGaVJQQonJFutYVuZXi7wVMNzGlorkHPoGEghnvnWvhjKbffdpLXvmqcDWMVUldhQta+zb4oC1ZlzQQmZfgFAGTgBQAEJi50vbZdtI7pk9HWtmJ8tK2AnkBCHf7y9nv589t6+I2tO3h+NaUltiaEqSkYGTFUFSMmV6LOO4LWMqxOGUZ6DIHq4/ZCUQkhnqEVWk4seLRWLH9x8Ch/c8jnv1Q3s28wpM/zXIxVrzF80cbM4rvKD1+0TZ5oTTO7/jgqXvE/uifIRXA8gj1VuHqTYXrSUu2H5RLcetKyryqmE7aNZmWrQtWZFUhVZY9cJldS+4olP9LOCta1R4ZgPLt/pA+n+LfAMXoGbwkWzkCzDhN9hrUlGNhjWDxn8SuGAcWEBZbhYsjJqYSv2cfpsQU2HZhgZCiHX7bYWeMSYuhvgy+1uZhBMWxiD9U3AlACYQ6wuSLJGUgT2NfY0HmqH7Wd+ldLZfRKtkbZzA6Yiw5ujwOt6/A1cr2KKtZw0fe+AFr7cycuTYCws71+U5/b40HjxEs97v1Cg4++o8lV1zQ4tvgAD3nddAdVaByhp3wl31m6hFGMy9vRTw67Ur6GuZZ18bOnl6E/hR3KITIwpQp8sTzMKX8WLzJZP0I5mma9cZrNXSnHn3yYN7/thfzEz77Yue3FGD6HSDpnrnQrfkuTsgKVLLG0pBqRGmkXp+daCwj4CNS7UG2x91JPb985F44JbXfmzJlnTFyarkHnr5i68XEVcHx2T+oyXW9vn+HXfukm4vgMf/IHX2LgqldxT22IagC3DLySX6wfZaU1RSnopZXfRH31fg6Vr6QRDjFKyroxXIHvNOAsKadMjr1+wAO2xfG0yYBf4JG0xTuCMocUlwvs8PJMx6uU/S62ly/lS41TvDA/xiNRgVZD/ibD9+Y9/msrptdarg0kuWJd+bB8yzBl5TYzDo+8bhyOLVmOLoiJ8vkxH1SV4mp8Z2PW5EoVoyPjbyOWbOoApypk9BmP6XiBsbDPqc9PL9/DpuIEl+fHWU0brjTZWdtk1ia80iVKNDmqZAC/Si5tEkaLDuSupi2kYydNCi+/jUSgrzXJYG6UX5/6HLvXrmSsuInxTTGrU9Zp8s3HsGoMz696pK0mRgZbwecyiHKZ9WdVD/ploGvGMThUsoLyzkiLXBMz0yN2CUKBiqolbwIuzVX4Y6uI9xN8+NOTXH7ZEMWCIZVbT8cTIBOlIsMsw+1AQzuzUWydvneAoP29ukTAoDNpe33V45GuJZiWz/Ezs7zt5w7z6n3fi5+knGqBt2SpWMPmcUP3bmjGkD5mKeyQBQe/29AlOZsQSn2wbg1FVTrZYikeDYiXp7nkgWPcVN3KfxofYl/RY52Um4uGD9UtvxFHjHmGU1huU2ZCssqP5/v5SGuKVwUD9EUBe3bFHF7FxbVJrLmaM5yYhKHtSqWOqfblGPYzQePlByHcK3epxV+CYIfBD6C6uS1novg6Y0kjSA9BsNdgNkoUDThgXV3TRFlEM5ZxxUKmUFSCzyws+vDkiuU7txuiSSiup7x41xh/9cRmPnV8Pz9w7bUU+7eiy/BH21p55zNmzWw32FNZdQczYLC6AdT/0nBRfKIAlUCcdAOFZtuuS9eP6l/1ufpMoKuT0CBQJ7ZOGbH6PXcMIbf2dtqv0+cOyImJbNt8x9q2gaW20b7aNcWNMT/nceQLDd72Y6vccr3PU3N/w1f9HQwEXayuPUKztIfh/DhXGNgQgO6FvZ4qhmQvHAurmUt+QykrNfh4Ew6p+IY1TPuWg0mTYrxCkKyx1pplQ7XIiYMP8qYf2cWv/cqr6VEZtwvsUmfYPheW3zKoU9C7sgIVKK/pQiP+XGiwi9eIy36WTp3rfz0p2sDO/d9mRN0D3/PcWBEAfCZMHWD6TMvW/ee0ra5ZWWED/UXe/cvfh7fwAT74iS9SvOxm3uD3sLvQw1v7X8b7pn6XpNmDV9yN53dxbPVeKn03s9147DS+c7fKpgyb0Om69doWywJ3WK7FZwnDI2nMLlPgAduky+SYxTKe1unPjXE0bfBH9Ul+uLiDcSVZyM7Ehp8x8JfpOsvNsgu43uvD/pblfAxr65abiobxVUNXXZUasoD853m+C/35Upo6PVa5iUdMwAieA2QH0ibYFvsE9pIaR22T+bCfvPTmGpN0Nad4MF5jb26E54UDLKQ1ZqJFQr9MzsQMqxasjVkLuhj3SkxH54niBTwXRe/hKSs22UTaOMS6xFSrLd71tc/ym3veRFoq0F9NOTVrGRfDGODco/hrepi6tFcTWqcL5pWMq64r0CUgwWLG1IhdVeKES5qQ8S5BsEkuSosqhZW7LC/z8vxxS4kGE/zO7z3GD3/XXq68ogKtNANxcqEqzk6/KQOt6j06rhggx9apjFjbXds21vreGW79r87Ot5//XmbIP/bpB6CxgRcWe7kqtPg16C0ZghiG9xryJYhPW9KSRGiti7naPGo4lk+RG33PvKG0U6+Khrjp8VA0x1tXZgh7CvxK/zADDcMSKalRCbqUu2zEAYlJC7HEDf5j3uPm8jBfXG2yubiPVxb66Q1jFs8Z58bP5wyT65YdPYbZ0zAzaimXEo6et2zNGfpK0KrBwlFL37Y2RtEbSI/Aa1af1Yo5Vb1Zxd1tydpH2Z0quebYL1W5WLOsrBvC/gxA53pg6bBlwhpai5YjD1nGJMznpfQtlNle3s6p+UeY8Zusn4SwZIkmjEsmUWk5O2czKRsBHmEeJWwIyAmoSXplgzJW2657sWvqHwE2mWj1k/pTAF3WvhMH2HGtKstWkjZ6wyu14ycdA9uOw9PLhfaTS1b7aLx0UENHzFi/49i+DOz6ice5R9b4rV9e48orc3x07na+Gpfo81PW1x4lV7qE1xS3M+GFXJ6HRgpzAYx2GVZblsk1yIstbcJ6y6KmfTTBxdUuEHFntIofL6rqLvVkjdGCx+nD+/nBmyf4X7/0A+5ZJo+Q6TCK/5wH5DNw32/J/Xrh9cntpvmZkNF48Rz/+ZmnX+9vT34JWF5edmyd2lbAvuNK7XxuNppOy7Cvr8+5Mp8JY6VzjReO8+fC5w6wK1fyXDu+l0fefz8H+6bpNVs50Qq4sTBKHwXuW/4yXUEvcTiIrR8kCYbYnR+lbFUWLmXAeFxvAvfyf9xGqMbq/rTFHq/IDuPzu0mNyDaoG4/Nxifv5Zhy8XAQ+EUORguEXouG30O/ZyUTRst6KMbtc8YS2sjZHWU0DqaZvMFSKgHbrPTTWSVSYBygkx0bDYwjEDybsN+2WDQe20xABcsLvCJftk1ytk6vl6NI6vSuFDdYzA2xkqzwVP0Iq0Ss+VVuDnsdyfFlYJMJ6MNjRO7c1jRGwsSFzQ6MpM2T2LSesXZpnTitMZTGTNkjXBbuVpolXUnK1JIqWkC/Zzgw12DsijVe/LyRDKiJPZM8SASB3p0WMwFeI4MsF6y+k4EvGhdcr8oF4cYsQ9LUjRPGHbU+cV2uyQWYXWWkO+XF149jcmnbndZmWQTUxN7IDdtxp3ZYF2E4NaTIGRlybatZ67Rs7+oHPkcPT/EdP/Fl3rz7Vby60sVKM3UKGX0Vw/g28AehuQoqcu9tgbTbEC1AsQz3NqB7CsaHDKWCz1la/O7+Wd742Tn2dm/hBeEYmzFcPmEw1nB/Az5aT+km4O60xoM24j09Xbyxp8jJVpMHai1eV+ojzBs2S9h4FCcEHAaG6XUYrhiHUebOgdcI2FCUgomhWFHlDbivARs2Qqmkuq8Grz8TGvYkQiwf4SwEA+BtB6NtVmDxMxBugCBnqCfGeTilqabHZW0W+rbjMjmXVzNibVu3obWqsBTDelrmrsYDfHYp5XtLOxjqDkjmLIEEkVUaQbF0euz2GVfFQ67hVONA1TskcSNQJTkSsYXqJ4koC/Bp1vt0p78ck5eBVPed3iOUlCGmUe5TXZsmHU+ASP9q2QFxYvi0TuND2bSa9Fm/ocQI96/H3KEWv/47yxxKPO6q3cnBOKAv6GYxnicqX87ewlZ+0PgM+4ZeH84rHlEvCMBxsaMm006WlqSA9p+srDBpLWUvz1PxAicSXaglqh+khyVmcmd50Xw/733/mxneU33OSJdkHfD3//6TQd3fP1T7re0f+uLiumdVCwgEaFbFCMXNCeCtrKx8fdY6iVMrJq2kJ+PTeHJvqE/j8/s/dWpqB5fIsCHPtl27+J33PsSD/TOEZpCdpupq9Ny59ghhukbsCV2kRNEM50q7qaURd6bNTPID4x7KRRMym6wy5eW43gvxTUDLptSNz2Hjc53JSnB/VUK+zbOsBl0s+V3s8Ra5vFLlRGQYdxJdhjkTMuZ5rKYJK0QcNh4q4X2T71EOjGO7dpeNi9GaUOxdCqqFLg+VjM3zTOhi6P7GJmz1PBczNxPPE3g5gqCfp+J5FuIFnu9XXOydPF5TXpHrcmOcjpdYqx/jEoyLrxt14sotFqPzbParFElYNwGxbeKRYNN1Z/esRLmSFRcoXsNSSGrcVT/LZcU9VOMutlQtcQqHWobDXSljM4qrG3Dgx8VNqeNVqkrhbo+DP2YydkZEjAiT7N3KFZW38xZP+m5ykZ6yFJQ96xuWz/vc0/CIhxssfeZJXvma3fROFEibKpPUZlcE6ATQxNzps5getZtK9cq6ap22vXCpc2uDOhGIqUDWx+/hs58f4Qe3Xwo9KXOh5QXjHmNiEbcLPBnmlRS8yxAIkEhbT6g2D4cXLduXPKpbfT5/dJk3fP4pPj7n85qRzbzULzOVGrq7DONVw0NzlmINxhM41YSfruZ4Q5J32m9n/Qa/OQvX5ErcEBr27Yaach7yMFwy1JZVxQIGBgwjg4qpTHgoaTGhKiNrlmK3oX8XjvVdPAWVMcgp/q8MpgG185CetARZ4Je7BrGXah5P7XMS6h4cPmLZqP3m4dwRkKOiWMpYu2bDEAXKgjacX4aZdVyJt4/WF9jkH+aHdl7H0FCBYEDafH8bs2Y2GJKFzL2r8nG6JdVPjl2dAVcOTksxgAJxAnQCY3LPS49Okz5rEhATS6twhVHFN2b1WI20BzXg9Ltys6qP9VnjQUyc2Do3RtrH1pWLvRNTm4AfeZy7v8W73lPnwfMpy807ONZKqOTGWDawqXwZl+WGGDR+FlYpICiWThImVQHGTMZFxTBmUjiZWh6IUp5IA8peyEpa48vRPL1pjXrtECGL9BVWWDo8z//407dxw6t7HaOdVc9pX+tzcPEvCuo6xv7iMgM9z9Z20E0ul6qSIOSOl7tSS61TPJpKhUmfTktPxtt7+rbHc/Ce/0cvWSyIrNOmS4vsvGSUj7/nEzQGIw6nvc4VmS9s4snlL1HWW3LQg9c6i+9XyRc2OuA27oXcZVOu1Iu/SmZ5BTabgOW0gcqJ5YzPF9OECS/HOJ6TFXmNCTgv1ywpl+jBneQo2xVeF1SYTQzHMC7W5nK3vecA3mPJuiv4XbOGJ02M4JRpeeQjS2KMAzRVD+oipZTUp3gz36PLGlZsxD3JGjM2dm/+QbxIkEbszI/z5WSFwLZ4gd/FFX6eFZMj8itO5+7haA4vXkKB+TVb50waMesFrqJG7Bcp2YhapOKgCcaT7oWyGXIO5MmKWs+n3DjCp9IcV/Xscy5cYywNa/ilZp0b8iu8ZO+AA2fOKCu+TjFZYrYUSL6j7VKTC1bGejZjiSRU7KlumrIlD0kM1rjams2ZlGJ/yFNry8zWI46dX+bG4RyX7BtxcXtGbKAMttgbTWLpxNC0bbjTHnOMTZvp0TYCA1onPCFDr+oMOZ+PfPYUt/zU3bz2uldTa+T4zFzMt3f5TFQMpRcYlznamhNIMvhilnRNSjrQIT1YOm04OVXnk+uT/OjBVb6ta4yfqoy6OM2dcqkDX2it8Lm5lNUoZKERMb7JcPPlHjcO+lR8wx/NJkwup04n8eq8x0TRuPg2FVZQ6c/qCHRvE2AzzEbWaeINdBu+Nm3ZGHoUcrC4AL01Q7UEjxyzhBsN/S2DRIPFyLUmM5eyv0Vl1TKgJzeoAHKoJAtlad4FA72GsydhtQZBJSPQ5C5fXIHVNcvpBPpjw8ElWI4sR02BWRNz5NS97KpeypUbeh2o8sS4CVjp95cyAWTV/DUSt1ayhFyuqgKiPhRA29ROpBC4F8BTdqySYATwdBzNbbetFQOsa9iU/YbL4FUyjKYLs1/bv+/Wqc8dgs1elty2nnUadH7qceauJu+6ZYU/rLRYqd/BpO0mF/bTlawyUd5H6Fe50QS8PFQ1GOOYaunQDfVDcdzgz0JPYFiLs+H4UArdiXFs/gM25rHWDN3xEs2m4rSbjBRnmTx0lA/+2Q/whjdsQ/eTe4a1LyO7mOfe339RUPfca77n7hULsMr12gF0+XyeC2eBPLkzO8D2udtSz6Arl0F32WKWvXu7GNtc5SPvu4Nmf5OlNKSUm2AwWeXo+oOE4RD4ZZr1A3SVLsEPeh0rkqNFxUacNp4T593khXw4WeUKE/BCP8dcWmfaJrS8kGsCn6oJXeKEXCqpV2Bn0M1HV89yy6hPJV9lvp464doNBSiLZUsMn02b9KXrHPVy/HViuT1q0RenLFjLYWPZUZDMhMd0Ctv8zBt1KIUn0iarwDbjM2BCbLLK8XiJYb/sWLgz0QILxtCwMWPGp+zlOGXkJg1JVT9WtWBbU7TiJaxfZCLoY9QYV55ImMgPegj8ClH9ODZewqiQp2isdN25ZnN+nmbtEYppH3vLW5kow1MxfGK+zkv3WW6sdLuMy3aBSmdApVvmYukU29Up+6QYrpMZA+NiqxSYL1ecshfPWFK5zHphqMtj5VSLv1pZYrDf8MRnjvDab9tJ12DehUtoZEpE2BlqGUIBAP3bMd6dpVaL/dEk9k7/+8oY9ZmZXuOXfuMvmVy4ipf1bKOQS3lzv8e2usfQTsj5WV3ahuTylrMSb1aJHWIehwytlsehh2Z53f33cudSwvf2TPCduX62+ikrSRbDv2Aj7m0u0p96jHlFXlSBG3f4bNruuTq+d5ywfLxm6TUBl2O4pge6B4yTRxnphq7NkEi8WSCmBGdrUGgacjmPj89Zxj3LZs9nfg2GBwxBGYbGDLXHoLpPzzlc9QiHdXTeSipQgoF05HqzChBqkig1VIcNYQGWz8JID/TtgZ6NkC8aJMlRCw2LNWguQs2HJ6wYZZ/FtMGZ+EF2rpd4yY27nKyMr1p5PW33ppjTBoix68RBWv0vdqtPADkrK+cEprWf+k6u1U4yhGLjBMr1QqCYPMWsKRZPsXYaA1qvWfsJ7GvWoJa8i0BhZ71Ania9ACZkgC70mDu4zrt/e5VPdSX0tO5kIe0lH1Ro6MWutJPtqixjfF4V5hxDLb1mhYPmPA1rQ9yynFrJqkToVFXD9SGbcMaIYEyZjucpJaus1Q7jmTrDJcvJg1/hve9/E29+0/Oz7NnnYKZruze+YXER1H1Dc1z855tpAT1INGkp4CZg1wF4+tyJTbsI6L6Z1nx6beNcezaTGLj6inEmdpT48z+4i9WBNUiKVEr7mIzmMI2jLrAoTdfoNyGt/GZWU+WZBpyUILHxXCjYorXExjidKVVr2Oq0InDs3OeSJkPGZ8KEsj+ctLFzAcep5VPpMq/KdfNCE3AkTjniJ+QTjy1F2JMPuCPKO5fObuPxcuPsFB8yhi+ncKyZcCJO6HNxOx466hfSBJUBGySgSMgp2ySPlP17sWEPT6UNXhF00RN0c0asgCR55JJOVlxeb2QTNpgQkxtgxa/gt6ZoNk45GZPYr7DFL9Prl5xun+98cjlsPI+Rq1o0TZsMk/zLw0t3UfQHGPa3smAN84spE5euctNwrwMdGYXR3iGAZL91+mmUjRrOudMETlxsnZObUHWGTLrB9Bq83eDnDPZIikeJ989P0RumHJzLsb1S56rrxzByr6Vtl66kTWTAHavTdt/JiHdmfSy05Tt06xd030Nj2vCzv3onH/nYGb5z78s40vR5XSHk6pwh1wc9AjcHLVFsKCjeLA/5XvD7fVZjw50P1Pmtu6f4ybtbXDI+yr+pTOATUvHgdJplOB/A8m4bs6KkGs/jP7+yxLXXGBdIf3TK8hePJgxGhhsT467pOtUO9o3LOC16UBrMMlBjlR6bhEoZvAXDibqlOgGzC4bDs56LhdtSNFRVqS2FoAGTs3K1QnePIZ62To/OE6iTO1OgTjV4xVxKX+54pjEXKht2DXpFiI7jqlcEdfBD4/B9Xxny64bAh939BoUNBDXLMYocTqYYm4p5+QuuoKc/A0xOK09uU2XdihWU61cly+QSXsmSJlRX1cneCKApvq3ZlmQRxSkQLvSklwK5hJUYcartdhUT2EmY6AA6DVUBNk06lsDdhUBO22lSXJ8v1RvF0DV41+9Nc9/JhKZ9kBONCN8vUwy6KBa30ecVOWdKvNQLXAWYHr0Ipdk9W1BmsQBxExqKr7N6+bI8hnGyJevEHG7NkSbLtFrnyJkWI+U6kwfu4AMfejO3/OBNWTjCczTTtd0b37C4COq+oTku/vPNtkAHsHWAXWfZYec6y2/2eBe3e/q0QAbssoy4yy/d6PTF/uqDD2KGAp6wA06gN1p7ELwAP+hnfv0xIr9KM+znkI24zsuTGo87bYs5Yq43ece2nbapS57oxfLXpFzm5fmihdf6PluCkC9FK7zYyzOU62P/esRabZGryt3sKsBsy3C6YSgVYYMXUG5EfCGe5ZxXYMwvYbyA1CaktsWC8dnlBZyThllc47CN2OLliB1kMBwmITU+iVei4hcYwLjYP2XuCrhJniT083hpA+sFnLdNhr0SW4JurG0SEbDud1G3DZaaZ1hvTlIlII9h0fikyRLGKzpCSxIvzkRaSZ7kqGOdxtrdy/fS8ofZW5ng3pVVtmxe4iXDQ1liwli7fqeMubLJc4bmaQh3ZOyWE/+VMK2YGEmczLQD5+UiPGYJVJVAIOYpCGYN3cbjM62EoQ1lPnrrAb7/u7cyuKlAWk8xoZeVvhI4ERunWfF1jrX7W/esY3zittiwXugCj7s/eIR3vvvjbL/yVZyKRvneXMAVZc9VaBjb2Y4zE5DwoGvUkN/kkfQaHr5/jZ+9o87Pfy7lymaJd27u455GiYrxucoPKBUN00nKf0siHm5F/JCB/7WxxJauPLuGPCrjhslTlgfOWG7wDZf2GoqR4bodhlwAh5Zg30bjMll9gbRugzcA9TVYnZGUSSYoXEgMy4sw1YJNJUNFGoktZd7C8gps3maoH9L+mRvVFsHfnFWZUMax2kRsmYSh64vgbcgyNpsnoHCJobmEA7/pMnhiu1R1YgWnm5dMPJCwAAAgAElEQVQG0GpCd2hIVhRfGbLQOsWdayf49v5L2VotYodS4jMGT+BMbS42VszdbNuN2iMga2gdzLQKxbQZATX5xgX61acCdQJi6gcdR2NKxxGbpxrAAqfKeBaQ67B5GgNi9TRwO0uBP/2vuR1apxi6ucca/MxvrvC+/fOY/JOcTCoU0jWC8l6C/Jh7YXueX+TfqYqKZ1lOLadSsXBZ6bBt5QzMzTTgqFsvYBczYjzHvu2X8HmyTlw/jOoxD1ZqnDzwNd7z+6/nlje/4iKgy4bGN/z9VwZ1yorU657GWRbhq3DsC9cJDGgQpi5II9vu/9i6b2iKi//8U1rA9ZXu9XbyhI7RWfdPOd7FfZ4eLaA+1D2r5dWXTrBlKODP/uQ+tg1XqZthUr+CrT0OXhnPCwmiaYYrVzOftthmmwR+VeITjJuAfgyTaZ3DtkmFgEv8giOBJBz8ItXnJHFqELeTZ1DZqdK9C0I+0ZqjL03oL3TTjCyDHvRJ3mA9pdsP2SPskzRYcy5c62L9rpOj03g8lUacTqVhlo3HRy0ci1c5hiXv5V32rcqZiTmTrluvWD0H+ix3G58XSYBY0iVegapXIu+JMVzgZNrku8J+9vg5bNBHMT/GctDDfLzIdP0IJppxgM7F1vkVB+SElKzL2MusYoOYXs/ngZW7WQn7sfEYOzZOc9OOcRcXJ4kKF/AuhkTxqNKXkyaaXGmiNWRgZXjF3pyw2KDNosnYLmTuQk8MVbvO7IauKn9wpkFPPmF1JmbjhoQbnj/iyrI57TnF7wkA6PeyxzWm5Dnj/3W2TnFa+s5YvMDngQcmeeEv/gH9fTtYzV/Pf81XeU3ZZ3sPLiYqV4B6DcJeQ2HEZ3k+Zf9cnT9+cpHvv7VBNQn49dEqN3bn2J0zfKXW4oj18NMmf11f5BOtlHdU4Jd357nZ5tne51FseDx0yBJNW+ZnYaQGg6FhOYbR6yGnOq0B1GdVjsxQX4V8NctIrZ2FspiuAg74qa6vCMoj61Bswr7AsKSar42MVewezIBcXMNp0pWVyarMHQEjubtVwkygrQpn7oNSl6G0IwOyNjQ0Z63rKzF+DkgJWKWg8xCWWsgZqgmUcoZKLyzXDCdjj8Prt/H6cA8bVUC2K8Uq87UBnkBZbzsbWWydWNsEkuPWAUcH/DTWlXAgUK5ZtlVCw+o3ydWczkqteVe14+zEyilGs+1KdbF5GqIaZwJzGg9igjXLRasYTx3W03Ua5h6L+JlfW+SDp46yufAk01GOnJd3gM7mN7paz1vxWfU8ej3YH2cxroeUg2GMq9U8nMB8C/arrJo1XOUZIhNw1MbcmSw76SArcfJkiZFKxMkDd/OB3/y33PLWb3OC151n1NPjqfn0OIt/PVDXAXFtY++MvrvmbwyadyPuwm00EDUK/9XXPT064OJZXGyBp2sLfB3YBYbLL93CSDXmQ3/+MGMbRmh5W4giVU444qooNFvTxF6BNBwgZwLmbYNl43OJl6NL7te0To2UTX4XD2B5vqsNa+jBco+1/HFquN7znUt0WFyYKRCbgE81ptmSlNkT5tkfpw7w7c0ZSiE8Guc54AU0k3WWk3XHuHUZn8dtxLF03WGfK7wKVZPjVhsxazw2moCTacMZlZ1ejh14hMYygeHaHE6Y+BICF3t3O6nzhI7ZlIpXwE9qzCTLRGmTczZ20ia9YR9jXoFC2Of2Ma0zWLVLoug9GUBZcx8TdGduWAV2kdJAxcg9Vur38WQ9ZM9QLy+4bJxwxiIXqhGA0KQ4sAr4qkYhN1u/wZ6zWfS9kiMkSiwRXwGWPARKRJCGnQfBOJRTyMVwbsHn7toq42PDfPxjB3j1K8bYsLmETaQK13a9CghknmJZ2AwM6MuOgVfOse/zwEOnue7NfwpnI64dfxX/T3mCF3VZ1mSgJVMSGAbHDPmix/n+iNuPLvPTd53nPz/U4vzZKj+0rcRPdBXZkjOkaymzNbGs83y8dpT5oMoLgZ8ud/HDO/Ls6groEuskQen5lD9csmxJDTfdYCgo41lyKKNQ7IazZxVYDwW1kZJjxKyJJUshlNZcQ7pnUJu39JWMq/hwZA42B8qYNny1Dl1DllHfkIjRqkKlBIvz4CsuTjIymhSzOQjpJER3WvwQiqrbu2pYOWRZm4NVSdZcnbGU5x/K+iXcbFiqQ7gMRVXTkFJN0VCoGtaWPaZbAfcv3sPzRi/nEn8Yvzel0Oc5Vk1xjpolQaLyYM7tq3M84Sr5OZewWDq72Nac08aKIxSbqzEh7uRI1o5yIbsSYwKKAoeyt2LpBOb0wqBEjLVMusXtKMBv2kk4Ok7dsP+OiP/+2yf4wPxBtthHOGlHCf0KYXEbzdwQTePxJk8H9Bw23JB6TjKo6uy7YYtc7MZSTsVuZyVllcQkfvuOtMYXWmcJmwrEXMS0zjFSWmTy0CO85z9/N7e8/duQvM9FQJcNx7/7958N6v7Bhm0DOhvXuPdLn+OTn/kiM6sxm7duJIjXuePzn+Ezt91OzZbYsmGY/4+994CS7KrPfX97n1CxuzrnnpnuyaMZaTSjNBLBgAwi2o+cTDAGbJyu7ef0ru1n/DBc32sDXg6AwY9g+5KMDCYJESSBkFCekTQ5z/TMdE6V65yz913/fbox5oGX5uIL5r6utXq6u6a6umpX9dnf+f5foFXhzh/Wdbd9jToF1o2I/8aden73mqx9v7YCayuwsgLfBnahYu/OjWwwMR++9TaGB9toBVcSR5PY1jm8oJ96/TC+38XFoJcpFZDTWYoYzqAp6oAtXp7HSXhFqFlnFCctDCnNbnHKKsWgMhy3MV+Ny1hTYd4rujSGupnllcOd9GU87ihblpRiVwauDERob5hNlsmbKudNlXtNjSXT5Mlega26QNW2eNC2uF4FvEgH9CuPs6ZOlw44gudctKLDEqg151nOJYY25TGsPW7RATVr+CoxY2jK2sNHcyZeZI6Ek6bOjGm4+5INetbUCLMbUH6P64A1jVPYaAprayjJfBA9Ib54WR3ySlDUydOf+TJ3HBjg5U+7iuE9PvFZg5aMMmFHVmIrItHSnQfZJwWwmSkZy4K/XTsw53ReklsnhzQBeKKvOpOO60TjFZ3x+ae4xjqvysxUTNC2xDNuHE2bNURbJ3u7WIUFg4rRQhZEHq4ACRkdhpK55vHA/vNc95MfgWqVPdufy/MzV7JLQFMM7SVNe1ETXmM43RHxkXvn+cj90/zhacWWbIFCR5HfLIU8IwhojxUP1SHJGm6tJ7yrVeYabfnFYISrbJ6d3ZrRHsXcYpoxd3bWkm8plgzERdhYUxQ3QLGUdqvqPsiJQ7QF/QOK88csFaUoSpuDkF2SydaE+iVoH1ZkM3Bh3vKexZhhT7MzpxgqwDHgikC524oztzIFeT8dmbr6MHGfip6tnI66JSaluCHVsMX7IdIu/YduuU1Foj4EoEMtC62FtCotm4Ny2bKwBJ0dioVLlm+UFX8eL9GoHeSMjnn1tq3OMe76U+UJlBWJmCXyoGXkLucMAthKkjO3kj0nxglpk5DEoSWQiEIxQgg76casfaC3rwB3YRDlNZbxq5hx5LUW8CagXt6ech8C7ERXJ6xkTaFji5VQ6Ntb7P1PZ3kkfJzh5CznbSfDXo5abiueX2JMZ1y8kbjSpRJ2O5oNGFcTeBzr3OzCzkqdXdYqkSZSVoYPWcMF4I5ohpKp0opm8EyVwWKZM0cP8LdveSE/9389y42gvyfuWDtquxWQl++yLrKYcmk1m8SJ5B2l41MJH129rH5192fex03P/Ckmps7wgmfexK23389dn/swT3/ei5iZPsXT9+3insfOcO8XP/hDu2529jRPu2EnX33wlDt9WR0Frz72tc9rK7C2Av96Bb4N7DrgNb/+E7znLc/g9OP30JuDfPdPpdEdyYJrVIiX7iTfukCJxJnmvmoTfKXYrkIuWOlebbEUW+dqE7nPHJZTGHah2O5LVVBAO6J3yxKbJn1hL3eoDt5x8Ry5esxLC5r5wHJ3y7AcRa4ySkrdj6qQKaSkPcTXOWe8kDiUzwuAw7JVB+SUYgnDPr/EG3TAm8EFHAtwkci0sxGcs9aJ8pumQSVepg3N03SWHu2xEU1D5yAzSiYcYDDscwzGA0mZo61LBK0LmKSGDvvxchvx229CZzY4RiGOp7CtiyAdUoLKvBJW5zAitrJbgc9w512f48TZWVTVQ8lYVdiWVir09/vTaA0j4zLZ2LcqtCygtmnv6jT/Uh0lIbmXwFzCjWZzecWOLZrXBb0kjSrrr+ziPX+1zIFHJiHQLjjaoUHZ1N3HCrgTQOenejDhWx548DTXvfqDoJtkNu7meeEutvsZYjFNeJoFv86Xa1M85/45fuH2WX7vPMz6vbyovZ0bMiXeFWTZqjUTdct9JuGL9Yh/rrbYkY25tTjAM/UWNkmvb4dluAsuNuDOc5ZvTKdGESESrxPgKgyftPjJNGcIcteLY1UR9yvinGj7LcNXKNoqcGHOUjWQ6QITQu8VEAzAvFzXlEgWixBwCZZmC5rTinnfMtALjSmoXoTcRmjOS2mF3D7VMFYfsrSmobhHuUoviZ6RWrdcF3SJQ1VMzxJu3JY6cbOCVmagespSX4SgoGhqxfSy5ZFFS5eCn/JzUOjnyNGzTDYbNCvaBReLCULAmTweIwBNgJtEk8jYXECYADIZkcr/idWoCclS6mx24cWrTSSio5PbyAhfOmU75I2/Mm4VgCdAsZq+51yQsQsVXhm5ivXX19z2tYjrfuss67YcYMjOcUH1sSe/nfHCLprKZ1DnyasAbeqcTsouukRq3R6zMd+S4gsUG7VyRm7fU1RDuNsq/tEkzCVVJkzNPbV6UseLZ+nOLHH68P2888bn8spf/cnUbSxtEe4s5F8fp9a+S1fgCTN1An5kIQ/cexvvfNeHuPPL/8h7PnEHNzz56egLD/D7f/VlnvbUPc4p5v7YbMTH/+BJXPfbX+Ydv/HzrC9/lvcea+PC+9/NK9/zaX7vl95M6exHuWNxkIN/915e9N8+yu//EK6T39t+4q18wzyV518vkaLpqHftDbG2Amsr8P1XYBXY6TbFnr3jjCU+H/7MbYyMbKYVbCeq7kd6VyW+I2lN09IZCvESYbxAXYcE0jShA24UB6lNCJxmL03p+KI15JWinCi60c61esA2CaQISPmO4ftieZIuHXFVZxcDiXX1QT/TaPCoTbjO7+LaoI11KmS/mCCEbLARraTi9qmneu2OfFjA0ItlJx4lYQY1TnP3qE73yAHZcJxDV/El02QJy1FrechU2aXgKi/nRsmy10U2cs/XJMuEXgEdzZIkVcfMSaKqNWWX06czw/hhP1kBqXLia2pY6ZVygjgpQm1RtlkG2tr41ENf4S++8Bi3XDdGb9zrNGLOQiybsFaYGRHqp+yZY1CEvauudJLKBu829/Q6x9BI1ZUwbD4UlmHpTI6HdINaPMtyo5uMPchTr9xIpug5Nu7b8SbyK+RJxhad8WjOKu7//HH2/eIHYLqfYNN69tpxtobDPJTUOZdMck9U5Zfm4LOLS0wvZLgiKvKktiJPVxk2qoB2nZI/hxYtt9fLPKISXj8S8OodGZ5yRcDmMc3RuvR5atpymvkFmFiwTLbgp/coskYxV4Ytw4pkLo0c6disqE8o13lbO2Xx+5RrqIgvQHabIgwgaSoKw9A+oNCidRONXAItYfUCxX1lzUAE7WE6wm61pGtUuYqzagRDN+HuR7SHrUnwPMX0/fL+g77nyH2mL6V08EpHrKTYRA2IFqGyBIsyHhWXqifVV24ohZVQ6bqYMSxzwq55imJD4WnLbbWTFHSdV3dfhx8FhAVLpk+ALHhlXMC0c7MKABO9es+KnlJGpmKSkPeAsJIzoK9QWHHKynmEjFyFwhHQLsyffC0COQF58r283s49+x21YBIOLMC3pTBVxYdvXeCl7z6K338PC/UpyiaHCodoDwdZEO1s0EVBpe0sMQm3+KKbkzpARVFrAqPYl1dsChSfq8Gihglt+UwccympsZAsU5a/ndYkqn6E/vwS548/yF/e/CLe8v5bCLesjFyFSV67fN8VeMKgLp2oKg4+8AU+czrD3/75O2iffpj98z7X7d5BV2cnI0O9KwaZVE155bN+nWdcvZFD3/o8t7zxXbz1N9/MY986zNN/6sVsG+tnYX6SX79ritkvTvO6t7yULRt+ONfNzS3zxwdy/J/Pv8ZNGlJLz/ddo7X/WFuBtRVwxEhqnhBgd9V1GxhoGv7+M19neHwnzaSIOGKVCLyTJdeusKg86qZK1cYctk1aVvynCRek7SFpscXLuJw3mQB93MYIXqkp353Fi87tsGlQ0xkiUyUndVq105xqhbTFJbeh3aADfrmQYcyDmUhxt1L06iwbkgoT0RxlyZoLutipPTdGvVvAmPLYpxVnLCxJ+4SGdaE8HjiXwE06JTHGVMiQn+OmQFFJYo5by/0mpmGblLx2eq3h8aRKm9eB1JA1SFw7BX47VmiSpO7MESZeIKsztPtdVL02l12nxEChc9ik5uaF0kRRNgGj7XliFnjfgfuYPGy4enCI0oaca3ZQOYtfUpjzYoSQVLJ0E5bjckPCfMdWpruSZbfK4vSspKksQ1SHU+egi4DPV0+wcWSIL31xlqeMBGy6oRvTMml2mYx3xSAjkcdaM3WhzO+++2F+4f/+NH57O9396yjXqiyGA0zHc67d4wsqz36bYW9Q5AWZDgayeXYEvouvkX6NKLFMNOEuY+nbkfDTY1leG4ZcvdensDENJNbacnBZ8Q9Hlev1FTKqw0BnAoEArS7FwHZFrluRNOD8smVkvcKLLNVHxSSRmjilQldaElqz1gE7P4HJQ5ZJCbbVkCmB1IXlE4g2wjsWYq6aha0DmoL0v8aKi3XLRM2y+VpFaSAFgMsyrr0EC0csHe2Kzn0KX8woAqbFjdoCGZHrosIsK4JxRaZfIXEmYnCRGiyJSpGxqOpxhk78hiKIFENFRalPMTMfs781QaInqS9sYvfmEr19lkYVsjLi7V/R1Z22KBmZb1JpXZzTQtq0D1jMEgloqRST0lRZj/WumiXNmxMwKOyegDj5w1sFeKszOwF28uEpjBgifKmiM/z9/TVe/9aj9AzdTSXWXBMOsatDmOj1HEPxgswAQypETpxu0gF7VYaGyiBOdyERRXSwW6UMneQPHrcwZxVnoohJsRDHC8ybCro1hTZl+ooLnD26n/e88ad487tuwR9b09A90U1IFBSXdYltnp979l4H9IfXbeXC0gJ+vpPr9gqPu6JRSxEgbUWfT7zvHbzsF/+Qf/7GIZ7/pA188cUvpmrSAW1l+iI3Dm0lQ5mKpEPKScYP47rJCfb0SUzl2mVtBdZW4HJWYJWxk8DTN/32zWRLHq//o9sYv3IfF5MbaS4JsOskqR8mCHrxMmMor905U2eiaaaCHqcVs3GVZjTtQFsc9OBFc9yqfHajeSwpk3g5itEipjXHlPyh+iWGs5u5tXLWxU9szQyyyQhIVIwpSymAvPE40Irxwn6qtkEZy4Jp8kkbOyduny4grsfTMlZUcLIFDwJPbsG4ToNxz3rSO2ld72qLBAGmT/UKzJuIc2g26gwP2Sbnlc9GL8/JeImcqeMJmDOx086psA+bVB1wM/Ec1drRFNDFs26W6kKJvTa0ymKiS9h4wYHgiVaGYn6U7tY8f3vsvUz/xgGuetMN7O6+lhvDboxKMLUm2at8gtCnbR4ksqMgNmDZpFfCg1k0KRMjFZkyVtPQtkXx5EctdqnEm/Ru/mbhEcZ2bOOWj5/l2NPWs3lTSLSUOKF8S3vU4yZfufcYH/mnr/DFzx2gd3w9zajIbPk8ueLV2Ow6HkmqbFGBi6y5JsySwWNU2CNr+EZiKdFkIfbYVNL8dIfm59o043t8wgVQG8H0Qdy0LJ2AYBrmDmtGZfwq/bo+tLdDW7eiPA1V6U8VJ2sn9BTh5EG4+A3L4IZ0ROxJ/pwAHzERFEGfB3vG0iaNCTOW4pQlI1lxLYU3kDpklw9ZSsuWb2QSqhcVOwNFLsB1v46LRm8IYjEOZODSectDZy1bhmHrEPjC+C1DS9ynRrmMXi1mjA5IYgTbp0UO0tIwCcvzFpOHkmjyRhXZrGVuDqIwbZVo1qGU8+irFzlTU3zAHOMXFkfxxqByFsRNnNmOA1oCxOIZixpQ+LLews7Ja++tVHuJHk4ux0CJhk6SfuV9IBcxTcg2LSJQ2XKF3ZOLkMfyHhJwJ3o8qXGT7OzI8oG7Grz594+yYeM9TFZjusL1bC5exTG/A2UD3uS3u9zJPcrSjs8j1vBMiUPWynkvtlqYNnBIov0SS5tVHAbuMzU0hiVTI0kW0Y0JNBH9hUVOH7qPD/7Wy3jNr/wkengN0KUv0hP797JBnbGGRkM4XGg26mi/DRM3OHF6ik2b1yOtjFa0Dlg++Re/z8t+7Z382YduZUMp4sjhC9z4pqfyzlu/xDWjAX/yux/izbd+Da/0OH/40du4eiTzQ7nuz//gk7zsH385ZaBlrLxG5z6xd8vardZW4DsZu0HFa3/1aTRUwi+89Zts2HUTlySGoHoG7RdJKg87ZbcydWzQTTbodro7IQKiIMM90TQBPjcBw34HXzIRh03FVX0VBa6Fvc7ZWkzKdOMx65dYn2nw/9ZO8bthngG/xCO1mIsYdvqaF+c8dhbgW0twbX6Mg0mZT0YRIzZipjVJLezjqqCfL5iYp2ifp4imT6JOgF5gXMEBiV0AF3y6V3k83DJMkDClNL0q4IjL2vNZr+X7PHebFg9Fc7S7EW/isu8kKy8WtV80i/La8MIMNlnESO6dbaDks3TESpadOB8kHsY0UMIo6gEa2R76g3Y+Gx/gs38jSqSr+W1eQGtsA0uD8ODdecbaizxnMmLHCDSvlrotD7/QxGY1QZJ1urJYmEhxzc4lJA1FvT2mslCmOw4hruOJ6OvADLf/yd34z38ak/cqLp2Fu/Qp7pq/jQMPHHTGjqHxq5htdpOYMmFmmLh4FZ61rPeLTr94n1Hs9GNytsV/bYQYr8lzMXTZgGd0+mzbqBjphTiSSA1LbQ6W2qF9EWrzMLnfMjiuuKLD8uWpiD1ByEIEGzcoAjEktKC0M2WWkjmL30xxSaVqsZOK3AD4orWTWrgWeGIgyCjsosRnwfhuzcXbLeemZNpo6Qsh2wX+kmWsoVjW8LCLx4G+BHoCyCwpjj8E6xJLJUhH2KM9UMzC+fOWMV+WUBG0KVrzFltUBOLSzUHYsKiWcq0Rqmwl9QeJ7xEZmBHt47LF61aE1nK4bllXgKGCRKoY7p+IkeTmFwSGWQk8vhq69gj7ZzHu7CbVw+kVV689jQNtqiMNR04EOUnAswQjb1UiYku1d3LkklGxbN0C6ISlk0wXAXjCNspn4VrKMhq2zowxf1Jz66emefPfPcr6kce5UI1IsjvpzG/lK6qdzUaz3ssihp8DNmYXvjM9PNlpW5XLlxSo8GAsmdnWuV89qziBQTIrRa4hulVPKM54Cd/z6MnMcfrwN3nP776E1/76TzpWc80UcXnbzmWDukwmTy5IfyzIZCgF7cyffYy3v/9e/ua//gqhK8QRinyB++86wA233MKX/uGv+Y3XfYW3/Okn+c+/+34ee+2z2Lrtd3jJL7+dFz/raahr+/nqi1/I1m3/+Ydy3St+7U94xc03pCu1Jri8vHfM2q3XVuA7gJ0wdm/+zZvJdmpe/5/uZWzHdi6YKlFzAaWamPpBdOFqTDTvQE0mHEJLhp2puwiEcUmdl5GryvAqLYCqwONxmTnlu9SFTX6IlyyzZGNUvETe7+aSyvBHi/t5Wv5KNgQdbJHw4AjiXstPeB57NZQblpLXRhbFUVujQw2Rt/DVeIFJYeBsgZYOXbVU1Vokx26DkBWu51yx2ZN4FVzMyUOR5xx8My6GBK6T/DqrXFNFLBRR2EPLy1NQnSwmIqBSFERDJ/pBGbGaJkaOi5I7EkdYoU1shJU5omiIpHHCyHXyvbTeb2bWDNBpNpLPHaLBOf7Evp0XxvtondvIeO4arqloJttzPHrRZ/RYRMm3PFAzHGzC9pwhcIyIpREr7pHvZeRcsMw0hb4zBEEXJ+YeYmC8n1/6/L288dNbKQzM8d/LjzBdPepu07dhHKOKTMV9BJ5B2TrN/A6uVz7nsZwVW2XSoEDEyXqRjQre2akZtnm6Q+jshN4snL4IjxyCjnHwD+HMDK3DkOmFQg9svEq5CqzeGEq+YaOALsEdJ9LxcTAChXZYngF5+HIO3l+Co00Y64NQMvRk+c6LzdWClNQLcxdJ72waJ5Ibg0fvtgz1Q3FSIUHCfk6zCcWIhkPW8FGb8Grr0xPCNesV+RDma1CvW4bGFbsKiqmLlgtlWH+1wl9Mma/WKWl3sIS7Fea0xZezg5a4hSGWx9uRav2kgSFRlubjkN0IfaOw8XjqaC52KJLphLw21HQfWsWcNTB1GIauENZsJXJE2LV6ep+il5SWCWd8kBy6c6nRQcw03uiKC1ZcrvIh2jn5vDJMcwYJAXhC1AnYkzWUrxOLzmtOHbO8423n+cCjdzK67gLnWx6bizeyFPRzSRKUvTae4Re4wio+iWUXqcnIJ+I5fsgRAyeqsODFHLSKQNzjpklehZywlsdMzZ0kqWiGqHnOxQr1hac4c+RxPvjuV/HaN9ycBmuvNUVc9n6j7Kqd9Qn+aBLHjokTVG0SGWsoPKWcEzZYAXvpXVmiKMIKPbySOSeur8D3SKIW1XqLQlvRhVnL7U3UovLDuq5YdEXSK1PiJ/jM1262tgJrK/DdK7B6Fi1C8vd95A5+4Q13sn7bCJOzB2lFTZQcITIb0dlxUZ3RHQ5gvTaWtE+g8/gqZNkvsQnL85QEhhgetYbTSZUztuE6WptJhcV4kZy1+DpPojS+bbAcL/Gzbde4BgthYIoZxTatnHPy0bdCwmsAACAASURBVEXrgmU3KMWCjflcElHRORd7IkxZOwlfVlk2KZ8bdJYlFDc7dx5MSkyX5JxZ2OjDx1rWbf4iV5NMrTEFNWM5guEuE7NeZzgi3Z22iVivsirkEVOjzRrq8RJxc4LQL8iREhsvYZN5rIttqaVmi0hsjWKXrKXHytwVyDYo8zEZ33rN83ShmDLiiqizLShRCEf52c1Pp2BKtNHN3o4ByoHm6/OGvT50BcqJ6k9WcPlxMl0bzsFDDcuno5iZeJ6j85/DT+q0eWV6vCmOt6QyIkd/uI6aNITEInfPurDmlqli/E5GinuoKY+iynJTkGdP4JGJUlYun1HsG7NU5sB0OSMn9hQcnoNGC3aFEHnWmTI68orcqIzVSLVn0jRwEd7xWMLPZjRFmQ4GUq8FnaJh82FqBnwLwkAu1i0PV+HqEcXOHvAFUYoCaBGMZLUJPpFy+OOW/BVpdt3BOywPnrBct07RU4AzC/C16TTU+jFreC+WN+FxldSL5eHsBAw+BZ5fVBTbU/2ejIKnJ2H0+YpMO8THRQ5mibWisC9lEwWjC6PYkv7ZRYs+Bt4GaMxA61JqaBFWrzAEDz1qOX0J+kLNdFLmJZe+QcGWyXoZfqXj+bxqSLNhV2q0kDGqFkPEoIQMr4BYeVOK21kMF8ICNi1cqdB9K4HBMoqV8awsiIxghU2UrwXcpQO39Ocl21AyCxPNA0eXeMGfHWPy0AOMDi4x1exCibtVTsiw3BBIq4zvchq7xdnuIkkEqFnWW3hqXrOQwCdaltgm7PE9clZxe9LicSu8nqUSzVOOl9HKQ8cT9PinOHf0Ad7z16/gzdIUIW0da4Duuw+3T+j7ywZ1T+he/40bffcLtYopRauzevlhXfedv3P1d699/v/nCsh7bvV9990rIO+TtffKd6/Kv3wv6ybrIxv0h//7t3j9q7/G2BbLpblzNGMfJSPH7BZ0bhtWzBLZTSv9SqIH6qCkPKZ1jmuVx4Dfzj02TpspdMgZFH2mTrlxnn6vyLwKnA7PNzW6TJPzNuHn8lu5Sdy2vqEsIMGDTq2oNCyJgWs2C0qDu+YNb6w32KYs62T0g6YgUFOFHEnq7NEBL9Ahh2zMsO8RGsVxI6PYhC40gyjZT118120m4bw7pVUO8E0mVbq1h9J5PBux30acjivs0lkyNuahaMqxhsY0seKAFQAnCbe24TR17rp4zvVIqew4SiJPZKcWBi9Zctq7vJgXVEjFsYEx1ERAJum7V/GS/Ba6POWiOrozY7yweydDnT4nFowDMJNoPnHmCEQnuITlgok4VTuEap7G89qIlSHw+8mE61yBuhIWLugitDG1pMrW3FZKhe3Eush6pQlNzLgXOBBtW4rHtWGLAIUEPtZI2Bl4bECxNwuXilBLYEhaJsJ0IhjEipnEMtqm0BtS3BHV4KPHLHNL8HIP8g3F2U5Ln4aeknKautYUNDPQ4cP+ZcPxTMTzejOOVYsk6qWMG3tKzIqwrsIzZAZxfbIzZ+DrlyzVuuGpV3p8thnz8MGYfcUMjxlDr1J41nDRWl4Y+vRJBlwetkifrhgrdFo1Jn2slQwM3KhQMiIdUJIM45zG4nI1R8QIYWmEkO9zRC1axvrCsjWhlUBooBrDA6cspxuW54z5fOtUmZef/QybskVONGJuu+pmntTRgdpkXByKgLHck6QqTrC3ws6mcTa2AvFJiyf9vGKYGVkxQYheTs4NZGsVYOccryvu6FWgJ4DQiMnGouY19903ww3vuQNOHWFw1DJvN+O17SWr81zj5ej22l37SsZa1gEnFOxxWNqKdBE5HenNg0yRjwmIFMLQU9ybGB6Iy7RLNqONqbYuUmteIB8U6NaPcObofj74d2/gta96shtTrx5T/uUos/bVE12Byx6/fr87/l4vwv93k0w3x3+5/l82yx/Vdd/v+axd/7//Cqy+Z8vlMjMzM//mE+7t7aWtTbIl1i7fvQIpoEuB3WtfeQPNVsjP/+xnGN7YTqtSZqbmoRsnMVIplhmFpOLChbXfgfXytOs8ysY8IDuAaTnm7EVa8zmXX6cxEhmS38zploSS1tFenkhiszLDDMeLfKB+gjP5rbwu8Zg3hmxLUfCskDYur25P09LRpti1oPgzP8PDWL5sYrYojx2u8l6jdcii0vy1iTiCYjxKkAlWFp8NSnPJGpd2P6408woHCmtWjBSKibjMRq/g2irebRrsUQFP0lnCwGfaKuRnJEplJl4kq1oQDmJNnUQKXaX6QEfpaFZGsjKOTSoOAAuz5wLHbB8qu5l67XGIJl0Fm0S2tJdKzigSxdN8cv4MCF0qG7fKcs9SjxPeN+qWMJQqNo8DS3Npaq2SklAfz+vEhoNuNCzAVgCnbLg6sw7ld+HpDLV4njAc5sriHk7hcQUJO5TPSWEDTcLhouL60GNTBYIAhvsVb895qEVFpqXo8WBTH0RLkN0DuXGFJ/EfEuVmFIHo44Q9aoNcFW7uhsc+YxnfrGgfga2SxSdxIEOKsACVr1q8XZANFPtqmt6vBeT2QWcO4naF7k5Htq5vVSJfxAEqYb0VmNMwmlfcGUTce1pxGzX6sqGLaxMmc9Eqlqzi75otdnVqfmLU4/Rpy6kErtgCtRkIPVxWnQC8849YCjXoGrDOqNCSWrReyfWzSJyhe2mbivkZ61ovQokmWUiNIVEEMwvQU4b2NkXNKo5UG5AtQTgEjSN4psbp6U4qvmXPeo3pX5HAnZXRsZXzoTS+5CR4Asw2ifNXXv+VUa1o5WSHFwOEPEG5CJMnblkZxcrI1rcu6iVZ1nzlIzPc8oHPwuAUfcM+82o7mcJWlvHYqjLs8Dpc4cS6VEzgIn8EAO9Es4xyptk2DXc34VGpMMO42q9vxXVua1UZ8gtcEClC44zTkHZ7TQJ7H2eOVnjvB352DdCtvEQ/6Kd/N1D3vZiM73WdPODvdf2P6rofdAHXfv7HdwXkPSfATgBdR0cHmUzmX1H+q6Cv2Wy62xSL0uX5L4zyj+8z//d/5KtrKZ/f+Lo9dOV7eenLb4eehxgqWS4uxujGcaxUickIJuhNt4bmRSbCfho6y1alndA6wLDNy/OAjfGtdQ0PQrad9NvY36zQZVo0lOZC8xL57CglG3OscZJKdiNbJUqj13BxDh5upD3miw3FdGTJxPDMosf1HvQueC70WLRhVRKkVeJhPK7PaH4GuK1hXMr9AjEt6zmG7gELU0jhuMTDGUTClDMRoQq4A8uQqXO1jZlyMQ4F18B01EZMmbrrv+1SIWUZ63ptJC7nTt5vETozkhol8DBGJCtSRVDDy23DMOAAoI7m8TpvcRtiUr4bo/MsCDUqLKhXpDOQOCnf6fgkNuZRoSgXmynIa8mubujK9NDSo0RkSGS0KyBQuseSZYyrHWihknm8/LPwgj7akwp1Qv5Lx1ZubMtxqppQwafXwk0mZFTqzHpT1ilbUuS7ISiBHtUYAVKnU3CTeFB9HNq0Qrdg7qylvaTokAwbeWgSjisCRmBrEUYHoacDWiVoG1N4C6nhwdQhvzfVz4lurD1QqEAztwD91yq8aZeL7DLownaHW7FLylVo2S7YnAevE67N5Ji93XLkbM4xsn09sG9B8/WmZUdW8SfZgPfPJvRULbsLPhdnLPl2mJ+EwZyiLQ8dTXhswXLdzYqwrGhNWbQwY8Lu7VPOEBFKnIiG/kBRn7c0O9Mu2pzUvmkY65RxqeLgMcvd5xOquTOwEOOLgUEyAsOG0yYuPmhpjloyDUXzZOrudaYGGVHKyLVq0VuUYwxdsayAZFlXAXW1Fau3aOokIFkqHuTctA6mbvHymun5hL/62zP80Qe/AePn6Wl2UAm2EoRDLOsCz/Lb6fYKnMYwajWRUjy6ItOT6e9pBbtXJrsSFfRgYomVop+ET0URndYyliwyaZYJonka8TIDmSoXqg/BhMeHP/pmfublooxdG7n+exyZ/6dBXZIkaxvcv8crsHYfP5IVWAVscZwKS7LZrHs/a70a2JQ+LCPMT1aOiCBfe57MLdYu32sFUmAnQnbLS142ymfVS3jvKw2fT+5ndL3H+dklbHU/On+FI5R00Jd2KkUzjn1a9Ap0KMOSzvIlmxCimVOwTdL+XeNEt8tPe6RxiqLfha8DyrVjbMhtpGKq/FHlJO8c3Mx5YQqalu1KMaDhyCVLlw/9WcVCw9KdgWf5cJeLWvDcuPT2pMZ9OkvQklDiLNcow1ZPU8bjn+MWBRUwoqwLN+7Bc27ZkpYgZc1nkgYvVr7rvJUYsKJpcCxZdvdzHYpTKqBLJSRhF3ONiy6gWEaeXnY9cf2Ea6Cw8aKLXXE5f5FBKsYIhykUdmBbU9RVJp2itT/ZoZWkej/WISGN1UWW3HxNfLvp+zPnBXh+gURchiKgCgZYcmp4OW4LgrIyzMWaprs/YQ6FnbPxPEnjOIWgj0viPsiOsjubZzBrGMpolkXjJm5QGUFKNVgGshJu3J6yQjJVlocgTRRuSlwDLUHAV6WaMnF85nMgtV5iVpAnJdhSAIiMTkW71spY156QHUpz2eKMQmpAM1ulA1dRPy9FGBabV3R2wKmDlk0jCj+ymB7lWCtnGhDEIZW78tikn3WdhBGDtwyVouVA06c/CxcqsLkIuyKN7k+r666YtvjdlmUfzmYtf3N/wi92+bQ1rOucNQpGQoUQpAUZH2eVizNRUysdvc00dkScu5E0r0hm3QBEByFbgFYV5iSHT2JDEsPRjoQFqZqIPCJZwFw3S+XEOXPbOhXnHrYMdoDphe7NCnvJkkzIeFjh3aRwtRhLwhiuGB+EiZP4EtflupJNJ/o60eIJespYvFgzvb/Ob9/6NT70T/vpW29pNoaoFXZR8wSd53ida00JXUezYE0xsEhCSoBlHEUomroA6glcSCzfQnGSxLl921VA2VSp2RYLUsXXnHB/L6P5hDOH7uYlrOfXPvYz3PDSzhVA970Jn+91rFm77vuvwGWDOjE/XLx4kXp9lc/9/ne+9j9rK/AffQUE3AUyN5JphKddNMR3PmYBKqvAT267dvm3V0CITCu7hjI876XtXL/9FfzmH8OHP34vG8YDzk1VsI2TWOU5w4SWIN6gh6A1zZTOkMjo0bY44Xc6IXZiLYcwbNUZp/Hu9DvZlt/ORLyEsoasZMWZJjcF/ZSt5f0zZ7gy3MA6X1M01lWZ2W746jx0N60LQJUNaCi0vM73uM8YHq4FjNPJPlru+181dYpK8UeeZj5S7PZ8np3T3F9LGEbxkLEcJ3Z5W2ed4tzjkrWubeJhEzEgz800WTYNSip12J6yLazRBLbFZp1lFs0CCdovYcVN6CIeclhP6gaaqKRC1DxLf9iPCvo5whTt0SyRtHF03JxWkS18jqb2seKylZ+T6BhR6UvQsyt/9Vxcig6H3P9rMWe4NgvJEEtPVIgXIRzBti64n5WGAG/5m8x5XfxEYZznBx1MV2GpArm8ZaoFV2Whu1sx3YRSLPVZUBZgJ25MKcmQ8Nsl8CX8WMaCMlUcg6gsujPlJoxWwF9Hyro5bCqjRA3VKcvZWdizA3zJs2uzzvWaEWf0eZksK+Shy3g10ZbBjTBzUjE9Yxm5QiEkp3Szulwa2aKkg1VuP58WMOga1PbDgxcS9nuWF8S+izHZ26vcuP6k0LAdsDGvGBrRnFuyPNkqRoY8p888mIH2eUtPQyJJ0uoufTXE8nwkomXRoiQzcWvK1snYOemGygVLh5hXmun6TEsUSdYyXYO2IGBsapk/m2sw2DFMIzMCrSXORDG7M9C5A8rHFcmNip5+S/wt6zwOnsger5M8npVmCHHIirnZXSwsyt9hCmydlVwKTTyLnJcapbn3vhrvfvudfGLha4yvKzGlriUobUJkBVfpkNfkhmkaqcYz9FtxkaeT3fMC7Axs9aBDFAIWvizh3Va6m1s0MCzahIWkSh3LbPM8oc6jjWUkd4LThx7kNS+/kj/9ndfQe1VRQtDcMWNtCLL62v1gny8b1AmgE2BXKpUcc/GD/fq1n15bgR/tCgj7JuNVAW9aiRD9X+s8v/v7H+2j/fH47e7gbLXrLu3dVeRP/+IV0BHw4fd9lZEtmoXpKap1hS87vowh3WFdkTcxQjbQOI3NZ8l4RfqV5ZJpccIGXK00iU3Ypgt4usXBaIq8VyJrWxyuHaQjHOa4l+WryTT/jy6x0cvxcD3hmf2aLVXL0SW4J7TcHClaBvr64DkZzRXn4E5lOVMP2actPaJtQ/OFyDrtnvAIuYalYjX9Cl4dwBcS3zkxxbWbSOirSjhhLFO2xSAhoQ45Ekv3ZZXOoJsxXaRuKlz08sypgIa4/kRbKGsgzlcHDk0Ktrwup6tT0RQXq4/j53eQ0XmWhdmsHycn8S757dQkLqX8TbxkwRkvdO4KdNCdVpHFi26kK7u6iWddNp4OR5zb1sWoSOuFrL3oGt04vImNpNBJ0+Z301j8Cs8rvoWdfgalIsZ7NYvLlp71iq4CXDprKWUVMxWL9hSNWrrJT58Ddc7StQ1ymwW0KaIFsBLSK4X0G9KgZC0MmlA/E9ZVeLiGuXkY2Kg4dg2UF6EhzJ1kvK1L40lalyytDkumKD246Vg1U4DximVqAsy4hO+uABlh+WRVpSJLwN0MNM5YamXINmBd5PGpDdAbKx6etDSr0C4xJx4seorR9XDhuCGS6i+luLZLsaQsncOKaD4FqI9VLItzUH0EhmPo3pKGEEsPrbBoQQbCHcr1lcZdoM6nDKSwhXmRznmKhbp148zZ1pJ0kGG8dmIxx8QzPLzQw0u3Q7aVNlJ4NXH1KmZOGLxxRd8zZNyb9sK6UasYE4Spk7y5Nnn2K4YJIXALK/l4Ep0yq/iHuy/y2t/+EvQ8xtCGbUzojZS8EjOmwfOzo6zXBRasYshLweilSFosoM8qLgTw00XNxUXroktOWcHQCeeI3ImM/A2EeMwkNUJTJzRNssrQ03aKU4/dw2vedCN/9rZX0NNbdBIM7U4CfzyObT8Oj/IJ14TJk5GR69TU1LcF46sb3trnFAisrcOP3zoI+yZMnJykyGhVxq+rr+PqH7AAPzFTrN5m9fq1z//GCri9VSGd0YVihifduI2lWj93fPE0qq9Cu6lRa0YoYZriMjrocc3xnpJYhSUEVg350kmgXH/rsApcBVEHloLSjOkcVR2QJGXqOkPL5Ugs0+u140WTHEzqbAhK7PVTB+jWNri2G041YCYxbCkpVwsl+2dvEUqyIUollYK7koi7TYO9ZBz7VpZgWgnaF/yhFHPK8qjVTid3pRLwKp2xAUXl83Q8rtQ+eeXxJOAMiWMS+728A09LSYWKjV3Ug59IiJmo1YWlErZYLJTyfQOlfJQwc14bkYQU+0WeH/SS8XJMyCjYa+OFuS3Ug15m4jKB6OLEjeuVUnexwDMBcWJCEepIGDqpQHC9VgKwpNFdJq8dLgTZddImyy4U2XoFIm1YrJR50vAWrujwmK1bhjcqenqksxTmlmFMWDetKPZCJoDC+lQrlh9TZGTCewqWz0KSTRsYpL3AzFhmHgVvUBFklRsFKtGhCS6WjyzcdRqCQ9AmzRAtCGJoVFKGT6L9PAEsSvLUlIsIEbLx2OOWMKPo2KgQZlZJrIiwdlVoLVhqj6UxIKL1C8YUM5cs6zKKoeH0NvMVRV8XjPYo107id8OjB2ChoAhrFs9XtHUpogupobTrasWmNjCLIKblllHO7do+rJAe2qw85hBOT0GXgMoKtCYkZkXiVSAnsoAzlqpRMnHl7fMnmNECKksYIhrVszQ7+3j93n46xZksY+kITh+1+EvQs0/jda02iazo58T6XVwxSciLK27b7ApbKvlzoebCccvf/Jf7+MX3f5qBkQk6M7tY9jYRBT1UvAKbMqPsEvYcS1sgeFtTi+GUSdvnpO6rU8O0APQGnLKW2y3cY2qcN00KotG0EUUJ4K4dJjFNOsIchgeZPHyCN7zl/+C/ve0F9PTk3bFByxngCv78N44ma/91GStwWaBO7ndubo4wDN3Gdxm/Z+2mayvwH3IFBLDJyUp7e/u3Ad0qsBPAJx+roE5us6apu7yXUQCyZFUW8gE379vAT3Ru4kOfOkSjtkxfV4NyRYrv/XRsKGypgA2/m4wYEUwN5RXY7OXpVppDNnKOuy7l04FxPa91mcXJBqJCml6BuWSJrM5xySZUkwXW+SGhCukQ3VYWwqbik3GL3SVNfUbjGCKJmNCKDSWFrsOysWzWAZNKc9QmLspETByjKOYtvC0qM2ENLR24ntoJlNOny97UoXxXYN6hNC2do+gVGFS+A6c5HZLz2nlp0ElVeVzyi2SFVlHBCsiScayo52WjE22nxjNVbGadk0fdEHThB90cUaELOb426OCS38GUVSSIyWEhLRVVOXTQ4dbOnaAEA3iZUQfwTFJGiVbOgUdx3cqsUsCk78Cf09R5Rdpz2zhV38+EGmS3HnBMS6uiiM6DGAuElVIxdJXS4npxfbaOQEFaHcTQKzv1gCL2oehDIPq5GCQKT7R2zUXIzkN8xKLPSSZzKuD3pCnhQcgtwdCmNJhXyMykSxEOy2xf4a0Tc0A64pVe1FDAn1Y0mtAtTJqMQhtQfcjidynKs5DpkVw8RX6PIvZg+gJ0idRQalGHFY/MWbwaVBrKhQbPXILRbuX6YrNtikVj+eyFdLq5KVRkpJGiR9EWpdErfTcrB0gv7YfF05a4XzkzSF4IuARMFsoz0NaZGlLLZVwzRrv2+PTSRT5ePsRAZpCalyfUOeqJZTxT5hVdo0STmpqBc9NQlry/l2t8EbaJIULMGcLE5Wya0yfvHXnriLRS8JVwzpJBN6t54PAiz3333Xzino/St0Hc6tfTldvFhMqwK+zjVWEPz9Ohm7GesRZh0B400G7SfEapkJ2RoG/RzlUTHlIijYgJTJUzpkmbqSMnLRJhFEdzThvYG0wzfeJ+mrNT/P3bfp5f+a3r6ez23TFBXrO1y7//ClwWqJPNbX5+nrwkM65d1lbgf4MVWGXqvhPUfefT+k5Qt8bUfefKPPGvHbAT7WIOxvd189xnb+P9d81SPXuGseGY+XIaUuziO9DosMf10khDepeXYxEZ0RoGVUCf8sgrzUM2Zj2WddrjDtm/bILwAUTTRDpLp9Ycjat8Lo54mtZYqc6yUgem2GZ8TlaV22wlGsP5CEKLlyiqVVjneQR4SK/CU5SmZi2tlcivUMCKbTixuECpBVOjR+c4h3VMxZLyOIClQcIDNkGeSVEFTq9+TMJWSejzCkxKpIg4gQXw2RY6nndsnNuG3UhaNuYCSXSJUHssZtdzwjSp6hwdSrnnPhnNscNvc7EtO/O7XPix17rkjBgqHERLP5Vs0liMFKWHfQ40Sves1JSlrRYVB6pdsWMkkSfCSULstdEf9HNw8X526VHabSe2keBpJfgPaTcrdab4QZoXBEOEo2BEw9aATEm5XtfCehlJKsdm+W0QbFYUhCXdrlzshxbnqzCVocLOgCeaM9/yyTMRw7HnQFCrABlhzHpTTZcuSt1Y6ubM7pb+U0X5Ijw+YemTaJUVs4KwbZlNisw6RTIP/lZQHXD6TktYgz4pxJCQ5GHF8DA8JiPdaeveA1JfNrRLuZiWIAvTLag14KOLlh3j0LEgU4mUDTt6DhZC2PwkRXuPopETBYGl+RgURM9XS5spinnJg0u9QfJ4WtKBWk742OS9nAzyhF6J2MvjJUs04gu01zK8anyU9k6P/Q+kzOTunYrMjRor6yxuVol8WXGzIsHE8kKIqVkyf1wVrGZ2SvH+T5znlX/8cWbnvsnIuqup+U+llhnlosrw4uwIW708V2jfSVAkCeY4lmVLajbKwo05RU6nrJ0w12e05VBSdic690m8jtJE0oFsKlg8WqbButwE54/cC+vh/s+/hVtetgUZl8sxda2a84kfPy/3lpcF6oTBqFarToMkbJ2APHmB1j7W1uDH9T2wytRJBt0qCycgRC7ynOTimLpKmVJ7OqJ1V679c1krkAI7YREswyOdPO8p23n/+0IWF/YzMhBTbcQYcWUKCDEtrG0SBL3MSXOEaVLTeZqmTqf2KCvJfpOAfOXitrpsTME2mDB1MipDy9SoKp9ODDm/jVuNYQTFs8Yy+DJWSmB7VvE1IawCEHPk0XkoBBJVAr0Z2OQprrPw9STmfhKaeJzButKCuo14hmyCyidQAYFS9KGoKI9jVjoY0gYKiYAQOHrGtuiXtgwxgMQLHIxmaaiQgpbypMRVKNXjRfe1PCAJGLYyCk3KeGE/zcZpNogTVudYVh5N5buWh3MuY8+4EapkyvVnRrhgY0w8jW5dRMao4jDWXjY1QXgFxGUrbJw4b3U47IKPHZCziRuBOwdssohtXSQOuohsjYnWSW7s2Up/JoeXscQziouz0CaSvinws9IDjmOPhFXz+hTBOpmqrbBI4oDdoFD9ikQAV0cay2EknmNAoQXgyWshjNsoVNssdx5LeHqbT7UCZ89bsiE0mxDOihnBSc5coPHShLB9sJiHoAa5TkX3Tyv0euUmzrVHIZTwYHnrSYBwDBcP4hzQfsaZgqlcgPY9ii4Z7dYVqgaxyD19RUcRLh6FE2W4vk+xIWc4W7GMWI2uQHZEkRlXnIthrEO5yJTOQUV3p6ISwqXz1qWISOiuRJkkGZDq3fOHDZWmzyO1S/zl8jEGClsoJ3VnevFtRFMntJ+f5dV7N5M0AwauhE3CMooBReJaOqTW5Dv0cwJyBb8nCtOQUatFNzwuHG/we5+8h3d87BO0jSzSXrqOeW8rcX6c7qCLl/idjHpZFsX0geHrqd+Da9AMyfvawrU9UGvCPXXDNJZPJ2UuuRMsKCfL7mG4aBx5ueMFtK3T7z3K2SP3Adfz4NdezTXXrcPqNF5l9fh6WQeQtRs/4RW4bKPE0NCQc78uLi5+e9N7wr9t7YZrK/AfbAUEsK3KCVZB3Hd+Xv3a/8cJUwAAIABJREFUsQn/wR77j9vDSQ0UqS3imt3dnDj+cm7/aB9v+YO/ZOO6FkstzXzS7UT9KujDSICvlycjDQzRLLXMMGcx+FZRQ3HWVFi0hrcGJaa8LAejBUKdYVjGnKbFY0mDNuXTb1q8qzXBuaPDvN4vuGiTcw1h06DZgkfFsSmLqaHTU7QalkhbklDx2pzPc03CP1ZafC7RLGnDTLzEhANqIaHXRg+aaWUZRNNhE+5NapxQiqLyOGJa9GrH77HB76BbZ7gzmsWTlgzRs3tFsipLVRid+kmU3FY0dKH0V0jfquDcEcpLX+c5PS/kfmVdqOtmFTLgFbnD1MmaKoPWkAl6uKLjZpp+GyeW7iKzfAdxa5Kg9FT83BhWQINo7MJsSrUJH6qzDjzidxJmx8l6Bcp+B0nlIVqISL6bQ7Xz/M7M1/nV7mfTaTU7Zahn4GwNekLFtIys5yyls4rONsjvg+oRCOuW+jLkd6SNC7LgAtoc+BD8LoWhBTBxyqDJ6FVeg9yMYtNyyHkxLhgoFRUH7rUsNmCwB+YbMN6nGO4Fuw3ar1d0luH+b6YgauC0IhiW3A0F46CG0nMFMQQvicbtWOrMrWVw49PqEuTOQV5A2no4NAvVOrQtQF1YQd+SyabTzauNxxerluXQEqBoXbAU22GvhbmvW7qeqrAyGi3B8JMVnevg2J0y0k9H0EPSeHHB0ul71PobfOrhxx3F2RAnt99BUQszLUvUzeTINGcfsex+TvocOZ3mBtsLAoYdYezqtNz4tU25/ls5afK0JqnCx+6Y5b1/+CXu5luMjQywGNzAdGaTazi5wWtjnwrpUYpvWuMIPw8P8aLu9hRXWpDMuUvK8q0pxRngn+VxWcM6NBUbMR2XWRZ2V+wOOkdianTYCUrqNMePHeE9b3sVz37ZzayXAhlh5wTkr01c/5cfti8b1En8w/r169diHv6XvzRrv+CHsQJikhBHd6PRcHl0chYpGrvVixyMxB0rF2Gq1y4/4Aq443oa+rxxk+bnf+dm9HDAz7/hY05ZPzAisRmj+MpL403CYWpIWLEPrUnmvRw9zv2aY59XYCqp8IWkwqBX4jlBD19JyvSLHkjn0M6QYCmTMGwNn2pN0Yp93qgGCWyGp2UN2sAx2XCBby4IQ2ddflh/Ap3dllxLM9jw6Mx5jPsJj1erHJSWdgWHlaFhY56ts/g6wxgeee0zrAr8nWm52jMZMz1qmpxVHrcIO6cy9AXdzMvkrDVDFcVVXtFFlUz6HWT9bky8gPJLeCqDaV0gxDKH5pGlr9Pe9TweMA12e9rpCruUohYtO5CWNS06dYApXssJApq1x8hFE8TikC1cidJFtN+FVsJXCpLKuFYOEy1iM4PE0TT12CfMbKAZL5E0JyiHQwx3PoUTCw9wpNbBy654EnbA8s0pQxApbsor5s5Yl3NsE8vmBPL3WZYuwXQDhiSoVzIvlhX+eBqBIoSN6oZyUdG2bMlLXEkkZgjIHIGshmwOPnzC8sxQsWEIRsYUVwo4a4E/mLJuZhI6alAXllVbF+Z74AGoTFg62xTBRlybhvDtqt065D5z0uIXIROlk0p/GBLpY61DdlC58ejmMfhqy3L3suVVszBSVPRouO+iZWNGsb4dDk7D3n4YHIYliSdJUgDqAJYcLiQ25KQlaxRX7rBMSh7eATh70rL+OjEvaN766FHubHoM5EeZixfxnUEhIdAZKvWDjDdjNj5L0XGNwsxb7ATo65Sjk+28RUmUiVBsNZmaW5QAp4bH8Qsxt9/6ML902xegf57R7E4uettoBl3sDHsZ1G3klERVW45YyyZXjyYSSMWC5CpKfp62DshljXa5c1+TkG4bc9G0iJIa9WSBGdMkjOaRcX4UV+jTZ7g4+RjT1YD3/u3redNrnuJG9Q7QrUw/fsCjx9qPP4EVuGxQt3qfvv8//aOrd7H2eW0FfiQrsHqQkc8ycu3s7HQGIHkwct3/YO89wCQ763PP3/edULk65+7pyUGj0UijLJRMMBiDwaQ1BmNsMPbCsrD3rh+v9+6ujdf3Ltf22sa+DgjWC1rAGIOJAhMEQjlOkCbnme7pnCue+O3z/2rGxgQbceEioernkXq6urrqnK9OnXrP+3/Dt35dGhXIfRzttK442yeob12iH+jfsq7ijNWe4lfffBtX9m3hg7/9ST546NNs3xBythETOcMQTlnRldZFqkkNk+SouzGLTsIrJazY6eCzaY390SIvdTu5XefYLxlv0jGrfeuOdWRkmR1nMJzm8+EC03GT/9A5TsnL4TUNGzzDIwm2NzZNpQEA4gSemgYva9CBYdxV/Krns9/JMYHLCZPncq8DZQK+lFTZLkyZzjGEyzrt8oeexos1Z0zKuPJYI+HhpM4m7VEyCQPKY1G0ddE8DTmevG4bnXEpBNi6USVex+209V3SwXownGRL5SFGCns4H69Sdjts9teL/EH2Beetti7rDzKkM7yifD0X/D4eX/kmfniBOJhEF/agMyPgSkCc5C8uENp5aZ6iRGmIREy0fRJ3ohyUiOCSNebdDkY69vAXJ55g1/XdvOWyy9leinHWK4JEUTkOkTRAxDaVg+YadHeCdxlkZmH+PkPgG7KT4OxW0tplTRYCRASkOeOKJBZQAmaTsk7a8RVDjwcvugJCMTp0Y92escR6SFCxzN/XQ/2YId5vrIM0F0FpAE4uwNXiahUGNpTcOWXB+/wBw1MPG35m/cVxrMzdexQDVxt8ydTrgeA4dF2puLIGXziaMjmrGLwKCn2toGTRx/WiuCAXe1WoXYCSPI6wexIxKDJGMSiI9FJki/2tGJfeJfCvMtydGGoLHieaZ7nz1AOMDOxkEQG2BUw4b6vqyl6XHaNm8j3kxhxbdZYuG/TOi/ElwmjO0KIOJcR5yaBFCOc5PHxmnhs/eA+cu5fxsR7W9M1MZLfR4/URCLurc2xRGjE9CFO9YgxF25AivLAQpYqsSfm/Ywn/VmxNlq3L/HQa4kTLZLXHhLzj42VyyidxirimyUb/CEeOPsJbd17NW/+v17PnZX1Wc2jf421DxA90jvxB/+hpaer+rSf59g/Ef+v+7d+3V+CZsAIyfs3lcjaEWC5WhI2W26RJolAo2Aox+S5ApM3W/fBeMVlPwdCONoxuK3PTz1/OXKPM3V99nHzHEjm3h4bUJUnKqoT32g+dlEQpiR2ziQ0V7TGiHJbTmCIJXTrDkPYJJNg3jfDcss22LZmUhnLY6vezN17i74J5OiPY5RfpLiiejAy1RFlgc86IiB3k8zQjLVpdsBTDZgMHY9iS7eKlJZ8n6zWOOFmGlcuYUnwpbTBiQsrKQdL070siBrVmRLmMoyhrlyUZs5mARhpw1ikigLMSzlLH4DoF0nDGulRFU2hDiQXYCaxN1ix7OV/dR14pVrIb6EhjurUwlxmKbtlmjE2Es4x53WzCaZkdnCwTaYNY++hoDmOaNloljhfJ4XJtdh2+OH1R5N2ynYxKzISEPiuTYNKKbbSI3F56Sx4fP3uOawsbWT+Qwxs1ZLQip6BLdHHSMDELw6OKnnWK4VQxeJlicFQxPgRFqfWqQ6FTUehVdNk+11Ysh3AEOX1RizegqEXQ0VQM7mxl4C1IGku3FC0oUqPEq2JBg1MDN76olZuD/pIiSqA0ppCu1ficrJ1t0eX8fYbMhKR+KAqbWv2zqluRlWgQ0f0tGpIKzJ2UKBbDYIei1KsY6AIvlUo4xd5Zw2iqOFCH4bIEJCtWpdc2B8tzthCCUtrS8+lRJXnQVB831I9C95DBbbrc8fAcv3PhS/R2DVLJjBE7BSQL0TUB2aRGgYQ11cFw1uONLxii0OdYp6vV0UkAs4ywm7JfBiKFPqeZPRByx93Hed1XPoKe3Mv68T2c928k44/yqswIPU6Z1zh5XKU4ZAxnbBIj7JFR7cVq2DyGZRORUZqYxLLJDyWrLKcNCuIu15I+p3CCSft+TNGUxekSPcTkqeO86b+/mT/4wBvYek0XEk0kIeRth+sP73z5/T7SDxXUyUm6/V97DZ5Nx4C8UWR7hbG7BOikA/Zb/xOQJ2Du0n59v2+u9v3+7RW4pLNLxCVa8rnlxsu4ubyT+z/7FFOVh1nfr1lqSDqWOPqk3DNHmqyJFIsTyiUnDlTlM6YcTpByrRAYcruTI58GnBerJtL4EFHUGcs+JFJvZDT3BDPU4jrVuJOduNyYN5xMDLOJsmkRUjp/IYXbBxTVhoz4lK1tmk40V/keL/SzdBhopNCjPTrRfDMNuEfcrIhj1mOzcqxxQj5Me3Cs0Fzuu6ZzbFaabu0zLaAMRWC1g4s2jLjV+GCsltBE83YcqyyDUyNuHKbbKfKkP0zeBOxRHru0zwVlOB0t0KFcRtwigZgm/GG6/WFcUpbiBVxpjkgqSJOHpANvcvLsdrupSHCsSW2IrLhycUvWXGEkJsW6i5u42qeBw/lDB3nZNZspV7I0FxOcjCKVyBFrPFCws1VHFZ40qEg8wq0YPj+FQi/4MjaUfN26wSyDzirCCYjFOJFK/DFMTMKhowbvHCzXJFZIke/CxpKkIuNaMjYjThoqhFgMO6FnDFsq//VDhoJkua1vtUu4ecXKFFQPGDZvVZQkTy8jpoZWTl6yYmykSmMejGTOXYB6DgYcxcbLYO6c6AOhqwsGfcWFZcOII9sFgyXpn4WFaisXb/+qYYtkxckBK2ZUiUmR+JIsZD3N5EH4TO1BTuoV8rkrCN0yOm1SjZeomIiS30dnZoT55jxdhZRfuXmMgqTGOa1KMEHdZta6GtCLmkRpvvLUIr/79c/xZ4/+A125QRoDN+FmrqInM8Ks18cb3RIdiA4VditNn9L2QkemxOKklqQY2xKBSBUMBySTkZSptEkmnLYGpcTrxQSTRI2j1l0kHcXrsqeZmniA8YWID733bbz1N19Ib3/OXhAI49ceaPzb578fxT1+qKDuR7GB7cdsr8CPcgUEqMmXfBfgJsDuEsCTf8ttbUD3o3wFWh9+wg4J05/Pa7bd3M+NL7uMD9yhWFnaz/iApNn71JMUuZ+E6aYScpoGFP1Bm7HmELNO55nB2Dy7TRLjITlxynDMxAxon6aJuJBWyWqfXr/X9s0+Ei3y1eY5tuscHW5e0hcQgDnmCnMCnzSQXYN1sXwow0Sk6Y2hELYAhE41N+CwaFKWlUegM7xIw2dCQR6tCLGSdlFGsYbhKq04hmaehGFhP0zIRl1gm/Y4FC1RNBGxOFTFxCCzPundEpZNnKs26EXK0l2W6kcYcMuc87opm5is8q3DtqDgocYZ1nSWjU6B0IQ0nAKB20tTojKiaSvw15khHK+bIxddwkFaZUS20ymyqDPEtnqsgIpWSFOpZAtpGk1/to8jLDB37Aw3bN5AT3+WZCVF1xVGMub6FaZiMIsKb6OyTQ0SIyKuEGmQ8NYrq51zBsEdU7Y2TNbJkXDdnLIOWHcOFmLFE/cbtvUoukJFcx4yK63MN6nkWpyA5SkodoAj4cd5xeyi9LAq8mtQmTEMSEhxpeUJ2XsyJZiC9ePa1prpgrzfwcwYklhcsIqJBSjmoGu3YnBcsXrecCpRbBxXtotW8u7yF7MOK014XHL6YkWvByuR4ZwwhLMwsANyvcpSycJEe0OQcxQzy/DxR+/ng07IaOlmG5jtkRKKVMAt4zklqtrDaJ/a2gzX9eV5za3D5MIUU9JSWkIyZ9AnwRlwOL6a8veff5A3fOH/5djMIdaN7ER1voit2V1sFIZa57lWu/aCYYNWZJVCWtCGlMgLFSVhZ6VNDGUjeY7aLldh6VLuS2sUoiULBEWDacf3ybLtEi47EX3+Sc4cPQzRNr706Bv4qdfuIJ9vyVPkPWpB7Y/4tNF++O++Am1Q993XpX3rc2gFLjFwl4Ddpe+XwNyl78+hJfmx7Kqsu5U0SuzJSBe/9Kbr2D7eycc+/jka1BguZ6jV6xjJnRMgYEKrQZq3LROe1bLNkHAWzf0SlE7MTVoiQgSAGSbTgO1KAnJDTiRrOKJv054dh/5jOM39qcOwLjGgHDq0oW5gg6xES4tOlBjuiGJe4mtbADERwkYXelzFmHG4zlOsmoQYqX7y6JKKMxNxIl5l1LJyUoPu0i9d68qxMSgNE9v2jEh7TDkFYq+HUPt40o+rJbFXasSqtvJLSY2duFetMcNnrfIQ406Z416XdQj3OiVrfjiUrLIQLTLrFulQGdu/uaw0TbeHLqdoq5xsjEm8gm8MkzpjNWLbnCJXaBnCGTJSHaYMgYzp0qZl90R1VVc+o8Ud3H9+keX+aW5aP06p5GMyKUL/SKerERlkACoUp3PL7apWrOQLEyjiBYNTlfw2he4BXTNEM62GLycVU4BUbBn2zkEjD/U6jJUVK3VYPgNLq7C6YuiUWA/PrpCYPsl2SluDwisopqYMG/a0emJP7DXc9Wm4bgw6OiBYhHoVClcpTEnChoWSNWRXIDMGbq7FjC3VoVgzDG5Uttt1cRqmFgw93YpAasE0hJGh7kB3Cc414HwDduxRuDIeHlTWwOBNa+Yq8LtP3M8fz59nR9d1hG4vs2mTRrREV3aU+OK6u2nD9qmunD7BH7xsgKs29pFKrojgpFmD03Son1E8cOos1z70Sb74zc8wNnglfudVLGQvp+AP4zkZrlCOHe13Kc2oxLlYICfu1ZaW7ggtbeAMMY+bkCUxz6R1DqHxSak0z1HVRVwnR9I8a2fdji7S604y29jLypmT/OUfvJb3/dVrufzyHlsJKC+2vIfbXz/eFWiDuh/v+ref/Rm0ApdOSPL9W//9DNrEn/hNsRf5opnC0N2tuPrazWzbto3mXXPsnX2c3tFOCxaCuIESEBOcxUtjq/eRSq66zrBdOdbJN05iWbodyrfjpAVitHIIdYbARCTxCjmJTcltoEu0Z8E009EKZWFKTI4FqZBTKeschdS7PpJENmBYwiwSoxC9uowUi7qVcdfnKq7XwsK53JXE7FIOVzlF9qV1jkSLlin8bBrTYYGmwwkTMW3LvVrMyQIJm7RLh1O0Y1phFFV2zDKRMjK1rJ3MLVXGauOU20VYP0gezUG3h5rTyrHb5has+WI2WmaGiDnl0BmvcIt2+dncNsZ0gf3RHH68aOvFdFKxLNE5kzBnQnq1AOQsqyQsOiVSXUCFkyjpnE0qVIzL+sERvnHfYeYKazxv+xhF42I6pKXgop5MdGopSBacE2Fz44QdkrGpNEM40lmqDaqgqMxALJP1pNWa5owo/CZ8Yzph26pi2NUshTIWxwK7bZcphndpSoOt1JL4QstMIbo8V7pmmzA1Z/CEOetS3H/CWL9NSSu6MlAUllByqmchWIN4BjIDLe2iClvbkAqIkgy8CQWbsMei9MMuoSyY8xzF3qrh66mx/bY9SSvYer+CxXMwUldkawYWNIsX4KMHHuK9J46wvf8qzpFnSTm80C1yTabIY3IMmcBWx8kst0M1WJo5xf/4mhHGSn0kboqzrEmnNUdm1/iPj/0j7/zsneRm6oyMXc1cZhdxdgPGH2FVu+xUihe7GbJG0VQpTxhDt1GIAVgMEsIhS33Ak6ScMhH7okUbkH02qdGI5pm2x6owxTVMcN5eaOXMAuXMEaaO7eellRF+9/97I2/+tRvp6dX/FFdy6Zz5E3+SeobvYBvUPcNfoPbmtVfgObcC8oF6yR3rwK4rRnnR665mLaxz75f3ofJ1+rI5alLkKYot6YpNG5YdE0efxKF0K0OPtDUoh0kZ1UpfrLQ+xCvUlUabhDStWW2RhP3WlIcvfbTK2Oy3nlSqCUqEuKwaw4AwfC4cxuGgMeSMZi8xEio7iBLjIWejVj/sprxiDz7HUsXfSbWZ0izpDEflOcILPJhU2SdRZtqzxgxh7lZNwLjOkJOuW6UpOCUWoyXE9YqTR0udgjRNyChW5eyYVJy3QVqj3jxFPlmlkkZUlct2t5NaWreZZ0njNE68yCsyo9zqdtHlQinTQ+wOMmFCarJf8TKrJrU5eyNunpPK4yiGLWiEW+yQyArlEkYLlr1T0SyrqcP6kR3cc9+TzAY1bu5bT7GuCacN0uUqAcRWzjgM6QKoGSPTcqtjs12vnZCeFnyqCDOKXD/4WxX0KQRYiRHg0FxCX1VbQHI+hHUF2DSiKGRaHbF1o8irVu2WVGH5MuaNW2xUuAz3H1CcWzHsW055Va/GE1OFMHpigNmiCFfBX4OcBCPnWuHGOlXMyph3Gqp1cUAbhjfKcypyJRiWQOEKeE2ItOJglNrjQ8wFQ27rGPh0ALeVobfk8kSlyTsefoz3rx5nY88uTjh9DHuGl+o8t+oCu90ck+ESp6UpJa0jBpaS57I8u8ybX9XP+JY+ZMQ/ezjgY3/7KC/9+N/x+ORj9K+7Bt17GyuS2+J2MeL1U1SaK4GrdY5jacS1jqaKshVzoukUxk50dF8CzqdNOkxsQ6yDNOBwvEKcVgnSJmF4ASVdymlAakJG9HGC5lEWzszwa++4jj+7803c8FPrUI5oHVsVbe1x6zPnLP20QZ3oXr4VkX/7z7JrQrlfImEvOWIv/c13u79cCkgp9qW/eeYsT3tL2ivQXoEf1wrYc8ZF1q7YneWnb7qan15/OR+4+5tUpk8xIp30YY2ALFrE/dLKkKwyS8oF5XOQlB4Us8LKKZ8FE9PUPmVpekibFihJin49XiGV++iMzXETE8Ij4kgN5+hxC3TIuBPDQ0bz6TTkViXhxJ4Fc+IabRr4bGy4zFEUfVgLsYXu1yqHK5TLDdpju3W/ZjlESkmidOJFJtM610k4skk54OTokhgRE+OTIOkVCyayH8quOH8TiVuWjlYx7DhoCVuT7yqDUi5R8wRpOEFV5SjpLNPxCtPxKhnpi3W7mDcpZSdLAU3RGMZ1kYI/QCla5WRapZuUqmna6rKq8olICUhlZVnnZFjVRWrpGjoWxJNHgN2aM8D40Hq++ci9zHeVed4NI3QMGBK5izhdN2Iz1QTEWeeydzGKRKzLotqXWqvYkJMqsVorPFdJ/ppWJEXF3Y9GTNc1uy/T7OhTdChFxzD4QjeVFbpu0BdHsLqsWJKO2Brk1inCNdg/HfOJhYBXd/j0pMpGj0ge8QWZJjrQKckuksNcUKQLhkkxRFSgQ7LrAqgvQnUQtnUpmZdbQ4Qsu9SGLUWGYuBwLFjj/jjk5dkC2YxhtmnYmoFq4nI4O8VvHfsKj8ytQcdmlp1O3pwZ5hfLZQomY9nC6RRmleasmBKSOnE0a+tclxfq/Np1YwyHndx99BRvufMj3PHUXRTXbaC7+zZqmR2s+f3Ebgdb3bJ19O7UGdty0qsVosC0fhK07Sf+zMWfj6VNPmtShjH0p02+ktRsC4lcHkk+Iok0QsQYp0yHrtKrD3Pu1JMEPSW+/J7f4G3vej6d67J23Co60Xbd14/r7Pi9n/dpgbpvBWSNRgNHXIEXr6jlBBw0mzhuy6nWbDRxPSmJbo2yoqBpNSmeK+V0kgcmAYwxQRi17vdtYPB7b3L7N+0VaK/Ac2YFLrJ2JjW4ec34tX28+oXXoU/08rV9d9HsrjPoetQjUbLJ2FaRsf9P8dOQibhi2a2MCNKFBUuqlJU0QeRYNgEZXUBL+XiySqfXTb/2mUHcrQMsKbg3nKE/jdiWy1L2MpgwYgoxJrjstDVdwt5J+n5CSUKME21L4Ddl4FQMsVLc5Gmu0j6jwI0S8KoL5J0MabzCwaTK/rSODQ5Wmn4nzzIOq+KmdYrUJITZ6YCkabWBJl5EORKv47V0blp0dE6rGcKEuI2DTETz9DhFOnSOQnaUEbeLA0mFuhBhOk/GGNZrww7X51XlPipxnSeasxTTKlMCHtPA6vw2uQUqIqIX80baRDllkM5YXUJGvzSOsKY6Ge8f4p4H9rKgytzYO2DHfMmyDT+zYi5bP+vLJXurZ9fSk/LTMDZzLp6HoK5QEpS2ALpHES0Z1g44FLXiip2KjBYNJSRiZha37CmDLzNEcYOKY9aD5dOtPDuvAvc8BW8Ol7nSDXlNucCGMZiQGBDb0GGYnIIFRzEgjRF2lxS9Q8p2/IohImjA5JKhOg+dRlmTTDotDlSIJlravYpk1c163K9XqcU+3Tjc3u1Qb7q8a+kMH37si8xVIui6nJe73bzN7+NWTwCxppYY+pRCpruflWy4eJkwFJtPavMbG3qGHvr5ygP38/Y7P85UEjAy9hqS3PXU3V5q2mU7mi1OkYxyOJUGbFWaQe1zuQdzxqHkCiMXM4FDg5TjJuRhUoaVw0pS5T4J9Y7X8MQAIRmQ8eLFUaqhT7RzEw+yMneO973ydbzv/a/juleM4eUV8l60F1xtFuYZeRr+vkHdJUAXrM7wl//5PXzgwx/jrvsOct01uzn+0Kf5w/f/PV/8xP/DAwcvcO7QI/zRf3oPJ9aKXH/1Nr7xiQ/w3r++k7s+9QnS7m1s6gp573t+l7u+fA8f+fBfMdUos2f3VpyL/q72sfKMPFbaG9VegR/bCsiHiDVRKMPAUJnbf34HO7YOcebOJU6uHLFhtxKFEcQBRkan0kstXbDiGnU7CUxg8+485eMol7L2yMeLBGlIqvM4bsn2VkYYEmG+tGPjUwZxuI+IxXCOStyg7BTZq1wGSWgql2VStinNBWFCjGHKKBsPsWYUfVpJPq6lqcRp2ZHRbPZz5I2LjA73SfCJRIvoHI20SRqc5qx1uzp2VBwIM6ezVExonbBSmyYVYmlwwb4OkuTfGslmW3NH2TeJJYku2G16QflG20TxRLJGj9vJ2WSNB0zIvWiRiREYjUo1ezJDHHM6OJMGlONlIhn3CnMZLSMGi36VsSPhhphKdJE4OIuyTKGLCU6z5m1gXd8I9xx5lHMPeNx8Wx8dZYdU+keFGpO6joa0OlwMF5Rs44v9r2ZZgJ4imTE0awrRxQlA87MtTV68CJuGFEmIjfUQJ6kS4CdaPcvUQjjXKrfv7INct2J5SfEchkjXAAAgAElEQVSn52XNNBE+15QduiTbb9lQayo2bFMUHMX9BwxuQ7X6ZNda9WNuAs48NB3ozyl6HYUUdki+ntuNzeKTimIngdyKmD00jxmY1y43yOWEX+Pjtce478JXGC6V8AtbaPgj/FJulGvcHA+nhgUM4m8WlL4kUT7AESM5eE3icIpGMEUhPcE3DjzIQzMTrNvyIpzS81nSvaRO1mYJBk6RW5wSReXztbTO5Tpvyc9hk/BkajhhDEtGEckUjJQ1FJ0mtTVg58NJmmnTumKjcIoknLaADl2gw3fJpYeYOXU/L+3axO9+8PX8ym/dRt+6loFGXqs2O/djOw1+X0/8/YM6ubJSii//3Z9wKL2Cv/iD/53C6klOV6Ay+QST7hW87//8d/zj79/KwCv+mPe86w186Pf+gKteejvH9h3nTe/4n7h62OW9d53i564e4hde9Ube++FP8c5ffiUffc+78Tb/FJtHOv/5KuD72vz2ndor0F6B58oKtJISWtEnUl96xa6N/Owrr6aRXeberx7G9asMdUbUqk1bjC4tqfKBpiQmQia1SQ1fyQd9ynLSoM+XurCIxXiZrA3rzdvRohuvMJ822OCUcE1MA8XJJOBQuMxTAnrsbZoKKV3Ko2Sxijyb4BMJMNY8FkdsRCwVMjKFxcRGs5ETR2kKG5TH64sluoxHWQTtwhCa1Dp6FyNpvVih6mSpmZQ4qWCUpqhdUmm7UtrSS0ZGoUQ4ThmTVDHJIqnbTcYb4Hw0RxhMWFBbzIzxJifLNcrlIRMjhfF3h7N8LQ2pGI9rHY+dfi8VXeJY44QNwDUCJgVIx0usOXmuczvpVRlmlGsr0ZLGMZTX02IImyeoeMOMF/t5YOUIiwZuvmaAfF2RZA1utyIRnH0Gq13TfRBL7MjpVt6cvDZeDrLbFHqdkjINO1atDMKTZw07PEVNQ6YHvM5W/pkz0GpskIw6Z1TRmMEGB4sr9o+eaFLB4Y1+hgnjMF5T6EU4JqxvouhahMWKoVdYOnHBXq4ob1JUDsHcKcMZeS5xSqxB53ow0me7XsbDijDbCiuWxBpvFMoZw+hahsNpwn3BBf6Pia+wf+EByoWNZErPY0NuMze7HSylKX20MvvuT1MCpTgtPysXkQJOmiaV2j5U4zBpTXp3fXr6dlEafAlTzjglt5uCk8PTWZooOqSjV3uESnSFjtWRXpB2k7RBXsC9XVp5fM2Kiawx6HhSZyaaxRdtJglB/VDr/aFcPO0ykJ9kevYb1KZX+PV3v4j3feAN3HDrenTG2OOuFSf0XDnbPHv38/sGdXJVoQi568Mf4rbX/Abj/Xk2Xr6HHev72f/YY+y65efYNNzF+QsVrn7xaxnuyXH08cfY/YIX4KzN8Zm7vszxY4+S7b+CW3d0spS9jF953fNRbp5cKWa2WuCKbaPtGqZn77HU3vL2Cvw3WYEWayc9nYaOoSwvuv0aXvrCnfz1Xz3E6uI5CkOQxaMZBhDP2mYGiWvQSAaXITENy5BpYdncMvW0ahk3z+1CSeOC14UnTtJ4lfMylnLyjLolOt0SkfJxw2kWmmesg/F6t9Nqom5D2aDhgui+LMmU2FpOARdShn4aY3V1YQoyiSwZrAZsAy63aJ8dTpZl7XJWMvi0j0maNmMultouCTIOZxh0SnYsKoG1IlyTMaxkiAXhBdK0jp/bjvaHbbF6xu1mMlpkvrbPmhzG3A52eJ30S1UUPnsk8iJe4X4nw5Lx8UxCoEt0uB2cDibpTFZtjl8qPbqkJLpgwW2320lVtjGasS0XSkbDkr4bzbFKifGOLu55eB9zjQo3bV5HueySzKQYYdPEBduhhBCiOSeJxMaCJdUj0Saqxd6JCWKwBdziLHxjf8we16FjTHIDFVrYvqyyFWCqX9E4D4GUGgTQTBWfqTV5csHwSj/DsmjspFdWw7Cr6DLQ36VwB2F2DUY6WhrIamTbwihnsEaNmhhXJmE5UZypGwqRjEchWwRXW+kmSd1QGtBksg4LSw3+YeZhHlz4MANKk8vvYDm/g6tymxlF2fy/O0zMdpMwJMBcKfZjWGcijkYrnGoeZ2XtfoLVuzFxy4zid/4UtdLz6HW72ex2M6ZzjGufJaStxLUdv0kaskU0mqK2TGocldBt5bJZO3Rql0UUE/GaZXqPxFUL0N1IeEJIxfrrFHDEFevM0YwOs3TqJJc3r+Cz9/4iv/LLt9HRJ0IGaRb55zSA/yZv8PaT/FetwPcN6mT0IQLdyeP3cmCpk1uu3MDXP/5+7j+9Sl5XiQsb2T7WwX13/R2De17MaDHh/nseo6cr5LVvuZM77vxTtnRFfP7AKi+7fiPve8/vcf1LX01/B3z4T36f8etewbZ1PXbEIift9ld7Bdor0F6B77UCrVNES9/j+pqx9X284ZeuZ+fOfj5151M0VxfZOBYShA5x0gAjw9mkFYMiPZYkrMQrtiGhzynR5+RwbKRDa7ZX0zm26SzbnBxCKBUlPkXMBBLv4eQt2xHGKzwVLZBz8pR1njUDvcpwlVKMK4eH0QygmCaxblfJO5xQCUvGMIzm0SRlTGHNEKeVIStAC80Op8AF5dhuTtHkianDj1dZipeJJak/XkKKUpXKWGZuJDtOxuuxhetKtG86g/EGyChX9phqbR9760eIxAHs9fK8bJ4JfAsWrtcuf2sMP5NT3JKBY1GGE9lRGmlI0DhCTvukXjcLwbQ1UqSiTZRgYl3EiBgtlYgVaa9voqILrKoS6weGuOeRgzYr70ZvhNJSlnAqRQ8rtN8ao/v9kCkqlIw0L+rrrENAEO9ya8QalRTnnlC2nk36XhsBZEoK6UEV+Z3cJq0QEt1RWKd4aCXi68cMb/KynAoM2z1sk8LbmyGv7dJc1aupr0GhBzo9hRfDTAXuPg+r0zAwrugbUpS7YWkOthUUMysgFWU9oaI4INl7hmBeqshcmisxX9h3iped/xRTtRMUy7exnBnH8/psRZunPFw0a8Rs0Fk+nNTJp1V7cfGFaIEDzbNMVx9hcenzhM0TaKfcYj51Ft35YjKZURuvU8bQqVwbN7NROayXqJuLUqV9Sc32scqFA7rVx14xMQ/FS1wwCZW0zqKwd2K2iZYuShMiS3MWVZVy/iSzx/eRrKV84IP/Hf/xfa9i++4eOwZva+e+19nnmX379w3qRBgiYGtk3Ra++Df/mbvuvo+vHavwzl99PUvnDqD7drJluMyJg48zdPktDJUUj93/KFc//4WU60/wubsf5ejhk5w4s8D2DYPMzR/m0b1H+OInPoq76Wf55de9EN++u9su2Gf2IdPeuvYKPHNWoMXayfZIKGyRq/ds45evv4lKc4av3/tNwopLV7ciiiPSJLIAxFg3aYOc18OifEg6OQZ0EV9qsJIqVQvfDANEDMnoVTm2xH3SFq9LNKtjx2YiFYuSFc5H80yR8gXTtG0MAlPkMeSDVwDeBqXYipaINo6m0sOuKGjNOWC/gDRtKDhwb2JYSZus11LdJYyYZtTEtt5s1cTkJGYiWcFEM0gTgWiQxb17mduJq3PMC9iUlUhWMUlNPL5Wk+e6Peh4nhNrj/JQUuX2cpFu8jyagpQu/LREl4h5wtN0pTIu9rk1W8RJHc42jlhAIHttdIZGvGDNcdrpsmPtpH7IGlEkM1BGwESLrKouxnsHuOfYY5w7OceNWzbSM5aFjhTrPhH029Vqj5CaMCsqE+2dCMsEb0yCHoKaA488ZNgl2rAK1M5DaVhZZ2pjDqQS2M8r5uYMH5+s8cn9mlc4HmO9rV7Wbi2aMjiSaBvrcdOgxh+C5bOwlBhmai3dnDB2f7JkyPXCeBPOToBfg66sohlAXxnKhVYGXnXBpeIlnMqe5U/3381vn96Pnx8jKV7FbYUr6fcHOSWvEQm7vC56dY6TGOpJjeF4ma81J9kbzpGtH8SpPIiqH7SVbUo2zJoOq+TdDkodt5J3S3japy8N6FE+jynH6uASBcfSOonthRAjR4LniOYtYTmpkIlXceNlC/51tIQTzZCEM/aIRFytXpOiOc78mSeozDf4n992Ox/9yJv4qZ/eQbGzJW+Qo7etnXvmnOeezpYocylz5Pv4q0tmCRPHrK6tkS934ruaKIrQjosjVvQ4QjmufeNHYYjn+6RRwFq1QaFUtKLY1bMP85t3PsWf/4e3kDYblDol7alle29zdN/HC9G+S3sF2ivwHSuQiivPStgUi4tVHnrkIH/4x5/m3rtPMbZ+kKbpYbWZxWS22mYGJ78No1w64lWSzCijyrXND1+Kq/SGk8zKMzhlenXWZsrNy6hKe9QlyyuuWMes45QsGFwJpugTI4NyWZcZIut28jLl2hw8GXd12IYLeNJguzalzkv6MU0a8M1UgFuWEUcxnSScjavc4pZZVK7V651IKnwtXsOJ5kmaZySDA2QEK+xhNG8z94yAPq/bRlEkIn6vHWyxeQK+nJKt+fLCGWpS3aDKvL6wkenCbor+KLe7JXqUYX8q/a+KRWX4eVcTxopvBlN8rPakHdWWMuPUhUUyAU4aEfu9mHCetHGsxdbJWVwoVBlZZ9Yx4vqcr87w6r4Mb9/6cm57cRlnLSEptcKIzVILZ9jIE4k4kZP/bKt/1bkGlqvwD3+W8jPdilwfuL2tui1XkGto1ZKcqKT8b1+v8cnE8PnOEht8Q1+PohHB0ophLoSHY9ii4HmjikoZOhwoNmF2BjoHQIjcJ+cM55swLhl22VY4b6kTMgkM5BRu3qHpxzyxeJ6/3v84n5Rqi84t9JR2U/L6cNI623AZkAuFNODe6uNscDvoUj6Hwllmhc1MGnSakLi2j1rzJMrtRmc3tcKlbZxIlSQ4y0DfLzLe+XwrRbpMNHEoRAUny7M3bdqWklGlWnpNYDqap2xSXJ1lLq2h0xATzrZEhibB9ve63fhOkUF/htPHxGhzjhtfspnff/druOH6jeRF2Cd6TRm3S/1E++tZuwJPC9TJXl4Cdpf2+Ft/lpPXdxwOrbntpbvb78HCCf72q6d44+tfQoswNu2x679YofYP7RVor8APugL2g+mihGN5pcon/+Eh3vaWu4EpOocHyWY7WawXIDOO8gZwlG9ZO6M9etH0up3U0zoHokXKSYUGrjVHjGTH6NY55gRcpQ3Wkob4V9Fen21aSKTWK63TjBbpdTvYltvINY6YISKyWlu91XRqCIyiV1L97eg15a40sUXqEjvRj2ItnOGAU2ZEKsFIbABxpy6QMTHn4jWkMSJNm6wmNZYxHA0ukAln8b1uIqdso1IQNjJetg5gxGErAW4mRqd1CqbJchpbRu6K3BgTuV28qTjIqNvD/tDQoVJOJiGvdjMsaYeHwzUOrD3I2WiJ/uKVLCVN4vpB6dUArw8TLSKmDeX1ojPj1pGb1A9jsusYcktcqB2G6U4++s4X87rNPThZQ1o2dhRLGSy4k5Hruta/bVdsE4JVeOBDKVvHFd15UIVW3VWcwvKs4olTFT6yphnOZrghq6xmMRH20ZNAYsVU3SDwpWzXGbb0QlkrmkUozoK3ATq74NEHDd0SLN2EXBF2y7ZlNd+YlsBexZAbcbx+ln8MDnNHdRGCAld17KbTG2VGZ5BXaSSpUVeujQtZilf4enCOevOcvShQyRq+8GjBGdJw0q67EoSqXAvspPrNJGsQThLXD7F53XsYyF9Gd9q0jlbp/i3gUkubPJLWGHBK1Gz2okdVKtyCaWJ5rZ0srgA6uS2as8ydMKqky/RmVlhrLlCdOMnP3LCFX3rX7bz4xdfS3SXO6e/8XP9B33vtv/vxr8DTBnWXDgD53ta+/fhfwPYWtFegvQLfuQJysWkzyWQEF8PeA4t89CN/y5//6YNIE+z6rSPUwgyVeITE6UZ7/Thuh21nuCKzjmZm1H5Ir8RrSB9nM62SSLCv120DesOkShotEaR1ytkNNO04TKGEMZHBZzBBlAbs8ft5VBd4uT/ETrfY+l2a2Lqpa8UBaVq1TaukzJiIe0X7Z0K2BmfJ+UNc7xTYKaXsSrNsxDEpwK7BVp3jtJJSds3eaJ7DwSSE0+Sle1aMC2KicIot7VW0RJpUrDjeCt+TCnnlWtYtFl2cWFHdHnYWdvH28h7Kbhd3GEVgYi5XsB2XLarJx+on+fvqcXYVd7HNKfPJ4IKNyAgaJ0kjKXzV6Mw6a9aw49/mcVLl0uNk8J0mF85m+Ou3X8lbh6/EYo1tKarZqoQwMq8Wy6a4FMp2akyzAvf+Wcp4XtE/AguLoFcVFyopH2tErPcCbvLzdkw8MghhF3S4Ld/G0gnDorRMJK0J0Fca8DJh6bpB/DPjfYrpDGSWYeW8IVOGhQBCV7O7T3NyJeHuxRU+v3CEjvAkX2jOQP4yhos76fFKlN1ONuDQaQL+MrhA2jxDn4zAhUE1KZ4APYkSaZ5mubYfomnQRbTk+4l2zulsxYgIPWaD/CT25SSJ28PG4bez0e3BTZuM6VbF3UQa25aQ08pH8parcYWmsKZJnTSaJ43msBZjt8OCeSSmRzuU3RWKeorTx04wQC/v++2f4wXvuJzeEQlSaYE5O2r9DjbmO99T7VueHSvwtEHdtzJzSSIm/vZXewWefSvgOJdCsP9lQ8qzb0/aW/yvrUDrfCX3UARhwr79J/jQnV/m/X9xP9BP59g42WwHS80yRudwBMCpDBID4jp5AomHkOgICf8Veb41Hxh6dd4ClmXRR7mdNjpCYkdkSJYmVetKFbBUl9iTeJl+p8iezDinlMcRt8x18qFvIk6b2BolJpXDaJpwnafJaY9qGDCjXbYpjz4M88ZY52WHp9lbr7BF5xjWInMxNix5Oo25kNT5x2CCclon1AXL0gWmaUGbbxLCeMmO6GRka+R2k9gRsLRxCJNUSQPGRZ9V3IOXGeE1dg3K7DeGnVozYwwrjeM83jiOk93KlNtlnz8J50jDaZLaE7aX1ilcZZ/biNNS9H12NNxHn+NydvUCd+x8Hq//6esp7tGkcYJNRRYmTtynoqmTZT5jaESKP/5ywsY67OrWBCEcrzb5VDNmBM3PZPIM9UG4ZOgWU4NryDdgQ7/i4ISxDRL5pqIRGz5Rg+eXYU+HYnUBhm7Atj9ceBCGux2qshmpYr6yRjgwxRcnpvjAzCEo9Uq+CutIaRQup9ctc1m8zLTyeTAN2BzNW2AujRxLaZNSfrvVFtarT5CpPo5UcCVS7SYuYWk9UZKWLCypB6ZhjSbWcCJ2keYZ0p5Xs7v31TZKZ9nEXKew49YxlSUk4YF4lQtJjZyMVdMKSXPCXgRIGLRlS0UCQER3pkozqLN87ghixfn1f/da3vqGm7nmirLtVZP3RRvM/Wtnjmfv754WqLsE6JrNJhMTEwRBYMexz97db2/5c3EFhGHOZDKMjY2RzWa/Q1LwXFyTn/R9Fr3dJa1QM4g5cOAEH/nYV/kv73vM1puv376FWlikkvbjeOuItVQhSUdsTCKuRUm7TYNWSDDGsmGCQYRlkfOidiVjs2Y1dSa92NHq9eIka7biqyrjtXjVLnNJ+1Sy6+nyhtiiNPlU87oejw05TccqzMfwwUgxRWhHilcqh68Zw88qTdXAXgFgSYVNOPR5RVyTMkFr/HjGpPx5soqEJkvIcr8JWE7qnG2coj+atQxhmjZszZeSfRT3qpaqMR/PLVvXq+35Cs5zY24LW71+qoXdDHr9FhCuap98ssiDKw+z3+2klN1MohzSeJGk8kQrkiOzwY6kZfybhheQOBnhyzx/kA5TZ3Z1htcWL+c333oj127vxRxJSDeJgVah5gxGKsYig+srvvhQyvkJ2NFjeJiY/2VylfcXOhjFp+EYrhpQJHUxrIDvGOpV6MtJjRg0GzC/ZvANHMsaPlWD944oMg3oKmrCbs2TpxLOrEnb0RR/s3iQhBqPhgZKGylJc4cBL21QiRfxvAHiNCAIp9mcGSXJjKF1hqLbyTnJDAwmqNb2YyT/LV7CZDbgZNa12j/ShgVdwsK2voShy7T0h2kNJ6kQpQm3jf0Wvj/IV9OIm3WGgnIIULa/+GS8hB9MoiW3UEa2ad0aYrTXQxpOoU2FsjNLIaM5e0wyYwLe+e9v4Q2/cAO7d60jm2l5ZoUcbE/ZLr4MP4Hfnhaok/0XQHfs2DH6+/vp7e1FbPrtr/YKPJtWIE1TFhYWmJubY9u2bRbYPZu2v72tP9gKWHLiootfHiEIYvYdOM6H7vwq7/8LGcv20DW2jmxumBW2S16GpY2aSYUxvx/lFK2mrT+t0oPDkWgOP6nYyiylS5Y5aQ37HAsIZf6bao/1mTGWo3mCZI1ESY8sFCUE2OvhZqfAelw25ztYdKDQtHiEc0aCi0UFCCMoHre5ZimblOagUkhzwGeSBo8pn+drCdkN6SHlWp1F+j2/ElfIKYcRp8gLnBxBGtFMqjxRP8pZFH6yTFA/ZlkeY6kxjRLBvw2mdezYVpyuwuaI63a7znNZ8UpWvQGkzcBVWb5Ue9LqAlXxGtJ4iaR2hDStYaJpC2Z0RspfU9J4FROctzpAybwbcB0mawv2MT52+yt5fp+4Y1PCTIpelggXRUbMFLHinkcCfulMwIgjq+bzGs9gtMMWWY8cbOxV1EOo1AylgqJhDLVVw4YBRRzCbA1Wm9LcoPjjZsrPe4rtGU2Uq5DEc3x1fp4/nTsIWuhBYa9c1mfHqLi9VMOZVh6fsLVOF0Fhhx0xF508BeXR4XQwmlaZrR/m0Oo9ZMIpQunjlaoLiR2R+Bt/0B43abTQGo9KDmIs7lgptHPsMaNEHxdMMdzzCnb0vIpJC/wMNTSTaWS1dhKCnY0rhDaZWXRwxl4syHHseF10uRM0qmdZmRQVYcRvvOsW3vzG27jyilEytqLtom7Omll+sPdP+6+eHSvwtEHdiRMnKBaLDA0NPTv2sL2V7RX4HiswPT1NtVply5Yt3+Me7Zt/EldAwJ0AmUsXpP+SuXvU7vL6rVuop+NUzCipKmEkqNUfJDGhdcv2u12sKY+KZN1J9piRfikJ4hWDv0/eZtmJI9GzQCOSVH8MmTS02WWS4L+choSpjOAiXmBCer1BbvC62Ats0C79yuEvkpjdWoKNYx5JYJdy6FcKGbIdlJErhiu1y7RJucek3Koc67iVrLJTySqP4fECMUg4GeZUhiipEOo8k7I9zbPoYAJtQUTaMjyI+SFtjWY9E5J1ilScHhIZGYbnQbZXZSjkt1uAkYYz6OxGtDdsI0/y2Q0k9Sep1Y/guAPo7Hq0P2jBY9I4RdI4ah+/z/WpRBVqC4a3XXs777npWgaDIqxLCXuhsmbY90TCf9o3w7KfY8ztYjeGXSiaytCZKHZ0QGdZkemG0+cNeXGqdmoWK5qyD8fr8FQjIU4CciblI40zmGiWjUT8jVRQrExBIUOH6xC7PUiFnHTcNgUoZcctwFNuB67OU1UOL/CHmVdwJm3QHy0R1vYTNI4x1ziGxMYkcoy4/S3AJmNQt9vWulmjil1jZbPi0vC8NTJgR+AGldZI8Ng49r9S8kdZTtYopRUOhXMU4jULmFPlWGezzKZl7CrHm6sSCn5C3lni7FEZs8YXmblb/omZk4PZusJtB/tP4ru5vU/fvgJPC9SJhu7IkSOW3fA8rz22+vbVbP/8jF6BS/KBS98likdY5x07dnBJY/eM3oH2xv1QV+C7M3eiuRPm7j4gQ8fQCKXSBhrpGDVnvXWjJson1Fm6dI4wldFdlUTYq2i+1cXqy6iy0AJLSgwH4MjIX2XoVy6xcqikMaJ3C0zKDu2zTzLegKuSNfY53fwPnsOg8vkUnXiElEyTJZXjNomtMLHNz7vCgbuTiDl8uqQFgRTPSByHoXrR/Xtv2qQ7DTkjfVg6RwfCwqXW4DGos6yGi6xFs9YBLGYK225vkWnUYpSEidQZ6/L1nTKJ7Gu8QJisWlOIkt+YutUSqtw2/MIeUglJFvAmoRuZMZzctotOT8myq5PUDpDEq1IMwaCqc3r5GG8cvZq33PASSv2D7D2neWLRsGdZcS6p8LByuN4p2fYEYS99pdgoTWkeDOYMyw54CTTqsJwPmVlbRJs636jGfDBaYHdwisVonma8jHCP0lU26hZxvD4qKktNaRy3m1hliQUsCQDLbbYB1TJKz7hd1oV6s7SSxMvcvfYg5fpTrAnQ9QfwnE7bAiLgXNzUmMD+p9x+q6HTXq8dkypXwoVzJNUnbHQJ8YKNtJF+M7f3FyiUb8SNKyxYJ+wMWSXpgKlcgWBEfyc5hSpL2UzgZWNm6nWYFC63wG+8aydvfsPzuHL3OBkb+No2QfxQTxbPogd72qDu8OHDXHbZZe0PwWfRi/xc31QBcWEYWh3dJUAnayIXKe3j+bl+dNjPzO/C3J3kox+/lz//k8exJaB0sWnnZczVuoi9LZaFkWZZYfykdaEoif/BBITzdhzrZIYJtc+Q02Fru6ZRDDtFUhNZHZYj41flWQZsIl6hW2q8dJYD4QKxCWlIBAmwRfs44TTLbg+zmSF+TQsIyRAoza2OwTWGd8Qhv4Ji+SIIkBaLR4Bf1+AbxRmRbynDf0ka1rQgrspFEkaVRGKEBMEEidSROSXrkhXtoMRsiObLiP5O5xDTg4QAamm1lzBAHNJomrRx3LJ3tmc3mESZBtqVQbDbYqxU1o515TFwcihHrK0lqzmz4Mcp0ydM4+KDUK1C18/zpr7r2exonpeDJ5Mmn0sM1zot7euGrMLxDOsdzdeWplhrnuRoGLLNEfMB3CFAJ5yyIc3ips3pIg1XxqEZ2+8lmXKZzHrq/iBGzAa2myIlMSm+zlEUwJ00KHg9lp2zINcYktWvEzZOQTTTYmydEp7EkliILP3CWcu+KXku+5gyzh6wo1cBe8KiSeSLkVG05PjFC6TBOUzjODq7AX/g10iSCnE4jYBnaT9JhZWLV+36K1Mh4xj68w1OHj2DAELo5p2/dTtveO12du8aIiv5fW1mrrUIz+H///ALYdEAABzvSURBVAhA3XdNq3sOL3F7139cK3AJwJ09e5bPfOYzvPvd7/4Xm9IGdf9iOZ7zP3wncxex96kqX/zc45y8/14+/o17hEujf9N2PG+EatLHahoz4g8w5HawT7Rq4gINplpGAeWS9QdsKHFowV+WWrxESTkcTUPWeb00xHxBiq888iairqSk3bXuRmUCZiUZN5yx0EGMGaGJeJ4/zPVemU7t4PkdPBk0OSDZZAKQCOlOY/4BeIeTYweKp9CWn5vCcNAESGdoj87avtvVpM5C2sSrHkBiMJQYP0TzJcxeGlqwojOjLUYpTSxAvdQ5K6NYIxl4ktUn48GkSlh/ygYiKyUhxU2MiPqtUULwjzjOndZIUv5KAKPoypwcnaZOQTVwiCg7fayoLB2qyFQaUjUpHTrLvIFr3JTYzTNvEi7UV1tgzS1AEtvtH9EyWw9R/ih1t4taGuCojAWsUbyIkdBnV8BWwQbtOlJvlhnE03k6lEPZhEwGk/RLX6+YHVbvae2DiVogzusjFcArY2hZL9l/E9v7iJZNaXG5ipVX22NAWEsBaGKc0P66VtBwWietPWmbQSSs2et+BcobJJXAYFl3q7UTAB6QzwxT8psW7M2enAYmefkLx3nF827ghS+/ir4rSuSFvpT1tAdwqzf3Of9mfg4vwH81qJPjSKmUh776WcqXv5idQ/l/OsC+3WEjV7XKXuW1DsBv//1z+HVo7/qPcAV+7/d+j9/5nd9h3759XHnllYhRQvRUbVD3I1z0Z/FDyzntWzV34hZcmKzw8MFD/NGffI5vfvWEFe2XtlxFtzvCfNhHpItWV2XrsuQDX0ZzGMJ4mTFhcaSGzOulZgK26gzzKGaTmmXqlpWmW0nRe9a6a2U8GjpFG2XiSBSJZJFh6HUKTIUzVAUoiLlBQo+V5lUm5lGTkPM6WFAZrkRT1DnWlMs9acAW5aBMzM3atdErn02aXK2zSKHD9YiJIOGL8Sq6foKmZKZF0ygJWFZeiynyehAjSCr1ZpJrp7BsoWgMi4UrLQBsGkXeKdAfL5Gr7mN/40gL8MQVTCJdtQLoPPu3Mm60oXQ2R0+1ctWk9kwAk1JIRIq1tAq7J+Nbt5skDSgRU7GP4ds2jR4ni5NIfEsGT7mIdq0u5gTtkKoc5mJmmzBy0pNrpL5LQJw0cThluvwBmiamaULyaWTB53L1ccuMJs1TKNNE+SMof8xGkki9nOw7Ko92xRgj6jsxcKQYCRXWRfvY0uAgbGULnFWtBlFYOaVl/11Eh4i8htL+UL7N3lcAnO3RdbvIaBeTrtGbT1kKVqmcFCfrKre/pI9//67buOHyrfSOtnLm5BhLxePR1sw9i884P9xNbxU6/MCP2fLwyHF+YfI4zpYXWtFvjIerpUPO1i5jUslIMmQzHkkUEOOS8RySOL5YKQZpImXboveQi64Ex23liP3Am9b+w+f0Clxi6b72ta9ZQCeL8cY3vpF7772X7u7u5/TatHf+X18BkaMJWPpn5g56x0q8bOwGbrrhCr741cM89Y39/NlfP8I57oGua8j1DNDFGEuNElp6PN0yJmnauBGp/UIHFMJpGk6RqnxwS3OEkydCMuByJCbhvLIxbfRKFEnatI5WrTzbiGBHoDpLtz9Er/JtEK0wQA3l8imjcJ0c3f9/e9cCY0d5Xs+87tx79+57vfba67Vx1wZjuzzCK3YSQXgoENQ0gUAaBVBLKiXqgyptQovUpKgptAm0VBFKWiH6CFGdJlRpCCIQaDBNeSdgY2MbjI3f3l3ver279zmv6nz/jL02BNnY2Oz6u9Lq3p07M/8/Z+beOfd833e++m4ESYQdLluDlWDHZZzHjgdWDq8kEV612uDDwSKpqo2xPtgnilVk2WhlK7HiIqn4LBYWw286F7RhSeq7xJbEsgIkTUvQbnmSYzde3y15Y3nm3DkFTDRGECR17IkbOKu4WDor7CivlhBzQgM6EiCL+WGUykx4UYgpiRsNeVkBGpVFxfNyc6QzhxNNIKy/KYpezu1Eze021jIS3rQw4XYhcUKEVAxzs+j0hoRhYbcZFnP24ipievHREw7GG85CKCFkKzdDCHSDoef6doyM/Z+oZA4JvZ2H6/eJbyHJqISi2dKLZ4u2NDZvm+xZwerVGpIkMPnl4skXiMmzNIEjkRTjvbpccFQyecckmaNPIJU7Oa/RmFjK2P5sdOZDDJZ3INg/ivKW7Sghwq1fuhDLll+EKz/afbADBPfDoBgsqAHFO3+eT7V3j1mpi9N8jUdX/j0eWlOGu3M1hpqX4Rt/fRvmlMbxN7ffjUoSYf1rI/jUNZdj8+onsfIXZfx45V3Ys/oxjLeejytX9OG2P/xz3PDVu7DIeg1/+71f4su33ICclJir1fWpdlEer+Mlsdu6dasoclTmmFdHK572dqaVa07d8cL5VNiPCW0d9PeiYLV5yy48/vzLuPXmhzGGYWOH0T8HyC3GaDgfod2JIKpK31XbmykhTFqgVJijhgQFvxcTjUGcmetEaOWwMWlgJkOlSYQz7IKERdckdTEgZu/QelzDDBZgIJLijCHpjWHDsX24fD8qY6K8DpHfY0KAjT2YJeFCH4ONPeingS4csECiKibIJSTsJhEMwvVmwGaYksbEcR3w2kgX0hww2ra0ArRkIXWyfCGM4pNGhYyFFPwBn8QI6Z3G3rB2M6xoFHF9m1HGuK+IvcCoVtF4lz/4JcyTqpr8nmf1Q0VUvQM5fZyPFGsUYRcWimrmFPrN9lQSnQKamANn+8Y82W1GFxysTWLstyzkwjEETlGMlqXLRX0bS1uR1HcgCgaQVNgjNycFHSxCkN61zAOkUbC0V2NOW8lYlJBESvstup/Q/JeKbEOOV7o5UAEUHEJYbrMxGo7p5RoCPHbmyJEoh8Ow8ovgtXwIOQRwUENbvixVsFvWb2JrCXT2zcTXv/ZBXLZ8ERYs6DnQdOLw6/BU+OzpMR4dAseo1GWDWSjvG0BLz0W4446v4qFv/yX++Uc/x9duuhCr/vWHuOe5F9Ax+hJ6ln4Tg5Wf4OOP/RMeXPUyPtk/C8+9tgnLT7Nx573/iItvvgV2dSu85hbQWieRBt1K6jKU9VkRUARODgJZqogowKQfTUD/0tk4bclsXHfFxdjw02347oYn8J27nwVAW5QuFPsWozM/G+NhO+pwUXRbUPJ7UaWNSFRGYDnIJwnWUTFj2NNtR49QthCronHp8VmEL6HXXFwTQrWtvhW9dgHtbgt2hmPid0cVrOYZ+4xc02IhHDQTjtnujH5poJLXjE3UsewCdqZtqfIMc5JoOK1i0cZWX+wpy2ONmGPHgKuoXMZGgyFli2FKEhO3TUCwpL9sTcxwLacI13KFZAr5c7sk147GuAz/svpWQowwBrxC3ER1c0XAI4mU4gzun6w5/T+bQ1ReA6u2WVQ9VtRyn7XGAHqdFnTaPjosBwNxHcsYhrVcPJOE6PBnYaS+XRrcR+zYUP6VycOT+fiw8yx6aTOqrFSXpvcb9mVlnqCQWPbboDC5RcghSSnVOYZShaAG+9PXOSAeM3lzXEZ7GIk90VOubPLwkiqc0kUoFnrRXAR2VoaBNzZjVBwJS/jCnyzE7171ISw8pw8tnXk4B25/2h/95Hzyp96ox4nUxYgL3fjtj1wqCFx29afw6P30Kqrj3GtvQv+cVsDtwB9//dOg23c0fwGCXzYw98zzET/+n3jimQj3/cu/YWDzWuwe34YLPvAJ2Q/tGQ9c01MPW53xSURAbr6WhU2bNuHyyy+XmSxZskTCrydxWjr0FEcgI3e8y8cJSZKFjt4iln/+DJxbPx2/99mrMbR1N75xz0NY9dQ2VPAk8qVZ6Oudg+FyM/Y5S2HbrUKKTPUlUGjsQeQU0ed2YF8SSNWla/vYFY1hES1QYGPEcjDT8tDrtmEwrmO4MYK5XhfG4nFUogrsxiC8tMNB1NiDhGFLy0GNvnGI0Oa1SUeM8SSW/DH2hg1yPaDxrW0xh6sMK2kVtdEuLBCyEte3wxLvPZZVMnxpql4RmnZXtDqxcj1SIEHlkXltQTCA2JsH2/akdZiQJqcTSbALSbDTqFei0FG5q4JEkMqZzCFhyz5HyJ8UV9DmhWFb9kq1GfYtStVotO8niMZ/Advvh9+0BFstD5tjH3ZoyOADbMUVjcKrvAoqnEntNSSNIQmJy3jMfeMfMZEESqqGzJdjkYFnGDtJXDgq4WmGW0kgEyG4DqzcTDAvMAmYK8g0o0DsXZKEljaRmA6LgkfCSi+6iDbCNeQ8D4WmZehobceWgZ0Y2/w0upBg6SVt+PIf3YgZ8+Zg2ZLZyEurEn5QTL6c6YSiBRBT/KvjhE3/GEldVulqA+Xd+Nmq53DBgivw/KqfodSzBPm8i30jEwj4mQkDDI+Oy++yMAhQrlZQKHVjfvw0rvn2Kxj60T14+I7fx1+80I61n/miAHDgC/SEwaEDTRcEsmvnsssuw+233y55dQ888IDk02WFEtPlWPU4TgICTA1LveAkJGZB2jCdf+484Nx5WH7xb2LD/4zhpy+twp0/fBmbNrDXLIvINqJjVh750pkYD09DPc4ZY1/bx97yehpZSAEEzY7ZL7ZgedgWV1G0PFQaA1IZOwJHjGf3s5jBbYFt1eHaBYSNXWAyP21DLK8TCMdFOWP/WmaEUV7c2xiCjTrioGoqdJOG5JyxGKERB7DZkJ5+bP58o5rVtqXdFmxDhBz2yGU1aZs0kWdvVxZRMBuazcBYxenF+xHHtjHfdZhj2I0kmg9U10lnCVGtGKZkAYkUCPAGwRxqyTgz4VcSTWnNRnPmKhIhXD4smhy7ERAOIK6tl+IH5qblvG7J8QsnnhP1M0zqCFmYR1XSbobdNCtVzSgTWGIvQlJKYknFTcKpHM8umKIKnoeY5Sns58u5maa0EpqubTZFEsylo6cc8+qE8GZK46h05GDBS9ENUWquod5wMLK3G5WhvRjGdrhnL8SdXzgf169YgfbzmtHWZgoMeSVPDrFqvtxJ+GxP8SGPkdRlOlqCjlnzsennj+K2tf+FNUMuvnXXH8CLJ9DR1SlpCo7torerJMqb7floKxoJfsGFV+BzVQtdpQIWfXAFrunsRkfelrBAVik7xTHW6Z8kBDK17sYbb0RLS4tUvnIqWSeBkzQtHXaaIZD9gJAQHVNGLKCtvYiLruHf9bj+hksxNPQ5PPjfL+M7DzyDkT3MhnsYOT+Pvnkt2FfpR63cjLploWHNhR3sRWIlqLhd2Em/NKcZzU4OcW4GhqMK+hMWB0Soua1i7eEzL85lYUdDFCOgLCHFOBwVbzXfKeFNFiNYNgpU0EQhM1Yk/AanETCLJVjwIAqW5Lw1kFCRy89FXN1iiBa7JUhRQCQmzAxbWpYLu7ETVdi4utCP18N92IgmaZkVkwyFDC27sD0Wb1yIxO9DuP8pIWOImZ9n2kzSwFjIlqhhObOcXTqounEdeuyBlbcMi5IMWtJZIQk2Imao2GKo04blz0fAXDi7A4704yVpS+8n9H+jwibBVMcUPUhVrs1uvkLiqLQlNvPlGC5lpSxf00y4ltq6pM88yUJIuf+ceW3nkXdjOEkN7V4Er6mKN7YMYHyAhSHd8PtsfPGzH8Ynr56L7u7TcMbCg7dffleJk4RNRS67r06zD4oezglB4JgLJbJZ0h6CMnFlfAJeoRk5z1S/hmEE1/PkyyYIY3gepf4IYZzAc10kUYTYAhyWoUex5FY4+vMkg1WfjwMC/MJU8+HjAKTu4ogRELWFVbSTEkgmyiGq1Soee2Id1qxZj+/+eBC711LBq0r3CiCPnnkzEMXAyFgFSW4RErcLDitx3U5px+V4HShGrOoMENt56cPalMRgf9pQ+raSJIXGa465X7QPSRpwvB4pdqDZrihl1KCYFyf2LY2UuASwqfLRCoQJ/qkhcVhZa0KRbpdRoNxWseWw/F5xO7DZW7YxgIVxFXsRYNRtgwtPupta4tE2Kqoc22Y5udlCkILRxxHX3jDKGBMURaHKcu18IUosSsgKKiSHTbpuMCnHFtJryJSUgKaKnG9y3nhQtHuR3r2h8d6TUmaqb6YFmlQecJnTJGFpQy4tUyDB4gdZLwacohmPJE+IJa1quJ8qLJI9qY610VFywPvW7jcngGQMEMOYZnzkA5di+cdasfSss3DFJXNRKDahdFCUO0SVO+KLS1dUBN4BgeNG6g4fI1NJDl/+6//PQrm/fg19RxE4FgQOvybVp+5Y0NRtjxQB5t5RhpmsEPPbbsNrYxga3IGHHxnD/d97CG2lEWxax7ZPNczta0Mu34xqrYGRsQYc10c9nimdGmKqXnYRTq4Prj9HOiNIrhg7FzDcyAR+Kj8hneiYpE+vuAmw24Hl0iw3I0IMN+YNWaEqxQpXqkQx88voUcdqVfqq7RQFLwlZGVuD7S8wyhTHiVn9Gchm7GprseUVx2OIlkSU+6ShMdXHiNv2wPZmSRVpOPEs4vLqtGeuZypfhSgZZU2IlURLU0WLKp2AzgIFYmq8AEmdxQOVBI0+dtJnNTIVryRdkpPHalRW1zJHj35yJK4pwUvqYikiWLEAgsRXMKIiybmYcUh0bTtG3mmIb157k4OCbyEIYmzbTC+5GmaeOQuLwh7cetPH0X5xB2Z3z8f8/oOKHKevvViP9JOj670bBI47qTv8xvluJqXbKALHG4G3uy6V1B1vlHV/74SA4VIMzx4aXitXgFqVdiRlrPzBk9gzUME939wB4CkAzQAYvrPQOduBn2N6ch0j44xqkMQwdMi3Hck3Y+GC6IPMDXOajGWJjOeI+kWrDVN1GhkiRBsTEi8SM64vBvGOqe7ka+p+tc3GfJe5bcGghHZJlKS4gq3P2ELMzsFlnl0SIo72i4LHVlh82LleIZhxY4fZ1vJBc18x4E3qYH9V6WvKggSGMuXBcCrnnFaYylHySA0hlcpYCd2ShJr2Z4hpJcI2ZFTsSM7EQsFUobLQQf5nu64eU7WaVCV/zpBgOvjaxqaE1iZitUx/OQu2FaO9mMC1q6g3YgzvpHky8wBZocsq11m45U+XY3ZPF6779MVoK5XQVixSeJ30MOFVOTeHnv5J6+hLReDYEThqUrd+/Xqcfvrp8DyWdb/1C+rYp6R7UATeWwSy6zYIAmzcuBGLFy/WXsbvLeS697dB4KCCx7v8wTt9rQG8vGYf7GQXXlm3C5//0sM4ZyHw0vMDpkoTTehb0Ao35yGoBRidqEsnBlbT1uokL+Q+DSFWQlhYOerOlEb2hhSFxlyZY5LUkQSRFJJAcWuqdwwci/WJBVqSmIQvWnaw7RcVwTwsslSnlOafVUylLPfAkC9z42hNwm05ju1IFauEU2Uc7t+076KPWxLw2BIT/pT5Zyocw8kkdnzOSF1qvCuEL10uIdOGycUWtTJbf9K26Tqcd1qSIdWuCfdDyzwkyOdo5ushikO0lxIUckAjiLF9C/Mg2cKrGeecPxsvvdrAfd/6BJYtnY84dnH22b+BvH/QMF9G0E4Pb3PV66L3GoGjInWczOuvv45SqYSeHv4i1IciMHUR2L17NyYmJrBw4cKpexA68ymPQKbg8UAOV/GYXzc8PCbWKY889iLWrduEMIxx199RxaPhMQkYlTxTUNAzr2BIEGLsrySokX/Z5ge4dPihgsWwo6F+aUN7+rNQeeIESJhM0YGoYeLHRsWMpJOkhX8kXBSr6NVGMa0tVb6YE0diyIUkVfzjv7QJoXoWpoocCzIKxiNPFLVA2oKJjUk6HitHZTvui+MI2aSKlxHgTIUT5mTGYhGHhI5zQlpZsWq6GskkZFsehinAY8gW8L0YrUUeLx8J9mwjNjw+4kESx78O/NlXroLr2ViypB9XXnEBojhGZ2crnPQQ0x0czJETDLKl+qwInDgEjprU1Wo1UTfozN/V1XVInsiJm7aOpAi8ewRoabJ3714MDg6K6pzPHxInefc71i0VgeOAAMkG1WSSjMl5eNmua/UIL6/eDFhV7Nlg4d6/+l/snLEatVEfWzZS8WIlqQ0v72Pe/E6EkSE5YxUL1Ybpe0yOFIcWagmVLBI1qmRpwYIQI1M0IHlnNM5layxR9UjG2NwsVcvYVUFMdlnJSvKXKn5CDHkcrGA1y8S+RA6OBQapGRvfY7stWc9GEgwB4g9HLzyOyRw5krM0nCqkLosQpWocK1alQ0VK9GyORyIbwbET+K7pdsHuR8yBa21iyDoBO1Fu3TqGoMYwKh8++hb2orsDeHFTE+67+yosO7MLUezgnLMXHaLEpRtIH2kSTZLxw6Lq2Sr6rAicUASOitRlYSsSu+3bt6NeZwuULMfhhM5bB1ME3jUC/AL2fR9z584FCV12Xb/rHeqGisB7iED2HXu4iidDJsC+4Tpiu46JiQj/8YMXUR7fCsdx8Q8rxzD26qpUbSIBYn4ef8CkifuzgJ58AUliCJlt+aiFLkbL9K4znmtU9aTNFZvRxwkSaaXFfdG3rogkqYpNyIEctjS0ahQ1EjqOZUijrEPiF1MJE8ksLbjg+9wnO1fQrLeSFmuwBZkjIp1hTMY7jlOTAgZhUeb+Q+eEtpKFfI7ecmZMy4pRbdQxujtT3qjyUU1kdSq3c1Ho/wi+cuMcRFGIptJMfOa6FWhp9hCGFro6m9+2r+o7no/38DrQXSsCR4LAUZE67nDyDZCJ5vpQBKYiAo7DG8mh1/NUPA6d86mHwGQl7wDBmQQD3yffWf96HUMDW+A6DTHR/f6Dr+D+lc9i4RxGQhPs/VWEndKeikSHpIqkx0P/6S0IYz813SUhYgGDi7BqYbRhI6ESxnWpjglpo2rH1yRdxpbELE9DujI37p/3C4ZQSaj4XradqXZFQi86rkMil4ZqZU6GAPK4aJXVKoLiQTGBy10nxKbXWJhBRc98tg157cOyczy4noWN2wLc/Dvn4fprLwIJXxjamDFzLhYvajUpgzz8Qx7s6MBxVIk7BBb9532NwFGTuvf10ejkFAFFQBE4xRAgqTFE6a05eZOhoE9epVIR9YndMPY/X8PKZ55GxR0RPpbLAWs3hPj+v78A4A1jDyLki2zHqHNG7SMxy0gg3+MEMrUtU8VIzvgel/OZf9kyrp9tw+dsm4wEchnXJ0HLlvF/PhgqZd4bH1yP8yCRnIvrb/gwlp5RQKORwHZiBLUmrDj3o7jw4iYJt5KgFYsFlJqyCtt0N5OeMhWOi95WGZ20rr5UBN6PCCipez+eFZ2TIqAIKALHiICpruVOSH5+fScVkkKhRxYwNJxg/YZtcO39JiwrdCiBa9nY+3gZ9z7yKGot++BErBVN1bYDxCubMAmYGfPgc0bKsnWy9yevm73H58nL2WM3wehoDisu+Riu/a3eQ8ieqG5RExYvno8ZnS6jxLK17OXwYbllVhSSrpC1e5s8ur5WBKYqAkrqpuqZ03krAoqAInAUCExW9GQzRjUP0J8j2FECNBqhqZA9sHpGzg4seM9esDWY63pIMyeOcJzMHy5bXQsaMiT0eXoioKRuep5XPSpFQBFQBI4aAVHthP29laxZ7Et6NCTwqEc/kg2yPLfD19W8t8MR0f9PTQSU1J2a512PWhFQBBQBRUARUASmGQLMMtWHIqAIKAKKgCKgCCgCisAUR0BJ3RQ/gTp9RUARUAQUAUVAEVAEiICSOr0OFAFFQBFQBBQBRUARmAYIKKmbBidRD0ERUAQUAUVAEVAEFAEldXoNKAKKgCKgCCgCioAiMA0QUFI3DU6iHoIioAgoAoqAIqAIKAJK6vQaUAQUAUVAEVAEFAFFYBogoKRuGpxEPQRFQBFQBBQBRUARUASU1Ok1oAgoAoqAIqAIKAKKwDRAQEndNDiJegiKgCKgCCgCioAioAgoqdNrQBFQBBQBRUARUAQUgWmAwP8Dl5rGJH1yXqIAAAAASUVORK5CYII='/>",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown"
] |
[
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
]
] |
cbdb5aeb9856b5c1d5fd2ca924b36285ec27d270
| 50,530 |
ipynb
|
Jupyter Notebook
|
workshop/4-Messaging/4.1-Pinpoint.ipynb
|
gs-aws/retail-demo-store
|
b1ffd5c1029030b7ec1cb614fde6fc67e5148254
|
[
"MIT-0"
] | 1 |
2020-12-12T03:43:56.000Z
|
2020-12-12T03:43:56.000Z
|
workshop/4-Messaging/4.1-Pinpoint.ipynb
|
gs-aws/retail-demo-store
|
b1ffd5c1029030b7ec1cb614fde6fc67e5148254
|
[
"MIT-0"
] | null | null | null |
workshop/4-Messaging/4.1-Pinpoint.ipynb
|
gs-aws/retail-demo-store
|
b1ffd5c1029030b7ec1cb614fde6fc67e5148254
|
[
"MIT-0"
] | null | null | null | 48.963178 | 963 | 0.631308 |
[
[
[
"# Retail Demo Store Messaging Workshop - Amazon Pinpoint\n\nIn this workshop we will use [Amazon Pinpoint](https://aws.amazon.com/pinpoint/) to add the ability to dynamically send personalized messages to the customers of the Retail Demo Store. We'll build out the following use-cases.\n\n- Send new users a welcome email after they sign up for a Retail Demo Store account\n- When users add items to their shopping cart but do not complete an order, send an email with a coupon code encouraging them to finish their order\n- Send users an email with product recommendations from the Amazon Personalize campaign we created in the Personalization workshop\n\nRecommended Time: 1 hour\n\n## Prerequisites\n\nSince this module uses Amazon Personalize to generate and associate personalized product recommendations for users, it is assumed that you have either completed the [Personalization](../1-Personalization/1.1-Personalize.ipynb) workshop or those resources have been pre-provisioned in your AWS environment. If you are unsure and attending an AWS managed event such as a workshop, check with your event lead.",
"_____no_output_____"
],
[
"## Architecture\n\nBefore diving into setting up Pinpoint to send personalize messages to our users, let's review the relevant parts of the Retail Demo Store architecture and how it uses Pinpoint to integrate with the machine learning campaigns created in Personalize.\n\n",
"_____no_output_____"
],
[
"### AWS Amplify & Amazon Pinpoint\n\nThe Retail Demo Store's Web UI leverages [AWS Amplify](https://aws.amazon.com/amplify/) to integrate with AWS services for authentication ([Amazon Cognito](https://aws.amazon.com/cognito/)), messaging and analytics ([Amazon Pinpoint](https://aws.amazon.com/pinpoint/)), and to keep our personalization ML models up to date ([Amazon Personalize](https://aws.amazon.com/personalize/)). AWS Amplify provides libraries for JavaScript, iOS, Andriod, and React Native for building web and mobile applications. For this workshop, we'll be focusing on how user information and events from the Retail Demo Store's Web UI are sent to Pinpoint. This is depicted as **(1)** and **(2)** in the architecture above. We'll also show how the user information and events synchronized to Pinpoint are used to create and send personalized messages.\n\nWhen a new user signs up for a Retail Demo Store account, views a product, adds a product to their cart, completes an order, and so on, the relevant function is called in [AnalyticsHandler.js](https://github.com/aws-samples/retail-demo-store/blob/master/src/web-ui/src/analytics/AnalyticsHandler.js) in the Retail Demo Store Web UI. The new user sign up event triggers a call to the `AnalyticsHandler.identify` function where user information from Cognito is used to [update an endpoint](https://docs.aws.amazon.com/pinpoint/latest/apireference/apps-application-id-endpoints.html) in Pinpoint. In Pinpoint, an endpoint represents a destination that you can send messages to, such as a mobile device, email address, or phone number.\n\n```javascript\n// Excerpt from src/web-ui/src/analytics/AnalyticsHandler.js\n\nexport const AnalyticsHandler = {\n identify(user) {\n Vue.prototype.$Amplify.Auth.currentAuthenticatedUser().then((cognitoUser) => {\n let endpoint = {\n userId: user.id,\n optOut: 'NONE',\n userAttributes: {\n Username: [ user.username ],\n ProfileEmail: [ user.email ],\n FirstName: [ user.first_name ],\n LastName: [ user.last_name ],\n Gender: [ user.gender ],\n Age: [ user.age.toString() ],\n Persona: user.persona.split(\"_\")\n }\n }\n\n if (user.addresses && user.addresses.length > 0) {\n let address = user.addresses[0]\n endpoint.location = {\n City: address.city,\n Country: address.country,\n PostalCode: address.zipcode,\n Region: address.state\n }\n }\n\n if (cognitoUser.attributes.email) {\n endpoint.address = cognitoUser.attributes.email\n endpoint.channelType = 'EMAIL'\n Amplify.Analytics.updateEndpoint(endpoint)\n }\n })\n }\n}\n```\n\nOnce an `EMAIL` endpoint is created for our user, we can update attributes on that endpoint based on actions the user takes in the web UI. For example, when the user adds an item to their shopping cart, we'll set the attribute `HasShoppingCart` to `true` to indicate that this endpoint has an active shopping cart. We can also set metrics such as the number of items in the endpoint's cart. As we'll see later, we can use these attributes when building Campaigns in Pinpoint to target endpoints based on their activity in the application.\n\n```javascript\n// Excerpt from src/web-ui/src/analytics/AnalyticsHandler.js\nproductAddedToCart(userId, cart, product, quantity, experimentCorrelationId) {\n Amplify.Analytics.updateEndpoint({\n attributes: {\n HasShoppingCart: ['true']\n },\n metrics: {\n ItemsInCart: cart.items.length\n }\n })\n}\n```\n\nWhen the user completes an order, we send revenue tracking events to Pinpoint, as shown below, and also update endpoint attributes and metrics. We'll see how these events, attributes, and metrics be can used later in this workshop.\n\n```javascript\n// Excerpt from src/web-ui/src/analytics/AnalyticsHandler.js\norderCompleted(user, cart, order) {\n // ...\n for (var itemIdx in order.items) {\n let orderItem = order.items[itemIdx]\n\n Amplify.Analytics.record({\n name: '_monetization.purchase', \n attributes: { \n userId: user ? user.id : null,\n cartId: cart.id,\n orderId: order.id.toString(),\n _currency: 'USD',\n _product_id: orderItem.product_id\n },\n metrics: {\n _quantity: orderItem.quantity,\n _item_price: +orderItem.price.toFixed(2)\n }\n })\n }\n\n Amplify.Analytics.updateEndpoint({\n attributes: {\n HasShoppingCart: ['false'],\n HasCompletedOrder: ['true']\n },\n metrics: {\n ItemsInCart: 0\n }\n })\n}\n```",
"_____no_output_____"
],
[
"### Integrating Amazon Pinpoint & Amazon Personalize - Pinpoint Recommenders\n\nWhen building a Campaign in Amazon Pinpoint, you can associate the Pinpoint Campaign with a machine learning model, or recommender, that will be used to retrieve item recommendations for each endpoint eligible for the campaign. A recommender is linked to an Amazon Personalize Campaign. As you may recall from the [Personalization workshop](../1-Personalization/1.1-Personalize.ipynb), a Personalize Campaign only returns a list of item IDs (which represent product IDs for Retail Demo Store products). In order to turn the list of item IDs into more useful information for building a personalized email, Pinpoint supports the option to associate an AWS Lambda function to a recommender. This function is called passing information about the endpoint and the item IDs from Personalize and the function can return metadata about each item ID. Then in your Pinpoint message template you can reference the item metadata to incorporate it into your messages.\n\nThe Retail Demo Store architecture already has a [Lambda function](https://github.com/aws-samples/retail-demo-store/blob/master/src/aws-lambda/pinpoint-recommender/pinpoint-recommender.py) deployed to use for our Pinpoint recommender. This function calls to Retail Demo Store's [Products](https://github.com/aws-samples/retail-demo-store/tree/master/src/products) microservice to retrieve useful information for each product (name, description, price, image URL, product URL, and so on). We will create a Pinpoint recommender in this workshop to tie it all together. This is depicted as **(3)**, **(4)**, and **(5)** in the architecture above.",
"_____no_output_____"
],
[
"## Setup\n\nBefore we can make API calls to setup Pinpoint from this notebook, we need to install and import the necessary dependencies.",
"_____no_output_____"
],
[
"### Import Dependencies\n\nNext, let's import the dependencies we'll need for this notebook. We also have to retrieve Uid from a SageMaker notebook instance tag.",
"_____no_output_____"
]
],
[
[
"# Import Dependencies\nimport boto3\nimport time\nimport json\nimport requests\n\nfrom botocore.exceptions import ClientError\n\n# Setup Clients\npersonalize = boto3.client('personalize')\nssm = boto3.client('ssm')\npinpoint = boto3.client('pinpoint')\nlambda_client = boto3.client('lambda')\niam = boto3.client('iam')\n\n# Service discovery will allow us to dynamically discover Retail Demo Store resources\nservicediscovery = boto3.client('servicediscovery')\n\nwith open('/opt/ml/metadata/resource-metadata.json') as f:\n data = json.load(f)\nsagemaker = boto3.client('sagemaker')\nsagemakerResponce = sagemaker.list_tags(ResourceArn=data[\"ResourceArn\"])\nfor tag in sagemakerResponce[\"Tags\"]:\n if tag['Key'] == 'Uid':\n Uid = tag['Value']\n break",
"_____no_output_____"
]
],
[
[
"### Determine Pinpoint Application/Project\n\nWhen the Retail Demo Store resources were deployed by the CloudFormation templates, a Pinpoint Application (aka Project) was automatically created with the name \"retaildemostore\". In order for us to interact with the application via API calls in this notebook, we need to determine the application ID.\n\nLet's lookup our Pinpoint application using the Pinpoint API.",
"_____no_output_____"
]
],
[
[
"pinpoint_app_name = 'retaildemostore'\npinpoint_app_id = None\n\nget_apps_response = pinpoint.get_apps()\nif get_apps_response['ApplicationsResponse'].get('Item'):\n for app in get_apps_response['ApplicationsResponse']['Item']:\n if app['Name'] == pinpoint_app_name:\n pinpoint_app_id = app['Id']\n break\n\nassert pinpoint_app_id is not None, 'Retail Demo Store Pinpoint project/application does not exist'\n\nprint('Pinpoint Application ID: ' + pinpoint_app_id)",
"_____no_output_____"
]
],
[
[
"### Get Personalize Campaign ARN\n\nBefore we can create a recommender in Pinpoint, we need the Amazon Personalize Campaign ARN for the product recommendation campaign. Let's look it up in the SSM parameter store where it was set by the Personalize workshop.",
"_____no_output_____"
]
],
[
[
"response = ssm.get_parameter(Name='retaildemostore-product-recommendation-campaign-arn')\npersonalize_campaign_arn = response['Parameter']['Value']\n\nassert personalize_campaign_arn != 'NONE', 'Personalize Campaign ARN not initialized - run Personalization workshop'\n\nprint('Personalize Campaign ARN: ' + personalize_campaign_arn)",
"_____no_output_____"
]
],
[
[
"### Get Recommendation Customizer Lambda ARN\n\nWe also need the ARN for our Lambda function that will return product metadata for the item IDs. This function has already been deployed for you. Let's lookup our function by its name.",
"_____no_output_____"
]
],
[
[
"response = lambda_client.get_function(FunctionName = 'RetailDemoStorePinpointRecommender')\nlambda_function_arn = response['Configuration']['FunctionArn']\nprint('Recommendation customizer Lambda ARN: ' + lambda_function_arn)",
"_____no_output_____"
]
],
[
[
"### Get IAM Role for Pinpoint to access Personalize\n\nIn order for Pinpoint to access our Personalize campaign to get recommendations, we need to provide it with an IAM Role. The Retail Demo Store deployment has already created a role with the necessary policies. Let's look it up by it's role name.",
"_____no_output_____"
]
],
[
[
"response = iam.get_role(RoleName = Uid+'-PinptP9e')\npinpoint_personalize_role_arn = response['Role']['Arn']\nprint('Pinpoint IAM role for Personalize: ' + pinpoint_personalize_role_arn)",
"_____no_output_____"
]
],
[
[
"## Create Pinpoint Recommender Configuration\n\nWith our environment setup and configuration info loaded, we can now create a recommender in Amazon Pinpoint. \n\n> We're using the Pinpoint API to create the Recommender Configuration in this workshop. You can also create a recommeder in the AWS Console for Pinpoint under the \"Machine learning models\" section.\n\nA few things to note in the recommender configuration below. \n\n- In the `Attributes` section, we're creating user-friendly names for the product information fields returned by our Lambda function. These names will be used in the Pinpoint console UI when designing message templates and can make it easier for template designers to select fields.\n- We're using `PINPOINT_USER_ID` for the `RecommendationProviderIdType` since the endpoint's `UserId` is where we set the ID for the user in the Retail Demo Store. Since this ID is what we use to represent each user when training the recommendation models in Personalize, we need Pinpoint to use this ID as well when retrieving recommendations.\n- We're limiting the number of recommendations per message to 4.",
"_____no_output_____"
]
],
[
[
"response = pinpoint.create_recommender_configuration(\n CreateRecommenderConfiguration={\n 'Attributes': {\n 'Recommendations.Name': 'Product Name',\n 'Recommendations.URL': 'Product Detail URL',\n 'Recommendations.Category': 'Product Category',\n 'Recommendations.Description': 'Product Description',\n 'Recommendations.Price': 'Product Price',\n 'Recommendations.ImageURL': 'Product Image URL'\n },\n 'Description': 'Retail Demo Store Personalize recommender for Pinpoint',\n 'Name': 'retaildemostore-recommender',\n 'RecommendationProviderIdType': 'PINPOINT_USER_ID',\n 'RecommendationProviderRoleArn': pinpoint_personalize_role_arn,\n 'RecommendationProviderUri': personalize_campaign_arn,\n 'RecommendationTransformerUri': lambda_function_arn,\n 'RecommendationsPerMessage': 4\n }\n)\n\nrecommender_id = response['RecommenderConfigurationResponse']['Id']\nprint('Pinpoint recommender configuration ID: ' + recommender_id)",
"_____no_output_____"
]
],
[
[
"### Verify Machine Learning Model / Recommender\n\nIf you open a web browser window/tab and browse to the Pinpoint service in the AWS console for the AWS account we're working with, you should see the ML Model / Recommender that we just created in Pinpoint.\n\n",
"_____no_output_____"
],
[
"## Create Personalized Email Templates\n\nWith Amazon Pinpoint we can create email templates that can be used to send to groups of our users based on criteria. We'll start by creating email templates for the following use-case then step through how we target and send emails to the right users at the appropriate time.\n\n- Welcome Email - sent to users shortly after creating a Retail Demo Store account\n- Abandoned Cart Email - sent to users who leave items in their cart without completing an order\n- Personalized Recommendations Email - includes recommendations from the recommeder we just created\n\n### Load Welcome Email Templates\n\nThe first email template will be a welcome email template that is sent to new users of the Retail Demo Store after they create an account. Our templates will support both HTML and plain text formats. We'll load both formats and create the template. You can find all templates used in this workshop in the `pinpoint-templates` directory where this notebook is located. They can also be found in the Retail Demo Store source code repository.\n\nLet's load the HTML version of our welcome template and then look at a snippet of it. Complete template is available for review at [pinpoint-templates/welcome-email-template.html](pinpoint-templates/welcome-email-template.html)",
"_____no_output_____"
]
],
[
[
"with open('pinpoint-templates/welcome-email-template.html', 'r') as html_file:\n html_template = html_file.read()\n ",
"_____no_output_____"
]
],
[
[
"```html \n// Excerpt from pinpoint-templates/welcome-email-template.html \n\n\n <table border=\"0\" cellpadding=\"0\" cellspacing=\"0\">\n <tr>\n <td>\n <h1>Thank you for joining the Retail Demo Store!</h1>\n <p><strong>Hi, {{User.UserAttributes.FirstName}}.</strong> We just wanted to send you a quick note thanking you for creating an account on the Retail Demo Store. We're excited to serve you. \n </p>\n <p>We pride ourselves in providing a wide variety of high quality products in our store and delivering exceptional customer service.\n </p>\n <p>Please drop-in and check out our store often to see what's new and for personalized recommendations we think you'll love.\n </p>\n <p>Cheers,<br/>Retail Demo Store team\n </p>\n </td>\n </tr>\n <tr>\n <td style=\"text-align: center; padding-top: 20px\">\n <small>Retail Demo Store © 2019-2020</small>\n </td>\n </tr>\n </table>\n\n\n\n```",
"_____no_output_____"
],
[
"Notice how we're using the mustache template tagging syntax, `{{User.UserAttributes.FirstName}}`, to display the user's first name. This will provide a nice touch of personalization to our welcome email.\n\nNext we'll load and display the text version of our welcome email.",
"_____no_output_____"
]
],
[
[
"with open('pinpoint-templates/welcome-email-template.txt', 'r') as text_file:\n text_template = text_file.read()\n \nprint('Text Template:')\nprint(text_template)",
"_____no_output_____"
]
],
[
[
"### Create Welcome Email Pinpoint Template\n\nNow let's take our HTML and text email template source and create a template in Amazon Pinpoint. We'll use a default substitution of \"there\" for the user's first name attribute if it is not set for some reason. This will result in the email greeting being \"Hi there,...\" rather than \"Hi ,...\" if we don't have a value for first name.",
"_____no_output_____"
]
],
[
[
"response = pinpoint.create_email_template(\n EmailTemplateRequest={\n 'Subject': 'Welcome to the Retail Demo Store',\n 'TemplateDescription': 'Welcome email sent to new customers',\n 'HtmlPart': html_template,\n 'TextPart': text_template,\n 'DefaultSubstitutions': json.dumps({\n 'User.UserAttributes.FirstName': 'there'\n })\n },\n TemplateName='RetailDemoStore-Welcome'\n)\n\nwelcome_template_arn = response['CreateTemplateMessageBody']['Arn']\nprint('Welcome email template ARN: ' + welcome_template_arn)",
"_____no_output_____"
]
],
[
[
"### Load Abandoned Cart Email Templates\n\nNext we'll create an email template that includes messaging for users who add items to their cart but fail to complete an order. The following is a snippet of the Abanoned Cart Email template. Notice how multiple style properties are bring set for email formatting. You can also see how the template refers to custom Attributes such as cart item properties like ```ShoppingCartItemTitle``` and ```ShoppingCartItemImageURL``` can be passed. Complete template available for review at [pinpoint-templates/abandoned-cart-email-template.html](pinpoint-templates/abandoned-cart-email-template.html)",
"_____no_output_____"
]
],
[
[
"with open('pinpoint-templates/abandoned-cart-email-template.html', 'r') as html_file:\n html_template = html_file.read()",
"_____no_output_____"
]
],
[
[
"```html\n\n// Excerpt from pinpoint-templates/abandoned-cart-email-template.html\n\n \n <tr>\n <td style=\"width:139px;\"> <img alt=\"{{Attributes.ShoppingCartItemTitle}}\" height=\"auto\" src=\"{{Attributes.ShoppingCartItemImageURL}}\" style=\"border:none;display:block;outline:none;text-decoration:none;height:auto;width:100%;font-size:13px;\" width=\"139\" /> </td>\n </tr>\n </tbody>\n </table>\n </td>\n </tr>\n </table>\n </div>\n <!--[if mso | IE]>\n </td>\n \n <td\n style=\"vertical-align:top;width:285px;\"\n >\n <![endif]-->\n <div class=\"mj-column-per-50 mj-outlook-group-fix\" style=\"font-size:0px;text-align:left;direction:ltr;display:inline-block;vertical-align:top;width:50%;\">\n <table border=\"0\" cellpadding=\"0\" cellspacing=\"0\" role=\"presentation\" style=\"vertical-align:top;\" width=\"100%\">\n <tr>\n <td align=\"center\" style=\"font-size:0px;padding:10px 25px;word-break:break-word;\">\n <div style=\"font-family:Ubuntu, Helvetica, Arial, sans-serif;font-size:18px;font-weight:bold;line-height:1;text-align:center;color:#000000;\">\n <p>{{Attributes.ShoppingCartItemTitle}}</p>\n </div>\n </td>\n </tr>\n <tr>\n <td align=\"center\" vertical-align=\"middle\" style=\"font-size:0px;padding:10px 25px;word-break:break-word;\">\n <table border=\"0\" cellpadding=\"0\" cellspacing=\"0\" role=\"presentation\" style=\"border-collapse:separate;line-height:100%;\">\n <tr>\n <td align=\"center\" bgcolor=\"#FF9900\" role=\"presentation\" style=\"border:none;border-radius:3px;cursor:auto;mso-padding-alt:10px 25px;background:#FF9900;\" valign=\"middle\"> <a href=\"{{Attributes.WebsiteCartURL}}\" style=\"display:inline-block;background:#FF9900;color:#ffffff;font-family:Ubuntu, Helvetica, Arial, sans-serif;font-size:9px;font-weight:normal;line-height:120%;margin:0;text-decoration:none;text-transform:none;padding:10px 25px;mso-padding-alt:0px;border-radius:3px;\"\n target=\"_blank\">\n BUY NOW\n </a> </td>\n </tr>\n \n\n```\n",
"_____no_output_____"
]
],
[
[
"with open('pinpoint-templates/abandoned-cart-email-template.txt', 'r') as text_file:\n text_template = text_file.read()\n \nprint('Text Template:')\nprint(text_template)",
"_____no_output_____"
]
],
[
[
"### Create Abandoned Cart Email Template\n\nNow we can create an email template in Pinpoint for our abandoned cart use-case.",
"_____no_output_____"
]
],
[
[
"response = pinpoint.create_email_template(\n EmailTemplateRequest={\n 'Subject': 'Retail Demo Store - Motivation to Complete Your Order',\n 'TemplateDescription': 'Abandoned cart email template',\n 'HtmlPart': html_template,\n 'TextPart': text_template,\n 'DefaultSubstitutions': json.dumps({\n 'User.UserAttributes.FirstName': 'there'\n }) \n },\n TemplateName='RetailDemoStore-AbandonedCart'\n)\n\nabandoned_cart_template_arn = response['CreateTemplateMessageBody']['Arn']\nprint('Abandoned cart email template ARN: ' + abandoned_cart_template_arn)",
"_____no_output_____"
]
],
[
[
"### Load Recommendations Email Templates\n\nNext we'll create an email template that includes recommendations from the Amazon Personalize product recommendation campaign that we created in the [Personalization workshop](../1-Personalization/1.1-Personalize.ipynb). If you haven't completed the personalization workshop, please do so now and come back to this workshop when complete.\n\nAs with the welcome email template, let's load and then view snippets of the HTML and text formats for our template. Complete template is available at [pinpoint-templates/recommendations-email-template.html](pinpoint-templates/recommendations-email-template.html)",
"_____no_output_____"
]
],
[
[
"with open('pinpoint-templates/recommendations-email-template.html', 'r') as html_file:\n html_template = html_file.read()",
"_____no_output_____"
]
],
[
[
" ``` html\n // Excerpt from pinpoint-templates/recommendations-email-template.html\n\n <table border=\"0\" cellpadding=\"0\" cellspacing=\"0\">\n <tr>\n <td>\n <h1>Hi, {{User.UserAttributes.FirstName}}. Greetings from the Retail Demo Store!</h1>\n <p>Here are a few products inspired by your shopping trends</p>\n <p> </p>\n </td>\n </tr>\n <tr>\n <td>\n <table border=\"0\" cellpadding=\"4\" cellspacing=\"0\">\n <tr valign=\"top\">\n <td style=\"text-align: left; width: 40%;\" width=\"40%\">\n <a href=\"{{Recommendations.URL.[0]}}\">\n <img src=\"{{Recommendations.ImageURL.[0]}}\" alt=\"{{Recommendations.Name.[0]}}\" style=\"min-width: 50px; max-width: 300px; border: 0; text-decoration:none; vertical-align: baseline;\"/>\n </a>\n </td>\n <td style=\"text-align: left;\">\n <h3>{{Recommendations.Name.[0]}}</h3>\n <p>{{Recommendations.Description.[0]}}</p>\n <p><strong>{{Recommendations.Price.[0]}}</strong></p>\n <p><a href=\"{{Recommendations.URL.[0]}}\"><strong>Buy Now!</strong></a></p>\n </td>\n </tr>\n```",
"_____no_output_____"
],
[
"Notice the use of several new mustache template tags in this template. For example, `{{Recommendations.Name.[0]}}` resolves to the product name of the first product recommended by Personalize. The product name came from our Lambda function which was called by Pinpoint after it called `get_recommendations` on our Personalize campaign. \n\nNext load the text version of our template.",
"_____no_output_____"
]
],
[
[
"with open('pinpoint-templates/recommendations-email-template.txt', 'r') as text_file:\n text_template = text_file.read()\n \nprint('Text Template:')\nprint(text_template)",
"_____no_output_____"
]
],
[
[
"### Create Recommendations Email Template\n\nThis time when we create the template in Pinpoint, we'll specify the `RecommenderId` for the machine learning model (Amazon Personalize) that we created earlier.",
"_____no_output_____"
]
],
[
[
"response = pinpoint.create_email_template(\n EmailTemplateRequest={\n 'Subject': 'Retail Demo Store - Products Just for You',\n 'TemplateDescription': 'Personalized recommendations email template',\n 'RecommenderId': recommender_id,\n 'HtmlPart': html_template,\n 'TextPart': text_template,\n 'DefaultSubstitutions': json.dumps({\n 'User.UserAttributes.FirstName': 'there'\n })\n },\n TemplateName='RetailDemoStore-Recommendations'\n)\n\nrecommendations_template_arn = response['CreateTemplateMessageBody']['Arn']\nprint('Recommendation email template ARN: ' + recommendations_template_arn)",
"_____no_output_____"
]
],
[
[
"### Verify Email Templates\n\nIf you open a web browser window/tab and browse to the Pinpoint service in the AWS console for the AWS account we're working with, you should see the message templates we just created.\n\n",
"_____no_output_____"
],
[
"## Enable Pinpoint Email Channel\n\nBefore we can setup Segments and Campaigns to send emails, we have to enable the email channel in Pinpoint and verify sending and receiving email addresses.\n\n> We'll be using the Pinpoint email channel in sandbox mode. This means that Pinpoint will only send emails from and to addresses that have been verified in the Pinpoint console.\n\nIn the Pinpoint console, click on \"All Projects\" and then the \"retaildemostore\" project.\n\n",
"_____no_output_____"
],
[
"### Email Settings\n\nFrom the \"retaildemostore\" project page, expand \"Settings\" in the left navigation and then click \"Email\". You will see that email has not yet been enabled as a channel. Click the \"Edit\" button to enable Pinpoint to send emails and to verify some email addresses.\n\n",
"_____no_output_____"
],
[
"### Verify Some Email Addresses\n\nOn the \"Edit email\" page, check the box to enable the email channel and enter a valid email address that you have the ability to check throughout the rest of this workshop.\n\n",
"_____no_output_____"
],
[
"### Verify Additional Email Addresses\n\nSo that we can send an email to more than one endpoint in this workshop, verify a couple more variations of your email address.\n\nAssuming your **valid** email address is `[email protected]`, add a few more variations using `+` notation such as...\n\n- `[email protected]`\n- `[email protected]`\n- `[email protected]`\n\nJust enter a variation, click the \"Verify email address\" button, and repeat until you've added a few more. Write down or commit to memory the variations you created--we'll need them later.\n\nBy adding these variations, we're able to create separate Retail Demo Store accounts for each email address and therefore separate endpoints in Pinpoint that we can target. Note that emails sent to the these variations should still be delivered to your same inbox.",
"_____no_output_____"
],
[
"### Check Your Inbox & Click Verification Links\n\nPinpoint should have sent verification emails to all of the email addresses you added above. Sign in to your email client and check your inbox for the verification emails. Once you receive the emails (it can take a few minutes), click on the verification link in **each email**. If after several minutes you don't receive the verification email or you want to use a different address, repeat the verification process above.\n\n> Your email address(es) must be verified before we can setup Campaigns in Pinpoint.\n\nAfter you click the verify link in the email sent to each variation of your email address, you should see a success page like the following.\n\n",
"_____no_output_____"
],
[
"## Let's Go Shopping - Create Retail Demo Store User Accounts & Pinpoint Endpoints\n\nNext let's create a few new user accounts in the Retail Demo Store Web UI using the email address(es) that we just verified. Based on the source code snippets we saw earlier, we know that the Retail Demo Store will create endpoints in Pinpoint for new accounts.\n\n<div class=\"alert alert-info\">\nIMPORTANT: each Retail Demo Store account must be created in an entirely separate web browser session in order for them to be created as separate endpoints in Pinpoint. Signing out and attempting to create a new account in the same browser will NOT work. The easiest way to do this successfully is to use Google Chrome and open new Incognito windows for each new account. Alternatively, you could use multiple browser types (i.e. Chrome, Firefox, Safari, IE) and/or separate devices to create accounts such as a mobile phone or tablet.\n</div>\n\n1. Open the Retail Demo Store Web UI in an new Incognito window. If you don't already have the Web UI open or need the URL, you can find it in the \"Outputs\" tab for the Retail Demo Store CloudFormation stack in your AWS account. Look for the \"WebURL\" output field, right click on the link, and select \"Open Link in Incognito Window\" (Chrome only).\n\n\n\n2. Click the \"Sign In\" button in the top navigation (right side) and then click on the \"Create account\" link in the \"Sign in\" form.\n\n\n\n3. A few seconds after creating your account you should receive an email with a six digit confirmation code. Enter this code on the confirmation page.\n\n\n\n4. Once confirmed you can sign in to your account with your user name and password. At this point you should have a endpoint in Pinpoint for this user.\n\n5. Close your Incognito window(s).\n\n6. Open a new Incognito window and **repeat the process for SOME (but not all) of your remaining email address variations** you verified in Pinpoint above. **As a reminder, it's important that you create each Retail Demo Store account in a separate/new Incognito window, browser application, or device. Otherwise, your accounts will overwrite the same endpoint in Pinpoint.**\n\n<div class=\"alert alert-info\">\nBe sure to hold back one or two of your verified email addresses until after we create a welcome email campaign below so the sign up events fall within the time window of the campaign.\n</div>",
"_____no_output_____"
],
[
"### Shopping Behavior\n\nWith your Retail Demo Store accounts created, perform some activities with some of your accounts. \n\n- For one of your users add some items to the shopping cart but do not checkout to simulate an abandoned cart scenario. \n- For another user, add some items to the cart and complete an order or two so that revenue events are sent all the way through to Pinpoint. \n- Also be sure to view a few products by clicking through to the product detail view. Select products that would indicate an affinity for a product type (e.g. shoes or electronics) so you can see how product recommendations are tailored in the product recommendations email.",
"_____no_output_____"
],
[
"## Create Pinpoint Segments\n\nWith our Recommender and message templates in place and a few test users created in the Retail Demo Store, let's turn to creating Segments in Pinpoint. After our Segments are created, we'll create some Campaigns.\n\n1. Start by browsing to the Amazon Pinpoint service page in the AWS account where the Retail Demo Store was deployed. Click on \"All Projects\" and you should see the \"retaildemostore\" project. Click on the \"retaildemostore\" project and then \"Segments\" in the left navigation. Click on the \"Create a segment\" button.\n\n",
"_____no_output_____"
],
[
"2. Then click on the \"Create segment\" button. We will be building a dynamic segment based on the endpoints that were automatically created when we created our Retail Demo Store user accounts. We'll include all endpoints that have an email address by adding a filter by channel type with a value of `EMAIL`. Name your segment \"AllEmailUsers\" and scroll down and click the \"Create segment\" button at the bottom of the page.\n\n",
"_____no_output_____"
],
[
"3. Create another segment that is based on the \"AllEmailUsers\" segment you just created but has an additional filter on the `HasShoppingCart` endpoint attribute and has a value of `true`. This represents all users that have a shopping cart and will be used for our abandoned cart campaign. If you don't see this endpoint attribute or don't see `true` as an option, switch to another browser tab/window and add items to the shopping cart for one of your test users.\n\n",
"_____no_output_____"
],
[
"## Create Campaigns\n\nWith our segments created for all users and for users with shopping carts, let's create campaigns for our welcome email, product recommendations, and abandoned cart use-cases.",
"_____no_output_____"
],
[
"### Welcome Email Campaign\n\nLet's start with with the welcome email campaign. For the \"retaildemostore\" project in Pinpoint, click \"Campaigns\" in the left navigation and then the \"Create a campaign\" button.\n\n1. For Step 1, give your campaign a name such as \"WelcomeEmail\", select \"Standard campaign\" as the campaign type, and \"Email\" as the channel. Click \"Next\" to continue.\n\n\n\n2. For Step 2, we will be using our \"AllEmailUsers\" dynamic segment. Click \"Next\" to continue.\n\n\n\n3. For Step 3, choose the \"RetailDemoStore-Welcome\" email template, scroll to the bottom of the page, and click \"Next\".\n\n\n\n4. For Step 4, we want the campaign to be sent when the `UserSignedUp` event occurs. Set the campaign start date to be today's date so that it begins immediately and the end date to be a few days into the future. **Be sure to adjust to your current time zone.**\n\n\n\n5. Scroll to the bottom of the page, click \"Next\".\n\n6. Click \"Launch campaign\" to launch your campaign.",
"_____no_output_____"
],
[
"<div class=\"alert alert-info\">\n<strong>Given that the welcome campaign is activated based on sign up events that occur between the campaign start and end times, to test this campaign you must wait until after the camapign starts and then use one of your remaining verified email addresses to create a new Retail Demo Store account.</strong>\n</div>",
"_____no_output_____"
],
[
"### Abandoned Cart Campaign\n\nTo create an abandoned cart campaign, repeat the steps you followed for the Welcome campaign above but this time select the `UsersWithCarts` segment, the `RetailDemoStore-AbandonedCart` email template, and the `Session Stop` event. This will trigger the abandoned cart email to be sent when users end their session while still having a shopping cart. Launch the campaign, wait for the campaign to start, and then close out some browser sessions for user(s) with items still in their cart. This can take some trial and error and waiting given the how browsers and devices trigger end of session events.",
"_____no_output_____"
],
[
"### Recommendations Campaign\n\nFinally, create a recommendations campaign that targets the `AllEmailUsers` segment and uses the `RetailDemoStore-Recommendations` message template. This time, however, rather than trigger the campaign based on an event, we'll send the campaign immediately. Click \"Next\", launch the campaign, and check the email inbox for your test accounts after a few moments.\n\n",
"_____no_output_____"
],
[
"## Bonus - Pinpoint Journeys\n\nWith Amazon Pinpoint journeys, you can create custom experiences for your customers using an easy to use, drag-and-drop interface. When you build a journey, you choose the activities that you want to add to the journey. These activities can perform a variety of different actions, like sending an email to journey participants, waiting a defined period of time, or splitting users based on a certain action, such as when they open or click a link in an email.\n\nUsing the segments and message templates you've already created, experiment with creating a journey that guides users through a messaging experience. For example, start a journey by sending all users the Recommendations message template. Then add a pause/wait step followed by a Multivariate Split that directs users down separate paths whether they've completed an order (hint: create a `OrderCompleted` segment), opened the Recommendations email, or done nothing. Perhaps users who completed an order might receive a message asking them to refer a friend to the Retail Demo Store and users who just opened the email might be sent a message with a coupon to motivate them to get shopping (you'll need to create a new message templates for these).",
"_____no_output_____"
],
[
"## Workshop Complete\n\nCongratulations! You have completed the Retail Demo Store Pinpoint Workshop.\n\n### Cleanup\n\nIf you launched the Retail Demo Store in your personal AWS account **AND** you're done with all workshops & your evaluation of the Retail Demo Store, you can remove all provisioned AWS resources and data by deleting the CloudFormation stack you used to deploy the Retail Demo Store. Although deleting the CloudFormation stack will delete the entire \"retaildemostore\" project in Pinpoint, including all endpoint data, it will not delete resources we created directly in this workshop (i.e. outside of the \"retaildemostore\" Pinpoint project). The following cleanup steps will remove the resources we created outside the \"retaildemostore\" Pinpoint project.\n\n> If you are participating in an AWS managed event such as a workshop and using an AWS provided temporary account, you can skip the following cleanup steps unless otherwise instructed.",
"_____no_output_____"
],
[
"#### Delete Recommeder Configuration",
"_____no_output_____"
]
],
[
[
"response = pinpoint.delete_recommender_configuration(RecommenderId=recommender_id)\nprint(json.dumps(response, indent=2))",
"_____no_output_____"
]
],
[
[
"#### Delete Email Message Templates",
"_____no_output_____"
]
],
[
[
"response = pinpoint.delete_email_template(TemplateName='RetailDemoStore-Welcome')\nprint(json.dumps(response, indent=2))",
"_____no_output_____"
],
[
"response = pinpoint.delete_email_template(TemplateName='RetailDemoStore-AbandonedCart')\nprint(json.dumps(response, indent=2))",
"_____no_output_____"
],
[
"response = pinpoint.delete_email_template(TemplateName='RetailDemoStore-Recommendations')\nprint(json.dumps(response, indent=2))",
"_____no_output_____"
]
],
[
[
"Other resources allocated for the Retail Demo Store will be deleted when the CloudFormation stack is deleted.\n\nEnd of workshop",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown"
] |
[
[
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
]
] |
cbdb6034291948e743e9fa5d65ab90cdb9d5aff9
| 14,145 |
ipynb
|
Jupyter Notebook
|
notebooks/ParametrizeYourQuery.ipynb
|
vijayraavi/jupyter-Kqlmagic
|
87a73b98eb114f6765477c1a300fd4e8f8f02481
|
[
"MIT"
] | null | null | null |
notebooks/ParametrizeYourQuery.ipynb
|
vijayraavi/jupyter-Kqlmagic
|
87a73b98eb114f6765477c1a300fd4e8f8f02481
|
[
"MIT"
] | null | null | null |
notebooks/ParametrizeYourQuery.ipynb
|
vijayraavi/jupyter-Kqlmagic
|
87a73b98eb114f6765477c1a300fd4e8f8f02481
|
[
"MIT"
] | null | null | null | 24.472318 | 161 | 0.536232 |
[
[
[
" # Kqlmagic - __parametrization__ features\n***\nExplains how to emebed python values in kql queries\n***\n***\n\n",
"_____no_output_____"
],
[
"## Make sure that you have the lastest version of Kqlmagic\nDownload Kqlmagic from PyPI and install/update\n(if latest version ims already installed you can skip this step)",
"_____no_output_____"
]
],
[
[
"#!pip install Kqlmagic --no-cache-dir --upgrade",
"_____no_output_____"
]
],
[
[
"## Add Kqlmagic to notebook magics",
"_____no_output_____"
]
],
[
[
"%reload_ext Kqlmagic",
"_____no_output_____"
]
],
[
[
"## Authenticate to get access to data",
"_____no_output_____"
]
],
[
[
"%kql azure-data-explorer://code;cluster='help';database='Samples'",
"_____no_output_____"
]
],
[
[
"## Use python user namespace as source of parameters\n- prefix query with **kql let statements** to parametrize the query\n- beware to the mapping: \n - int -> long \n - float -> real\n - str -> string\n - bool -> bool\n - datetime -> datetime\n - timedelta -> timespan\n - dict, list, set, tuple -> dynamic (only if can be serialized to json)\n - **pandas dataframe -> view table**\n - None -> null\n - unknown, str(value) == 'nan' -> real(null)\n - unknown, str(value) == 'NaT' -> datetime(null)\n - unknown str(value) == 'nat' -> time(null)\n - other -> string",
"_____no_output_____"
]
],
[
[
"from datetime import datetime, timedelta\nmy_limit = 10\nmy_not_state = 'TEXAS'\nmy_start_datetime = datetime(2007, 8, 29)\nmy_timespan = timedelta(days=100)\nmy_dict = {\"a\":1}\nmy_list = [\"x\", \"y\", \"z\"]\nmy_tuple = (\"t\", 44, my_limit)\nmy_set = {6,7,8}",
"_____no_output_____"
],
[
"%%kql \n let _dict_ = my_dict;\n let _list_ = my_list;\n let _tuple_ = my_tuple;\n let _set_ = my_set;\n let _start_time_ = my_start_datetime;\n let _timespan_ = my_timespan;\n let _limit_ = my_limit;\n let _not_val_ = my_not_state;\n StormEvents\n | where StartTime >= _start_time_\n | where EndTime <= _start_time_ + _timespan_\n | where State != _not_val_\n | summarize count() by State\n | extend d = _dict_\n | extend l = _list_\n | extend t = _tuple_\n | extend s = _set_\n | sort by count_ \n | limit _limit_",
"_____no_output_____"
]
],
[
[
"## Dataframe prameter as a kql table\n- prefix query with **kql let statement** that assigns a kql var to the dataframe\n- beware to the mapping of the dataframe to kql table columns types : \n - int8,int16,int32,int64,uint8,uint16,uint32,uint64 -> long \n - float16,float32,float64 -> real\n - character -> string\n - bytes -> string\n - void -> string\n - category -> string\n - datetime,datetime64,datetime64[ns],datetime64[ns,tz] -> datetime\n - timedelta,timedelta64,timedelta64[ns] -> timespan\n - bool -> bool\n - record -> dynamic\n - complex64,complex128 -> dynamic([real, imag])\n - object -> if all objects of type:\n - dict,list,tuple,set -> dynamic (only if can be serialized to json)\n - bool or nan -> bool\n - float or nan -> float\n - int or nan -> long\n - datetime or 'NaT' -> datetime\n - timedeltae or 'NaT' -> timespan\n - other -> string",
"_____no_output_____"
]
],
[
[
"my_df =_kql_raw_result_.to_dataframe()",
"_____no_output_____"
],
[
"my_df",
"_____no_output_____"
],
[
"%%kql \nlet _my_table_ = my_df;\n_my_table_ | project State, s, t | limit 3",
"_____no_output_____"
],
[
"_kql_raw_result_.parametrized_query",
"_____no_output_____"
]
],
[
[
"## Parametrize the whole query string",
"_____no_output_____"
]
],
[
[
"sort_col = 'count_'\nmy_query = \"\"\"StormEvents \n | where State != 'OHIO'\n | summarize count() by State\n | sort by {0} \n | limit 5\"\"\".format(sort_col)",
"_____no_output_____"
],
[
"%kql -query my_query",
"_____no_output_____"
]
],
[
[
"## Use python dictionary as source of parameters\n- set option -params_dict with the name of a python variable that refer to the dictionary\n- prefix query with kql let statements to parametrize the query",
"_____no_output_____"
]
],
[
[
"p_dict = {'p_limit':20, 'p_not_state':'IOWA'}",
"_____no_output_____"
],
[
"%%kql \n -params_dict p_dict\n let _limit_ = p_limit;\n let _not_val_ = p_not_state;\n StormEvents \n | where State != _not_val_\n | summarize count() by State\n | sort by count_ \n | limit _limit_",
"_____no_output_____"
]
],
[
[
"## Use python dictionary expression as source of parameters\n- set option -params_dict with a dictionary string (python format)\n- prefix query with kql let statements to parametrize the query\n- **make sure that the dictionary expression is without spaces**",
"_____no_output_____"
]
],
[
[
"%%kql \n -params_dict {'p_limit':5,'p_not_state':'OHIO'}\n let _limit_ = p_limit;\n let _not_val_ = p_not_state;\n StormEvents \n | where State != _not_val_\n | summarize count() by State\n | sort by count_ \n | limit _limit_",
"_____no_output_____"
]
],
[
[
"## get query string\n- shows the original query, as in the input cell",
"_____no_output_____"
]
],
[
[
"_kql_raw_result_.query",
"_____no_output_____"
]
],
[
[
"## get parametrized query string\n- shows the parametrized query, that was submited to kusto",
"_____no_output_____"
]
],
[
[
"_kql_raw_result_.parametrized_query",
"_____no_output_____"
]
],
[
[
" - ### <span style=\"color:#82CAFA\">*Note - additional let statements were added to the original query, one let statement for each parameter*</span>",
"_____no_output_____"
]
],
[
[
"p_dict = {'p_limit':5,'p_not_state':'OHIO'}",
"_____no_output_____"
],
[
"%%kql \n -params_dict p_dict\n let _limit_ = p_limit;\n let _not_val_ = p_not_state;\n StormEvents \n | where State != _not_val_\n | summarize count() by State\n | sort by count_ \n | limit _limit_",
"_____no_output_____"
]
],
[
[
"## parameters dictionary is modified",
"_____no_output_____"
]
],
[
[
"p_dict = {'p_limit': 5, 'p_not_state': 'IOWA'}",
"_____no_output_____"
]
],
[
[
"## refresh use original parameters\n- the same parameter values are used",
"_____no_output_____"
]
],
[
[
"_kql_raw_result_.refresh()",
"_____no_output_____"
]
],
[
[
" - ### <span style=\"color:#82CAFA\">*Note - the refresh method use the original parameter values, as they were set*</span>",
"_____no_output_____"
],
[
"## submit use the current python values as parameters\n- a new query is created and parametrized with the current python values",
"_____no_output_____"
]
],
[
[
"_kql_raw_result_.submit()",
"_____no_output_____"
]
],
[
[
" - ### <span style=\"color:#82CAFA\">*Note - the submit method cretes a new query and parametrize with the current parameter values*</span>",
"_____no_output_____"
],
[
"## Parametrize option\nall options can be parametrized.\ninstead of providing a quoted parameter value, specify the python variable or python expression\n\n- beware, that python expression must not have spaces !!! \n - valid expression examples: ```my_var```, ```str(type(x))```, ```[a,1,2]```\n - invalid expressions: ```str( type ( x ) )```, ```[a, 1, 2]```",
"_____no_output_____"
]
],
[
[
"table_package = 'pandas'\nmy_popup_state = True",
"_____no_output_____"
],
[
"%%kql -tp=table_package -pw=my_popup_state -f=table_package!='pandas'\n StormEvents \n | where State != 'OHIO'\n | summarize count() by State\n | sort by count_ \n | limit 5",
"_____no_output_____"
]
],
[
[
"## Parametrize commands\nall commands can be parametrized.\ninstead of providing a quoted parameter value, specify the python variable or python expression.\n\n- **note**, if instead of the python expression, you specify a variable that starts with $, it will be retreived from the environment variables.<br><br>\n- **beware**, that python expression must not have spaces !!!",
"_____no_output_____"
]
],
[
[
"my_topic = \"kql\"",
"_____no_output_____"
],
[
"%kql --help my_topic",
"_____no_output_____"
]
],
[
[
"## Parametrize connection string\nall values in connection string can be parametrized.\ninstead of providing a quoted parameter value, specify the python variable or python expression\n\n- **note**, if you don't specify the credential's secret you will be prompted.\n- **note**, if instead of the python expression, you specify a variable that starts with $, it will be retreived from the environment variables.<br><br>\n\n- beware, that python expression must not have spaces !!!",
"_____no_output_____"
]
],
[
[
"my_appid = \"DEMO_APP\"\nmy_appkey = \"DEMO_KEY\"",
"_____no_output_____"
],
[
"%kql appinsights://appid=my_appid;appkey=my_appkey",
"_____no_output_____"
]
],
[
[
"## Parametrize the whold connection string",
"_____no_output_____"
]
],
[
[
"my_connection_str = \"\"\"\nloganalytics://workspace='DEMO_WORKSPACE';appkey='DEMO_KEY';alias='myworkspace'\n\"\"\"",
"_____no_output_____"
],
[
"%kql -conn=my_connection_str",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
]
] |
cbdb7353e11213dfc7b67fc157f59afb617e4cc7
| 23,588 |
ipynb
|
Jupyter Notebook
|
notebooks/lab13_cluster_analysis.ipynb
|
luitelgaurav/Data-Mining-21-22
|
f58069df37dd06cae260cc7369f92a2f787714b7
|
[
"MIT"
] | 10 |
2021-09-14T13:48:55.000Z
|
2022-02-07T15:00:02.000Z
|
notebooks/lab13_cluster_analysis.ipynb
|
luitelgaurav/Data-Mining-21-22
|
f58069df37dd06cae260cc7369f92a2f787714b7
|
[
"MIT"
] | 3 |
2021-09-14T19:37:47.000Z
|
2021-12-04T11:11:52.000Z
|
notebooks/lab13_cluster_analysis.ipynb
|
luitelgaurav/Data-Mining-21-22
|
f58069df37dd06cae260cc7369f92a2f787714b7
|
[
"MIT"
] | 144 |
2021-09-14T13:59:26.000Z
|
2022-02-16T14:47:18.000Z
| 31.325365 | 163 | 0.576988 |
[
[
[
"# !pip install graphviz",
"_____no_output_____"
]
],
[
[
"To produce the decision tree visualization you should install the graphviz package into your system: \nhttps://stackoverflow.com/questions/35064304/runtimeerror-make-sure-the-graphviz-executables-are-on-your-systems-path-aft",
"_____no_output_____"
]
],
[
[
"# Run one of these in case you have problems with graphviz\n\n# All users: try this first\n# ! conda install graphviz\n\n# If that doesn't work:\n# Ubuntu/Debian users only\n# ! sudo apt-get update && sudo apt-get install graphviz\n\n# Mac users only (assuming you have homebrew installed)\n# ! brew install graphviz\n\n# Windows users, check the stack overflow link. Sorry!",
"_____no_output_____"
],
[
"from collections import Counter\nfrom os.path import join\nimport matplotlib.pyplot as plt\nimport seaborn as sns\nimport numpy as np\nimport pandas as pd\nfrom sklearn.cluster import DBSCAN, KMeans, AgglomerativeClustering\nfrom sklearn.base import clone\nfrom sklearn.metrics import pairwise_distances\nfrom scipy.cluster.hierarchy import dendrogram\nfrom sklearn.manifold import TSNE\nfrom sklearn.tree import DecisionTreeClassifier, export_graphviz\nfrom sklearn.model_selection import train_test_split\nimport graphviz\n\nsns.set()",
"_____no_output_____"
]
],
[
[
"## Import preprocessed data",
"_____no_output_____"
]
],
[
[
"df = pd.read_csv(join('..', 'data', 'tugas_preprocessed.csv'))",
"_____no_output_____"
],
[
"# Splitting feature names into groups\nnon_metric_features = df.columns[df.columns.str.startswith('x')]\npc_features = df.columns[df.columns.str.startswith('PC')]\nmetric_features = df.columns[~df.columns.str.startswith('x') & ~df.columns.str.startswith('PC')]",
"_____no_output_____"
]
],
[
[
"# Before we proceed\n\n- Consider applying the outlier filtering method discussed last class.\n - We manually filtered the dataset's outliers based on a univariate analysis\n- Consider dropping/transforming the variable \"rcn\". Why?\n - Very little correlation with any other variables\n - Remember the Component planes: the SOM's units were indistinguishable on this variable",
"_____no_output_____"
]
],
[
[
"# Based on the hyperparameters found in the previous class\ndbscan = DBSCAN(eps=1.9, min_samples=20, n_jobs=4)\ndbscan_labels = dbscan.fit_predict(df[metric_features])\nCounter(dbscan_labels)",
"_____no_output_____"
],
[
"# Save the newly detected outliers (they will be classified later based on the final clusters)\ndf_out = # CODE HERE\n\n# New df without outliers and 'rcn'\ndf = # CODE HERE\n\n# Update metric features list\nmetric_features = # CODE HERE",
"_____no_output_____"
]
],
[
[
"# Clustering by Perspectives\n- Demographic/Behavioral Perspective:\n- Product Perspective:",
"_____no_output_____"
]
],
[
[
"# Split variables into perspectives (example, requires critical thinking and domain knowledge)\ndemographic_features = [\n 'income',\n 'frq',\n 'per_net_purchase',\n 'spent_online'\n]\n\npreference_features = [\n 'clothes', \n 'kitchen', \n 'small_appliances',\n 'toys', \n 'house_keeping', \n]\n\ndf_dem = df[demographic_features].copy()\ndf_prf = df[preference_features].copy()",
"_____no_output_____"
]
],
[
[
"## Testing on K-means and Hierarchical clustering\nBased on (1) our previous tests and (2) the context of this problem, the optimal number of clusters is expected to be between 3 and 7.",
"_____no_output_____"
]
],
[
[
"def get_ss(df):\n \"\"\"Computes the sum of squares for all variables given a dataset\n \"\"\"\n ss = np.sum(df.var() * (df.count() - 1))\n return ss # return sum of sum of squares of each df variable\n\ndef r2(df, labels):\n sst = get_ss(df)\n ssw = np.sum(df.groupby(labels).apply(get_ss))\n return 1 - ssw/sst\n \ndef get_r2_scores(df, clusterer, min_k=2, max_k=10):\n \"\"\"\n Loop over different values of k. To be used with sklearn clusterers.\n \"\"\"\n r2_clust = {}\n for n in range(min_k, max_k):\n clust = clone(clusterer).set_params(n_clusters=n)\n labels = clust.fit_predict(df)\n r2_clust[n] = r2(df, labels)\n return r2_clust\n\n\n# Set up the clusterers (try out a KMeans and a AgglomerativeClustering)\nkmeans = # CODE HERE\n\nhierarchical = # CODE HERE",
"_____no_output_____"
]
],
[
[
"### Finding the optimal clusterer on demographic variables",
"_____no_output_____"
]
],
[
[
"# Obtaining the R² scores for each cluster solution on demographic variables\nr2_scores = {}\nr2_scores['kmeans'] = get_r2_scores(df_dem, kmeans)\n\nfor linkage in ['complete', 'average', 'single', 'ward']:\n r2_scores[linkage] = get_r2_scores(\n df_dem, hierarchical.set_params(linkage=linkage)\n )\n\npd.DataFrame(r2_scores)",
"_____no_output_____"
],
[
"# Visualizing the R² scores for each cluster solution on demographic variables\npd.DataFrame(r2_scores).plot.line(figsize=(10,7))\n\nplt.title(\"Demographic Variables:\\nR² plot for various clustering methods\\n\", fontsize=21)\nplt.legend(title=\"Cluster methods\", title_fontsize=11)\nplt.xlabel(\"Number of clusters\", fontsize=13)\nplt.ylabel(\"R² metric\", fontsize=13)\nplt.show()",
"_____no_output_____"
]
],
[
[
"### Repeat the process for product variables",
"_____no_output_____"
]
],
[
[
"# Obtaining the R² scores for each cluster solution on product variables\nr2_scores = {}\nr2_scores['kmeans'] = get_r2_scores(df_prf, kmeans)\n\nfor linkage in ['complete', 'average', 'single', 'ward']:\n r2_scores[linkage] = get_r2_scores(\n df_prf, hierarchical.set_params(linkage=linkage)\n )\n\n# Visualizing the R² scores for each cluster solution on product variables\npd.DataFrame(r2_scores).plot.line(figsize=(10,7))\n\nplt.title(\"Product Variables:\\nR2 plot for various clustering methods\\n\", fontsize=21)\nplt.legend(title=\"Cluster methods\", title_fontsize=11)\nplt.xlabel(\"Number of clusters\", fontsize=13)\nplt.ylabel(\"R2 metric\", fontsize=13)\nplt.show()",
"_____no_output_____"
]
],
[
[
"## Merging the Perspectives\n- How can we merge different cluster solutions?",
"_____no_output_____"
]
],
[
[
"# Applying the right clustering (algorithm and number of clusters) for each perspective\nkmeans_prod = # CODE HERE\nprod_labels = kmeans_prod.fit_predict(df_prf)\n\nkmeans_behav = # CODE HERE\nbehavior_labels = kmeans_behav.fit_predict(df_dem)\n\n# Setting new columns\ndf['product_labels'] = prod_labels\ndf['behavior_labels'] = behavior_labels",
"_____no_output_____"
],
[
"# Count label frequencies (contigency table)\n# CODE HERE",
"_____no_output_____"
]
],
[
[
"### Manual merging: Merge lowest frequency clusters into closest clusters",
"_____no_output_____"
]
],
[
[
"# Clusters with low frequency to be merged:\nto_merge = # CODE HERE\n\ndf_centroids = df.groupby(['behavior_labels', 'product_labels'])\\\n [metric_features].mean()\n\n# Computing the euclidean distance matrix between the centroids\neuclidean = # CODE HERE\ndf_dists = pd.DataFrame(\n euclidean, columns=df_centroids.index, index=df_centroids.index\n)\n\n# Merging each low frequency clustering (source) to the closest cluster (target)\nsource_target = {}\nfor clus in to_merge:\n if clus not in source_target.values():\n source_target[clus] = df_dists.loc[clus].sort_values().index[1]\n\nsource_target",
"_____no_output_____"
],
[
"df_ = df.copy()\n\n# Changing the behavior_labels and product_labels based on source_target\nfor source, target in source_target.items():\n mask = # CODE HERE (changing the behavior and product labels of each source based on target)\n df_.loc[mask, 'behavior_labels'] = target[0]\n df_.loc[mask, 'product_labels'] = target[1]\n\n# New contigency table\ndf_.groupby(['product_labels', 'behavior_labels'])\\\n .size()\\\n .to_frame()\\\n .reset_index()\\\n .pivot('behavior_labels', 'product_labels', 0)",
"_____no_output_____"
]
],
[
[
"### Merging using Hierarchical clustering",
"_____no_output_____"
]
],
[
[
"# Centroids of the concatenated cluster labels\ndf_centroids = # CODE HERE (group by both on behavior and product label)\ndf_centroids",
"_____no_output_____"
],
[
"# Using Hierarchical clustering to merge the concatenated cluster centroids\nhclust = AgglomerativeClustering(\n linkage='ward', \n affinity='euclidean', \n distance_threshold=0, \n n_clusters=None\n)\nhclust_labels = hclust.fit_predict(df_centroids)",
"_____no_output_____"
],
[
"# Adapted from:\n# https://scikit-learn.org/stable/auto_examples/cluster/plot_agglomerative_dendrogram.html#sphx-glr-auto-examples-cluster-plot-agglomerative-dendrogram-py\n\n# create the counts of samples under each node (number of points being merged)\ncounts = np.zeros(hclust.children_.shape[0])\nn_samples = len(hclust.labels_)\n\n# hclust.children_ contains the observation ids that are being merged together\n# At the i-th iteration, children[i][0] and children[i][1] are merged to form node n_samples + i\nfor i, merge in enumerate(hclust.children_):\n # track the number of observations in the current cluster being formed\n current_count = 0\n for child_idx in merge:\n if child_idx < n_samples:\n # If this is True, then we are merging an observation\n current_count += 1 # leaf node\n else:\n # Otherwise, we are merging a previously formed cluster\n current_count += counts[child_idx - n_samples]\n counts[i] = current_count\n\n# the hclust.children_ is used to indicate the two points/clusters being merged (dendrogram's u-joins)\n# the hclust.distances_ indicates the distance between the two points/clusters (height of the u-joins)\n# the counts indicate the number of points being merged (dendrogram's x-axis)\nlinkage_matrix = np.column_stack(\n [hclust.children_, hclust.distances_, counts]\n).astype(float)\n\n# Plot the corresponding dendrogram\nsns.set()\nfig = plt.figure(figsize=(11,5))\n# The Dendrogram parameters need to be tuned\ny_threshold = 2.3\ndendrogram(linkage_matrix, truncate_mode='level', labels=df_centroids.index, p=5, color_threshold=y_threshold, above_threshold_color='k')\nplt.hlines(y_threshold, 0, 1000, colors=\"r\", linestyles=\"dashed\")\nplt.title(f'Hierarchical Clustering - {linkage.title()}\\'s Dendrogram', fontsize=21)\nplt.xlabel('Number of points in node (or index of point if no parenthesis)')\nplt.ylabel(f'Euclidean Distance', fontsize=13)\nplt.show()",
"_____no_output_____"
],
[
"# Re-running the Hierarchical clustering based on the correct number of clusters\nhclust = # CODE HERE\nhclust_labels = hclust.fit_predict(df_centroids)\ndf_centroids['hclust_labels'] = hclust_labels\n\ndf_centroids # centroid's cluster labels",
"_____no_output_____"
],
[
"# Mapper between concatenated clusters and hierarchical clusters\ncluster_mapper = df_centroids['hclust_labels'].to_dict()\n\ndf_ = df.copy()\n\n# Mapping the hierarchical clusters on the centroids to the observations\ndf_['merged_labels'] = df_.apply(# CODE HERE)\n\n# Merged cluster centroids\ndf_.groupby('merged_labels').mean()[metric_features]",
"_____no_output_____"
],
[
"#Merge cluster contigency table\n# Getting size of each final cluster\ndf_counts = df_.groupby('merged_labels')\\\n .size()\\\n .to_frame()\n\n# Getting the product and behavior labels\ndf_counts = df_counts\\\n .rename({v:k for k, v in cluster_mapper.items()})\\\n .reset_index()\n\ndf_counts['behavior_labels'] = df_counts['merged_labels'].apply(lambda x: x[0])\ndf_counts['product_labels'] = df_counts['merged_labels'].apply(lambda x: x[1])\ndf_counts.pivot('behavior_labels', 'product_labels', 0)",
"_____no_output_____"
],
[
"# Setting df to have the final product, behavior and merged clusters\ndf = df_.copy()",
"_____no_output_____"
]
],
[
[
"## Cluster Analysis",
"_____no_output_____"
]
],
[
[
"def cluster_profiles(df, label_columns, figsize, compar_titles=None):\n \"\"\"\n Pass df with labels columns of one or multiple clustering labels. \n Then specify this label columns to perform the cluster profile according to them.\n \"\"\"\n if compar_titles == None:\n compar_titles = [\"\"]*len(label_columns)\n \n sns.set()\n fig, axes = plt.subplots(nrows=len(label_columns), ncols=2, figsize=figsize, squeeze=False)\n for ax, label, titl in zip(axes, label_columns, compar_titles):\n # Filtering df\n drop_cols = [i for i in label_columns if i!=label]\n dfax = df.drop(drop_cols, axis=1)\n \n # Getting the cluster centroids and counts\n centroids = dfax.groupby(by=label, as_index=False).mean()\n counts = dfax.groupby(by=label, as_index=False).count().iloc[:,[0,1]]\n counts.columns = [label, \"counts\"]\n \n # Setting Data\n pd.plotting.parallel_coordinates(centroids, label, color=sns.color_palette(), ax=ax[0])\n sns.barplot(x=label, y=\"counts\", data=counts, ax=ax[1])\n\n #Setting Layout\n handles, _ = ax[0].get_legend_handles_labels()\n cluster_labels = [\"Cluster {}\".format(i) for i in range(len(handles))]\n ax[0].annotate(text=titl, xy=(0.95,1.1), xycoords='axes fraction', fontsize=13, fontweight = 'heavy') \n ax[0].legend(handles, cluster_labels) # Adaptable to number of clusters\n ax[0].axhline(color=\"black\", linestyle=\"--\")\n ax[0].set_title(\"Cluster Means - {} Clusters\".format(len(handles)), fontsize=13)\n ax[0].set_xticklabels(ax[0].get_xticklabels(), rotation=-20)\n ax[1].set_xticklabels(cluster_labels)\n ax[1].set_xlabel(\"\")\n ax[1].set_ylabel(\"Absolute Frequency\")\n ax[1].set_title(\"Cluster Sizes - {} Clusters\".format(len(handles)), fontsize=13)\n \n plt.subplots_adjust(hspace=0.4, top=0.90)\n plt.suptitle(\"Cluster Simple Profilling\", fontsize=23)\n plt.show()",
"_____no_output_____"
],
[
"# Profilling each cluster (product, behavior, merged)\ncluster_profiles(\n df = df[metric_features.to_list() + ['product_labels', 'behavior_labels', 'merged_labels']], \n label_columns = ['product_labels', 'behavior_labels', 'merged_labels'], \n figsize = (28, 13), \n compar_titles = [\"Product clustering\", \"Behavior clustering\", \"Merged clusters\"]\n)",
"_____no_output_____"
]
],
[
[
"## Cluster visualization using t-SNE",
"_____no_output_____"
]
],
[
[
"# This is step can be quite time consuming\ntwo_dim = # CODE HERE (explore the TSNE class and obtain the 2D coordinates)",
"_____no_output_____"
],
[
"# t-SNE visualization\npd.DataFrame(two_dim).plot.scatter(x=0, y=1, c=df['merged_labels'], colormap='tab10', figsize=(15,10))\nplt.show()",
"_____no_output_____"
]
],
[
[
"## Assess feature importance and reclassify outliers",
"_____no_output_____"
],
[
"### Using the R²\nWhat proportion of each variables total SS is explained between clusters?",
"_____no_output_____"
]
],
[
[
"def get_ss_variables(df):\n \"\"\"Get the SS for each variable\n \"\"\"\n ss_vars = df.var() * (df.count() - 1)\n return ss_vars\n\ndef r2_variables(df, labels):\n \"\"\"Get the R² for each variable\n \"\"\"\n sst_vars = get_ss_variables(df)\n ssw_vars = np.sum(df.groupby(labels).apply(get_ss_variables))\n return 1 - ssw_vars/sst_vars",
"_____no_output_____"
],
[
"# We are essentially decomposing the R² into the R² for each variable\n# CODE HERE (obtain the R² for each variable using the functions above)",
"_____no_output_____"
]
],
[
[
"### Using a Decision Tree\nWe get the normalized total reduction of the criterion (gini or entropy) brought by that feature (also known as Gini importance).",
"_____no_output_____"
]
],
[
[
"# Preparing the data\nX = df.drop(columns=['product_labels','behavior_labels','merged_labels'])\ny = df.merged_labels\n\n# Splitting the data\nX_train, X_test, y_train, y_test = train_test_split(\n X, y, test_size=0.2, random_state=42\n)\n\n# Fitting the decision tree\ndt = # CODE HERE (set a simple decision tree with max depth of 3)\ndt.fit(X_train, y_train)\nprint(\"It is estimated that in average, we are able to predict {0:.2f}% of the customers correctly\".format(dt.score(X_test, y_test)*100))",
"_____no_output_____"
],
[
"# Assessing feature importance\npd.Series(dt.feature_importances_, index=X_train.columns)",
"_____no_output_____"
],
[
"# Predicting the cluster labels of the outliers\ndf_out['merged_labels'] = # CODE HERE\ndf_out.head()",
"_____no_output_____"
],
[
"# Visualizing the decision tree\ndot_data = export_graphviz(dt, out_file=None, \n feature_names=X.columns.to_list(),\n filled=True,\n rounded=True,\n special_characters=True) \ngraphviz.Source(dot_data)",
"_____no_output_____"
]
]
] |
[
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code"
]
] |
cbdb7c925dbd643cbdaae41329e60f393d0dbf21
| 45,525 |
ipynb
|
Jupyter Notebook
|
9.tcga-classify/interpret-all-feature-ensemble-model.ipynb
|
ctb/BioBombe
|
8e4da86914d3d45d57f549b057d17e42a712ed62
|
[
"BSD-3-Clause"
] | 1 |
2018-12-04T21:35:59.000Z
|
2018-12-04T21:35:59.000Z
|
9.tcga-classify/interpret-all-feature-ensemble-model.ipynb
|
ctb/BioBombe
|
8e4da86914d3d45d57f549b057d17e42a712ed62
|
[
"BSD-3-Clause"
] | 33 |
2018-03-26T19:33:58.000Z
|
2018-12-07T14:49:36.000Z
|
9.tcga-classify/interpret-all-feature-ensemble-model.ipynb
|
ctb/BioBombe
|
8e4da86914d3d45d57f549b057d17e42a712ed62
|
[
"BSD-3-Clause"
] | 3 |
2018-03-22T18:21:41.000Z
|
2018-09-23T10:35:45.000Z
| 34.410431 | 212 | 0.406216 |
[
[
[
"## Interpreting Ensemble Compressed Features\n\n**Gregory Way, 2019**\n\nThe following notebook will assign biological knowledge to the compressed features using the network projection approach. I use the model previously identified that was used to predict TP53 inactivation.\nI observe the BioBombe gene set enrichment scores for the features with high coefficients in this model.",
"_____no_output_____"
]
],
[
[
"import os\nimport sys\nimport pandas as pd",
"_____no_output_____"
]
],
[
[
"## Load the `All Feature` Ensemble Model",
"_____no_output_____"
]
],
[
[
"model_file = os.path.join(\"results\", \"top_model_ensemble_all_features_tp53_feature_for_followup.tsv\")\ntop_model_df = pd.read_table(model_file)\ntop_model_df",
"_____no_output_____"
],
[
"coef_file = os.path.join(\"results\",\n \"mutation_ensemble_all\",\n \"TP53\",\n \"TP53_ensemble_all_features_coefficients.tsv.gz\")\ncoef_df = pd.read_table(coef_file).drop(['signal', 'z_dim', 'seed', 'algorithm'], axis='columns')\ncoef_df.head()",
"_____no_output_____"
],
[
"full_coef_id_df = (\n pd.DataFrame(coef_df.feature.str.split(\"_\").values.tolist(),\n columns=['algorithm', 'individual_feature', 'seed', 'k', 'signal'])\n)\n\nfull_coef_id_df = pd.concat([full_coef_id_df, coef_df], axis='columns')\nfull_coef_id_df = full_coef_id_df.query(\"abs > 0\").query(\"signal == 'signal'\")\n\nprint(full_coef_id_df.shape)\nfull_coef_id_df.head()",
"(317, 9)\n"
]
],
[
[
"## Load Network Projection Results",
"_____no_output_____"
]
],
[
[
"gph_dir = os.path.join(\"..\",\n \"6.biobombe-projection\",\n \"results\",\n \"tcga\",\n \"gph\",\n \"signal\")\ngph_files = os.listdir(gph_dir)",
"_____no_output_____"
],
[
"all_scores_list = []\nfor file in gph_files:\n file = os.path.join(gph_dir, file)\n scores_df = pd.read_table(file)\n all_scores_list.append(scores_df)",
"_____no_output_____"
],
[
"all_scores_df = pd.concat(all_scores_list, axis='rows')\n\nprint(all_scores_df.shape)\nall_scores_df.head()",
"(1542500, 8)\n"
],
[
"all_scores_df = all_scores_df.assign(big_feature_id=all_scores_df.algorithm + \"_\" +\n all_scores_df.feature.astype(str) + \"_\" +\n all_scores_df.seed.astype(str) + \"_\" +\n all_scores_df.z.astype(str) + \"_signal\")\nall_scores_df = all_scores_df.assign(abs_z_score=all_scores_df.z_score.abs())",
"_____no_output_____"
],
[
"all_coef_scores_df = (\n full_coef_id_df\n .merge(all_scores_df,\n how='left',\n left_on=\"feature\",\n right_on=\"big_feature_id\")\n .sort_values(by=['abs', 'abs_z_score'], ascending=False)\n .reset_index(drop=True)\n)\n\nall_coef_scores_df.head()",
"_____no_output_____"
],
[
"# Explore the biobombe scores for specific DAE features\ntop_n_features = 5\n\nbiobombe_df = (\n all_coef_scores_df\n .groupby('big_feature_id')\n .apply(func=lambda x: x.abs_z_score.nlargest(top_n_features))\n .reset_index()\n .merge(all_coef_scores_df\n .reset_index(),\n right_on=['index', 'abs_z_score', 'big_feature_id'],\n left_on=['level_1', 'abs_z_score', 'big_feature_id'])\n .drop(['level_1', 'index', 'feature_x',\n 'algorithm_x', 'seed_x',\n 'model_type', 'algorithm_y',\n 'feature_y', 'seed_y', 'z'], axis='columns')\n .sort_values(by=['abs', 'abs_z_score'], ascending=False)\n .reset_index(drop=True)\n)\n\nprint(biobombe_df.shape)\nbiobombe_df.head(20)",
"(1585, 11)\n"
],
[
"# Output biobombe scores applied to the all feature ensemble model\nfile = os.path.join('results', 'tcga_tp53_classify_top_biobombe_scores_all_feature_ensemble_model_table.tsv')\nbiobombe_df.to_csv(file, sep='\\t', index=False)",
"_____no_output_____"
]
],
[
[
"## Detect the highest contributing variables",
"_____no_output_____"
]
],
[
[
"neg_biobombe_df = biobombe_df.query(\"weight < 0\")\npos_biobombe_df = biobombe_df.query(\"weight > 0\")\n\ntop_neg_variables_df = neg_biobombe_df.groupby(\"variable\")['weight'].sum().sort_values(ascending=True)\ntop_pos_variables_df = pos_biobombe_df.groupby(\"variable\")['weight'].sum().sort_values(ascending=False)",
"_____no_output_____"
],
[
"full_result_df = pd.DataFrame(pd.concat([top_pos_variables_df, top_neg_variables_df]))\nfull_result_df = (\n full_result_df\n .assign(abs_weight=full_result_df.weight.abs())\n .sort_values(by='abs_weight', ascending=False)\n)\n\nfull_result_df.head()",
"_____no_output_____"
],
[
"# Output biobombe scores applied to the all feature ensemble model\nfile = os.path.join('results', 'tcga_tp53_classify_aggregate_biobombe_scores_all_feature_ensemble.tsv')\nfull_result_df.to_csv(file, sep='\\t', index=False)",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
]
] |
cbdb8eb9016eae510b2317fbea0b1dec76ab025e
| 20,991 |
ipynb
|
Jupyter Notebook
|
notebooks/Plot Antarctic Tidal Currents.ipynb
|
bludka/pyTMD
|
2f866508f743b8b130b21c5379c400b39f73d3f0
|
[
"MIT"
] | null | null | null |
notebooks/Plot Antarctic Tidal Currents.ipynb
|
bludka/pyTMD
|
2f866508f743b8b130b21c5379c400b39f73d3f0
|
[
"MIT"
] | null | null | null |
notebooks/Plot Antarctic Tidal Currents.ipynb
|
bludka/pyTMD
|
2f866508f743b8b130b21c5379c400b39f73d3f0
|
[
"MIT"
] | null | null | null | 43.73125 | 125 | 0.58506 |
[
[
[
"Plot Antarctic Tidal Currents\n=============================\n\nDemonstrates plotting hourly tidal currents around Antarctica\n\nOTIS format tidal solutions provided by Ohio State University and ESR \n- http://volkov.oce.orst.edu/tides/region.html \n- https://www.esr.org/research/polar-tide-models/list-of-polar-tide-models/\n- ftp://ftp.esr.org/pub/datasets/tmd/ \n\nFinite Element Solution (FES) provided by AVISO \n- https://www.aviso.altimetry.fr/en/data/products/auxiliary-products/global-tide-fes.html\n\n#### Python Dependencies\n - [numpy: Scientific Computing Tools For Python](https://www.numpy.org) \n - [scipy: Scientific Tools for Python](https://www.scipy.org/) \n - [pyproj: Python interface to PROJ library](https://pypi.org/project/pyproj/) \n - [netCDF4: Python interface to the netCDF C library](https://unidata.github.io/netcdf4-python/) \n - [matplotlib: Python 2D plotting library](http://matplotlib.org/) \n - [cartopy: Python package designed for geospatial data processing](https://scitools.org.uk/cartopy/docs/latest/) \n\n#### Program Dependencies\n\n- `calc_astrol_longitudes.py`: computes the basic astronomical mean longitudes \n- `calc_delta_time.py`: calculates difference between universal and dynamic time \n- `convert_ll_xy.py`: convert lat/lon points to and from projected coordinates \n- `load_constituent.py`: loads parameters for a given tidal constituent \n- `load_nodal_corrections.py`: load the nodal corrections for tidal constituents \n- `infer_minor_corrections.py`: return corrections for minor constituents \n- `read_tide_model.py`: extract tidal harmonic constants from OTIS tide models \n- `read_netcdf_model.py`: extract tidal harmonic constants from netcdf models \n- `read_FES_model.py`: extract tidal harmonic constants from FES tide models \n- `predict_tide.py`: predict tidal elevation at a single time using harmonic constants \n\nThis notebook uses Jupyter widgets to set parameters for calculating the tidal maps. \nThe widgets can be installed as described below. \n```\npip3 install --user ipywidgets\njupyter nbextension install --user --py widgetsnbextension\njupyter nbextension enable --user --py widgetsnbextension\njupyter-notebook\n```",
"_____no_output_____"
],
[
"#### Load modules",
"_____no_output_____"
]
],
[
[
"import os\nimport pyproj\nimport datetime\nimport numpy as np\nimport matplotlib\nmatplotlib.rcParams['axes.linewidth'] = 2.0\nmatplotlib.rcParams[\"animation.html\"] = \"jshtml\"\nimport matplotlib.pyplot as plt\nimport matplotlib.animation as animation\nimport cartopy.crs as ccrs\nimport ipywidgets as widgets\nfrom IPython.display import HTML\n\nimport pyTMD.time\nfrom pyTMD.calc_delta_time import calc_delta_time\nfrom pyTMD.read_tide_model import extract_tidal_constants\nfrom pyTMD.read_netcdf_model import extract_netcdf_constants\nfrom pyTMD.read_GOT_model import extract_GOT_constants\nfrom pyTMD.read_FES_model import extract_FES_constants\nfrom pyTMD.infer_minor_corrections import infer_minor_corrections\nfrom pyTMD.predict_tide import predict_tide\n#-- autoreload\n%load_ext autoreload\n%autoreload 2",
"_____no_output_____"
]
],
[
[
"#### Set parameters for program\n\n- Model directory \n- Tide model \n- Date to run ",
"_____no_output_____"
]
],
[
[
"#-- set the directory with tide models\ndirText = widgets.Text(\n value=os.getcwd(),\n description='Directory:',\n disabled=False\n)\n\n#-- dropdown menu for setting tide model\nmodel_list = ['CATS0201','CATS2008','TPXO9-atlas','TPXO9-atlas-v2',\n 'TPXO9-atlas-v3','TPXO9-atlas-v4','TPXO9.1','TPXO8-atlas',\n 'TPXO7.2','FES2014']\nmodelDropdown = widgets.Dropdown(\n options=model_list,\n value='CATS2008',\n description='Model:',\n disabled=False,\n)\n\n#-- date picker widget for setting time\ndatepick = widgets.DatePicker(\n description='Date:',\n value = datetime.date.today(),\n disabled=False\n)\n\n#-- display widgets for setting directory, model and date\nwidgets.VBox([dirText,modelDropdown,datepick])",
"_____no_output_____"
]
],
[
[
"#### Setup tide model parameters",
"_____no_output_____"
]
],
[
[
"#-- directory with tide models\ntide_dir = os.path.expanduser(dirText.value)\nMODEL = modelDropdown.value\nif (MODEL == 'CATS0201'):\n grid_file = os.path.join(tide_dir,'cats0201_tmd','grid_CATS')\n model_file = os.path.join(tide_dir,'cats0201_tmd','UV0_CATS02_01')\n model_format = 'OTIS'\n EPSG = '4326'\n TYPES = ['u','v']\nelif (MODEL == 'CATS2008'):\n grid_file = os.path.join(tide_dir,'CATS2008','grid_CATS2008')\n model_file = os.path.join(tide_dir,'CATS2008','uv.CATS2008.out')\n model_format = 'OTIS'\n EPSG = 'CATS2008'\n TYPES = ['u','v']\nelif (MODEL == 'TPXO9-atlas'):\n model_directory = os.path.join(tide_dir,'TPXO9_atlas')\n grid_file = os.path.join(model_directory,'grid_tpxo9_atlas.nc.gz')\n model_files = {}\n model_files['u'] = ['u_q1_tpxo9_atlas_30.nc.gz','u_o1_tpxo9_atlas_30.nc.gz',\n 'u_p1_tpxo9_atlas_30.nc.gz','u_k1_tpxo9_atlas_30.nc.gz',\n 'u_n2_tpxo9_atlas_30.nc.gz','u_m2_tpxo9_atlas_30.nc.gz',\n 'u_s2_tpxo9_atlas_30.nc.gz','u_k2_tpxo9_atlas_30.nc.gz',\n 'u_m4_tpxo9_atlas_30.nc.gz','u_ms4_tpxo9_atlas_30.nc.gz',\n 'u_mn4_tpxo9_atlas_30.nc.gz','u_2n2_tpxo9_atlas_30.nc.gz']\n model_files['v'] = ['v_q1_tpxo9_atlas_30.nc.gz','v_o1_tpxo9_atlas_30.nc.gz',\n 'v_p1_tpxo9_atlas_30.nc.gz','v_k1_tpxo9_atlas_30.nc.gz',\n 'v_n2_tpxo9_atlas_30.nc.gz','v_m2_tpxo9_atlas_30.nc.gz',\n 'v_s2_tpxo9_atlas_30.nc.gz','v_k2_tpxo9_atlas_30.nc.gz',\n 'v_m4_tpxo9_atlas_30.nc.gz','v_ms4_tpxo9_atlas_30.nc.gz',\n 'v_mn4_tpxo9_atlas_30.nc.gz','v_2n2_tpxo9_atlas_30.nc.gz']\n model_format = 'netcdf'\n TYPES = ['u','v']\n SCALE = 1.0/100.0\n GZIP = True\nelif (MODEL == 'TPXO9-atlas-v2'):\n model_directory = os.path.join(tide_dir,'TPXO9_atlas_v2')\n grid_file = os.path.join(model_directory,'grid_tpxo9_atlas_30_v2.nc.gz')\n model_files = {}\n model_files['u'] = ['u_q1_tpxo9_atlas_30_v2.nc.gz','u_o1_tpxo9_atlas_30_v2.nc.gz',\n 'u_p1_tpxo9_atlas_30_v2.nc.gz','u_k1_tpxo9_atlas_30_v2.nc.gz',\n 'u_n2_tpxo9_atlas_30_v2.nc.gz','u_m2_tpxo9_atlas_30_v2.nc.gz',\n 'u_s2_tpxo9_atlas_30_v2.nc.gz','u_k2_tpxo9_atlas_30_v2.nc.gz',\n 'u_m4_tpxo9_atlas_30_v2.nc.gz','u_ms4_tpxo9_atlas_30_v2.nc.gz',\n 'u_mn4_tpxo9_atlas_30_v2.nc.gz','u_2n2_tpxo9_atlas_30_v2.nc.gz']\n model_files['v'] = ['v_q1_tpxo9_atlas_30_v2.nc.gz','v_o1_tpxo9_atlas_30_v2.nc.gz',\n 'v_p1_tpxo9_atlas_30_v2.nc.gz','v_k1_tpxo9_atlas_30_v2.nc.gz',\n 'v_n2_tpxo9_atlas_30_v2.nc.gz','v_m2_tpxo9_atlas_30_v2.nc.gz',\n 'v_s2_tpxo9_atlas_30_v2.nc.gz','v_k2_tpxo9_atlas_30_v2.nc.gz',\n 'v_m4_tpxo9_atlas_30_v2.nc.gz','v_ms4_tpxo9_atlas_30_v2.nc.gz',\n 'v_mn4_tpxo9_atlas_30_v2.nc.gz','v_2n2_tpxo9_atlas_30_v2.nc.gz']\n model_file = {}\n for key,val in model_files.items():\n model_file[key] = [os.path.join(model_directory,m) for m in val]\n model_format = 'netcdf'\n TYPES = ['u','v']\n SCALE = 1.0/100.0\n GZIP = True\nelif (MODEL == 'TPXO9-atlas-v3'):\n model_directory = os.path.join(tide_dir,'TPXO9_atlas_v3')\n grid_file = os.path.join(model_directory,'grid_tpxo9_atlas_30_v3.nc.gz')\n model_files = {}\n model_files['u'] = ['u_q1_tpxo9_atlas_30_v3.nc.gz','u_o1_tpxo9_atlas_30_v3.nc.gz',\n 'u_p1_tpxo9_atlas_30_v3.nc.gz','u_k1_tpxo9_atlas_30_v3.nc.gz',\n 'u_n2_tpxo9_atlas_30_v3.nc.gz','u_m2_tpxo9_atlas_30_v3.nc.gz',\n 'u_s2_tpxo9_atlas_30_v3.nc.gz','u_k2_tpxo9_atlas_30_v3.nc.gz',\n 'u_m4_tpxo9_atlas_30_v3.nc.gz','u_ms4_tpxo9_atlas_30_v3.nc.gz',\n 'u_mn4_tpxo9_atlas_30_v3.nc.gz','u_2n2_tpxo9_atlas_30_v3.nc.gz']\n model_files['v'] = ['v_q1_tpxo9_atlas_30_v3.nc.gz','v_o1_tpxo9_atlas_30_v3.nc.gz',\n 'v_p1_tpxo9_atlas_30_v3.nc.gz','v_k1_tpxo9_atlas_30_v3.nc.gz',\n 'v_n2_tpxo9_atlas_30_v3.nc.gz','v_m2_tpxo9_atlas_30_v3.nc.gz',\n 'v_s2_tpxo9_atlas_30_v3.nc.gz','v_k2_tpxo9_atlas_30_v3.nc.gz',\n 'v_m4_tpxo9_atlas_30_v3.nc.gz','v_ms4_tpxo9_atlas_30_v3.nc.gz',\n 'v_mn4_tpxo9_atlas_30_v3.nc.gz','v_2n2_tpxo9_atlas_30_v3.nc.gz']\n model_file = {}\n for key,val in model_files.items():\n model_file[key] = [os.path.join(model_directory,m) for m in val]\n model_format = 'netcdf'\n TYPES = ['u','v']\n SCALE = 1.0/100.0\n GZIP = True\nelif (MODEL == 'TPXO9-atlas-v4'):\n model_directory = os.path.join(tide_dir,'TPXO9_atlas_v4')\n grid_file = os.path.join(tide_dir,'grid_tpxo9_atlas_30_v4')\n model_files = ['u_q1_tpxo9_atlas_30_v4','u_o1_tpxo9_atlas_30_v4',\n 'u_p1_tpxo9_atlas_30_v4','u_k1_tpxo9_atlas_30_v4',\n 'u_n2_tpxo9_atlas_30_v4','u_m2_tpxo9_atlas_30_v4',\n 'u_s2_tpxo9_atlas_30_v4','u_k2_tpxo9_atlas_30_v4',\n 'u_m4_tpxo9_atlas_30_v4','u_ms4_tpxo9_atlas_30_v4',\n 'u_mn4_tpxo9_atlas_30_v4','u_2n2_tpxo9_atlas_30_v4']\n model_file = [os.path.join(model_directory,m) for m in model_files]\n model_format = 'OTIS'\n EPSG = '4326'\n TYPES = ['u','v']\nelif (MODEL == 'TPXO9.1'):\n grid_file = os.path.join(tide_dir,'TPXO9.1','DATA','grid_tpxo9')\n model_file = os.path.join(tide_dir,'TPXO9.1','DATA','u_tpxo9.v1')\n model_format = 'OTIS'\n EPSG = '4326'\n TYPES = ['u','v']\nelif (MODEL == 'TPXO8-atlas'):\n grid_file = os.path.join(tide_dir,'tpxo8_atlas','grid_tpxo8atlas_30_v1')\n model_file = os.path.join(tide_dir,'tpxo8_atlas','uv.tpxo8_atlas_30_v1')\n model_format = 'ATLAS'\n EPSG = '4326'\n TYPES = ['u','v']\nelif (MODEL == 'TPXO7.2'):\n grid_file = os.path.join(tide_dir,'TPXO7.2_tmd','grid_tpxo7.2')\n model_file = os.path.join(tide_dir,'TPXO7.2_tmd','u_tpxo7.2')\n model_format = 'OTIS'\n EPSG = '4326'\n TYPES = ['u','v']\nelif (MODEL == 'FES2014'):\n model_directory = {}\n model_directory['u'] = os.path.join(tide_dir,'fes2014','eastward_velocity')\n model_directory['v'] = os.path.join(tide_dir,'fes2014','northward_velocity')\n model_files = ['2n2.nc.gz','eps2.nc.gz','j1.nc.gz','k1.nc.gz',\n 'k2.nc.gz','l2.nc.gz','la2.nc.gz','m2.nc.gz','m3.nc.gz','m4.nc.gz',\n 'm6.nc.gz','m8.nc.gz','mf.nc.gz','mks2.nc.gz','mm.nc.gz',\n 'mn4.nc.gz','ms4.nc.gz','msf.nc.gz','msqm.nc.gz','mtm.nc.gz',\n 'mu2.nc.gz','n2.nc.gz','n4.nc.gz','nu2.nc.gz','o1.nc.gz','p1.nc.gz',\n 'q1.nc.gz','r2.nc.gz','s1.nc.gz','s2.nc.gz','s4.nc.gz','sa.nc.gz',\n 'ssa.nc.gz','t2.nc.gz']\n model_file = {}\n for key,val in model_directory.items():\n model_file[key] = [os.path.join(val,m) for m in model_files]\n c = ['2n2','eps2','j1','k1','k2','l2','lambda2','m2','m3','m4','m6',\n 'm8','mf','mks2','mm','mn4','ms4','msf','msqm','mtm','mu2','n2',\n 'n4','nu2','o1','p1','q1','r2','s1','s2','s4','sa','ssa','t2']\n model_format = 'FES'\n TYPES = ['u','v']\n SCALE = 1.0\n GZIP = True",
"_____no_output_____"
]
],
[
[
"#### Setup coordinates for calculating tidal currents",
"_____no_output_____"
]
],
[
[
"#-- create an image around Antarctica\nxlimits = [-560.*5e3,560.*5e3]\nylimits = [-560.*5e3,560.*5e3]\nspacing = [20e3,-20e3]\n#-- x and y coordinates\nx = np.arange(xlimits[0],xlimits[1]+spacing[0],spacing[0])\ny = np.arange(ylimits[1],ylimits[0]+spacing[1],spacing[1])\nxgrid,ygrid = np.meshgrid(x,y)\n#-- x and y dimensions\nnx = int((xlimits[1]-xlimits[0])/spacing[0])+1\nny = int((ylimits[0]-ylimits[1])/spacing[1])+1\n#-- convert image coordinates from polar stereographic to latitude/longitude\ncrs1 = pyproj.CRS.from_string(\"epsg:{0:d}\".format(3031))\ncrs2 = pyproj.CRS.from_string(\"epsg:{0:d}\".format(4326))\ntransformer = pyproj.Transformer.from_crs(crs1, crs2, always_xy=True)\nlon,lat = transformer.transform(xgrid.flatten(), ygrid.flatten())",
"_____no_output_____"
]
],
[
[
"#### Calculate tide map",
"_____no_output_____"
]
],
[
[
"#-- convert from calendar date to days relative to Jan 1, 1992 (48622 MJD)\nYMD = datepick.value\ntide_time = pyTMD.time.convert_calendar_dates(YMD.year, YMD.month,\n YMD.day, hour=np.arange(24))\n#-- delta time (TT - UT1) file\ndelta_file = pyTMD.utilities.get_data_path(['data','merged_deltat.data'])\n\n#-- save tide currents\ntide = {}\n#-- iterate over u and v currents\nfor TYPE in TYPES:\n #-- read tidal constants and interpolate to grid points\n if model_format in ('OTIS','ATLAS'):\n amp,ph,D,c = extract_tidal_constants(lon, lat, grid_file, model_file,\n EPSG, TYPE=TYPE, METHOD='spline', GRID=model_format)\n DELTAT = np.zeros_like(tide_time)\n elif (model_format == 'netcdf'):\n amp,ph,D,c = extract_netcdf_constants(lon, lat, grid_file,\n model_file[TYPE], TYPE=TYPE, METHOD='spline',\n SCALE=SCALE, GZIP=GZIP)\n DELTAT = np.zeros_like(tide_time)\n elif (model_format == 'GOT'):\n amp,ph,c = extract_GOT_constants(lon, lat, model_file,\n METHOD='spline', SCALE=SCALE)\n #-- interpolate delta times from calendar dates to tide time\n DELTAT = calc_delta_time(delta_file, tide_time)\n elif (model_format == 'FES'):\n amp,ph = extract_FES_constants(lon, lat, model_file[TYPE],\n TYPE=TYPE, VERSION=MODEL, METHOD='spline', SCALE=SCALE,\n GZIP=GZIP)\n #-- interpolate delta times from calendar dates to tide time\n DELTAT = calc_delta_time(delta_file, tide_time)\n \n #-- calculate complex phase in radians for Euler's\n cph = -1j*ph*np.pi/180.0\n #-- calculate constituent oscillation\n hc = amp*np.exp(cph)\n\n #-- allocate for tide current map calculated every hour\n tide[TYPE] = np.ma.zeros((ny,nx,24))\n for hour in range(24):\n #-- predict tidal elevations at time and infer minor corrections\n TIDE = predict_tide(tide_time[hour], hc, c, DELTAT=DELTAT[hour],\n CORRECTIONS=model_format)\n MINOR = infer_minor_corrections(tide_time[hour], hc, c,\n DELTAT=DELTAT[hour], CORRECTIONS=model_format)\n #-- add major and minor components and reform grid\n tide[TYPE][:,:,hour] = np.reshape((TIDE+MINOR),(ny,nx))",
"_____no_output_____"
]
],
[
[
"#### Create animation of hourly tidal oscillation",
"_____no_output_____"
]
],
[
[
"#-- output Antarctic Tidal Current Animation\nprojection = ccrs.Stereographic(central_longitude=0.0,\n central_latitude=-90.0,true_scale_latitude=-71.0)\n#-- figure axis and image objects\nax1,im = ({},{})\nfig, (ax1['u'],ax1['v']) = plt.subplots(num=1, ncols=2,\n figsize=(11.5,7), subplot_kw=dict(projection=projection))\nvmin = np.min([tide['u'].min(),tide['v'].min()])\nvmax = np.max([tide['u'].max(),tide['v'].max()])\nextent = (xlimits[0],xlimits[1],ylimits[0],ylimits[1])\nfor TYPE,ax in ax1.items():\n #-- plot tidal currents\n im[TYPE] = ax.imshow(np.zeros((ny,nx)),\n interpolation='nearest', vmin=vmin, vmax=vmax,\n transform=projection, extent=extent, origin='upper',\n animated=True)\n #-- add high resolution cartopy coastlines\n ax.coastlines('10m')\n #-- set x and y limits\n ax.set_xlim(xlimits)\n ax.set_ylim(ylimits)\n # stronger linewidth on frame\n ax.spines['geo'].set_linewidth(2.0)\n ax.spines['geo'].set_capstyle('projecting')\n\n#-- Add colorbar with a colorbar axis\n#-- Add an axes at position rect [left, bottom, width, height]\ncbar_ax = fig.add_axes([0.085, 0.075, 0.83, 0.035])\n#-- extend = add extension triangles to upper and lower bounds\n#-- options: neither, both, min, max\ncbar = fig.colorbar(im['u'], cax=cbar_ax, extend='both',\n extendfrac=0.0375, drawedges=False, orientation='horizontal')\n#-- rasterized colorbar to remove lines\ncbar.solids.set_rasterized(True)\n#-- Add label to the colorbar\ncbar.ax.set_xlabel('{0} Tidal Velocity'.format(MODEL), fontsize=13)\ncbar.ax.set_ylabel('cm/s', fontsize=13, rotation=0)\ncbar.ax.yaxis.set_label_coords(1.04, 0.15)\n#-- ticks lines all the way across\ncbar.ax.tick_params(which='both', width=1, length=18,\n labelsize=13, direction='in')\n\n#-- add title (date and time)\nttl = fig.suptitle(None, fontsize=13)\n# adjust subplot within figure\nfig.subplots_adjust(left=0.02,right=0.98,bottom=0.1,top=0.98,wspace=0.04)\n \n#-- animate each map\ndef animate_maps(hour):\n #-- set map data iterating over u and v currents\n for TYPE in TYPES:\n im[TYPE].set_data(tide[TYPE][:,:,hour])\n #-- set title\n args = (YMD.year,YMD.month,YMD.day,hour)\n ttl.set_text('{0:4d}-{1:02d}-{2:02d}T{3:02d}:00:00'.format(*args))\n\n#-- set animation\nanim = animation.FuncAnimation(fig, animate_maps, frames=24)\n%matplotlib inline\nHTML(anim.to_jshtml())",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
]
] |
cbdb96eb1e2a8ded0f226a8f7cf64603245305dd
| 99,921 |
ipynb
|
Jupyter Notebook
|
_notebooks/2021-04-02-otimizacao.ipynb
|
hereismari/codando-redes-neurais
|
7b573c56c7691c9a6c3b08aa5069217496793195
|
[
"Apache-2.0"
] | 13 |
2021-09-11T17:03:45.000Z
|
2022-02-06T17:08:41.000Z
|
_notebooks/2021-04-02-otimizacao.ipynb
|
hereismari/codando-redes-neurais
|
7b573c56c7691c9a6c3b08aa5069217496793195
|
[
"Apache-2.0"
] | 3 |
2021-09-11T18:32:18.000Z
|
2021-09-16T11:22:48.000Z
|
_notebooks/2021-04-02-otimizacao.ipynb
|
hereismari/codando-redes-neurais
|
7b573c56c7691c9a6c3b08aa5069217496793195
|
[
"Apache-2.0"
] | 1 |
2021-09-18T20:51:43.000Z
|
2021-09-18T20:51:43.000Z
| 254.90051 | 21,056 | 0.919576 |
[
[
[
"# Algoritmos de Otimização",
"_____no_output_____"
],
[
"No Deep Learning temos como propósito que nossas redes neurais aprendam a aproximar uma função de interesse, como o preço de casas numa regressão, ou a função que classifica objetos numa foto, no caso da classificação.",
"_____no_output_____"
],
[
"No último notebook, nós programos nossa primeira rede neural. Além disso, vimos também a fórmula de atualização dos pesos. Se você não lembra, os pesos e os bias foram atualizados da seguinte forma:\n\n$$w_i = w_i - \\lambda * \\partial w $$\n\n$$b_i = b_i - \\lambda * \\partial b$$\n\nMas, você já parou pra pensar da onde vêm essas fórmulas? Além disso, será que existem melhores formas de atualizar os pesos? É isso que vamos ver nesse notebook.",
"_____no_output_____"
],
[
"## Descida de Gradiente Estocástica (SGD)\nNa descida de gradiente estocástica separamos os nossos dados de treino em vários subconjuntos, que chamamos de mini-batches. No começo eles serão pequenos, como 32-128 exemplos, para aplicações mais avançadas eles tendem a ser muito maiores, na ordem de 1024 e até mesmo 8192 exemplos por mini-batch.\n\nComo na descida de gradiente padrão, computamos o gradiente da função de custo em relação aos exemplos, e subtraímos o gradiente vezes uma taxa de apredizado dos parâmetros da rede. Podemos ver o SGD como tomando um passo pequeno na direção de maior redução do valor da loss.\n\n### Equação\n$w_{t+1} = w_t - \\eta \\cdot \\nabla L$\n\n### Código",
"_____no_output_____"
]
],
[
[
"import jax\ndef sgd(weights, gradients, eta):\n return jax.tree_util.tree_multimap(lambda w, g: w - eta*g, weights, gradients)",
"_____no_output_____"
]
],
[
[
"Vamos usar o SGD para otimizar uma função simples",
"_____no_output_____"
]
],
[
[
"#hide\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n%matplotlib inline\n\ndef f(x):\n return x**2 - 25\nx = np.linspace(-10, 10, num=100)\ny = f(x)\n\nplt.plot(x, y)\nplt.ylim(-50)",
"_____no_output_____"
],
[
"x0 = 9.0\nf_g = jax.value_and_grad(f)\nx_ = []\ny_ = []\nfor i in range(10):\n y0, grads = f_g(x0)\n x_.append(x0)\n y_.append(y0)\n x0 = sgd(x0, grads, 0.9)\n\nplt.plot(x, y)\nplt.plot(x_, y_, color='red', marker='o');",
"WARNING:absl:No GPU/TPU found, falling back to CPU. (Set TF_CPP_MIN_LOG_LEVEL=0 and rerun for more info.)\n"
]
],
[
[
"## Momentum\nUm problema com o uso de mini-batches é que agora estamos **estimando** a direção que diminui a função de perda no conjunto de treino, e quão menor o mini-batch mais ruidosa é a nossa estimativa. Para consertar esse problema do ruído nós introduzimos a noção de momentum. O momentm faz sua otimização agir como uma bola pesada descendo uma montanha, então mesmo que o caminho seja cheio de montes e buracos a direção da bola não é muito afetada. De um ponto de vista mais matemático as nossas atualizações dos pesos vão ser uma combinação entre os gradientes desse passo e os gradientes anteriores, estabilizando o treino.\n\n### Equação\n$v_{t} = \\gamma v_{t-1} + \\nabla L \\quad \\text{o gamma serve como um coeficiente ponderando entre usar os updates anteriores e o novo gradiente} \\\\\nw_{t+1} = w_t - \\eta v_t\n$\n\n### Código",
"_____no_output_____"
]
],
[
[
"def momentum(weights, gradients, eta, mom, gamma):\n mom = jax.tree_util.tree_multimap(\n lambda v, g: gamma*v + (1 - gamma)*g, weights, gradients)\n weights = jax.tree_util.tree_multimap(\n lambda w, v: w - eta*mom, weights, mom)\n return weights, mom",
"_____no_output_____"
],
[
"x0 = 9.0\nmom = 0.0\nx_ = []\ny_ = []\nfor i in range(10):\n y0, grads = f_g(x0)\n x_.append(x0)\n y_.append(y0)\n x0, mom = momentum(x0, grads, 0.9, mom, 0.99)\n\nplt.plot(x, y)\nplt.plot(x_, y_, color='red', marker='o');",
"_____no_output_____"
]
],
[
[
"## RMSProp\nCriado por Geoffrey Hinton durante uma aula, esse método é o primeiro **método adaptivo** que estamos vendo. O que isso quer dizer é que o método tenta automaticamente computar uma taxa de aprendizado diferente para cada um dos pesos da nossa rede neural, usando taxas pequenas para parâmetros que sofrem atualização frequentemente e taxas maiores para parâmetros que são atualizados mais raramente, permitindo uma otimização mais rápida. \n\nMais especificamente, o RMSProp divide o update normal do SGD pela raiz da soma dos quadrados dos gradientes anteriores (por isso seu nome Root-Mean-Square Proportional), assim reduzindo a magnitude da atualização de acordo com as magnitudes anteriores.\n\n### Equação\n$\n\\nu_{t} = \\gamma \\nu_{t-1} + (1 - \\gamma) (\\nabla L)^2 \\\\\nw_{t+1} = w_t - \\frac{\\eta \\nabla L}{\\sqrt{\\nu_t + \\epsilon}} \n$\n\n### Código",
"_____no_output_____"
]
],
[
[
"def computa_momento(updates, moments, decay, order):\n return jax.tree_multimap(\n lambda g, t: (1 - decay) * (g ** order) + decay * t, updates, moments)\ndef rmsprop(weights, gradients, eta, nu, gamma):\n nu = computa_momento(gradients, nu, gamma, 2)\n updates = jax.tree_multimap(\n lambda g, n: g * jax.lax.rsqrt(n + 1e-8), gradients, nu)\n weights = jax.tree_util.tree_multimap(lambda w, g: w - eta*g, weights, updates)\n return weights, nu",
"_____no_output_____"
],
[
"x0 = 9.0\nnu = 0.0\nx_ = []\ny_ = []\nfor i in range(10):\n y0, grads = f_g(x0)\n x_.append(x0)\n y_.append(y0)\n x0, nu = rmsprop(x0, grads, 0.9, nu, 0.99)\n\nplt.plot(x, y)\nplt.plot(x_, y_, color='red', marker='o');",
"_____no_output_____"
]
],
[
[
"## Adam\nPor fim o Adam usa ideias semelhantes ao Momentum e ao RMSProp, mantendo médias exponenciais tanto dos gradientes passados, quanto dos seus quadrados.\n\n### Equação\n$\nm_t = \\beta_1 m_{t-1} + (1 - \\beta_1) \\nabla L \\\\\nv_t = \\beta_2 v_{t-1} + (1 - \\beta_2) (\\nabla L)^2 \\\\\nw_{t+1} = w_t - \\frac{\\eta m_t}{\\sqrt{v_t} \\epsilon} \n$\n\n### Código",
"_____no_output_____"
]
],
[
[
"import jax.numpy as jnp\ndef adam(weights, gradients, eta, mu, nu, b1, b2):\n mu = computa_momento(gradients, mu, b1, 1)\n nu = computa_momento(gradients, nu, b2, 2)\n updates = jax.tree_multimap(\n lambda m, v: m / (jnp.sqrt(v + 1e-6) + 1e-8), mu, nu)\n weights = jax.tree_util.tree_multimap(lambda w, g: w - eta*g, weights, updates)\n return weights, mu, nu",
"_____no_output_____"
],
[
"x0 = 9.0\nmu = 0.0\nnu = 0.0\nx_ = []\ny_ = []\nfor i in range(10):\n y0, grads = f_g(x0)\n x_.append(x0)\n y_.append(y0)\n x0, mu, nu = adam(x0, grads, 0.8, mu, nu, 0.9, 0.999)\n\nplt.plot(x, y)\nplt.plot(x_, y_, color='red', marker='o');",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
]
] |
cbdb96f9b49d0292e269ceeb581baa273bd8647d
| 4,799 |
ipynb
|
Jupyter Notebook
|
notebooks/ensemble_ex_03.ipynb
|
miykael/scikit-learn-mooc
|
06840c79db1c891e36715175c9053535bf886094
|
[
"CC-BY-4.0"
] | null | null | null |
notebooks/ensemble_ex_03.ipynb
|
miykael/scikit-learn-mooc
|
06840c79db1c891e36715175c9053535bf886094
|
[
"CC-BY-4.0"
] | null | null | null |
notebooks/ensemble_ex_03.ipynb
|
miykael/scikit-learn-mooc
|
06840c79db1c891e36715175c9053535bf886094
|
[
"CC-BY-4.0"
] | null | null | null | 28.064327 | 101 | 0.610335 |
[
[
[
"# -*- coding: utf-8 -*-",
"_____no_output_____"
]
],
[
[
"# 📝 Exercise M6.03\n\nThe aim of this exercise is to:\n\n* verifying if a random forest or a gradient-boosting decision tree overfit\n if the number of estimators is not properly chosen;\n* use the early-stopping strategy to avoid adding unnecessary trees, to\n get the best generalization performances.\n\nWe will use the California housing dataset to conduct our experiments.",
"_____no_output_____"
]
],
[
[
"from sklearn.datasets import fetch_california_housing\nfrom sklearn.model_selection import train_test_split\n\ndata, target = fetch_california_housing(return_X_y=True, as_frame=True)\ntarget *= 100 # rescale the target in k$\ndata_train, data_test, target_train, target_test = train_test_split(\n data, target, random_state=0, test_size=0.5)",
"_____no_output_____"
]
],
[
[
"<div class=\"admonition note alert alert-info\">\n<p class=\"first admonition-title\" style=\"font-weight: bold;\">Note</p>\n<p class=\"last\">If you want a deeper overview regarding this dataset, you can refer to the\nAppendix - Datasets description section at the end of this MOOC.</p>\n</div>",
"_____no_output_____"
],
[
"Create a gradient boosting decision tree with `max_depth=5` and\n`learning_rate=0.5`.",
"_____no_output_____"
]
],
[
[
"# Write your code here.",
"_____no_output_____"
]
],
[
[
"\nAlso create a random forest with fully grown trees by setting `max_depth=None`.",
"_____no_output_____"
]
],
[
[
"# Write your code here.",
"_____no_output_____"
]
],
[
[
"\nFor both the gradient-boosting and random forest models, create a validation\ncurve using the training set to assess the impact of the number of trees on\nthe performance of each model. Evaluate the list of parameters `param_range =\n[1, 2, 5, 10, 20, 50, 100]` and use the mean absolute error.",
"_____no_output_____"
]
],
[
[
"# Write your code here.",
"_____no_output_____"
]
],
[
[
"Both gradient boosting and random forest models will always improve when\nincreasing the number of trees in the ensemble. However, it will reach a\nplateau where adding new trees will just make fitting and scoring slower.\n\nTo avoid adding new unnecessary tree, unlike random-forest gradient-boosting\noffers an early-stopping option. Internally, the algorithm will use an\nout-of-sample set to compute the generalization performance of the model at\neach addition of a tree. Thus, if the generalization performance is not\nimproving for several iterations, it will stop adding trees.\n\nNow, create a gradient-boosting model with `n_estimators=1_000`. This number\nof trees will be too large. Change the parameter `n_iter_no_change` such\nthat the gradient boosting fitting will stop after adding 5 trees that do not\nimprove the overall generalization performance.",
"_____no_output_____"
]
],
[
[
"# Write your code here.",
"_____no_output_____"
]
],
[
[
"Estimate the generalization performance of this model again using\nthe `sklearn.metrics.mean_absolute_error` metric but this time using\nthe test set that we held out at the beginning of the notebook.\nCompare the resulting value with the values observed in the validation\ncurve.",
"_____no_output_____"
]
],
[
[
"# Write your code here.",
"_____no_output_____"
]
]
] |
[
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
]
] |
cbdb9c09276e09fe5899b174972c5baabcfa9d02
| 46,683 |
ipynb
|
Jupyter Notebook
|
Data Science/Numpy/Untitled4.ipynb
|
manash-ranjan-khamari/notebook
|
c5f6caac62ae89995065e74de460ba251c3ac42a
|
[
"MIT"
] | null | null | null |
Data Science/Numpy/Untitled4.ipynb
|
manash-ranjan-khamari/notebook
|
c5f6caac62ae89995065e74de460ba251c3ac42a
|
[
"MIT"
] | null | null | null |
Data Science/Numpy/Untitled4.ipynb
|
manash-ranjan-khamari/notebook
|
c5f6caac62ae89995065e74de460ba251c3ac42a
|
[
"MIT"
] | null | null | null | 111.681818 | 32,208 | 0.87805 |
[
[
[
"import numpy as np",
"_____no_output_____"
],
[
"inp = int(input(\"Enter matrix dimension: \"))\no3 = np.zeros((inp, inp), dtype=np.int64)",
"Enter matrix dimension: 7\n"
],
[
"o3",
"_____no_output_____"
],
[
"o3[[0,-1]] = o3[:,[0,-1]] = 1\no3",
"_____no_output_____"
],
[
"np.unravel_index(99, (11, 13))",
"_____no_output_____"
],
[
"np.unravel_index(99, (11,12))",
"_____no_output_____"
],
[
"np.reshape([[1, 5],\n [3, 7],\n [4, 9]], -1)",
"_____no_output_____"
],
[
"i1 = np.array([[1, 2, 3, 4, 5],\n [6, 7, 8, 9, 10],\n [11, 12, 13, 14, 15],\n [16, 17, 18, 19, 20]])\ni1",
"_____no_output_____"
],
[
"i1.reshape(-1)[0::2].reshape(5, 2)",
"_____no_output_____"
],
[
"i1[i1 %2 != 0]",
"_____no_output_____"
],
[
"import matplotlib.pyplot as plt",
"_____no_output_____"
],
[
"y = np.random.randint(1,100, 50)\nplt.plot(y)",
"_____no_output_____"
],
[
"plt.plot(y, 'ro')",
"_____no_output_____"
],
[
"?plt.plot",
"_____no_output_____"
],
[
"i5 = [48.49, 67.54, 57.47, 68.17, 51.18, 68.31, 50.33, 66.7, 45.62, 43.59, 53.64, 70.08, 47.69, 61.27, 44.14, 51.62, 48.72, 65.11]",
"_____no_output_____"
],
[
"np.percentile(i5, 25)",
"_____no_output_____"
],
[
"np.percentile(i5, 75)",
"_____no_output_____"
],
[
"np.percentile(i5, 75) - np.percentile(i5, 25)",
"_____no_output_____"
],
[
"np.unravel_index(49, (8, 8))",
"_____no_output_____"
]
]
] |
[
"code"
] |
[
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
]
] |
cbdb9f23a20a8208dcdd45ee16eac73af562ecc0
| 6,468 |
ipynb
|
Jupyter Notebook
|
colab/BoW.ipynb
|
cateto/python4NLP
|
1d2d5086f907bf75be01762bf0b384c76d8f704e
|
[
"MIT"
] | 2 |
2021-12-16T22:38:27.000Z
|
2021-12-17T13:09:49.000Z
|
colab/BoW.ipynb
|
cateto/python4NLP
|
1d2d5086f907bf75be01762bf0b384c76d8f704e
|
[
"MIT"
] | null | null | null |
colab/BoW.ipynb
|
cateto/python4NLP
|
1d2d5086f907bf75be01762bf0b384c76d8f704e
|
[
"MIT"
] | null | null | null | 47.558824 | 223 | 0.560451 |
[
[
[
"<a href=\"https://colab.research.google.com/github/cateto/python4NLP/blob/main/colab/BoW.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>",
"_____no_output_____"
]
],
[
[
"pip install konlpy",
"Collecting konlpy\n\u001b[?25l Downloading https://files.pythonhosted.org/packages/85/0e/f385566fec837c0b83f216b2da65db9997b35dd675e107752005b7d392b1/konlpy-0.5.2-py2.py3-none-any.whl (19.4MB)\n\u001b[K |████████████████████████████████| 19.4MB 1.5MB/s \n\u001b[?25hRequirement already satisfied: lxml>=4.1.0 in /usr/local/lib/python3.7/dist-packages (from konlpy) (4.2.6)\nCollecting beautifulsoup4==4.6.0\n\u001b[?25l Downloading https://files.pythonhosted.org/packages/9e/d4/10f46e5cfac773e22707237bfcd51bbffeaf0a576b0a847ec7ab15bd7ace/beautifulsoup4-4.6.0-py3-none-any.whl (86kB)\n\u001b[K |████████████████████████████████| 92kB 10.1MB/s \n\u001b[?25hRequirement already satisfied: tweepy>=3.7.0 in /usr/local/lib/python3.7/dist-packages (from konlpy) (3.10.0)\nRequirement already satisfied: numpy>=1.6 in /usr/local/lib/python3.7/dist-packages (from konlpy) (1.19.5)\nCollecting colorama\n Downloading https://files.pythonhosted.org/packages/44/98/5b86278fbbf250d239ae0ecb724f8572af1c91f4a11edf4d36a206189440/colorama-0.4.4-py2.py3-none-any.whl\nCollecting JPype1>=0.7.0\n\u001b[?25l Downloading https://files.pythonhosted.org/packages/cd/a5/9781e2ef4ca92d09912c4794642c1653aea7607f473e156cf4d423a881a1/JPype1-1.2.1-cp37-cp37m-manylinux2010_x86_64.whl (457kB)\n\u001b[K |████████████████████████████████| 460kB 35.1MB/s \n\u001b[?25hRequirement already satisfied: requests[socks]>=2.11.1 in /usr/local/lib/python3.7/dist-packages (from tweepy>=3.7.0->konlpy) (2.23.0)\nRequirement already satisfied: six>=1.10.0 in /usr/local/lib/python3.7/dist-packages (from tweepy>=3.7.0->konlpy) (1.15.0)\nRequirement already satisfied: requests-oauthlib>=0.7.0 in /usr/local/lib/python3.7/dist-packages (from tweepy>=3.7.0->konlpy) (1.3.0)\nRequirement already satisfied: typing-extensions; python_version < \"3.8\" in /usr/local/lib/python3.7/dist-packages (from JPype1>=0.7.0->konlpy) (3.7.4.3)\nRequirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.7/dist-packages (from requests[socks]>=2.11.1->tweepy>=3.7.0->konlpy) (2020.12.5)\nRequirement already satisfied: idna<3,>=2.5 in /usr/local/lib/python3.7/dist-packages (from requests[socks]>=2.11.1->tweepy>=3.7.0->konlpy) (2.10)\nRequirement already satisfied: chardet<4,>=3.0.2 in /usr/local/lib/python3.7/dist-packages (from requests[socks]>=2.11.1->tweepy>=3.7.0->konlpy) (3.0.4)\nRequirement already satisfied: urllib3!=1.25.0,!=1.25.1,<1.26,>=1.21.1 in /usr/local/lib/python3.7/dist-packages (from requests[socks]>=2.11.1->tweepy>=3.7.0->konlpy) (1.24.3)\nRequirement already satisfied: PySocks!=1.5.7,>=1.5.6; extra == \"socks\" in /usr/local/lib/python3.7/dist-packages (from requests[socks]>=2.11.1->tweepy>=3.7.0->konlpy) (1.7.1)\nRequirement already satisfied: oauthlib>=3.0.0 in /usr/local/lib/python3.7/dist-packages (from requests-oauthlib>=0.7.0->tweepy>=3.7.0->konlpy) (3.1.0)\nInstalling collected packages: beautifulsoup4, colorama, JPype1, konlpy\n Found existing installation: beautifulsoup4 4.6.3\n Uninstalling beautifulsoup4-4.6.3:\n Successfully uninstalled beautifulsoup4-4.6.3\nSuccessfully installed JPype1-1.2.1 beautifulsoup4-4.6.0 colorama-0.4.4 konlpy-0.5.2\n"
],
[
"from konlpy.tag import Okt\nimport re\nokt = Okt()\n\ntoken=re.sub(\"(\\.)\",\"\",\"정부가 발표하는 물가상승률과 소비자가 느끼는 물가상승률은 다르다.\")\ntoken=okt.morphs(token)\n\nword2index={}\nbow=[]\nfor voca in token:\n if voca not in word2index.keys():\n word2index[voca]=len(word2index)\n# token을 읽으면서, word2index에 없는 (not in) 단어는 새로 추가하고, 이미 있는 단어는 넘깁니다. \n bow.insert(len(word2index)-1,1)\n# BoW 전체에 전부 기본값 1을 넣어줍니다. 단어의 개수는 최소 1개 이상이기 때문입니다.\n else:\n index=word2index.get(voca)\n# 재등장하는 단어의 인덱스를 받아옵니다. \n bow[index]=bow[index]+1\n# 재등장한 단어는 해당하는 인덱스의 위치에 1을 더해줍니다. (단어의 개수를 세는 것입니다.) \nprint(word2index)",
"{'정부': 0, '가': 1, '발표': 2, '하는': 3, '물가상승률': 4, '과': 5, '소비자': 6, '느끼는': 7, '은': 8, '다르다': 9}\n"
],
[
"",
"_____no_output_____"
]
]
] |
[
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code",
"code",
"code"
]
] |
cbdba32590c8858135c20f8ebb9eee48b45939ff
| 2,139 |
ipynb
|
Jupyter Notebook
|
03_nbtemplate.ipynb
|
hallmx/nbd_colab
|
b17b65b4f84a8f8eb734dd10852355971819bc86
|
[
"Apache-2.0"
] | 5 |
2020-03-24T07:01:10.000Z
|
2021-01-27T06:21:00.000Z
|
03_nbtemplate.ipynb
|
hallmx/nbd_colab
|
b17b65b4f84a8f8eb734dd10852355971819bc86
|
[
"Apache-2.0"
] | 1 |
2021-09-28T00:31:29.000Z
|
2021-09-28T00:31:29.000Z
|
03_nbtemplate.ipynb
|
hallmx/nbd_colab
|
b17b65b4f84a8f8eb734dd10852355971819bc86
|
[
"Apache-2.0"
] | 1 |
2021-02-12T09:57:00.000Z
|
2021-02-12T09:57:00.000Z
| 19.27027 | 113 | 0.526882 |
[
[
[
"#default_exp nbtemplate",
"_____no_output_____"
],
[
"#all_flag\n# remove this cell to use with tests",
"_____no_output_____"
],
[
"# #hide\n# optional user defined test_flag here\nfrom google.colab import drive\ndrive.mount('/content/drive')",
"_____no_output_____"
],
[
"# #hide\n%load_ext autoreload\n%autoreload 2",
"_____no_output_____"
],
[
"# #hide\n!pip install nbdev\n!pip install fastcore",
"_____no_output_____"
],
[
"# #hide\n# add repo name then remove this comment\n%cd /content/drive/My drive/<repo name>",
"_____no_output_____"
],
[
"# #hide\n#not deps but we need them to use nbdev and run tests\nfrom nbdev import * \nfrom nbdev.showdoc import *\nfrom fastcore.test import *",
"_____no_output_____"
]
],
[
[
"# Notebook template\n\n> A minimal template for a new nbdev-Colab notebook ",
"_____no_output_____"
]
],
[
[
"# #hide\n# optional build script code - not required if you manage your project from another dedicated notebook. \nfrom nbdev.export import notebook2script\nnotebook2script()",
"_____no_output_____"
]
]
] |
[
"code",
"markdown",
"code"
] |
[
[
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
]
] |
cbdbb4ac0f8d077b6d5b27ca52e512cf71695b9c
| 142,115 |
ipynb
|
Jupyter Notebook
|
movie-recommendation/recommendation.ipynb
|
ztanis/experiments
|
d298563958fbcdf7bec6b44bf4c6a18b90f729d8
|
[
"MIT"
] | null | null | null |
movie-recommendation/recommendation.ipynb
|
ztanis/experiments
|
d298563958fbcdf7bec6b44bf4c6a18b90f729d8
|
[
"MIT"
] | null | null | null |
movie-recommendation/recommendation.ipynb
|
ztanis/experiments
|
d298563958fbcdf7bec6b44bf4c6a18b90f729d8
|
[
"MIT"
] | null | null | null | 105.036955 | 30,644 | 0.789065 |
[
[
[
"# Movie recommendation\n\nObjective: Based on movies list and user ratings estimate how user would rate other movies \n\nReferences: \nhttps://medium.com/@jdwittenauer/deep-learning-with-keras-recommender-systems-e7b99cb29929",
"_____no_output_____"
]
],
[
[
"import pandas as pd\nimport numpy as np\nfrom pandas_profiling import ProfileReport\nimport seaborn as sns\nimport matplotlib.pyplot as plt\nimport tensorflow as tf\nfrom sklearn.preprocessing import LabelEncoder\nfrom sklearn.model_selection import train_test_split\n\nfrom tensorflow.keras import Model\nfrom tensorflow.keras.layers import Input, Reshape, Dot, Add, Activation, Lambda, Concatenate, Dropout, Dense\nfrom tensorflow.keras.layers import Embedding\nfrom tensorflow.keras.optimizers import Adam\nfrom tensorflow.keras.regularizers import l2",
"_____no_output_____"
],
[
"movies = pd.read_csv(\"data/movies.csv\")",
"_____no_output_____"
],
[
"ratings = pd.read_csv(\"data/ratings.csv\")",
"_____no_output_____"
],
[
"movies.head()",
"_____no_output_____"
],
[
"ratings.head()",
"_____no_output_____"
],
[
"# Figure out how top 15 users interacts with top 15 movies",
"_____no_output_____"
],
[
"top_users = (\n ratings\n .groupby('userId')['rating']\n .count()\n .sort_values(ascending=False)[:15]\n)\n\ntop_movies = (\n ratings\n .groupby('movieId')['rating']\n .count()\n .sort_values(ascending=False)[:15]\n)\n\ntop_r = ratings.join(top_users, rsuffix='_r', how='inner', on='userId')\ntop_r = top_r.join(top_movies, rsuffix='_r', how='inner', on='movieId')\npd.crosstab(top_r.userId, top_r.movieId, top_r.rating, aggfunc=np.max)",
"_____no_output_____"
],
[
"rates_per_user = ratings.groupby('movieId')['rating'].count()\nsns.distplot(rates_per_user)",
"_____no_output_____"
],
[
"sns.distplot(ratings['rating'])",
"_____no_output_____"
],
[
"# encoding\n\nuser_enc = LabelEncoder()\nmovie_enc = LabelEncoder()\n\nratings['user'] = user_enc.fit_transform(ratings['userId'].values)\nratings['movie'] = movie_enc.fit_transform(ratings['movieId'].values)\nratings['rating'] = ratings['rating'].values.astype(np.float32)\n\nn_users = ratings['user'].nunique()\nn_movies = ratings['movie'].nunique()\nmin_rating = min(ratings['rating'])\nmax_rating = max(ratings['rating'])\n\n(n_users, n_movies)",
"_____no_output_____"
],
[
"# Uncomment to print profile report\n#profile = ProfileReport(ratings, title='Movies report', html={'style':{'full_width':True}})\n#profile",
"_____no_output_____"
],
[
"X = ratings[['user', 'movie']].values\ny = ratings['rating'].values\nX_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)\nX_train.shape, X_test.shape, y_train.shape, y_test.shape",
"_____no_output_____"
],
[
"n_factors = 50\nX_train_array = [X_train[:, 0], X_train[:, 1]]\nX_test_array = [X_test[:, 0], X_test[:, 1]]\n",
"_____no_output_____"
],
[
"def RecommenderV1(n_users, n_movies, n_factors):\n user = Input(shape=(1,))\n u = Embedding(n_users, n_factors, embeddings_initializer='he_normal',\n embeddings_regularizer=l2(1e-6))(user)\n u = Reshape((n_factors,))(u)\n \n movie = Input(shape=(1,))\n m = Embedding(n_movies, n_factors, embeddings_initializer='he_normal',\n embeddings_regularizer=l2(1e-6))(movie)\n m = Reshape((n_factors,))(m)\n \n x = Dot(axes=1)([u, m])\n model = Model(inputs=[user, movie], outputs=x)\n opt = Adam(lr=0.001)\n model.compile(loss='mean_squared_error', optimizer=opt)\n return model",
"_____no_output_____"
],
[
"model = RecommenderV1(n_users, n_movies, n_factors)\nmodel.summary()",
"Model: \"model\"\n__________________________________________________________________________________________________\nLayer (type) Output Shape Param # Connected to \n==================================================================================================\ninput_1 (InputLayer) [(None, 1)] 0 \n__________________________________________________________________________________________________\ninput_2 (InputLayer) [(None, 1)] 0 \n__________________________________________________________________________________________________\nembedding (Embedding) (None, 1, 50) 30500 input_1[0][0] \n__________________________________________________________________________________________________\nembedding_1 (Embedding) (None, 1, 50) 486200 input_2[0][0] \n__________________________________________________________________________________________________\nreshape (Reshape) (None, 50) 0 embedding[0][0] \n__________________________________________________________________________________________________\nreshape_1 (Reshape) (None, 50) 0 embedding_1[0][0] \n__________________________________________________________________________________________________\ndot (Dot) (None, 1) 0 reshape[0][0] \n reshape_1[0][0] \n==================================================================================================\nTotal params: 516,700\nTrainable params: 516,700\nNon-trainable params: 0\n__________________________________________________________________________________________________\n"
],
[
"history = model.fit(x=X_train_array, y=y_train, batch_size=64, epochs=5,\n verbose=1, validation_data=(X_test_array, y_test))",
"Train on 90752 samples, validate on 10084 samples\nEpoch 1/5\n"
],
[
"def draw_history(history):\n plt.plot(history.history['loss'], label='train')\n plt.plot(history.history['val_loss'], label='test')\n plt.legend();\n \ndraw_history(history) ",
"_____no_output_____"
],
[
"class EmbeddingLayer:\n def __init__(self, n_items, n_factors):\n self.n_items = n_items\n self.n_factors = n_factors\n \n def __call__(self, x):\n x = Embedding(self.n_items, self.n_factors, embeddings_initializer='he_normal',\n embeddings_regularizer=l2(1e-6))(x)\n x = Reshape((self.n_factors,))(x)\n return x",
"_____no_output_____"
],
[
"def RecommenderNet(n_users, n_movies, n_factors, min_rating, max_rating):\n user = Input(shape=(1,))\n u = EmbeddingLayer(n_users, n_factors)(user)\n \n movie = Input(shape=(1,))\n m = EmbeddingLayer(n_movies, n_factors)(movie)\n \n x = Concatenate()([u, m])\n x = Dropout(0.05)(x)\n \n x = Dense(10, kernel_initializer='he_normal')(x)\n x = Activation('relu')(x)\n x = Dropout(0.5)(x)\n \n x = Dense(1, kernel_initializer='he_normal')(x)\n x = Activation('sigmoid')(x)\n x = Lambda(lambda x: x * (max_rating - min_rating) + min_rating)(x)\n model = Model(inputs=[user, movie], outputs=x)\n opt = Adam(lr=0.001)\n model.compile(loss='mean_squared_error', optimizer=opt)\n return model",
"_____no_output_____"
],
[
"model = RecommenderNet(n_users, n_movies, n_factors, min_rating, max_rating)\nmodel.summary()",
"Model: \"model_1\"\n__________________________________________________________________________________________________\nLayer (type) Output Shape Param # Connected to \n==================================================================================================\ninput_3 (InputLayer) [(None, 1)] 0 \n__________________________________________________________________________________________________\ninput_4 (InputLayer) [(None, 1)] 0 \n__________________________________________________________________________________________________\nembedding_2 (Embedding) (None, 1, 50) 30500 input_3[0][0] \n__________________________________________________________________________________________________\nembedding_3 (Embedding) (None, 1, 50) 486200 input_4[0][0] \n__________________________________________________________________________________________________\nreshape_2 (Reshape) (None, 50) 0 embedding_2[0][0] \n__________________________________________________________________________________________________\nreshape_3 (Reshape) (None, 50) 0 embedding_3[0][0] \n__________________________________________________________________________________________________\nconcatenate (Concatenate) (None, 100) 0 reshape_2[0][0] \n reshape_3[0][0] \n__________________________________________________________________________________________________\ndropout (Dropout) (None, 100) 0 concatenate[0][0] \n__________________________________________________________________________________________________\ndense (Dense) (None, 10) 1010 dropout[0][0] \n__________________________________________________________________________________________________\nactivation (Activation) (None, 10) 0 dense[0][0] \n__________________________________________________________________________________________________\ndropout_1 (Dropout) (None, 10) 0 activation[0][0] \n__________________________________________________________________________________________________\ndense_1 (Dense) (None, 1) 11 dropout_1[0][0] \n__________________________________________________________________________________________________\nactivation_1 (Activation) (None, 1) 0 dense_1[0][0] \n__________________________________________________________________________________________________\nlambda (Lambda) (None, 1) 0 activation_1[0][0] \n==================================================================================================\nTotal params: 517,721\nTrainable params: 517,721\nNon-trainable params: 0\n__________________________________________________________________________________________________\n"
],
[
"history = model.fit(x=X_train_array, y=y_train, batch_size=64, epochs=10,\n verbose=1, validation_data=(X_test_array, y_test))",
"Train on 90752 samples, validate on 10084 samples\nEpoch 1/10\n"
],
[
"draw_history(history) ",
"_____no_output_____"
],
[
"top_r",
"_____no_output_____"
],
[
"estimated = model.predict(X_test_array)\nestimated",
"_____no_output_____"
],
[
"esimated_rounded = map(lambda x: round(x * 2) / 2, estimated[:, 0])\nexpected_actual = pd.DataFrame({'actual':y_test, 'estimated': esimated_rounded})\nexpected_actual_grouped = expected_actual.groupby(['actual', 'estimated']).size().reset_index(name=\"size\")",
"_____no_output_____"
],
[
"ax = sns.scatterplot(x=\"actual\", y=\"estimated\",\n hue=\"size\", size=\"size\",\n sizes=(0, 1000), data=expected_actual_grouped)",
"_____no_output_____"
]
]
] |
[
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
]
] |
cbdbbfe58d53eb4aae4079ac557f5318b9084564
| 158,907 |
ipynb
|
Jupyter Notebook
|
Chapter_LatentVariable2/TEP_data_explore.ipynb
|
ML-PSE/Machine_Learning_for_PSE
|
b53578d7cc0e0eca4907527b188a60de06d6710e
|
[
"Apache-2.0"
] | 2 |
2022-02-20T18:57:46.000Z
|
2022-03-03T07:07:12.000Z
|
Chapter_LatentVariable2/TEP_data_explore.ipynb
|
ML-PSE/Machine_Learning_for_PSE
|
b53578d7cc0e0eca4907527b188a60de06d6710e
|
[
"Apache-2.0"
] | null | null | null |
Chapter_LatentVariable2/TEP_data_explore.ipynb
|
ML-PSE/Machine_Learning_for_PSE
|
b53578d7cc0e0eca4907527b188a60de06d6710e
|
[
"Apache-2.0"
] | null | null | null | 863.625 | 97,516 | 0.952972 |
[
[
[
"# Chapter: Dimension Reduction and Latent Variable Methods (Part 2)¶\n\n# Topic: TEP data explore",
"_____no_output_____"
]
],
[
[
"##%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%\n## exploration of Etch data\n## %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%",
"_____no_output_____"
],
[
"# import required packages\nimport numpy as np\nimport matplotlib.pyplot as plt",
"_____no_output_____"
],
[
"# fetch TEP data\nTEdata_noFault_train = np.loadtxt('d00.dat').T # data arrnagement in d00.dat is different than that in other files\nTEdata_Fault_train = np.loadtxt('d10.dat')",
"_____no_output_____"
],
[
"# quick visualize\nplt.figure()\nplt.plot(TEdata_noFault_train[:,17])\nplt.xlabel('sample #')\nplt.ylabel('Stripper Tempearture')\nplt.title('Normal operation')\nplt.show()\n\nplt.figure()\nplt.plot(TEdata_Fault_train[:,17])\nplt.xlabel('sample #')\nplt.ylabel('Stripper Tempearture')\nplt.title('Faulty operation') \nplt.show()",
"_____no_output_____"
],
[
"##%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%\n## Visualize normal and faulty data in PC space\n## %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%",
"_____no_output_____"
],
[
"# scale data\nfrom sklearn.preprocessing import StandardScaler\n\nscaler = StandardScaler()\nTEdata_noFault_scaled = scaler.fit_transform(TEdata_noFault_train)\nTEdata_Fault_scaled = scaler.transform(TEdata_Fault_train)",
"_____no_output_____"
],
[
"#%% build PCA model and copmute PC scores\nfrom sklearn.decomposition import PCA\npca = PCA(n_components = 3).fit(TEdata_noFault_scaled)\nTEdata_noFault_scores = pca.transform(TEdata_noFault_scaled)\nTEdata_Fault_scores = pca.transform(TEdata_Fault_scaled)",
"_____no_output_____"
],
[
"#%% visualize in 3D plot\nfrom mpl_toolkits.mplot3d import Axes3D\nfig = plt.figure()\nax = Axes3D(fig)\nax.scatter(TEdata_noFault_scores[:,0],TEdata_noFault_scores[:,1],TEdata_noFault_scores[:,2], c='blue', alpha=0.1, label='Normal operation')\nax.scatter(TEdata_Fault_scores[:,0],TEdata_Fault_scores[:,1],TEdata_Fault_scores[:,2], c='red', marker = '*', label='Faulty operation')\nax.set_xlabel('PC1 scores')\nax.set_ylabel('PC2 scores')\nax.set_zlabel('PC3 scores')\nplt.show()",
"_____no_output_____"
]
]
] |
[
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
]
] |
cbdbc29f2ba31b3355badab62df1e6b5b2f9b71d
| 614,223 |
ipynb
|
Jupyter Notebook
|
tutorials/Certification_Trainings/Healthcare/3.Clinical_Entity_Resolvers_v2.5.ipynb
|
richardclarus/spark-nlp-workshop
|
64abb0ca9db5385d5ba38343bd3312ad4cc57845
|
[
"Apache-2.0"
] | 1 |
2021-01-23T15:24:45.000Z
|
2021-01-23T15:24:45.000Z
|
tutorials/Certification_Trainings/Healthcare/3.Clinical_Entity_Resolvers_v2.5.ipynb
|
richardclarus/spark-nlp-workshop
|
64abb0ca9db5385d5ba38343bd3312ad4cc57845
|
[
"Apache-2.0"
] | null | null | null |
tutorials/Certification_Trainings/Healthcare/3.Clinical_Entity_Resolvers_v2.5.ipynb
|
richardclarus/spark-nlp-workshop
|
64abb0ca9db5385d5ba38343bd3312ad4cc57845
|
[
"Apache-2.0"
] | 1 |
2021-01-23T15:24:52.000Z
|
2021-01-23T15:24:52.000Z
| 163.270335 | 263,998 | 0.822947 |
[
[
[
"",
"_____no_output_____"
],
[
"[](https://colab.research.google.com/github/JohnSnowLabs/spark-nlp-workshop/blob/master/tutorials/Certification_Trainings/Healthcare/3.Clinical_Entity_Resolvers_v2.5.ipynb)",
"_____no_output_____"
],
[
"# Clinical Entity Resolvers (v 2.5)",
"_____no_output_____"
],
[
"## Colab Setup",
"_____no_output_____"
]
],
[
[
"import json\n\nwith open('workshop_license_keys.json') as f_in:\n license_keys = json.load(f_in)\n\nlicense_keys.keys()",
"_____no_output_____"
],
[
"# template for license_key.json\n\n{'secret':\"xxx\",\n'SPARK_NLP_LICENSE': 'aaa',\n'JSL_OCR_LICENSE': 'bbb',\n'AWS_ACCESS_KEY_ID':\"ccc\",\n'AWS_SECRET_ACCESS_KEY':\"ddd\",\n'JSL_OCR_SECRET':\"eee\"}",
"_____no_output_____"
],
[
"import os\n\n# Install java\n! apt-get install -y openjdk-8-jdk-headless -qq > /dev/null\nos.environ[\"JAVA_HOME\"] = \"/usr/lib/jvm/java-8-openjdk-amd64\"\nos.environ[\"PATH\"] = os.environ[\"JAVA_HOME\"] + \"/bin:\" + os.environ[\"PATH\"]\n! java -version\n\n# Install pyspark\n! pip install --ignore-installed -q pyspark==2.4.4\n\nsecret = license_keys['secret']\nos.environ['SPARK_NLP_LICENSE'] = license_keys['SPARK_NLP_LICENSE']\nos.environ['JSL_OCR_LICENSE'] = license_keys['JSL_OCR_LICENSE']\nos.environ['AWS_ACCESS_KEY_ID']= license_keys['AWS_ACCESS_KEY_ID']\nos.environ['AWS_SECRET_ACCESS_KEY'] = license_keys['AWS_SECRET_ACCESS_KEY']\n\n! python -m pip install --upgrade spark-nlp-jsl==2.5.2 --extra-index-url https://pypi.johnsnowlabs.com/$secret\n\n# Install Spark NLP\n! pip install --ignore-installed -q spark-nlp==2.5.2\n\nimport sparknlp\n\nprint (sparknlp.version())\n\nimport json\nimport os\nfrom pyspark.ml import Pipeline\nfrom pyspark.sql import SparkSession\n\n\nfrom sparknlp.annotator import *\nfrom sparknlp_jsl.annotator import *\nfrom sparknlp.base import *\nimport sparknlp_jsl\n\n\n\ndef start(secret):\n builder = SparkSession.builder \\\n .appName(\"Spark NLP Licensed\") \\\n .master(\"local[*]\") \\\n .config(\"spark.driver.memory\", \"16G\") \\\n .config(\"spark.serializer\", \"org.apache.spark.serializer.KryoSerializer\") \\\n .config(\"spark.kryoserializer.buffer.max\", \"2000M\") \\\n .config(\"spark.jars.packages\", \"com.johnsnowlabs.nlp:spark-nlp_2.11:2.5.2\") \\\n .config(\"spark.jars\", \"https://pypi.johnsnowlabs.com/\"+secret+\"/spark-nlp-jsl-2.5.2.jar\")\n \n return builder.getOrCreate()\n\n\nspark = start(secret) # if you want to start the session with custom params as in start function above\n# sparknlp_jsl.start(secret)",
"E: Could not open lock file /var/lib/dpkg/lock-frontend - open (13: Permission denied)\nE: Unable to acquire the dpkg frontend lock (/var/lib/dpkg/lock-frontend), are you root?\nopenjdk version \"1.8.0_252\"\nOpenJDK Runtime Environment (build 1.8.0_252-8u252-b09-1ubuntu1-b09)\nOpenJDK 64-Bit Server VM (build 25.252-b09, mixed mode)\n\u001b[33mWARNING: You are using pip version 20.1; however, version 20.1.1 is available.\nYou should consider upgrading via the '/home/myilmaz/devel/mag/notebooks/3.7env/bin/python -m pip install --upgrade pip' command.\u001b[0m\nLooking in indexes: https://pypi.org/simple, https://pypi.johnsnowlabs.com/8zvTuUjWPt\nRequirement already up-to-date: spark-nlp-jsl==2.5.2 in /home/myilmaz/devel/mag/notebooks/3.7env/lib/python3.7/site-packages (2.5.2)\nRequirement already satisfied, skipping upgrade: spark-nlp==2.5.2 in /home/myilmaz/devel/mag/notebooks/3.7env/lib/python3.7/site-packages (from spark-nlp-jsl==2.5.2) (2.5.2)\nRequirement already satisfied, skipping upgrade: pyspark==2.4.4 in /home/myilmaz/devel/mag/notebooks/3.7env/lib/python3.7/site-packages (from spark-nlp-jsl==2.5.2) (2.4.4)\nRequirement already satisfied, skipping upgrade: py4j==0.10.7 in /home/myilmaz/devel/mag/notebooks/3.7env/lib/python3.7/site-packages (from pyspark==2.4.4->spark-nlp-jsl==2.5.2) (0.10.7)\n\u001b[33mWARNING: You are using pip version 20.1; however, version 20.1.1 is available.\nYou should consider upgrading via the '/home/myilmaz/devel/mag/notebooks/3.7env/bin/python -m pip install --upgrade pip' command.\u001b[0m\n\u001b[33mWARNING: You are using pip version 20.1; however, version 20.1.1 is available.\nYou should consider upgrading via the '/home/myilmaz/devel/mag/notebooks/3.7env/bin/python -m pip install --upgrade pip' command.\u001b[0m\n2.5.2\n"
],
[
"sparknlp_jsl.version()",
"_____no_output_____"
],
[
"spark",
"_____no_output_____"
]
],
[
[
"# Clinical Resolvers",
"_____no_output_____"
],
[
"",
"_____no_output_____"
],
[
"## Entity Resolvers for ICD-10",
"_____no_output_____"
],
[
"A common NLP problem in biomedical aplications is to identify the presence of clinical entities in a given text. This clinical entities could be diseases, symptoms, drugs, results of clinical investigations or others.\n\nOther than providing the code in the \"result\" field it provides more metadata about the matching process:\n\n- target_text -> Text to resolve\n- resolved_text -> Best match text\n- confidence -> Relative confidence for the top match (distance to probability)\n- confidence_ratio -> Relative confidence for the top match. TopMatchConfidence / SecondMatchConfidence\n- alternative_codes -> List of other plausible codes (in the KNN neighborhood)\n- alternative_confidence_ratios -> Rest of confidence ratios\n- all_k_results -> All resolved codes for metrics calculation purposes\n- sentence -> SentenceId\n- chunk -> ChunkId",
"_____no_output_____"
]
],
[
[
"import sys, os, time\nfrom sparknlp.base import *\nfrom sparknlp.annotator import *\nfrom sparknlp.util import *\nfrom sparknlp_jsl.annotator import *\n\nfrom sparknlp.pretrained import ResourceDownloader\n\nfrom pyspark.sql import functions as F\nfrom pyspark.ml import Pipeline, PipelineModel",
"_____no_output_____"
]
],
[
[
"### Clinical NER Pipeline creation",
"_____no_output_____"
]
],
[
[
"# Annotator that transforms a text column from dataframe into an Annotation ready for NLP\n\ndocumentAssembler = DocumentAssembler()\\\n .setInputCol(\"text\")\\\n .setOutputCol(\"document\")\n\n# Sentence Detector annotator, processes various sentences per line\n\nsentenceDetector = SentenceDetector()\\\n .setInputCols([\"document\"])\\\n .setOutputCol(\"sentence\")\n\n# Tokenizer splits words in a relevant format for NLP\n\ntokenizer = Tokenizer()\\\n .setInputCols([\"sentence\"])\\\n .setOutputCol(\"token\")",
"_____no_output_____"
]
],
[
[
"The fourth annotator in the pipeline is \"WordEmbeddingsModel\". We will download a pretrained model available from \"clinical/models\" named \"embeddings_clinical\".\n\nWhen running this cell your are advised to be patient.\n\nFirst time you call this pretrained model it needs to be downloaded in your local.\n\nThe model size is about will download the embeddings_clinical corpus it takes a while.\n\nThe size is about 1.7Gb and will be saved typically in your home folder as\n\n~HOMEFOLDER/cached_models/embeddings_clinical_en_2.0.2_2.4_1558454742956.zip\n\nNext times you call it the model is loaded from your cached copy but even in that case it needs to be indexed each time so expect waiting up to 5 minutes (depending on your machine)",
"_____no_output_____"
]
],
[
[
"# WordEmbeddingsModel pretrained \"embeddings_clinical\" includes a model of 1.7Gb that needs to be downloaded\n\nword_embeddings = WordEmbeddingsModel.pretrained(\"embeddings_clinical\", \"en\", \"clinical/models\")\\\n .setInputCols([\"sentence\", \"token\"])\\\n .setOutputCol(\"embeddings\")",
"embeddings_clinical download started this may take some time.\nApproximate size to download 1.6 GB\n[OK!]\n"
]
],
[
[
"The fifth and final annotator in our NER pipeline is the pretrained `ner_clinical` NerDLModel avaliable from \"clinical/models\". It requires as input the \"sentence\", \"token\" and \"embeddings\" (clinical embeddings pretrained model) and will classify each token in four categories:\n\n- PROBLEM: for patient problems\n\n- TEST: for tests, labs, etc.\n\n- TREATMENT: for treatments, medicines, etc.\n\n- OTHER: for the rest of tokens.\n\nIn order to split those identified NER that are consecutive, the B prefix (as B-PROBLEM) will be used at the first token of each NER. The I prefix (as I-PROBLEM) will be used for the rest of tokens inside the NER.",
"_____no_output_____"
]
],
[
[
"# Named Entity Recognition for clinical concepts.\n\nclinical_ner = NerDLModel.pretrained(\"ner_clinical\", \"en\", \"clinical/models\") \\\n .setInputCols([\"sentence\", \"token\", \"embeddings\"]) \\\n .setOutputCol(\"ner\")",
"ner_clinical download started this may take some time.\nApproximate size to download 13.8 MB\n[OK!]\n"
]
],
[
[
"### Define the NER pipeline\n\nNow we will define the actual pipeline that puts together the annotators we have created.",
"_____no_output_____"
]
],
[
[
"# Build up the pipeline\n\npipeline_ner = Pipeline(\n stages = [\n documentAssembler,\n sentenceDetector,\n tokenizer,\n word_embeddings,\n clinical_ner\n ])",
"_____no_output_____"
]
],
[
[
"### Create a SparkDataFrame with the content\n\nNow we will create a sample Spark dataframe with our clinical note example.\n\nIn this example we are working over a unique clinical note. In production environments a table with several of those clinical notes could be distributed in a cluster and be run in large scale systems.",
"_____no_output_____"
]
],
[
[
"\nclinical_note = (\n 'A 28-year-old female with a history of gestational diabetes mellitus diagnosed eight years '\n 'prior to presentation and subsequent type two diabetes mellitus (T2DM), one prior '\n 'episode of HTG-induced pancreatitis three years prior to presentation, associated '\n 'with an acute hepatitis, and obesity with a body mass index (BMI) of 33.5 kg/m2, '\n 'presented with a one-week history of polyuria, polydipsia, poor appetite, and vomiting. '\n 'Two weeks prior to presentation, she was treated with a five-day course of amoxicillin '\n 'for a respiratory tract infection. She was on metformin, glipizide, and dapagliflozin '\n 'for T2DM and atorvastatin and gemfibrozil for HTG. She had been on dapagliflozin for six months '\n 'at the time of presentation. Physical examination on presentation was significant for dry oral mucosa; '\n 'significantly, her abdominal examination was benign with no tenderness, guarding, or rigidity. Pertinent '\n 'laboratory findings on admission were: serum glucose 111 mg/dl, bicarbonate 18 mmol/l, anion gap 20, '\n 'creatinine 0.4 mg/dL, triglycerides 508 mg/dL, total cholesterol 122 mg/dL, glycated hemoglobin (HbA1c) '\n '10%, and venous pH 7.27. Serum lipase was normal at 43 U/L. Serum acetone levels could not be assessed '\n 'as blood samples kept hemolyzing due to significant lipemia. The patient was initially admitted for '\n 'starvation ketosis, as she reported poor oral intake for three days prior to admission. However, '\n 'serum chemistry obtained six hours after presentation revealed her glucose was 186 mg/dL, the anion gap '\n 'was still elevated at 21, serum bicarbonate was 16 mmol/L, triglyceride level peaked at 2050 mg/dL, and '\n 'lipase was 52 U/L. The β-hydroxybutyrate level was obtained and found to be elevated at 5.29 mmol/L - '\n 'the original sample was centrifuged and the chylomicron layer removed prior to analysis due to '\n 'interference from turbidity caused by lipemia again. The patient was treated with an insulin drip '\n 'for euDKA and HTG with a reduction in the anion gap to 13 and triglycerides to 1400 mg/dL, within '\n '24 hours. Her euDKA was thought to be precipitated by her respiratory tract infection in the setting '\n 'of SGLT2 inhibitor use. The patient was seen by the endocrinology service and she was discharged on '\n '40 units of insulin glargine at night, 12 units of insulin lispro with meals, and metformin 1000 mg '\n 'two times a day. It was determined that all SGLT2 inhibitors should be discontinued indefinitely. She '\n 'had close follow-up with endocrinology post discharge.'\n)\n\ndata_ner = spark.createDataFrame([[clinical_note]]).toDF(\"text\")",
"_____no_output_____"
],
[
"data_ner.show(truncate = 100)\n",
"+----------------------------------------------------------------------------------------------------+\n| text|\n+----------------------------------------------------------------------------------------------------+\n|A 28-year-old female with a history of gestational diabetes mellitus diagnosed eight years prior ...|\n+----------------------------------------------------------------------------------------------------+\n\n"
],
[
"model = pipeline_ner.fit(data_ner)",
"_____no_output_____"
]
],
[
[
"### Transform / annotate the clinical note using the model.\n\nIn order to process the data with the new created model we have two options.\n\nThe first one would be to use the model to transform our clinical note by the command:\n\n`output = model_ner.transform(data_ner)`\n\nThat would save in a Spakr DataFrame (output) the resuls of running the model over the clinical note.\n\nHowever for small tests like this or for real-time request a LightPipelines is a simpler way of managing the data. It will return a dictionary (instead of a Spark DataFrame) with the results of the transformation\n\nWe will create a light_pipeline_ner using our model_ner and then will annotate the clinical_note using this light_pipeline.",
"_____no_output_____"
]
],
[
[
"light_pipeline = LightPipeline(model)\nlight_data = light_pipeline.annotate(clinical_note)",
"_____no_output_____"
]
],
[
[
"Now we have a dictionaty (light_data_ner) that contains the results of running the NER pipeline over our clinical note.\n\nIt contains the original document:",
"_____no_output_____"
]
],
[
[
"light_data['document'][0][0:100]\n",
"_____no_output_____"
],
[
"print(\"Number of sentences: {}\".format(len(light_data['sentence'])))\nprint(\"\")\nfor i in range(5):\n print(\"Sentence {}: {}\".format(i, light_data['sentence'][i][0:80]))",
"Number of sentences: 17\n\nSentence 0: A 28-year-old female with a history of gestational diabetes mellitus diagnosed e\nSentence 1: Two weeks prior to presentation, she was treated with a five-day course of amoxi\nSentence 2: She was on metformin, glipizide, and dapagliflozin for T2DM and atorvastatin and\nSentence 3: She had been on dapagliflozin for six months at the time of presentation.\nSentence 4: Physical examination on presentation was significant for dry oral mucosa;\n"
],
[
"print(\"Number of tokens: {}\".format(len(light_data['token'])))\nprint(\"\")\nfor i in range(25):\n print(\"Token {}: {} ({})\".format(i, light_data['token'][i][0:20], light_data['ner'][i]))\nprint(\"...\")",
"Number of tokens: 436\n\nToken 0: A (O)\nToken 1: 28-year-old (O)\nToken 2: female (O)\nToken 3: with (O)\nToken 4: a (O)\nToken 5: history (O)\nToken 6: of (O)\nToken 7: gestational (B-PROBLEM)\nToken 8: diabetes (I-PROBLEM)\nToken 9: mellitus (I-PROBLEM)\nToken 10: diagnosed (O)\nToken 11: eight (O)\nToken 12: years (O)\nToken 13: prior (O)\nToken 14: to (O)\nToken 15: presentation (O)\nToken 16: and (O)\nToken 17: subsequent (O)\nToken 18: type (B-PROBLEM)\nToken 19: two (I-PROBLEM)\nToken 20: diabetes (I-PROBLEM)\nToken 21: mellitus (I-PROBLEM)\nToken 22: ( (O)\nToken 23: T2DM (B-PROBLEM)\nToken 24: ), (I-PROBLEM)\n...\n"
]
],
[
[
"Lets apply some HTML formating to see the results of the pipeline in a nicer layout:\n\n",
"_____no_output_____"
]
],
[
[
"import random\nfrom IPython.core.display import display, HTML\n\n\ndef get_color():\n r = lambda: random.randint(100,255)\n return '#%02X%02X%02X' % (r(),r(),r())\n\ndef get_NER_html (light_data):\n\n html_output=''\n \n problem_flag = False\n new_problem = []\n problem_list = []\n \n label_list = list(set([i.split('-')[1] for i in light_data['ner'] if i!='O']))\n \n \n label_color={}\n \n for l in label_list:\n \n label_color[l]=get_color()\n \n for index, this_token in enumerate(light_data['token']):\n\n try:\n ent = light_data['ner'][index].split('-')[1]\n except:\n ent = light_data['ner'][index]\n \n \n if ent in label_list:\n color = label_color[ent]\n html_output+='<SPAN style=\"background-color: {}\">'.format(color) + this_token + \" </SPAN>\"\n else:\n html_output+=this_token + \" \"\n \n\n html_output += '</div>'\n html_output += '<div>Color codes:'\n \n\n for l in label_list:\n \n html_output += '<SPAN style=\"background-color: {}\">{}</SPAN>, '.format(label_color[l],l)\n \n \n return display(HTML(html_output))\n",
"_____no_output_____"
],
[
"get_NER_html (light_data)",
"_____no_output_____"
]
],
[
[
"## ICD10 coding Pipeline creation.\n\nWe will now create a new pipeline that from each of these problems will try to assign an ICD10 base on the content, the wordembeddings and some pretrained models for ICD10 annotation.\n\nThe architecture of this new pipeline will be as follows:\n\n- DocumentAssembler (text -> document)\n\n- SentenceDetector (document -> sentence)\n\n- Tokenizer (sentence -> token)\n\n- WordEmbeddingsModel ([sentence, token] -> embeddings)\n\n- NerDLModel ([sentence, token, embeddings] -> ner)\n\n- NerConverter ([\"sentence, token, ner] -> ner_chunk\n\n- ChunkTokenizer (ner_chunk -> ner_chunk_tokenized)\n\n- ICD10CMEntityResolverModel ([ner_chunk_tokenized, embeddings] -> resolution)\n\n- ICD10PCSEntityResolverModel ([ner_chunk_tokenized, embeddings] -> resolution)\n\nSo from a text we end having a list of Named Entities (ner_chunk) and their ICD10 codes (resolution)\n\nMost of the annotators in this pipeline have been already created for the previous pipeline, but we need to create four additional annotators: NerConverter, ChunkEmbeddigns, EntityResolverModel for ICD10CM and EntityResolverModel for ICD10PCS.\n\nNow we define the new pipeline\n\n\n",
"_____no_output_____"
]
],
[
[
"# Named Entity Recognition concepts parser, transforms entities into CHUNKS (required for next step: assertion status)\n\nner_converter = NerConverter() \\\n .setInputCols([\"sentence\", \"token\", \"ner\"]) \\\n .setOutputCol(\"ner_chunk\")\\\n .setWhiteList(['PROBLEM'])\n\nchunk_embeddings = ChunkEmbeddings()\\\n .setInputCols(\"ner_chunk\", \"embeddings\")\\\n .setOutputCol(\"chunk_embeddings\")\n\nchunk_tokenizer = ChunkTokenizer()\\\n .setInputCols(\"ner_chunk\")\\\n .setOutputCol(\"ner_token\")\n\n# ICD resolution model\n\nicd10cm_resolution = ChunkEntityResolverModel.pretrained(\"chunkresolve_icd10cm_clinical\", \"en\", \"clinical/models\") \\\n .setInputCols([\"ner_token\", \"chunk_embeddings\"]) \\\n .setOutputCol(\"icd10cm_code\") \\\n .setDistanceFunction(\"COSINE\") \\\n .setNeighbours(5)",
"chunkresolve_icd10cm_clinical download started this may take some time.\nApproximate size to download 166.3 MB\n[OK!]\n"
],
[
"\npipeline_icd10 = Pipeline(\n stages = [\n documentAssembler,\n sentenceDetector,\n tokenizer,\n word_embeddings,\n clinical_ner,\n ner_converter,\n chunk_embeddings,\n chunk_tokenizer,\n icd10cm_resolution\n ])\n\nmodel_icd10 = pipeline_icd10.fit(data_ner)\n",
"_____no_output_____"
],
[
"light_pipeline_icd10 = LightPipeline(model_icd10)\n",
"_____no_output_____"
],
[
"text = light_data['document'][0]\n\ntext",
"_____no_output_____"
],
[
"import pandas as pd\n\n\nlight_result = light_pipeline_icd10.annotate(text)\n\ndf = pd.DataFrame(list(zip(light_result['ner_chunk'], light_result['icd10cm_code'])),\n columns = ['Problem','ICD10-CM-Code'])\n\ndf",
"_____no_output_____"
],
[
"def get_icd10_codes (light_model, text):\n\n full_light_result = light_model.fullAnnotate(text)\n\n chunks = []\n codes = []\n resolved= []\n begin = []\n end = []\n resolutions=[]\n\n for chunk, code in zip(full_light_result[0]['ner_chunk'], full_light_result[0]['icd10cm_code']):\n \n begin.append(chunk.begin)\n end.append(chunk.end)\n chunks.append(chunk.result)\n codes.append(code.result) \n resolved.append(code.metadata['resolved_text'])\n resolutions.append(code.metadata['all_k_resolutions'])\n \n import pandas as pd\n\n df = pd.DataFrame({'chunks':chunks, 'begin': begin, 'end':end, \n 'icd10_cm_code':codes, 'resolved_text':resolved,\n 'target_text':resolutions})\n\n return df\n\n\n",
"_____no_output_____"
],
[
"df = get_icd10_codes (light_pipeline_icd10, text)\n\ndf",
"_____no_output_____"
],
[
"output = model_icd10.transform(data_ner).cache()\n\noutput.select(F.explode(F.arrays_zip(\"ner_chunk.result\",\"ner_chunk.metadata\",\n \"icd10cm_code.result\",\"icd10cm_code.metadata\")).alias(\"icd10cm_result\")) \\\n.select(F.expr(\"icd10cm_result['0']\").alias(\"chunk\"),\n F.expr(\"icd10cm_result['1'].entity\").alias(\"entity\"),\n F.expr(\"icd10cm_result['3'].resolved_text\").alias(\"resolved_text\"),\n F.expr(\"icd10cm_result['2']\").alias(\"code\"),\n F.expr(\"icd10cm_result['3'].all_k_resolutions\").alias(\"cms\"))\\\n.distinct() \\\n.toPandas()\n",
"_____no_output_____"
],
[
"text = 'He has a brain damage and lung cancer'\n\ndf = get_icd10_codes(light_pipeline_icd10, text)\n\ndf",
"_____no_output_____"
]
],
[
[
"# RxNorm Resolver",
"_____no_output_____"
]
],
[
[
"data_ner.show(truncate=100)",
"+----------------------------------------------------------------------------------------------------+\n| text|\n+----------------------------------------------------------------------------------------------------+\n|A 28-year-old female with a history of gestational diabetes mellitus diagnosed eight years prior ...|\n+----------------------------------------------------------------------------------------------------+\n\n"
],
[
"ner_converter = NerConverter() \\\n .setInputCols([\"sentence\", \"token\", \"ner\"]) \\\n .setOutputCol(\"ner_chunk\")\\\n .setWhiteList(['TREATMENT'])\n\nchunk_embeddings = ChunkEmbeddings()\\\n .setInputCols(\"ner_chunk\", \"embeddings\")\\\n .setOutputCol(\"chunk_embeddings\")\n\nchunk_tokenizer = ChunkTokenizer()\\\n .setInputCols(\"ner_chunk\")\\\n .setOutputCol(\"ner_token\")\n\nrxnormResolver = EnsembleEntityResolverModel()\\\n .pretrained('ensembleresolve_rxnorm_small_clinical', 'en', \"clinical/models\")\\\n .setEnableLevenshtein(True)\\\n .setNeighbours(200).setAlternatives(5).setDistanceWeights([1,3,2,0,0,1])\\\n .setInputCols('ner_token', 'chunk_embeddings')\\\n .setOutputCol('rxnorm_resolution')\\\n\n\npipeline_rx = Pipeline(\n stages = [\n documentAssembler,\n sentenceDetector,\n tokenizer,\n word_embeddings,\n clinical_ner,\n ner_converter,\n chunk_embeddings,\n chunk_tokenizer,\n rxnormResolver\n ])\n\nmodel_rxnorm = pipeline_rx.fit(data_ner)\n",
"ensembleresolve_rxnorm_small_clinical download started this may take some time.\nApproximate size to download 159.7 MB\n[OK!]\n"
],
[
"output = model_rxnorm.transform(data_ner)\n\noutput.show()",
"+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+\n| text| document| sentence| token| embeddings| ner| ner_chunk| chunk_embeddings| ner_token| rxnorm_resolution|\n+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+\n|A 28-year-old fem...|[[document, 0, 24...|[[document, 0, 42...|[[token, 0, 0, A,...|[[word_embeddings...|[[named_entity, 0...|[[chunk, 499, 509...|[[word_embeddings...|[[token, 499, 509...|[[entity, 499, 50...|\n+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+\n\n"
],
[
"output.select(F.explode(F.arrays_zip(\"ner_chunk.result\",\"ner_chunk.metadata\",\"rxnorm_resolution.result\",\"rxnorm_resolution.metadata\")).alias(\"rxnorm_result\")) \\\n .select(F.expr(\"rxnorm_result['0']\").alias(\"chunk\"),\n F.expr(\"rxnorm_result['1'].entity\").alias(\"entity\"),\n F.expr(\"rxnorm_result['3'].all_k_resolutions\").alias(\"target_text\"),\n F.expr(\"rxnorm_result['2']\").alias(\"code\"),\n F.expr(\"rxnorm_result['3'].confidence\").alias('confidence')).show(truncate = 100)",
"+--------------------+---------+----------------------------------------------------------------------------------------------------+-------+----------+\n| chunk| entity| target_text| code|confidence|\n+--------------------+---------+----------------------------------------------------------------------------------------------------+-------+----------+\n| amoxicillin|TREATMENT|ciprofloxacin kit:::ciprofloxacin kit:::amoxicillin 5 0 0 mg oral tablet:::amoxicillin 2 5 0 mg o...| 309310| 0.5000|\n| metformin|TREATMENT|metformin hydrochloride tablets:::metformin hydrochloride oral solution:::pioglitazone tablets:::...|2054650| 0.9931|\n| glipizide|TREATMENT|glipizide 5 mg oral tablet [ glucotrol ] :::glimepiride 3 mg oral tablet:::glimepiride 8 mg oral ...| 205828| 0.9093|\n| dapagliflozin|TREATMENT|dapagliflozin tablets:::linagliptin tablets:::canagliflozin tablets:::empagliflozin tablets:::sax...|1488568| 0.9999|\n| atorvastatin|TREATMENT|candesartan cilexetil 8 mg oral tablet:::candesartan cilexetil 4 mg oral tablet:::simvastatin 5 m...| 153823| 0.4998|\n| gemfibrozil|TREATMENT|gemfibrozil 6 0 0 mg oral tablet [ gemfibrozil ] :::fenofibrate 4 8 mg oral tablet [ tricor ] :::...| 310459| 1.0000|\n| dapagliflozin|TREATMENT|dapagliflozin tablets:::linagliptin tablets:::canagliflozin tablets:::empagliflozin tablets:::sax...|1488568| 0.9999|\n| an insulin drip|TREATMENT|glucagon 3 mg nasal powder [ baqsimi ] :::insulin , regular , human 8 unt inhalation powder [ afr...|2180671| 0.5282|\n| SGLT2 inhibitor|TREATMENT|anti - inhibitor coagulant complex 1 unt injection:::bumetanide 2 mg oral tablet:::bumetanide 1 m...|1659998| 1.0000|\n| insulin glargine|TREATMENT|insulin glargine 1 0 0 units / ml subcutaneous solution:::insulin glargine 3 0 0 units / ml subcu...| 311041| 0.4944|\n| insulin lispro|TREATMENT|insulin lispro 1 0 0 units / ml subcutaneous solution:::insulin lispro 2 0 0 units / ml subcutane...| 242120| 0.4790|\n| metformin|TREATMENT|metformin hydrochloride tablets:::metformin hydrochloride oral solution:::pioglitazone tablets:::...|2054650| 0.9931|\n|all SGLT2 inhibitors|TREATMENT|haegarda c 1 esterase inhibitor subcutaneous ( human ) kit:::haegarda c 1 esterase inhibitor subc...|1926827| 0.5000|\n+--------------------+---------+----------------------------------------------------------------------------------------------------+-------+----------+\n\n"
],
[
"text = 'The patient was prescribed 1 capsule of Advil for 5 days after meals. He was seen by the endocrinology service and she was discharged on 40 units of insulin glargine at night , 12 units of insulin lispro with meals , and metformin 1000 mg two times a day.'\ntext",
"_____no_output_____"
]
],
[
[
"",
"_____no_output_____"
]
],
[
[
"\ndata_ner = spark.createDataFrame([[text]]).toDF(\"text\")",
"_____no_output_____"
],
[
"posology_ner = NerDLModel.pretrained(\"ner_posology\", \"en\", \"clinical/models\") \\\n .setInputCols([\"sentence\", \"token\", \"embeddings\"]) \\\n .setOutputCol(\"ner\")\n\nner_converter = NerConverter() \\\n .setInputCols([\"sentence\", \"token\", \"ner\"]) \\\n .setOutputCol(\"ner_chunk\")\\\n .setWhiteList(['DRUG'])\n\nposology_rx = Pipeline(\n stages = [\n documentAssembler,\n sentenceDetector,\n tokenizer,\n word_embeddings,\n posology_ner,\n ner_converter,\n chunk_embeddings,\n chunk_tokenizer,\n rxnormResolver\n ])\n\nmodel_rxnorm = posology_rx.fit(data_ner)\n\noutput = model_rxnorm.transform(data_ner)\n\noutput.select(F.explode(F.arrays_zip(\"ner_chunk.result\",\"ner_chunk.metadata\",\"rxnorm_resolution.result\",\"rxnorm_resolution.metadata\")).alias(\"rxnorm_result\")) \\\n .select(F.expr(\"rxnorm_result['0']\").alias(\"chunk\"),\n F.expr(\"rxnorm_result['1'].entity\").alias(\"entity\"),\n F.expr(\"rxnorm_result['3'].all_k_resolutions\").alias(\"target_text\"),\n F.expr(\"rxnorm_result['2']\").alias(\"code\"),\n F.expr(\"rxnorm_result['3'].confidence\").alias(\"confidence\")).show(truncate = 100)",
"ner_posology download started this may take some time.\nApproximate size to download 13.7 MB\n[OK!]\n+----------------+------+----------------------------------------------------------------------------------------------------+-------+----------+\n| chunk|entity| target_text| code|confidence|\n+----------------+------+----------------------------------------------------------------------------------------------------+-------+----------+\n| Advil| DRUG|moxetumomab pasudotox - tdfk 1 mg injection [ lumoxiti ] :::phenolphthalein 9 0 mg oral tablet [ ...|2099314| 1.0000|\n|insulin glargine| DRUG|insulin glargine 1 0 0 units / ml subcutaneous solution:::insulin glargine 3 0 0 units / ml subcu...| 311041| 0.4944|\n| insulin lispro| DRUG|insulin lispro 1 0 0 units / ml subcutaneous solution:::insulin lispro 2 0 0 units / ml subcutane...| 242120| 0.4790|\n| metformin| DRUG|metformin hydrochloride tablets:::metformin hydrochloride oral solution:::pioglitazone tablets:::...|2054650| 0.9931|\n+----------------+------+----------------------------------------------------------------------------------------------------+-------+----------+\n\n"
],
[
"model_rxnorm.stages",
"_____no_output_____"
],
[
"model_rxnorm.save('posology_rxNorm_resolver')",
"_____no_output_____"
],
[
"from pyspark.ml import PipelineModel\n\nloaded_rxnorm = PipelineModel.load('posology_rxNorm_resolver')\n\nloaded_rxnorm.stages",
"_____no_output_____"
],
[
"loaded_rxnorm.transform(data_ner).select(\"ner_chunk.result\",\"rxnorm_resolution.result\").show(truncate=100)",
"+----------------------------------------------------+----------------------------------+\n| result| result|\n+----------------------------------------------------+----------------------------------+\n|[Advil, insulin glargine, insulin lispro, metformin]|[2099314, 311041, 242120, 2054650]|\n+----------------------------------------------------+----------------------------------+\n\n"
]
],
[
[
"# ICD10 + RxNorm with multiple NERs",
"_____no_output_____"
]
],
[
[
"notes = [\n'Pentamidine 300 mg IV q . 36 hours , Pentamidine nasal wash 60 mg per 6 ml of sterile water q.d . , voriconazole 200 mg p.o . b.i.d . , acyclovir 400 mg p.o . b.i.d . , cyclosporine 50 mg p.o . b.i.d . , prednisone 60 mg p.o . q.d . , GCSF 480 mcg IV q.d . , Epogen 40,000 units subcu q . week , Protonix 40 mg q.d . , Simethicone 80 mg p.o . q . 8 , nitroglycerin paste 1 \" ; q . 4 h . p.r.n . , flunisolide nasal inhaler , 2 puffs q . 8 , OxyCodone 10-15 mg p.o . q . 6 p.r.n . , Sudafed 30 mg q . 6 p.o . p.r.n . , Fluconazole 2% cream b.i.d . to erythematous skin lesions , Ditropan 5 mg p.o . b.i.d . , Tylenol 650 mg p.o . q . 4 h . p.r.n . , Ambien 5-10 mg p.o . q . h.s . p.r.n . , Neurontin 100 mg q . a.m . , 200 mg q . p.m . , Aquaphor cream b.i.d . p.r.n . , Lotrimin 1% cream b.i.d . to feet , Dulcolax 5-10 mg p.o . q.d . p.r.n . , Phoslo 667 mg p.o . t.i.d . , Peridex 0.12% , 15 ml p.o . b.i.d . mouthwash , Benadryl 25-50 mg q . 4-6 h . p.r.n . pruritus , Sarna cream q.d . p.r.n . pruritus , Nystatin 5 ml p.o . q.i.d . swish and !',\n'Albuterol nebulizers 2.5 mg q.4h . and Atrovent nebulizers 0.5 mg q.4h . , please alternate albuterol and Atrovent ; Rocaltrol 0.25 mcg per NG tube q.d .; calcium carbonate 1250 mg per NG tube q.i.d .; vitamin B12 1000 mcg IM q . month , next dose is due Nov 18 ; diltiazem 60 mg per NG tube t.i.d .; ferrous sulfate 300 mg per NG t.i.d .; Haldol 5 mg IV q.h.s .; hydralazine 10 mg IV q.6h . p.r.n . hypertension ; lisinopril 10 mg per NG tube q.d .; Ativan 1 mg per NG tube q.h.s .; Lopressor 25 mg per NG tube t.i.d .; Zantac 150 mg per NG tube b.i.d .; multivitamin 10 ml per NG tube q.d .; Macrodantin 100 mg per NG tube q.i.d . x 10 days beginning on 11/3/00 .',\n'Tylenol 650 mg p.o . q . 4-6h p.r.n . headache or pain ; acyclovir 400 mg p.o . t.i.d .; acyclovir topical t.i.d . to be applied to lesion on corner of mouth ; Peridex 15 ml p.o . b.i.d .; Mycelex 1 troche p.o . t.i.d .; g-csf 404 mcg subcu q.d .; folic acid 1 mg p.o . q.d .; lorazepam 1-2 mg p.o . q . 4-6h p.r.n . nausea and vomiting ; Miracle Cream topical q.d . p.r.n . perianal irritation ; Eucerin Cream topical b.i.d .; Zantac 150 mg p.o . b.i.d .; Restoril 15-30 mg p.o . q . h.s . p.r.n . insomnia ; multivitamin 1 tablet p.o . q.d .; viscous lidocaine 15 ml p.o . q . 3h can be applied to corner of mouth or lips p.r.n . pain control .',\n'The patient\\'s incisions sternal and right leg were clean and healing well , normal sinus rhythm at 70-80 , with blood pressure 98-110/60 and patient was doing well , recovering , ambulating , tolerating regular diet and last hematocrit prior to discharge was 39% with a BUN and creatinine of 15 and 1.0 , prothrombin time level of 13.8 , chest X-ray prior to discharge showed small bilateral effusions with mild cardiomegaly and subsegmental atelectasis bibasilar and electrocardiogram showed normal sinus rhythm with left atrial enlargement and no acute ischemic changes on electrocardiogram .',\n'This is an 82 - year-old male with a history of prior tobacco use , hypertension , chronic renal insufficiency , COPD , gastritis , and TIA who initially presented to Braintree with a non-ST elevation MI and Guaiac positive stools , transferred to St . Margaret\\'s Center for Women & Infants for cardiac catheterization with PTCA to mid LAD lesion complicated by hypotension and bradycardia requiring Atropine , IV fluids and transient dopamine possibly secondary to vagal reaction , subsequently transferred to CCU for close monitoring , hemodynamically stable at the time of admission to the CCU .',\n'O2 95% on 3L NC mixed Quinn 82% genrl : in nad , resting comfortably heent : perrla ( 4->3 mm ) bilaterally , blind in right visual field , eomi , dry mm , ? thrush neck : no bruits cv : rrr , no m/r/g , faint s1/s2 pulm : cta bilaterally abd : midline scar ( from urostomy ) , nabs , soft , appears distended but patient denies , ostomy RLQ c/d/i , NT to palpation back : right flank urostomy tube , c/d/i , nt to palpation extr : no Gardner neuro : a , ox3 , wiggles toes bilaterally , unable to lift LE , 06-12 grip bilaterally w/ UE , decrease sensation to soft touch in left',\n'Is notable for an inferior myocardial infarction , restrictive and obstructive lung disease with an FEV1 of . 9 and FVC of 1.34 and a moderate at best response to bronchodilators , and a negative sestamibi scan in May , 1999 apart from a severe fixed inferolateral defect , systolic dysfunction with recent echocardiography revealing an LVID of 62 mm . and ejection fraction of 28 percent , moderate mitral regurgitation and mild-to-moderate aortic stenosis with a peak gradient of 33 and a mean gradient of 19 and a valve area of 1.4 cm . squared .',\n'This is a 47 - year-old male with a past medical history of type 2 diabetes , high cholesterol , hypertension , and coronary artery disease , status post percutaneous transluminal coronary angioplasty times two , who presented with acute coronary syndrome refractory to medical treatment and TNK , now status post Angio-Jet percutaneous transluminal coronary angioplasty and stent of proximal left anterior descending artery and percutaneous transluminal coronary angioplasty of first diagonal with intra-aortic balloon pump placement .',\n'Clinical progression of skin and sinus infection on maximal antimicrobial therapy continued , with emergence on November 20 of a new right-sided ptosis in association with a left homonymous hemianopsia , and fleeting confusion while febrile , prompting head MRI which revealed a large 5 x 2 x 4.3 cm region in the right occipital lobe of hemorrhage and edema , with dural and , likely , leptomeningeal enhancement in association with small foci in the right cerebellum and pons , concerning for early lesions of similar type .',\n'The patient had an echocardiogram on day two of admission , which revealed a mildly dilated left atrium , mild symmetric LVH , normal LV cavity size , mild region LV systolic dysfunction , arresting regional wall motion abnormality including focal apical hypokinesis , a normal right ventricular chamber size and free wall motion , a moderately dilated aortic root , a mildly dilated ascending aorta , normal aortic valve leaflet , normal mitral valve leaflet and no pericardial effusions .',\n'The patient is a 65-year-old man with refractory CLL , status post non-myeloblative stem cell transplant approximately nine months prior to admission , and status post prolonged recent Retelk County Medical Center stay for Acanthamoeba infection of skin and sinuses , complicated by ARS due to medication toxicity , as well as GVHD and recent CMV infection , readmitted for new fever , increasing creatinine , hepatomegaly and fluid surge spacing , in the setting of hyponatremia .',\n'Tylenol 650 mg p.o . q.4h . p.r.n . , Benadryl 25 mg p.o . q.h.s . p.r.n . , Colace 100 mg p.o . q.i.d . , Nortriptyline 25 mg p.o . q.h.s . , Simvastatin 10 mg p.o . q.h.s . , Metamucil one packet p.o . b.i.d . p.r.n . , Neurontin 300 mg p.o . t.i.d . , Levsinex 0.375 mg p.o . q.12h . , Lisinopril / hydrochlorothiazide 20/25 mg p.o . q.d . , hydrocortisone topical ointment to affected areas , MS Contin 30 mg p.o . b.i.d . , MSIR 15 to 30 mg p.o . q.4h . p.r.n . pain .',\n'Aspirin 325 q.d . ; albuterol nebs 2.5 mg q . 4h ; Colace 100 mg b.i.d . ; heparin 5,000 units subcu b.i.d . ; Synthroid 200 mcg q.d . ; Ocean Spray 2 sprays q . i.d . ; simvastatin 10 mg q . h.s . ; Flovent 220 mcg 2 puffs b.i.d . ; Zantac 150 b.i.d . ; nystatin ointment to the gluteal fold b.i.d . ; Lisinopril 20 mg q.d . ; Mestinon controlled release 180 q . h.s . ; Mestinon 30 mg q . 4h while awake ; prednisone 60 mg p.o . q . IM ; Atrovent nebs 0.5 mg q . i.d .',\n'An echocardiogram was obtained on 4-26 which showed concentric left ventricular hypertrophy with normal _____ left ventricular function , severe right ventricular dilatation with septal hypokinesis and flattening with a question of right ventricular apical clot raised with mild aortic stenosis , severe tricuspid regurgitation and increased pulmonary artery pressure of approximately 70 millimeters , consistent with fairly severe pulmonary hypertension .',\n'1 ) CV ( R ) finished amio IV load then started on po , agressive lytes ; although interrogation showed >100 episodes of VT ( as / x ) , pt prefers med therapy as opposed to ablation ( I ) enzymes mildly elevated but not actively ischemic ; lipids , ASA , statin , BB ; Adenosine thal 1/4 and echo 1/4 to look for signs of ischemia as active cause for VT ( P ) JVP at angle of jaw 1/4 -- > giving 20 Lasix ; dig level 1/4 1.3 -- > 1/2 dose as on Amio',\n'sodium 141 , potassium 3.5 , chloride 107 , bicarbonates 23.8 , BUN 23 , creatinine 1.1 , glucose 165 , PO2 377 , PCO2 32 , PH 7.50 , asomus 298 , toxic screen negative , white blood cell count 11.1 , hematocrit 39.6 , platelet count 137 , prothrombin time 25.2 , INR 4.3 , partial thromboplastin time 34.7 , urinalysis 1+ albumin , 0-5 high link caths , cervical spine negative , pelvis negative , lumbar spine ; negative , thoracic spine negative .',\n]",
"_____no_output_____"
],
[
"from IPython.core.display import display, HTML\n\nhtml_output=\"\"\nfor i, d in enumerate(notes):\n html_output += f'Note {i}:'\n html_output +='<div style=\"border:2px solid #747474; background-color: #e3e3e3; margin: 5px; padding: 10px\">'\n html_output += d\n html_output += '</div><br/>'\n\ndisplay(HTML(html_output))",
"_____no_output_____"
],
[
"data = spark.createDataFrame([(i,n.lower()) for i,n in enumerate(notes)]).toDF('doc_id', 'text_feed')\n\ndata.show(truncate=50)",
"+------+--------------------------------------------------+\n|doc_id| text_feed|\n+------+--------------------------------------------------+\n| 0|pentamidine 300 mg iv q . 36 hours , pentamidin...|\n| 1|albuterol nebulizers 2.5 mg q.4h . and atrovent...|\n| 2|tylenol 650 mg p.o . q . 4-6h p.r.n . headache ...|\n| 3|the patient's incisions sternal and right leg w...|\n| 4|this is an 82 - year-old male with a history of...|\n| 5|o2 95% on 3l nc mixed quinn 82% genrl : in nad ...|\n| 6|is notable for an inferior myocardial infarctio...|\n| 7|this is a 47 - year-old male with a past medica...|\n| 8|clinical progression of skin and sinus infectio...|\n| 9|the patient had an echocardiogram on day two of...|\n| 10|the patient is a 65-year-old man with refractor...|\n| 11|tylenol 650 mg p.o . q.4h . p.r.n . , benadryl ...|\n| 12|aspirin 325 q.d . ; albuterol nebs 2.5 mg q . 4...|\n| 13|an echocardiogram was obtained on 4-26 which sh...|\n| 14|1 ) cv ( r ) finished amio iv load then started...|\n| 15|sodium 141 , potassium 3.5 , chloride 107 , bic...|\n+------+--------------------------------------------------+\n\n"
]
],
[
[
"let's build a SparkNLP pipeline with the following stages:\n\n`DocumentAssembler`: Entry annotator for our pipelines; it creates the data structure for the Annotation Framework\n\n`SentenceDetector`: Annotator to pragmatically separate complete sentences inside each document\n\n`Tokenizer`: Annotator to separate sentences in tokens (generally words)\n\n`StopWordsCleaner`: Annotator to remove words defined as StopWords in SparkML\n\n`WordEmbeddings`: Vectorization of word tokens, in this case using word embeddings trained from PubMed, ICD10 and other clinical resources.\n\n`ChunkEmbeddings`: Aggregates the WordEmbeddings for each NER Chunk\n\n`BioNLP NER + NerConverter`: This annotators return Chunks related to Cancer and Genetics diseases\n\n`ChunkEntityResolver`: Annotator that performs search for the KNNs, in this case trained from ICDO Histology Behavior.",
"_____no_output_____"
]
],
[
[
"# preparation Annotators and Embeddings\n\ndocAssembler = DocumentAssembler()\\\n .setInputCol('text_feed')\\\n .setOutputCol('document')\n\nsentenceDetector = SentenceDetector()\\\n .setInputCols('document')\\\n .setOutputCol('sentence')\n\ntokenizer_chars = [\",\",\"\\/\",\" \",\".\",\"|\",\"@\",\"#\",\"%\",\"&\",\"\\\\$\",\"\\\\[\",\"\\\\]\",\"\\\\(\",\"\\\\)\",\"\\\\-\",\";\"]\n\ntokenizer = Tokenizer().setSplitChars(tokenizer_chars)\\\n .setInputCols('sentence')\\\n .setOutputCol('token')\n",
"_____no_output_____"
],
[
"\n# Annotators responsible for the Cancer Genetics Entity Recognition task\n\n\njslNer = NerDLModel.pretrained('ner_jsl', 'en', \"clinical/models\")\\\n .setInputCols('sentence', 'token', 'embeddings')\\\n .setOutputCol('ner_jsl')\n\ndrugNer = NerDLModel.pretrained('ner_drugs', 'en', \"clinical/models\")\\\n .setInputCols('sentence', 'token', 'embeddings')\\\n .setOutputCol('ner_drug')",
"ner_jsl download started this may take some time.\nApproximate size to download 14 MB\n[OK!]\nner_drugs download started this may take some time.\nApproximate size to download 13.8 MB\n[OK!]\n"
],
[
"\n#Converter annotators transform IOB tags into full chunks (sequence set of tokens) tagged with `entity` metadata\n\njslConverter = NerConverter()\\\n .setInputCols('sentence', 'token', 'ner_jsl')\\\n .setOutputCol('chunk_jsl')\\\n .setWhiteList([\"Diagnosis\"])\n\ndrugConverter = NerConverter()\\\n .setInputCols('sentence', 'token', 'ner_drug')\\\n .setOutputCol('chunk_drug')",
"_____no_output_____"
],
[
"\n#ChunkEmbeddings annotators aggregate embeddings for each token in the chunk\n\njslChunkEmbeddings = ChunkEmbeddings()\\\n .setInputCols('chunk_jsl', 'embeddings')\\\n .setOutputCol('chunk_embs_jsl')\n\ndrugChunkEmbeddings = ChunkEmbeddings()\\\n .setInputCols('chunk_drug', 'embeddings')\\\n .setOutputCol('chunk_embs_drug')",
"_____no_output_____"
],
[
"\n# ChunkTokenizer provides extra flexibility at the time of tokenizing a given chunk\n\njslChunkTokenizer = ChunkTokenizer()\\\n .setSplitChars(tokenizer_chars)\\\n .setInputCols('chunk_jsl')\\\n .setOutputCol('chunk_token_jsl')\n\ndrugChunkTokenizer = ChunkTokenizer()\\\n .setSplitChars(tokenizer_chars)\\\n .setInputCols('chunk_drug')\\\n .setOutputCol('chunk_token_drug')",
"_____no_output_____"
],
[
"# Entity Resolution Pretrained Models\n\nicd10cmResolver = ChunkEntityResolverModel.pretrained('chunkresolve_icd10cm_clinical', 'en', \"clinical/models\")\\\n .setEnableLevenshtein(True)\\\n .setNeighbours(200).setAlternatives(5).setDistanceWeights([1,3,2,0,0,1])\\\n .setInputCols('chunk_token_jsl', 'chunk_embs_jsl')\\\n .setOutputCol('icd10cm_resolution')\n\nrxnormResolver = EnsembleEntityResolverModel()\\\n .pretrained('ensembleresolve_rxnorm_small_clinical', 'en', \"clinical/models\")\\\n .setEnableLevenshtein(True)\\\n .setNeighbours(200).setAlternatives(5).setDistanceWeights([1,3,2,0,0,1])\\\n .setInputCols('chunk_token_drug', 'chunk_embs_drug')\\\n .setOutputCol('rxnorm_resolution')\\",
"chunkresolve_icd10cm_clinical download started this may take some time.\nApproximate size to download 166.3 MB\n[OK!]\nensembleresolve_rxnorm_small_clinical download started this may take some time.\nApproximate size to download 159.7 MB\n[OK!]\n"
],
[
"pipelineFull = Pipeline().setStages([\n docAssembler, \n sentenceDetector, \n tokenizer, \n word_embeddings, \n jslNer,\n drugNer,\n jslConverter,\n drugConverter,\n jslChunkEmbeddings, \n drugChunkEmbeddings,\n jslChunkTokenizer,\n drugChunkTokenizer,\n rxnormResolver,\n icd10cmResolver\n])",
"_____no_output_____"
],
[
"# Persisiting temporarily to keep DAG size and resource usage low (Ensmeble Resolvers are Resource Intensive)\npipelineModelFull = pipelineFull.fit(data)\n\noutput = pipelineModelFull.transform(data)\n\noutput.write.mode(\"overwrite\").save(\"temp\")\n\noutput = spark.read.load(\"temp\")",
"_____no_output_____"
],
[
"output.show()",
"+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+\n| text| document| sentence| token| embeddings| ner| ner_chunk| chunk_embeddings| ner_token| rxnorm_resolution|\n+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+\n|The patient was p...|[[document, 0, 25...|[[document, 0, 68...|[[token, 0, 2, Th...|[[word_embeddings...|[[named_entity, 0...|[[chunk, 40, 44, ...|[[word_embeddings...|[[token, 40, 44, ...|[[entity, 40, 44,...|\n+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+\n\n"
],
[
"def quick_metadata_analysis(df, doc_field, chunk_field, code_fields):\n code_res_meta = \", \".join([f\"{cf}.metadata\" for cf in code_fields])\n expression = f\"explode(arrays_zip({chunk_field}.begin, {chunk_field}.end, {chunk_field}.result, {chunk_field}.metadata, \"+code_res_meta+\")) as a\"\n top_n_rest = [(f\"float(a['{i+4}'].confidence) as {(cf.split('_')[0])}_conf\",\n f\"arrays_zip(split(a['{i+4}'].all_k_results,':::'),split(a['{i+4}'].all_k_resolutions,':::')) as {cf.split('_')[0]+'_opts'}\")\n for i, cf in enumerate(code_fields)]\n top_n_rest_args = []\n for tr in top_n_rest:\n for t in tr:\n top_n_rest_args.append(t)\n return df.selectExpr(doc_field, expression) \\\n .orderBy('doc_id', F.expr(\"a['0']\"), F.expr(\"a['1']\"))\\\n .selectExpr(f\"concat_ws('::',{doc_field},a['0'],a['1']) as coords\", \"a['2'] as chunk\",\"a['3'].entity as entity\", *top_n_rest_args)",
"_____no_output_____"
],
[
"icd10cm_analysis = quick_metadata_analysis(output, 'doc_id', 'chunk_jsl',['icd10cm_resolution']).toPandas()",
"_____no_output_____"
],
[
"rxnorm_analysis = \\\nquick_metadata_analysis(output, 'doc_id', 'chunk_drug',['rxnorm_resolution']).toPandas()",
"_____no_output_____"
],
[
"pd.set_option('display.max_colwidth', 250)\npd.set_option('display.max_rows', 500)",
"_____no_output_____"
],
[
"icd10cm_analysis[icd10cm_analysis.icd10cm_conf>0.25].head(20)",
"_____no_output_____"
],
[
"rxnorm_analysis[rxnorm_analysis.rxnorm_conf>0.5].head(20)",
"_____no_output_____"
]
],
[
[
"## Snomed Resolver",
"_____no_output_____"
]
],
[
[
"data.show(3)",
"+------+--------------------+\n|doc_id| text_feed|\n+------+--------------------+\n| 0|pentamidine 300 m...|\n| 1|albuterol nebuliz...|\n| 2|tylenol 650 mg p....|\n+------+--------------------+\nonly showing top 3 rows\n\n"
],
[
"docAssembler = DocumentAssembler()\\\n .setInputCol('text_feed')\\\n .setOutputCol('document')\n\nsentenceDetector = SentenceDetector()\\\n .setInputCols('document')\\\n .setOutputCol('sentence')\n\n\ntokenizer = Tokenizer()\\\n .setInputCols('sentence')\\\n .setOutputCol('token')\n\nner_converter = NerConverter() \\\n .setInputCols([\"sentence\", \"token\", \"ner\"]) \\\n .setOutputCol(\"ner_chunk\")\n\nchunk_embeddings = ChunkEmbeddings()\\\n .setInputCols(\"ner_chunk\", \"embeddings\")\\\n .setOutputCol(\"chunk_embeddings\")\n\nchunk_tokenizer = ChunkTokenizer()\\\n .setInputCols(\"ner_chunk\")\\\n .setOutputCol(\"ner_token\")\n\n\nsnomedResolver = \\\n EnsembleEntityResolverModel.pretrained(\"ensembleresolve_snomed_small_clinical\",\"en\",\"clinical/models\")\\\n .setInputCols(\"ner_token\",\"chunk_embeddings\").setOutputCol(\"snomed_resolution\")\n\n\npipeline_snomed = Pipeline(\n stages = [\n docAssembler,\n sentenceDetector,\n tokenizer,\n word_embeddings,\n clinical_ner,\n ner_converter,\n chunk_embeddings,\n chunk_tokenizer,\n snomedResolver\n ])\n\nmodel_snomed = pipeline_snomed.fit(data)\n",
"ensembleresolve_snomed_small_clinical download started this may take some time.\nApproximate size to download 440.3 MB\n[OK!]\n"
],
[
"\noutput = model_snomed.transform(data)\n\noutput.write.mode(\"overwrite\").save(\"temp\")\n\noutput = spark.read.load(\"temp\")",
"_____no_output_____"
],
[
"\noutput.select(F.explode(F.arrays_zip(\"ner_chunk.result\",\"ner_chunk.metadata\",\"snomed_resolution.result\",\"snomed_resolution.metadata\")).alias(\"snomed_result\")) \\\n .select(F.expr(\"snomed_result['0']\").alias(\"chunk\"),\n F.expr(\"snomed_result['1'].entity\").alias(\"entity\"),\n F.expr(\"snomed_result['3'].all_k_resolutions\").alias(\"target_text\"),\n F.expr(\"snomed_result['2']\").alias(\"code\"),\n F.expr(\"snomed_result['3'].confidence\").alias(\"confidence\")).show(truncate = 100)",
"+-------------------------+---------+----------------------------------------------------------------------------------------------------+---------+----------+\n| chunk| entity| target_text| code|confidence|\n+-------------------------+---------+----------------------------------------------------------------------------------------------------+---------+----------+\n| pentamidine|TREATMENT|pentamidine:::pentamidine isethionate:::pentamidine diisothionate:::eprazinone:::chloroquine phos...|372699006| 0.9962|\n| pentamidine nasal wash|TREATMENT|pentamidine isetionate:::pentamidine ( substance ) :::pentamidine diisothionate ( substance ) :::...| 16826009| 0.9281|\n| voriconazole|TREATMENT| voriconazole:::trifluridine:::amikacin liposome:::lymecycline:::luliconazole|385469007| 1.0000|\n| acyclovir|TREATMENT| acyclovir:::acyclovir sodium:::trolamine salicylate:::trifluridine:::lymecycline|372729009| 0.9981|\n| cyclosporine|TREATMENT|cyclosporine and cyclosporine derivative:::doxercalciferol:::trolamine salicylate:::trifluridine:...|412039006| 1.0000|\n| prednisone|TREATMENT| prednisone:::hydrocortisone valerate:::doxercalciferol:::mycophenolate sodium:::busulphan|116602009| 1.0000|\n| gcsf|TREATMENT| ephelis:::ketoacidaemia:::parafilariosis:::gasterophilosis:::carnosinuria|403536009| 0.2000|\n| epogen|TREATMENT| polyonychia:::micronychia:::cocainism:::colporrhexis:::choledochocoele|403280008| 0.2000|\n| protonix|TREATMENT| diastematocrania:::trachonychia:::silicosiderosis:::melanoglossia:::hydromeningomyelocele| 18355000| 0.2000|\n| simethicone|TREATMENT|simethicone only product:::magaldrate and simethicone only product:::loperamide and simethicone o...|777535005| 0.9708|\n| nitroglycerin paste|TREATMENT|silicone paste:::fish paste:::polyantibiotic paste:::dental prophylaxis paste:::bite registration...|261246006| 0.3221|\n| p.r.n|TREATMENT| cocainism:::colporrhexis:::proctodynia:::gooseflesh:::polyalgia|403748005| 0.2000|\n|flunisolide nasal inhaler|TREATMENT| flunisolide:::nasal mucus:::nasal fluid:::nasal contents:::nasal flap ( substance ) |116588001| 0.8640|\n| oxycodone|TREATMENT|oxycodone hydrochloride:::oxycodone terephthalate:::netupitant:::sufentanil citrate:::morphine di...|387024006| 0.5000|\n| sudafed|TREATMENT| diastematocrania:::trachonychia:::silicosiderosis:::melanoglossia:::hydromeningomyelocele| 18355000| 0.2000|\n| p.r.n|TREATMENT| cocainism:::colporrhexis:::proctodynia:::gooseflesh:::polyalgia|403748005| 0.2000|\n| fluconazole|TREATMENT| fluconazole:::griseofulvin microsize:::amikacin liposome:::lymecycline:::luliconazole|387174006| 1.0000|\n|erythematous skin lesions| PROBLEM|achromic skin lesions of pinta:::achromic and hyperchromic skin lesions of pinta:::skin depigment...| 82074004| 0.8271|\n| ditropan|TREATMENT| diastematocrania:::trachonychia:::silicosiderosis:::melanoglossia:::hydromeningomyelocele| 18355000| 0.2000|\n| tylenol|TREATMENT| polyonychia:::micronychia:::aesthesioneuroblastoma:::cocainism:::colporrhexis|403280008| 0.2000|\n+-------------------------+---------+----------------------------------------------------------------------------------------------------+---------+----------+\nonly showing top 20 rows\n\n"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code",
"code",
"code",
"code",
"code"
],
[
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code"
]
] |
cbdbc95d54005b58298232a262ddde7fe5378d22
| 99,273 |
ipynb
|
Jupyter Notebook
|
Session 6/gdelt_example.ipynb
|
Josedbarnes/applications
|
bd14211aca6b49ad2ebfb8f4f3513807b0f64f1e
|
[
"MIT"
] | null | null | null |
Session 6/gdelt_example.ipynb
|
Josedbarnes/applications
|
bd14211aca6b49ad2ebfb8f4f3513807b0f64f1e
|
[
"MIT"
] | null | null | null |
Session 6/gdelt_example.ipynb
|
Josedbarnes/applications
|
bd14211aca6b49ad2ebfb8f4f3513807b0f64f1e
|
[
"MIT"
] | null | null | null | 40.569268 | 914 | 0.486658 |
[
[
[
"Adopted from GDELT Data Wrangle by James Houghton https://nbviewer.jupyter.org/github/JamesPHoughton/Published_Blog_Scripts/blob/master/GDELT%20Wrangler%20-%20Clean.ipynb",
"_____no_output_____"
],
[
"Additional GDELT resources: \n \n GDELT library overview: https://colab.research.google.com/drive/1rnKEHKV1StOwGtFPsCctKDPTBB_kHOc_?usp=sharing \n \n GDELT with big data: https://github.com/linwoodc3/gdeltPyR/wiki/Pulling-Large-GDELT-Data\n ",
"_____no_output_____"
],
[
"# PART I: Get GDELT DATA FOR NIGER\n\n\n### Get the GDELT index files",
"_____no_output_____"
]
],
[
[
"#Pike trying to create a graph of conflict related news reports on the current Armenia Azerbaijan war\n#to highlight in increase in bellicose statements from the principle participant Azerbaijan\nimport requests\nimport lxml.html as lh\n\ngdelt_base_url = 'http://data.gdeltproject.org/events/'\n\n# get the list of all the links on the gdelt file page\npage = requests.get(gdelt_base_url+'index.html') #Grab GDELT reference list which is by day\ndoc = lh.fromstring(page.content)\nlink_list = doc.xpath(\"//*/ul/li/a/@href\") #Returns all the possible CSV files of GDELT data as a references list\n\n# separate out those links that begin with four digits \n'''\nWill extract just the days resulting in list like: \n['20200617.export.CSV.zip',\n '20200616.export.CSV.zip',\n '20200615.export.CSV.zip',...]\n Until 2015\n'''\n\nfile_list = [x for x in link_list if str.isdigit(x[0:4])]\nfile_list",
"_____no_output_____"
],
[
"#Counters to help assess how many files are coming and going out\ninfilecounter = 0\noutfilecounter = 0",
"_____no_output_____"
]
],
[
[
"### Uses GDELT Index file list to download GDELT data for that day for that country",
"_____no_output_____"
]
],
[
[
"\nimport os.path #To help navigate the file directories\nimport urllib #To request from GDELT\nimport zipfile #TO unzip the files we downlaod\nimport glob #To go through multiple files in a directory\nimport operator \n\nlocal_path = './results/' # Will save to empy results folder to help keep file clean\n#Pike pulled the country codes and fought a tuple error\nfips_country_code = 'AJ'\n\n\n\n#Adjust list number to get days wanted \nfor compressed_file in file_list[:10]:\n print(compressed_file,)\n \n # if we dont have the compressed file stored locally, go get it. Keep trying if necessary.\n while not os.path.isfile(local_path+compressed_file): \n print('downloading,'),\n urllib.request.urlretrieve(url=gdelt_base_url+compressed_file, \n filename=local_path+compressed_file)\n \n # extract the contents of the compressed file to a temporary directory \n print('extracting,'),\n z = zipfile.ZipFile(file=local_path+compressed_file, mode='r') \n z.extractall(path=local_path+'tmp/')\n \n # parse each of the csv files in the working directory, \n print('parsing,'),\n for infile_name in glob.glob(local_path+'tmp/*'):\n outfile_name = local_path+fips_country_code+'%04i.tsv'%outfilecounter\n \n # open the infile and outfile\n with open(infile_name, mode='r', encoding=\"ISO-8859-1\") as infile, open(outfile_name, mode='w') as outfile:\n for line in infile:\n # extract lines with our interest country code\n if fips_country_code in operator.itemgetter(51, 37, 44)(line.split('\\t')): \n outfile.write(line)\n outfilecounter +=1\n \n # delete the temporary file\n os.remove(infile_name)\n infilecounter +=1\n print('done', infilecounter)\n ",
"_____no_output_____"
]
],
[
[
"# PART II: PARSE DATA AGAIN\n\n### Read in the data",
"_____no_output_____"
]
],
[
[
"import pandas as pd\n\n# Get the GDELT field names from a helper file\ncolnames = pd.read_csv('CSV.header.fieldids.csv')['Field Name']\n\n\n# Build DataFrames from each of the intermediary files\nfiles = glob.glob(local_path+fips_country_code+'*')\nDFlist = []\nfor active_file in files:\n print(active_file)\n DFlist.append(pd.read_csv(active_file, sep='\\t', header=None, dtype=str,\n names=colnames, index_col=['GLOBALEVENTID'], encoding='iso-8859-1'))\n\n# Merge the file-based dataframes and save a pickle\nDF = pd.concat(DFlist)\nDF.to_pickle(local_path+'backup'+fips_country_code+'.pickle') \n \n# once everythin is safely stored away, remove the temporary files\nfor active_file in files:\n os.remove(active_file)",
"./results/AJ0000.tsv\n./results/AJ0001.tsv\n./results/AJ0002.tsv\n./results/AJ0003.tsv\n./results/AJ0004.tsv\n./results/AJ0005.tsv\n./results/AJ0006.tsv\n./results/AJ0007.tsv\n"
],
[
"import pickle\n#Pike changed the name from Niger_Data to Conflict_Data\nConflict_Data = pd.read_pickle(r\"./results/backupAJ.pickle\")",
"_____no_output_____"
]
],
[
[
"### See top 5 lines of data",
"_____no_output_____"
]
],
[
[
"Conflict_Data.head()",
"_____no_output_____"
]
],
[
[
"### Helper Function to turn codebooks into look up tables",
"_____no_output_____"
]
],
[
[
"def ref_dict(df):\n cols = list(df)\n ref_dict = {}\n for row in df.iterrows(): \n ref_dict[row[1][cols[0]]] = row[1][cols[1]]\n \n return ref_dict",
"_____no_output_____"
]
],
[
[
"### Convert each codebook and store in object",
"_____no_output_____"
]
],
[
[
"#Read in event codes\neventCodes = ref_dict(pd.read_csv(\"./Ref Codes/CAMEO.eventcodes.txt\", sep='\\t'))\n#Read in Goldsteinscale\ngoldScale = ref_dict(pd.read_csv(\"./Ref Codes/CAMEO.goldsteinscale.txt\", sep='\\t'))\n#Read in ethnic groups\nethnicCodes =ref_dict(pd.read_csv(\"./Ref Codes/CAMEO.ethnic.txt\", sep='\\t'))\n#Read in known Groups\nknownGroups = ref_dict(pd.read_csv(\"./Ref Codes/CAMEO.knowngroup.txt\", sep='\\t'))\n#Read in relgion\nreligionCodes = ref_dict(pd.read_csv(\"./Ref Codes/CAMEO.religion.txt\", sep='\\t'))\n#Read in type\ntypeCodes = ref_dict(pd.read_csv(\"./Ref Codes/CAMEO.type.txt\", sep='\\t'))\n\neventCodes",
"_____no_output_____"
],
[
"# Turn colnames into list for ref\n\ncross_ref = list(colnames)\n",
"_____no_output_____"
],
[
"# Create look up table to get values instead of numbers\n\nlook_up_code = {\"eventCodes\": [26,27,28], \"goldScale\":[30], \"ethnicCodes\":[9,19], \"knownGroups\":[8,18], \n \"religionCodes\":[10,11,20,21], \"typeCodes\":[12,13,14,22,23,24]}",
"_____no_output_____"
],
[
"'''\nHelper function to user can reorient data based on interest from codes\n\ndata: Conflict_Data - pandas dataframe\nref: key value from look_look_code - string\ncodebook: reference \n'''\n\nimport math\n\ndef search_dict(data,ref, codebook):\n res = {}\n look_up = look_up_code[ref]\n col_names = []\n for i in look_up: \n col_names.append(cross_ref[i])\n \n for col in col_names: \n for row in data.iterrows(): \n if isinstance(row[1][col],float):\n #print (type(row[1][col]), col)\n pass\n else: \n #print (col)\n var = codebook[row[1][col]].upper()\n #print (var, row[1][col])\n if var in res.keys(): \n #print(row[1][col])\n res[var].append(dict(row[1]))\n else: \n res[var] = [dict(row[1])]\n return res\n \n",
"_____no_output_____"
],
[
"#Pike attempting to graph news statements over time leading up to the current conflict\n#Code adapted from http://linwoodc3.github.io/2017/10/15/Taking-the-Pulse-of-worldwide-news-Media-using-gdelt.html\n#and https://nbviewer.jupyter.org/github/dmasad/GDELT_Intro/blob/master/Getting_Started_with_GDELT.ipynb\nimport matplotlib.pyplot as plt\nimport seaborn as sns\n\nf,ax = plt.subplots(figsize=(15,5))\n\nplt.title('Distribution of Azerbaijian-Armenia News Production')\n\nsns.kdeplot(conflict_data.assign(provider=conflict_data.sourceurl.\\\n apply(lambda x: s.search(x).group() if s.search(x) else np.nan))['provider']\\\n.value_counts(),bw=0.4,shade=True,lable='No. of articles written',ax=ax)\n\nsns.kdeplot(conflict_data.assign(provider=conflict_data.sourceurl.\\\n apply(lambda x: s.search(x_.group(_ if s.search else np.nan))['provider']\\\n.value_counts(),bw=0.4,shade=True,label='Cumulative',cumulative=True,ax=ax)\n\nplt.show()",
"_____no_output_____"
],
[
"#verfication to ensure code is working properly\nfor k,v in res.items(): \n print (k, \": \", len(v))",
"AZANDE : 120\nPAHARI RAJPUT : 1\nKURD : 2\n"
],
[
"#Put each collection of articles in a Dataframe\nlist_res = []\n\nfor cat in res.values(): \n #print(cat)\n list_res.append(pd.DataFrame(cat))\n ",
"_____no_output_____"
],
[
"list_res[0] #access the group you are interested in by changing the variables",
"_____no_output_____"
]
],
[
[
"### Homework 4: Do some type of analysis with GDELT data. It can be country focused (e.g. Guatemala) or topic focused (e.g. attacks or bilateral agreements)\n\n### Must write in the first cell what you are interested in. Code must work but results can be garabage. Update the GDELT parameters to get the information you want and then include some type of plot can be a graph or can be a map. \n\n### Total Points Possible 19\n ",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown"
] |
[
[
"markdown",
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
]
] |
cbdbd9355cc8d1a49da4fa6186dea23740ed8534
| 21,351 |
ipynb
|
Jupyter Notebook
|
.ipynb_checkpoints/master_script-checkpoint.ipynb
|
ferdinandwp/CEBD1260
|
cd70eaf59433c9215b630a0422e094da522710d4
|
[
"MIT"
] | null | null | null |
.ipynb_checkpoints/master_script-checkpoint.ipynb
|
ferdinandwp/CEBD1260
|
cd70eaf59433c9215b630a0422e094da522710d4
|
[
"MIT"
] | null | null | null |
.ipynb_checkpoints/master_script-checkpoint.ipynb
|
ferdinandwp/CEBD1260
|
cd70eaf59433c9215b630a0422e094da522710d4
|
[
"MIT"
] | null | null | null | 52.459459 | 1,505 | 0.503255 |
[
[
[
"# This is master script\n# Import library\nimport numpy as np \nimport pandas as pd\nimport os\nimport glob\nimport sys\nfrom statistics import mode \n\nsys.path.append('..')\nfrom processed_script.ipay import preprocessing_ipay\nfrom processed_script.cc_balance import preprocessing_cc_balance\nfrom processed_script.prev_app import preprocessing_prev_app\nfrom processed_script.POS_CASH_balance import preprocessing_POS_CASH_balance\nfrom processed_script.main import preprocessing_main\n\n# ipay_df = preprocessing_ipay()\ncc_balance_df = preprocessing_cc_balance()\nprev_app_df = preprocessing_prev_app()\n# POS_CASH_balance_df = preprocessing_POS_CASH_balance()\nmain_df = preprocessing_main()\n",
"_____no_output_____"
],
[
"# application train dataset\nmain_df.SK_ID_CURR.nunique(), main_df.shape",
"_____no_output_____"
],
[
"# first join for prev application\n# prev_app_df.SK_ID_CURR.nunique(), prev_app_df.shape\n# for f in prev_app_df:\n# print(f)\n \nprev_app_df.head()",
"_____no_output_____"
],
[
"# find common ID\n\nmain_ids = list(main_df['SK_ID_CURR'].unique())\nprev_ids = list(prev_app_df['SK_ID_CURR'].unique())\n\ncommon_ids = set(main_ids).intersection(set(prev_ids))\n\nlen(common_ids)\n\n# identify dataset in prev_application dataset\nfiltered_prev_df = prev_app_df.loc[prev_app_df.SK_ID_CURR.isin(main_ids)]\n\ngp = filtered_prev_df.groupby('SK_ID_CURR')",
"_____no_output_____"
],
[
"# do prev app table first\n\nagg_dict = {\n # int/float features\n 'AMT_ANNUITY':['mean'],\n 'AMT_APPLICATION':['mean'],\n 'AMT_CREDIT':['mean'],\n 'AMT_DOWN_PAYMENT':['mean'],\n 'AMT_GOODS_PRICE':['mean'],\n 'HOUR_APPR_PROCESS_START':['mean'],\n 'NFLAG_LAST_APPL_IN_DAY':['mean'],\n 'RATE_DOWN_PAYMENT':['mean'],\n 'RATE_INTEREST_PRIMARY':['mean'],\n 'RATE_INTEREST_PRIVILEGED':['mean'],\n 'DAYS_DECISION':['mean'],\n 'SELLERPLACE_AREA':['mean'],\n 'CNT_PAYMENT':['mean'],\n 'DAYS_FIRST_DRAWING':['mean'],\n 'DAYS_FIRST_DUE':['mean'],\n 'DAYS_LAST_DUE_1ST_VERSION':['mean'],\n 'DAYS_LAST_DUE':['mean'],\n 'DAYS_TERMINATION':['mean'],\n 'NFLAG_INSURED_ON_APPROVAL':['mean']\n}\n\nagg_df = prev_app_df.groupby('SK_ID_CURR').agg(agg_dict)\nagg_df.columns = ['PREV_APP_{}_{}'.format(x[0],x[1]) for x in agg_df.columns.tolist()]\n\n\n\n\n",
"_____no_output_____"
],
[
"\n",
"_____no_output_____"
]
]
] |
[
"code"
] |
[
[
"code",
"code",
"code",
"code",
"code",
"code"
]
] |
cbdbdcb9f1b1df7f747837495da50541ac14d9cb
| 12,404 |
ipynb
|
Jupyter Notebook
|
07-prediction-serving.ipynb
|
tgaillard1/ucaip-labs
|
44b2d8ec017793e40ae1a26b6b6a505d18bdf002
|
[
"Apache-2.0"
] | 52 |
2021-04-22T18:28:54.000Z
|
2021-12-11T23:12:55.000Z
|
07-prediction-serving.ipynb
|
tgaillard1/ucaip-labs
|
44b2d8ec017793e40ae1a26b6b6a505d18bdf002
|
[
"Apache-2.0"
] | 2 |
2021-05-23T21:49:18.000Z
|
2021-08-25T09:45:10.000Z
|
07-prediction-serving.ipynb
|
tgaillard1/ucaip-labs
|
44b2d8ec017793e40ae1a26b6b6a505d18bdf002
|
[
"Apache-2.0"
] | 12 |
2021-05-03T16:38:19.000Z
|
2022-03-03T20:16:12.000Z
| 26.004193 | 110 | 0.557401 |
[
[
[
"# 07 - Serving predictions\n\nThe purpose of the notebook is to show how to use the deployed model for online and batch prediction.\nThe notebook covers the following tasks:\n\n1. Test the `Endpoint` resource for online prediction.\n2. Use the custom model uploaded as a `Model` resource for batch prediciton.\n3. Run a the batch prediction pipeline using `Vertex Pipelines`.",
"_____no_output_____"
],
[
"## Setup",
"_____no_output_____"
],
[
"### Import libraries",
"_____no_output_____"
]
],
[
[
"import os\nimport time\nfrom datetime import datetime\nimport tensorflow as tf\n\nfrom google.cloud import aiplatform as vertex_ai",
"_____no_output_____"
]
],
[
[
"### Setup Google Cloud project",
"_____no_output_____"
]
],
[
[
"PROJECT_ID = '[your-project-id]' # Change to your project id.\nREGION = 'us-central1' # Change to your region.\nBUCKET = '[your-bucket-name]' # Change to your bucket name.\n\nif PROJECT_ID == '' or PROJECT_ID is None or PROJECT_ID == '[your-project-id]':\n # Get your GCP project id from gcloud\n shell_output = !gcloud config list --format 'value(core.project)' 2>/dev/null\n PROJECT_ID = shell_output[0]\n \nif BUCKET == '' or BUCKET is None or BUCKET == '[your-bucket-name]':\n # Set your bucket name using your GCP project id\n BUCKET = PROJECT_ID\n # Try to create the bucket if it doesn'exists\n ! gsutil mb -l $REGION gs://$BUCKET\n print('')\n \nprint('Project ID:', PROJECT_ID)\nprint('Region:', REGION)\nprint('Bucket name:', BUCKET)",
"_____no_output_____"
]
],
[
[
"### Set configurations",
"_____no_output_____"
]
],
[
[
"VERSION = 'v1'\nDATASET_DISPLAY_NAME = 'chicago-taxi-tips'\nMODEL_DISPLAY_NAME = f'{DATASET_DISPLAY_NAME}-classifier-{VERSION}'\nENDPOINT_DISPLAY_NAME = f'{DATASET_DISPLAY_NAME}-classifier'\n\nSERVE_BQ_DATASET_NAME = 'playground_us' # Change to your serving BigQuery dataset name.\nSERVE_BQ_TABLE_NAME = 'chicago_taxitrips_prep' # Change to your serving BigQuery table name.",
"_____no_output_____"
]
],
[
[
"## 1. Making an online prediciton\n",
"_____no_output_____"
]
],
[
[
"vertex_ai.init(\n project=PROJECT_ID,\n location=REGION,\n staging_bucket=BUCKET\n)\n\nendpoint_name = vertex_ai.Endpoint.list(\n filter=f'display_name={ENDPOINT_DISPLAY_NAME}', \n order_by='update_time')[-1].gca_resource.name\n\nendpoint = vertex_ai.Endpoint(endpoint_name)",
"_____no_output_____"
],
[
"test_instances = [ \n {\n 'dropoff_grid': ['POINT(-87.6 41.9)'],\n 'euclidean': [2064.2696],\n 'loc_cross': [''],\n 'payment_type': ['Credit Card'],\n 'pickup_grid': ['POINT(-87.6 41.9)'],\n 'trip_miles': [1.37],\n 'trip_day': [12],\n 'trip_hour': [16],\n 'trip_month': [2],\n 'trip_day_of_week': [4],\n 'trip_seconds': [555]\n }\n]",
"_____no_output_____"
],
[
"predictions = endpoint.predict(test_instances).predictions\n\nfor prediction in predictions:\n print(prediction)",
"_____no_output_____"
],
[
"# TODO {for Khalid, get error saying model does not support explanations}\n\nexplanations = endpoint.explain(test_instances).explanations\n\nfor explanation in explanations:\n print(explanation)",
"_____no_output_____"
]
],
[
[
"## 2. Make a batch prediction",
"_____no_output_____"
]
],
[
[
"WORKSPACE = f'gs://{BUCKET}/{DATASET_DISPLAY_NAME}/'\nSERVING_DATA_DIR = os.path.join(WORKSPACE, 'serving_data')\nSERVING_INPUT_DATA_DIR = os.path.join(SERVING_DATA_DIR, 'input_data')\nSERVING_OUTPUT_DATA_DIR = os.path.join(SERVING_DATA_DIR, 'output_predictions')",
"_____no_output_____"
],
[
"if tf.io.gfile.exists(SERVING_DATA_DIR):\n print('Removing previous serving data...')\n tf.io.gfile.rmtree(SERVING_DATA_DIR)\nprint('Creating serving data directory...')\ntf.io.gfile.mkdir(SERVING_DATA_DIR)\nprint('Serving data directory is ready.')",
"_____no_output_____"
]
],
[
[
"### Extract serving data to Cloud Storage as JSONL",
"_____no_output_____"
]
],
[
[
"\nfrom src.model_training import features as feature_info\nfrom src.preprocessing import etl\nfrom src.common import datasource_utils",
"_____no_output_____"
],
[
"LIMIT = 10000\n\nsql_query = datasource_utils.create_bq_source_query(\n dataset_display_name=DATASET_DISPLAY_NAME, \n missing=feature_info.MISSING_VALUES,\n limit=LIMIT\n)\n\nprint(sql_query)",
"_____no_output_____"
],
[
"args = {\n #'runner': 'DataflowRunner',\n 'sql_query': sql_query,\n 'exported_data_prefix': os.path.join(SERVING_INPUT_DATA_DIR, 'data-'),\n 'temporary_dir': os.path.join(WORKSPACE, 'tmp'),\n 'gcs_location': os.path.join(WORKSPACE, 'bq_tmp'),\n 'project': PROJECT_ID,\n 'region': REGION,\n 'setup_file': './setup.py'\n}",
"_____no_output_____"
],
[
"tf.get_logger().setLevel('ERROR')\n\nprint('Data extraction started...')\netl.run_extract_pipeline(args)\nprint('Data extraction completed.')",
"_____no_output_____"
],
[
"! gsutil ls {SERVING_INPUT_DATA_DIR}",
"_____no_output_____"
]
],
[
[
"### Submit the batch prediction job",
"_____no_output_____"
]
],
[
[
"model_name = vertex_ai.Model.list(\n filter=f'display_name={MODEL_DISPLAY_NAME}',\n order_by='update_time')[-1].gca_resource.name",
"_____no_output_____"
],
[
"job_resources = {\n 'machine_type': 'n1-standard-2',\n #'accelerator_count': 1,\n #'accelerator_type': 'NVIDIA_TESLA_T4'\n 'starting_replica_count': 1,\n 'max_replica_coun': 10,\n}\n\njob_display_name = f'{MODEL_DISPLAY_NAME}-prediction-job-{datetime.now().strftime('%Y%m%d%H%M%S')}'\n\nvertex_ai.BatchPredictionJob.create(\n job_display_name=job_display_name,\n model_name=model_name,\n gcs_source=SERVING_INPUT_DATA_DIR + '/*.jsonl',\n gcs_destination_prefix=SERVING_OUTPUT_DATA_DIR,\n instances_format='jsonl',\n predictions_format='jsonl',\n sync=True,\n **job_resources,\n)",
"_____no_output_____"
]
],
[
[
"## 3. Run the batch prediction pipeline using `Vertex Pipelines`",
"_____no_output_____"
]
],
[
[
"WORKSPACE = f'{BUCKET}/{DATASET_DISPLAY_NAME}/'\nMLMD_SQLLITE = 'mlmd.sqllite'\nARTIFACT_STORE = os.path.join(WORKSPACE, 'tfx_artifacts')\nPIPELINE_NAME = f'{MODEL_DISPLAY_NAME}-predict-pipeline'",
"_____no_output_____"
],
[
"os.environ['PROJECT'] = PROJECT_ID\nos.environ['REGION'] = REGION\nos.environ['MODEL_DISPLAY_NAME'] = MODEL_DISPLAY_NAME\nos.environ['PIPELINE_NAME'] = PIPELINE_NAME\nos.environ['ARTIFACT_STORE_URI'] = ARTIFACT_STORE\nos.environ['BATCH_PREDICTION_BQ_DATASET_NAME'] = SERVE_BQ_DATASET_NAME\nos.environ['BATCH_PREDICTION_BQ_TABLE_NAME'] = SERVE_BQ_TABLE_NAME\nos.environ['SERVE_LIMIT'] = '1000'\nos.environ['BEAM_RUNNER'] = 'DirectRunner'\nos.environ['TFX_IMAGE_URI'] = f'gcr.io/{PROJECT_ID}/{DATASET_DISPLAY_NAME}:{VERSION}'",
"_____no_output_____"
],
[
"import importlib\nfrom src.tfx_pipelines import config\nimportlib.reload(config)\n\nfor key, value in config.__dict__.items():\n if key.isupper(): print(f'{key}: {value}')",
"_____no_output_____"
],
[
"from src.tfx_pipelines import runner\n\npipeline_definition_file = f'{config.PIPELINE_NAME}.json'\npipeline_definition = runner.compile_prediction_pipeline(pipeline_definition_file)",
"_____no_output_____"
],
[
"from kfp.v2.google.client import AIPlatformClient\n\npipeline_client = AIPlatformClient(\n project_id=PROJECT_ID, region=REGION)\n \npipeline_client.create_run_from_job_spec(\n job_spec_path=pipeline_definition_file\n)",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code"
]
] |
cbdbea2da8955dd7ac2aa9460a565c8dd15bf0d7
| 561,911 |
ipynb
|
Jupyter Notebook
|
Wildfire Damage Assessment (binary classification)-SantaRosa.ipynb
|
aalten77/wildfireassessment
|
0d4f1639b443350b50068e154e845f1abafcb49f
|
[
"MIT"
] | null | null | null |
Wildfire Damage Assessment (binary classification)-SantaRosa.ipynb
|
aalten77/wildfireassessment
|
0d4f1639b443350b50068e154e845f1abafcb49f
|
[
"MIT"
] | null | null | null |
Wildfire Damage Assessment (binary classification)-SantaRosa.ipynb
|
aalten77/wildfireassessment
|
0d4f1639b443350b50068e154e845f1abafcb49f
|
[
"MIT"
] | null | null | null | 99.541364 | 84,612 | 0.45297 |
[
[
[
"# Wildfire Damage Assessment (binary classification)",
"_____no_output_____"
]
],
[
[
"#First rescale data to minmax and convert to uint8\n#!gdal_translate -scale 0 13016 0 255 -ot Byte data/SantaRosa/Sent2/post/B08_clip_post.tif data/SantaRosa/Sent2/post/B08_clip_scale_post.tif",
"Input file size is 876, 871\r\n0...10...20...30...40...50...60...70...80...90...100 - done.\r\n"
],
[
"#!gdal_translate -scale 0 13016 0 255 -ot Byte data/SantaRosa/Sent2/pre/B08_clip_pre.tif data/SantaRosa/Sent2/pre/B08_clip_scale_pre.tif",
"Input file size is 876, 871\n0...10...20...30...40...50...60...70...80...90...100 - done.\n"
],
[
"%load_ext autoreload",
"_____no_output_____"
],
[
"from wildfireassessment.ops import * #my package\nfrom wildfireassessment.plots import * #my package\n#from wildfireassessment.test import *\nimport numpy as np\nimport numpy.ma as ma\nimport matplotlib.pyplot as plt\nfrom pathlib import Path\nfrom skimage import morphology\nfrom skimage.transform import resize\nimport pandas as pd\nimport geopandas as gpd\nimport pickle\nfrom sklearn.impute import SimpleImputer\nfrom sklearn.ensemble import RandomForestClassifier\nfrom sklearn.linear_model import LogisticRegression\nfrom sklearn.naive_bayes import GaussianNB, BernoulliNB\nfrom sklearn import linear_model\nfrom sklearn.neighbors import KNeighborsClassifier\nfrom sklearn.tree import DecisionTreeClassifier\nfrom sklearn import svm\nfrom sklearn.neural_network import MLPClassifier\nfrom sklearn.metrics import confusion_matrix, recall_score, precision_score, accuracy_score, f1_score\nfrom sklearn.externals import joblib\nfrom rasterstats import zonal_stats\nimport fiona\n%matplotlib inline",
"/home/ec2-user/anaconda3/lib/python3.7/site-packages/sklearn/externals/joblib/__init__.py:15: DeprecationWarning: sklearn.externals.joblib is deprecated in 0.21 and will be removed in 0.23. Please import this functionality directly from joblib, which can be installed with: pip install joblib. If this warning is raised when loading pickled models, you may need to re-serialize those models with scikit-learn 0.21+.\n warnings.warn(msg, category=DeprecationWarning)\n"
],
[
"#read in filepaths for data\n\n#filepath_post = \n#filepath_pre = Path(\"./data/SantaRosa/WorldView/pre\")\n\n#WorldView Post/Pre\nfps_wv_post = Path(\"./data/SantaRosa/WorldView/post/0221213.tif\")\nfps_wv_pre = Path(\"./data/SantaRosa/WorldView/pre/0221213_2.tif\")\n\n#Sent2 Post/Pre\nfps_sent2_post = Path(\"./data/SantaRosa/Sent2/post/B08_clip_scale_post.tif\")\nfps_sent2_pre = Path(\"./data/SantaRosa/Sent2/pre/B08_clip_scale_pre.tif\")",
"_____no_output_____"
],
[
"raster_src_post, rgb_post = readRGBImg(fps_wv_post)\nraster_src_pre, rgb_pre = readRGBImg(fps_wv_pre)",
"_____no_output_____"
],
[
"raster_src_post_b08, b08_post = readOneImg(fps_sent2_post)\nraster_src_pre_b08, b08_pre = readOneImg(fps_sent2_pre)",
"_____no_output_____"
],
[
"b08_upscaled_post = resize(b08_post, raster_src_post.shape, anti_aliasing=True)\nb08_upscaled_post = b08_upscaled_post * 255\nb08_upscaled_post = b08_upscaled_post.astype(rasterio.uint8)",
"_____no_output_____"
],
[
"b08_upscaled_pre = resize(b08_pre, raster_src_pre.shape, anti_aliasing=True)\nb08_upscaled_pre = b08_upscaled_pre * 255\nb08_upscaled_pre = b08_upscaled_pre.astype(rasterio.uint8)",
"_____no_output_____"
],
[
"plt.imshow(b08_upscaled_post, cmap='gray')",
"_____no_output_____"
]
],
[
[
"## Extract pixels",
"_____no_output_____"
],
[
"#### unravel",
"_____no_output_____"
]
],
[
[
"rgb_rav_post = {0 : rgb_post[:,:,0].ravel().astype(float),\n 1 : rgb_post[:,:,1].ravel().astype(float),\n 2 : rgb_post[:,:,2].ravel().astype(float)}\nrgb_rav_pre = {0 : rgb_pre[:,:,0].ravel().astype(float),\n 1 : rgb_pre[:,:,1].ravel().astype(float),\n 2 : rgb_pre[:,:,2].ravel().astype(float)}",
"_____no_output_____"
],
[
"b08_rav_post = b08_upscaled_post.ravel().astype(float)\nb08_rav_pre = b08_upscaled_pre.ravel().astype(float)",
"_____no_output_____"
],
[
"b08_rav_pre.shape[0]",
"_____no_output_____"
],
[
"b08_upscaled_post = None\nb08_upscaled_pre = None\nb08_post = None\nb08_pre = None\nrgb_pre = None\nrgb_post = None",
"_____no_output_____"
],
[
"def computeSI(b1, b2):\n return (b1-b2)/(b1+b2)\n\ndef changedSI(SI_pre, SI_post):\n return SI_pre - SI_post",
"_____no_output_____"
],
[
"def makeChunkX(b, g, r, n, b_p, g_p, r_p, n_p):\n SI_gb = (computeSI(g, b), computeSI(g_p, b_p)) #(post, pre)\n SI_rb = (computeSI(r, b), computeSI(r_p, b_p))\n SI_rg = (computeSI(r, g), computeSI(r_p, g_p))\n SI_nb = (computeSI(n, b), computeSI(n_p, b_p))\n SI_ng = (computeSI(n, g), computeSI(n_p, g_p))\n SI_nr = (computeSI(n, r), computeSI(n_p, r_p))\n \n dSI_gb = changedSI(SI_gb[1], SI_gb[0])\n dSI_rb = changedSI(SI_rb[1], SI_rb[0])\n dSI_rg = changedSI(SI_rg[1], SI_rg[0])\n dSI_nb = changedSI(SI_nb[1], SI_nb[0])\n dSI_ng = changedSI(SI_ng[1], SI_ng[0])\n dSI_nr = changedSI(SI_nr[1], SI_nr[0])\n \n return np.dstack((b, b_p, g, g_p, r, r_p, n, n_p,\n SI_gb[0], SI_rb[0], SI_rg[0], SI_nb[0], SI_ng[0], SI_nr[0],\n SI_gb[1], SI_rb[1], SI_rg[1], SI_nb[1], SI_ng[1], SI_nr[1],\n dSI_nb, dSI_rg, dSI_rb, dSI_gb, dSI_nr, dSI_ng))[0]",
"_____no_output_____"
],
[
"X_chunk = makeChunkX(rgb_rav_post[2][0:100], rgb_rav_post[1][0:100], rgb_rav_post[0][0:100], b08_rav_post[0:100],\n rgb_rav_pre[2][0:100], rgb_rav_pre[1][0:100], rgb_rav_pre[0][0:100], b08_rav_pre[0:100])",
"_____no_output_____"
],
[
"m = ma.masked_invalid(X_chunk)\nma.set_fill_value(m, -999)\nm",
"_____no_output_____"
],
[
"imp = SimpleImputer(missing_values=np.nan, strategy='mean')\nimp.fit(X_chunk)\nX_chunk_imp = imp.transform(X_chunk)",
"_____no_output_____"
],
[
"rf_model = joblib.load(open(\"models/rf_grid_bin_precision.pkl\", 'rb'))",
"_____no_output_____"
],
[
"pred_y = rf_model.predict(X_chunk_imp)",
"_____no_output_____"
]
],
[
[
"### Test Random Forest",
"_____no_output_____"
]
],
[
[
"from joblib import Parallel, delayed\nimport multiprocessing\n\ndef processInParallel(i):\n X_chunk = makeChunkX(rgb_rav_post[2][i:i+100], rgb_rav_post[1][i:i+100], rgb_rav_post[0][i:i+100], b08_rav_post[i:i+100],\n rgb_rav_pre[2][i:i+100], rgb_rav_pre[1][i:i+100], rgb_rav_pre[0][i:i+100], b08_rav_pre[i:i+100])\n \n return rf_model.predict(ma.masked_invalid(X_chunk))",
"_____no_output_____"
],
[
"%%time\nnum_cores = multiprocessing.cpu_count()\npred_y = Parallel(n_jobs=num_cores, backend=\"multiprocessing\")(delayed(processInParallel)(i) for i in range(0, len(b08_rav_post), 100))\npred_y",
"CPU times: user 1min 48s, sys: 15.6 s, total: 2min 4s\nWall time: 11min 29s\n"
],
[
"pred_y_rf = np.hstack(pred_y).reshape(raster_src_post.shape)",
"_____no_output_____"
],
[
"plt.imshow(pred_y_rf)",
"_____no_output_____"
],
[
"pred_y_rf_clean = morphology.remove_small_holes(pred_y_rf==1, 500) \npred_y_rf_clean = morphology.remove_small_objects(pred_y_rf_clean, 500)\nplt.imshow(pred_y_rf_clean)",
"_____no_output_____"
],
[
"pred_y_rf_clean.astype(np.uint8)",
"_____no_output_____"
],
[
"raster_src_post.meta",
"_____no_output_____"
],
[
"metadata = {\n 'driver': 'GTiff',\n 'dtype': 'uint8',\n 'width': raster_src_post.meta['width'], \n 'height': raster_src_post.meta['height'],\n 'count': 1,\n 'crs': raster_src_post.meta['crs'], \n 'transform': raster_src_post.meta['transform']\n}\nmetadata",
"_____no_output_____"
],
[
"with rasterio.open(\"results/predict_mask_rf_SantaRosa_0221213.tif\", 'w', **metadata) as dst:\n dst.write(pred_y_rf_clean.astype(np.uint8), 1)",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
]
] |
cbdc076aa20c4ee977865072971b57c7dd28c954
| 101,710 |
ipynb
|
Jupyter Notebook
|
source/sample_ml/Chapter03/3-3 NMF.ipynb
|
alphajayGithub/ai.online
|
3e440d88111627827456aa8672516eb389a68e98
|
[
"MIT"
] | null | null | null |
source/sample_ml/Chapter03/3-3 NMF.ipynb
|
alphajayGithub/ai.online
|
3e440d88111627827456aa8672516eb389a68e98
|
[
"MIT"
] | null | null | null |
source/sample_ml/Chapter03/3-3 NMF.ipynb
|
alphajayGithub/ai.online
|
3e440d88111627827456aa8672516eb389a68e98
|
[
"MIT"
] | null | null | null | 420.289256 | 51,838 | 0.927421 |
[
[
[
"## NMF\n[2.5. Decomposing signals in components (matrix factorization problems) — scikit-learn 1.0.2 documentation](https://scikit-learn.org/stable/modules/decomposition.html?highlight=nmf#non-negative-matrix-factorization-nmf-or-nnmf)\n",
"_____no_output_____"
]
],
[
[
"from sklearn.decomposition import NMF\nfrom sklearn.datasets import make_blobs\nimport numpy as np\n\ncenters = [[5, 10, 5], [10, 4, 10], [6, 8, 8]]\nX, _ = make_blobs(centers=centers) # 以centers为中心生成数据\nn_components = 2 # 潜在变量的个数\nmodel = NMF(n_components=n_components)\nmodel.fit(X)\nW = model.transform(X) # 分解后的矩阵\nH = model.components_\n\nprint(X.shape)\nprint(W.shape)\nprint(H.shape)\nprint(H)\n#print(W)\n\nV = np.dot(W,H)\n\nfor i in range(10):\n print('V - ', V[i:(i+1),:])\n print('X - ', X[i:(i+1),:])\n\nprint('reconstruction_err_', model.reconstruction_err_) # 损失函数值\nprint('n_iter_', model.n_iter_) # 实际迭代次数\n\n\n",
"(100, 3)\n(100, 2)\n(2, 3)\n[[7.2218991 1.9664158 7.5930157 ]\n [0. 8.54594098 0.67940793]]\nV - [[ 4.02444789 10.53787994 4.98190669]]\nX - [[ 3.97607259 10.53414786 5.02885065]]\nV - [[ 4.28984302 10.31787803 5.23770459]]\nX - [[ 3.64857704 10.26837202 5.86041649]]\nV - [[11.08805509 5.50690252 11.85562539]]\nX - [[11.41653497 5.53226337 11.53662362]]\nV - [[10.01053152 3.5896554 10.59363297]]\nX - [[ 9.77021692 3.57110228 10.82700371]]\nV - [[5.10985528 9.74851337 6.03683982]]\nX - [[6.7056925 9.87172165 4.48706309]]\nV - [[ 3.38946804 10.00635625 4.28578553]]\nX - [[ 3.40076885 10.00723133 4.27477823]]\nV - [[4.35793099 9.89882675 5.2745027 ]]\nX - [[4.97472468 9.94644853 4.67549163]]\nV - [[8.87580932 5.8164594 9.60219557]]\nX - [[ 7.72009432 5.72723395 10.72451907]]\nV - [[5.99988805 7.38242583 6.76523739]]\nX - [[5.61328876 7.35258027 7.14065017]]\nV - [[9.31123449 4.82752722 9.97194989]]\nX - [[ 9.27993843 4.82511168 10.00233377]]\nreconstruction_err_ 10.32946029062523\nn_iter_ 200\n"
]
],
[
[
"## olivetti_faces MNF",
"_____no_output_____"
]
],
[
[
"from time import time\nfrom numpy.random import RandomState\nimport matplotlib.pyplot as plt\n#from sklearn.datasets import fetch_olivetti_faces\nfrom sklearn import decomposition\nimport scipy.io as spio\n\nn_row, n_col = 2, 3\nn_components = n_row * n_col\nimage_shape = (64, 64)\nrng = RandomState(0)\n\n# #############################################################################\n# Load faces data\n# dataset = fetch_olivetti_faces('./', True,random_state=rng)\n\ndatafile = '../resource/data/olivettifaces/olivettifaces.mat'\ndataset = spio.loadmat(datafile)\n# print(dataset.keys())\n# dict_keys(['__header__', '__version__', '__globals__', 'faces', 'p', 'u', 'v'])\nfaces = np.transpose(dataset['faces'])\nprint(dataset['faces'].shape)\n\nn_samples,n_features= faces.shape\n\nprint(\"Dataset consists of %d faces, features is %s\" % (n_samples, n_features))\n\ndef plot_gallery(title, images, n_col=n_col, n_row=n_row, cmap=plt.cm.gray):\n plt.figure(figsize=(2. * n_col, 2.26 * n_row))\n plt.suptitle(title, size=16)\n for i, comp in enumerate(images):\n plt.subplot(n_row, n_col, i + 1)\n vmax = max(comp.max(), -comp.min())\n plt.imshow(comp.reshape(image_shape), cmap=cmap,\n interpolation='nearest',\n vmin=-vmax, vmax=vmax)\n plt.xticks(())\n plt.yticks(())\n plt.subplots_adjust(0.01, 0.05, 0.99, 0.93, 0.04, 0.)\n\n\n# #############################################################################\nestimators = [\n ('Non-negative components - NMF',\n decomposition.NMF(n_components=n_components, init='nndsvda', tol=5e-3))\n]\n\n# #############################################################################\n# Plot a sample of the input data\n\nplot_gallery(\"First centered Olivetti faces\", faces[:n_components])\n\n# #############################################################################\n# Do the estimation and plot it\n\nfor name, estimator in estimators:\n print(\"Extracting the top %d %s...\" % (n_components, name))\n t0 = time()\n data = faces\n estimator.fit(data)\n train_time = (time() - t0)\n print(\"done in %0.3fs\" % train_time)\n\n components_ = estimator.components_\n print('components_:', components_.shape, '\\n**\\n', components_)\n plot_gallery('%s - Train time %.1fs' % (name, train_time),\n components_)\nplt.show()",
"(4096, 400)\nDataset consists of 400 faces, features is 4096\nExtracting the top 6 Non-negative components - NMF...\ndone in 0.133s\ncomponents_: (6, 4096) \n**\n [[ 0. 0. 0. ... 8.47930895 9.17286304\n 9.50178305]\n [307.44670826 331.3063659 352.37269434 ... 0. 0.\n 0. ]\n [ 0. 0. 0. ... 49.03347129 48.58511386\n 53.3622545 ]\n [ 0. 0. 0. ... 490.53895214 511.87157392\n 488.21654134]\n [ 31.76317671 29.984939 27.59062498 ... 98.52633072 97.14706167\n 92.77534811]\n [ 86.48513011 87.3752808 88.43348544 ... 61.76492362 62.71729237\n 62.94962925]]\n"
]
]
] |
[
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
]
] |
cbdc0dc1f320977c6b24ef9fb6e5a5585a455747
| 164,218 |
ipynb
|
Jupyter Notebook
|
notebooks/semisupervised/cifar10/learned-metric/old/augmented-old/cifar10-aug-1024ex-learned-0.8conf.ipynb
|
timsainb/ParametricUMAP_paper
|
00b4d676647e45619552aec8f2663c0903a83e3f
|
[
"MIT"
] | 124 |
2020-09-27T23:59:01.000Z
|
2022-03-22T06:27:35.000Z
|
notebooks/semisupervised/cifar10/learned-metric/old/augmented-old/cifar10-aug-1024ex-learned-0.8conf.ipynb
|
kiminh/ParametricUMAP_paper
|
00b4d676647e45619552aec8f2663c0903a83e3f
|
[
"MIT"
] | 2 |
2021-02-05T18:13:13.000Z
|
2021-11-01T14:55:08.000Z
|
notebooks/semisupervised/cifar10/learned-metric/old/augmented-old/cifar10-aug-1024ex-learned-0.8conf.ipynb
|
kiminh/ParametricUMAP_paper
|
00b4d676647e45619552aec8f2663c0903a83e3f
|
[
"MIT"
] | 16 |
2020-09-28T07:43:21.000Z
|
2022-03-21T00:31:34.000Z
| 218.665779 | 123,556 | 0.894269 |
[
[
[
"# reload packages\n%load_ext autoreload\n%autoreload 2",
"_____no_output_____"
]
],
[
[
"### Choose GPU",
"_____no_output_____"
]
],
[
[
"%env CUDA_DEVICE_ORDER=PCI_BUS_ID\n%env CUDA_VISIBLE_DEVICES=0",
"env: CUDA_DEVICE_ORDER=PCI_BUS_ID\nenv: CUDA_VISIBLE_DEVICES=0\n"
],
[
"import tensorflow as tf\ngpu_devices = tf.config.experimental.list_physical_devices('GPU')\nif len(gpu_devices)>0:\n tf.config.experimental.set_memory_growth(gpu_devices[0], True)\nprint(gpu_devices)\ntf.keras.backend.clear_session()",
"[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]\n"
]
],
[
[
"### Load packages",
"_____no_output_____"
]
],
[
[
"import tensorflow as tf\nimport numpy as np\nimport matplotlib.pyplot as plt\nfrom tqdm.autonotebook import tqdm\nfrom IPython import display\nimport pandas as pd\nimport umap\nimport copy\nimport os, tempfile\nimport tensorflow_addons as tfa",
"/mnt/cube/tsainbur/conda_envs/tpy3/lib/python3.6/site-packages/tqdm/autonotebook/__init__.py:14: TqdmExperimentalWarning: Using `tqdm.autonotebook.tqdm` in notebook mode. Use `tqdm.tqdm` instead to force console mode (e.g. in jupyter console)\n \" (e.g. in jupyter console)\", TqdmExperimentalWarning)\n"
]
],
[
[
"### parameters",
"_____no_output_____"
]
],
[
[
"dataset = \"cifar10\"\nlabels_per_class = 1024 # 'full'\nn_latent_dims = 1024\nconfidence_threshold = 0.8 # minimum confidence to include in UMAP graph for learned metric\nlearned_metric = True # whether to use a learned metric, or Euclidean distance between datapoints\n\naugmented = True #\nmin_dist= 0.001 # min_dist parameter for UMAP\nnegative_sample_rate = 5 # how many negative samples per positive sample \nbatch_size = 128 # batch size \noptimizer = tf.keras.optimizers.Adam(1e-3) # the optimizer to train\noptimizer = tfa.optimizers.MovingAverage(optimizer)\nlabel_smoothing = 0.2 # how much label smoothing to apply to categorical crossentropy\nmax_umap_iterations = 50 # how many times, maximum, to recompute UMAP\nmax_epochs_per_graph = 50 # how many epochs maximum each graph trains for (without early stopping)\numap_patience = 5 # how long before recomputing UMAP graph",
"_____no_output_____"
]
],
[
[
"#### Load dataset",
"_____no_output_____"
]
],
[
[
"from tfumap.semisupervised_keras import load_dataset",
"_____no_output_____"
],
[
"(\n X_train,\n X_test,\n X_labeled,\n Y_labeled,\n Y_masked,\n X_valid,\n Y_train,\n Y_test,\n Y_valid,\n Y_valid_one_hot,\n Y_labeled_one_hot,\n num_classes,\n dims\n) = load_dataset(dataset, labels_per_class)",
"_____no_output_____"
]
],
[
[
"### load architecture",
"_____no_output_____"
]
],
[
[
"from tfumap.semisupervised_keras import load_architecture",
"_____no_output_____"
],
[
"encoder, classifier, embedder = load_architecture(dataset, n_latent_dims)",
"_____no_output_____"
]
],
[
[
"### load pretrained weights",
"_____no_output_____"
]
],
[
[
"from tfumap.semisupervised_keras import load_pretrained_weights",
"_____no_output_____"
],
[
"encoder, classifier = load_pretrained_weights(dataset, augmented, labels_per_class, encoder, classifier)",
"WARNING: Logging before flag parsing goes to stderr.\nW0812 09:59:18.808132 140536619722496 base.py:272] Inconsistent references when loading the checkpoint into this object graph. Either the Trackable object references in the Python program have changed in an incompatible way, or the checkpoint was generated in an incompatible program.\n\nTwo checkpoint references resolved to different objects (<tensorflow_addons.layers.wrappers.WeightNormalization object at 0x7fd037b0b4a8> and <tensorflow.python.keras.layers.advanced_activations.LeakyReLU object at 0x7fd037b0b940>).\nW0812 09:59:18.810333 140536619722496 base.py:272] Inconsistent references when loading the checkpoint into this object graph. Either the Trackable object references in the Python program have changed in an incompatible way, or the checkpoint was generated in an incompatible program.\n\nTwo checkpoint references resolved to different objects (<tensorflow_addons.layers.wrappers.WeightNormalization object at 0x7fd037af09b0> and <tensorflow.python.keras.layers.advanced_activations.LeakyReLU object at 0x7fd0379bee80>).\nW0812 09:59:18.837200 140536619722496 base.py:272] Inconsistent references when loading the checkpoint into this object graph. Either the Trackable object references in the Python program have changed in an incompatible way, or the checkpoint was generated in an incompatible program.\n\nTwo checkpoint references resolved to different objects (<tensorflow_addons.layers.wrappers.WeightNormalization object at 0x7fd0385ac898> and <tensorflow.python.keras.layers.normalization_v2.BatchNormalization object at 0x7fd0385aca20>).\nW0812 09:59:18.842061 140536619722496 base.py:272] Inconsistent references when loading the checkpoint into this object graph. Either the Trackable object references in the Python program have changed in an incompatible way, or the checkpoint was generated in an incompatible program.\n\nTwo checkpoint references resolved to different objects (<tensorflow.python.keras.layers.normalization_v2.BatchNormalization object at 0x7fd0385aca20> and <tensorflow.python.keras.layers.advanced_activations.LeakyReLU object at 0x7fd038422a20>).\nW0812 09:59:18.848726 140536619722496 base.py:272] Inconsistent references when loading the checkpoint into this object graph. Either the Trackable object references in the Python program have changed in an incompatible way, or the checkpoint was generated in an incompatible program.\n\nTwo checkpoint references resolved to different objects (<tensorflow_addons.layers.wrappers.WeightNormalization object at 0x7fd03844a198> and <tensorflow.python.keras.layers.normalization_v2.BatchNormalization object at 0x7fd03844a4a8>).\nW0812 09:59:18.853579 140536619722496 base.py:272] Inconsistent references when loading the checkpoint into this object graph. Either the Trackable object references in the Python program have changed in an incompatible way, or the checkpoint was generated in an incompatible program.\n\nTwo checkpoint references resolved to different objects (<tensorflow.python.keras.layers.normalization_v2.BatchNormalization object at 0x7fd03844a4a8> and <tensorflow.python.keras.layers.advanced_activations.LeakyReLU object at 0x7fd03844add8>).\nW0812 09:59:18.860227 140536619722496 base.py:272] Inconsistent references when loading the checkpoint into this object graph. Either the Trackable object references in the Python program have changed in an incompatible way, or the checkpoint was generated in an incompatible program.\n\nTwo checkpoint references resolved to different objects (<tensorflow_addons.layers.wrappers.WeightNormalization object at 0x7fd037d725f8> and <tensorflow.python.keras.layers.normalization_v2.BatchNormalization object at 0x7fd037d72908>).\nW0812 09:59:18.864767 140536619722496 base.py:272] Inconsistent references when loading the checkpoint into this object graph. Either the Trackable object references in the Python program have changed in an incompatible way, or the checkpoint was generated in an incompatible program.\n\nTwo checkpoint references resolved to different objects (<tensorflow.python.keras.layers.normalization_v2.BatchNormalization object at 0x7fd037d72908> and <tensorflow.python.keras.layers.advanced_activations.LeakyReLU object at 0x7fd037d72a58>).\nW0812 09:59:18.876319 140536619722496 base.py:272] Inconsistent references when loading the checkpoint into this object graph. Either the Trackable object references in the Python program have changed in an incompatible way, or the checkpoint was generated in an incompatible program.\n\nTwo checkpoint references resolved to different objects (<tensorflow_addons.layers.wrappers.WeightNormalization object at 0x7fd038424d68> and <tensorflow.python.keras.layers.normalization_v2.BatchNormalization object at 0x7fd038429390>).\nW0812 09:59:18.880830 140536619722496 base.py:272] Inconsistent references when loading the checkpoint into this object graph. Either the Trackable object references in the Python program have changed in an incompatible way, or the checkpoint was generated in an incompatible program.\n\nTwo checkpoint references resolved to different objects (<tensorflow.python.keras.layers.normalization_v2.BatchNormalization object at 0x7fd038429390> and <tensorflow.python.keras.layers.advanced_activations.LeakyReLU object at 0x7fd0384295f8>).\nW0812 09:59:18.887790 140536619722496 base.py:272] Inconsistent references when loading the checkpoint into this object graph. Either the Trackable object references in the Python program have changed in an incompatible way, or the checkpoint was generated in an incompatible program.\n\nTwo checkpoint references resolved to different objects (<tensorflow_addons.layers.wrappers.WeightNormalization object at 0x7fd03840e1d0> and <tensorflow.python.keras.layers.normalization_v2.BatchNormalization object at 0x7fd03840e400>).\nW0812 09:59:18.892295 140536619722496 base.py:272] Inconsistent references when loading the checkpoint into this object graph. Either the Trackable object references in the Python program have changed in an incompatible way, or the checkpoint was generated in an incompatible program.\n\nTwo checkpoint references resolved to different objects (<tensorflow.python.keras.layers.normalization_v2.BatchNormalization object at 0x7fd03840e400> and <tensorflow.python.keras.layers.advanced_activations.LeakyReLU object at 0x7fd03840e668>).\nW0812 09:59:18.899184 140536619722496 base.py:272] Inconsistent references when loading the checkpoint into this object graph. Either the Trackable object references in the Python program have changed in an incompatible way, or the checkpoint was generated in an incompatible program.\n\nTwo checkpoint references resolved to different objects (<tensorflow_addons.layers.wrappers.WeightNormalization object at 0x7fd037ca6128> and <tensorflow.python.keras.layers.normalization_v2.BatchNormalization object at 0x7fd037ca6438>).\nW0812 09:59:18.903781 140536619722496 base.py:272] Inconsistent references when loading the checkpoint into this object graph. Either the Trackable object references in the Python program have changed in an incompatible way, or the checkpoint was generated in an incompatible program.\n\nTwo checkpoint references resolved to different objects (<tensorflow.python.keras.layers.normalization_v2.BatchNormalization object at 0x7fd037ca6438> and <tensorflow.python.keras.layers.advanced_activations.LeakyReLU object at 0x7fd037ca66a0>).\nW0812 09:59:18.914460 140536619722496 base.py:272] Inconsistent references when loading the checkpoint into this object graph. Either the Trackable object references in the Python program have changed in an incompatible way, or the checkpoint was generated in an incompatible program.\n\nTwo checkpoint references resolved to different objects (<tensorflow_addons.layers.wrappers.WeightNormalization object at 0x7fd03855a828> and <tensorflow.python.keras.layers.normalization_v2.BatchNormalization object at 0x7fd03855ad68>).\nW0812 09:59:18.919230 140536619722496 base.py:272] Inconsistent references when loading the checkpoint into this object graph. Either the Trackable object references in the Python program have changed in an incompatible way, or the checkpoint was generated in an incompatible program.\n\nTwo checkpoint references resolved to different objects (<tensorflow.python.keras.layers.normalization_v2.BatchNormalization object at 0x7fd03855ad68> and <tensorflow.python.keras.layers.advanced_activations.LeakyReLU object at 0x7fd038540128>).\n"
]
],
[
[
"#### compute pretrained accuracy",
"_____no_output_____"
]
],
[
[
"# test current acc\npretrained_predictions = classifier.predict(encoder.predict(X_test, verbose=True), verbose=True)\npretrained_predictions = np.argmax(pretrained_predictions, axis=1)\npretrained_acc = np.mean(pretrained_predictions == Y_test)\nprint('pretrained acc: {}'.format(pretrained_acc))",
"\r 1/313 [..............................] - ETA: 0s"
]
],
[
[
"### get a, b parameters for embeddings",
"_____no_output_____"
]
],
[
[
"from tfumap.semisupervised_keras import find_a_b",
"_____no_output_____"
],
[
"a_param, b_param = find_a_b(min_dist=min_dist)",
"_____no_output_____"
]
],
[
[
"### build network",
"_____no_output_____"
]
],
[
[
"from tfumap.semisupervised_keras import build_model",
"_____no_output_____"
],
[
"model = build_model(\n batch_size=batch_size,\n a_param=a_param,\n b_param=b_param,\n dims=dims,\n encoder=encoder,\n classifier=classifier,\n negative_sample_rate=negative_sample_rate,\n optimizer=optimizer,\n label_smoothing=label_smoothing,\n embedder = None,\n)",
"_____no_output_____"
]
],
[
[
"### build labeled iterator",
"_____no_output_____"
]
],
[
[
"from tfumap.semisupervised_keras import build_labeled_iterator",
"_____no_output_____"
],
[
"labeled_dataset = build_labeled_iterator(X_labeled, Y_labeled_one_hot, augmented, dims)",
"_____no_output_____"
]
],
[
[
"### training",
"_____no_output_____"
]
],
[
[
"from livelossplot import PlotLossesKerasTF\nfrom tfumap.semisupervised_keras import get_edge_dataset\nfrom tfumap.semisupervised_keras import zip_datasets",
"_____no_output_____"
]
],
[
[
"#### callbacks",
"_____no_output_____"
]
],
[
[
"# early stopping callback\nearly_stopping = tf.keras.callbacks.EarlyStopping(\n monitor='val_classifier_acc', min_delta=0, patience=15, verbose=0, mode='auto',\n baseline=None, restore_best_weights=False\n)\n\n# plot losses callback\ngroups = {'acccuracy': ['classifier_accuracy', 'val_classifier_accuracy'], 'loss': ['classifier_loss', 'val_classifier_loss']}\nplotlosses = PlotLossesKerasTF(groups=groups)",
"_____no_output_____"
],
[
"history_list = []\ncurrent_validation_acc = 0\nbatches_per_epoch = np.floor(len(X_train)/batch_size).astype(int)\nepochs_since_last_improvement = 0\nfor current_umap_iterations in tqdm(np.arange(max_umap_iterations)):\n \n # make dataset\n edge_dataset = get_edge_dataset(\n model,\n classifier,\n encoder,\n X_train,\n Y_masked,\n batch_size,\n confidence_threshold,\n labeled_dataset,\n dims,\n learned_metric = learned_metric\n )\n \n # zip dataset\n zipped_ds = zip_datasets(labeled_dataset, edge_dataset, batch_size)\n \n # train dataset\n history = model.fit(\n zipped_ds,\n epochs=max_epochs_per_graph,\n validation_data=(\n (X_valid, tf.zeros_like(X_valid), tf.zeros_like(X_valid)),\n {\"classifier\": Y_valid_one_hot},\n ),\n callbacks = [early_stopping, plotlosses],\n max_queue_size = 100,\n steps_per_epoch = batches_per_epoch,\n #verbose=0\n )\n history_list.append(history)\n # get validation acc\n pred_valid = classifier.predict(encoder.predict(X_valid))\n new_validation_acc = np.mean(np.argmax(pred_valid, axis = 1) == Y_valid)\n \n # if validation accuracy has gone up, mark the improvement\n if new_validation_acc > current_validation_acc:\n epochs_since_last_improvement = 0\n current_validation_acc = copy.deepcopy(new_validation_acc)\n else:\n epochs_since_last_improvement += 1\n \n if epochs_since_last_improvement > umap_patience:\n print('No improvement in {} UMAP iterators'.format(umap_patience))\n break",
"_____no_output_____"
],
[
"# umap loss 0.273",
"_____no_output_____"
],
[
"class_pred = classifier.predict(encoder.predict(X_test))\nclass_acc = np.mean(np.argmax(class_pred, axis=1) == Y_test)\nprint(class_acc)",
"0.8795\n"
]
]
] |
[
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code"
]
] |
cbdc18387ad27e7a546e972d3ed33106e7acacab
| 3,704 |
ipynb
|
Jupyter Notebook
|
Create_dfs.ipynb
|
guimarais/dd_df
|
01f9b0a078cf55be79a1652317e7bbcfbb9a0ee3
|
[
"MIT"
] | null | null | null |
Create_dfs.ipynb
|
guimarais/dd_df
|
01f9b0a078cf55be79a1652317e7bbcfbb9a0ee3
|
[
"MIT"
] | null | null | null |
Create_dfs.ipynb
|
guimarais/dd_df
|
01f9b0a078cf55be79a1652317e7bbcfbb9a0ee3
|
[
"MIT"
] | null | null | null | 28.060606 | 136 | 0.567765 |
[
[
[
"from dd_df import create_dd_df\nimport pandas as pd\nimport matplotlib.pyplot as plt\nimport numpy as np\n%load_ext autoreload\n%autoreload 2\n#\nplt.style.use('helvet2')",
"The autoreload extension is already loaded. To reload it, use:\n %reload_ext autoreload\n"
]
],
[
[
"#### Creates and saves the dataframes",
"_____no_output_____"
]
],
[
[
"df_35156 = create_dd_df(shotnr=35156, tBegin=0.5, tEnd=6.5, dt=0.02, dne_data=['FVL', 'Ne', 18, 'AUGD', 21])\ndf_35157 = create_dd_df(shotnr=35157, tBegin=0.5, tEnd=6.5, dt=0.02, dne_data=['FVL', 'Ne', 18, 'AUGD', 21])\ndf_35158 = create_dd_df(shotnr=35158, tBegin=0.5, tEnd=6.5, dt=0.02, dne_data=['FVL', 'Ne', 18, 'AUGD', 21])\ndf_35165 = create_dd_df(shotnr=35165, tBegin=0.5, tEnd=5.7, dt=0.02, dne_data=['FVL', 'Ne', 18, 'AUGD', 21])\ndf_35156.to_csv('./Support_Files/df_35156.csv')\ndf_35157.to_csv('./Support_Files/df_35157.csv')\ndf_35158.to_csv('./Support_Files/df_35158.csv')\ndf_35165.to_csv('./Support_Files/df_35165.csv')",
"/afs/ipp/aug/ads-diags/common/python/lib/dd_20180216.py:117: RuntimeWarning: Warning by DDcsgnl (8.1): no calibration of data\n warnings.warn(text.replace('\\x00','').strip(), RuntimeWarning)\n/afs/ipp/aug/ads-diags/common/python/lib/dd_20180216.py:117: RuntimeWarning: Warning by DDcsgrp (8.1): No PARAM_SET found\n warnings.warn(text.replace('\\x00','').strip(), RuntimeWarning)\n"
],
[
"df_35167 = create_dd_df(shotnr=35167, tBegin=0.5, tEnd=6.5, dt=0.02, dne_data=['FVL', 'Ne', 18, 'AUGD', 21])\ndf_35167.to_csv('./Support_Files/df_35167.csv')",
"_____no_output_____"
],
[
"df_35356 = create_dd_df(shotnr=35356, tBegin=0.5, tEnd=6.0, dt=0.02, dne_data=['FVL', 'Ne', 18, 'AUGD', 21])\ndf_35356.to_csv('./Support_Files/df_35356.csv')",
"_____no_output_____"
],
[
"df_35358 = create_dd_df(shotnr=35358, tBegin=0.5, tEnd=6.5, dt=0.02, dne_data=['FVL', 'Ne', 18, 'AUGD', 21])\ndf_35358.to_csv('./Support_Files/df_35358.csv')",
"_____no_output_____"
],
[
"df_35381 = create_dd_df(shotnr=35381, tBegin=0.5, tEnd=5.0, dt=0.02, dne_data=['FVL', 'Ne', 18, 'AUGD', 21])\ndf_35381.to_csv('./Support_Files/df_35381.csv')",
"_____no_output_____"
]
]
] |
[
"code",
"markdown",
"code"
] |
[
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code"
]
] |
cbdc1f294bf22aa822dcd4e5ad094af39ce7c44c
| 135,897 |
ipynb
|
Jupyter Notebook
|
Experiment_2.ipynb
|
anhinga/synapses
|
c60e2ebefce5357009a795e29f6cf69f378f8576
|
[
"MIT"
] | 2 |
2019-04-01T13:47:57.000Z
|
2019-05-27T15:34:32.000Z
|
Experiment_2.ipynb
|
anhinga/synapses
|
c60e2ebefce5357009a795e29f6cf69f378f8576
|
[
"MIT"
] | null | null | null |
Experiment_2.ipynb
|
anhinga/synapses
|
c60e2ebefce5357009a795e29f6cf69f378f8576
|
[
"MIT"
] | 1 |
2019-03-18T18:12:14.000Z
|
2019-03-18T18:12:14.000Z
| 210.366873 | 15,428 | 0.905178 |
[
[
[
"# MNIST With SET\nThis is an example of training an SET network on the MNIST dataset using synapses, pytorch, and torchvision.",
"_____no_output_____"
]
],
[
[
"#Import torch libraries and get SETLayer from synapses\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nimport torch.optim as optim\nfrom torchvision import datasets, transforms\n\nfrom synapses import SETLayer\n\n#Some extras for visualizations\nimport numpy as np\nimport matplotlib.pyplot as plt\nimport seaborn as sns\nfrom IPython.display import clear_output\nprint(\"done\")",
"done\n"
]
],
[
[
"## SET Layer\nThe SET layer is a pytorch module that works with a similar API to a standard fully connected layer; to initialize, specify input and output dimensions.<br><br>\nNOTE: one condition mentioned in the paper is that epsilon (a hyperparameter controlling layer sparsity) be much less than the input dimension and much less than the output dimension. The default value of epsilon is 11. Keep dimensions much bigger than epsilon! (epsilon can be passed in as an init argument to the layer).",
"_____no_output_____"
]
],
[
[
"#initialize the layer\nsprs = SETLayer(128, 256)\n\n#We can see the layer transforms inputs as we expect\ninp = torch.randn((2, 128))\nprint('Input batch shape: ', tuple(inp.shape))\nout = sprs(inp)\nprint('Output batch shape: ', tuple(out.shape))",
"Input batch shape: (2, 128)\nOutput batch shape: (2, 256)\n"
]
],
[
[
"In terms of behavior, the SETLayer transforms an input vector into the output space as would a fcl.\n## Initial Connection Distribution\nThe intialized layer has randomly assigned connections between input nodes and output nodes; each connection is associated with a weight, drawn from a normal distribution.",
"_____no_output_____"
]
],
[
[
"#Inspect init weight distribution\nplt.hist(np.array(sprs.weight.data), bins=40)\nplt.title('Weights distribution on initialization')\nplt.xlabel('Weight Value')\nplt.ylabel('Number of weights')\nplt.show()\n\nvec = sprs.connections[:, 0]\nvec = np.array(vec)\nvalues, counts = np.unique(vec, return_counts=True)\nplt.title('Connections to inputs')\nplt.bar(values, counts)\nplt.xlabel('Input vector index')\nplt.ylabel('Number of connections')\nplt.show()\nprint(\"done\")",
"_____no_output_____"
]
],
[
[
"The weights are sampled from a normal distribution, as is done with a standard fcl. The connections to the inputs are uniformly distributed.<br><br>\n\n## Killing Connections\nWhen connections are reassigned in SET, some proportion (defined by hyperparameter zeta) of the weights closest to zero are removed. We can set these to zero using the zero_connections method on the layer. (This method leaves the connections unchanged.)",
"_____no_output_____"
]
],
[
[
"sprs.zero_connections()\n\n#Inspect init weight distribution\nplt.hist(np.array(sprs.weight.data), bins=40)\nplt.title('Weights distribution after zeroing connections')\nplt.xlabel('Weight Value')\nplt.ylabel('Number of weights')\nplt.show()\nprint(\"done\")",
"_____no_output_____"
]
],
[
[
"## Evolving Connections\n\nThe evolve_connections() method will reassign these weights to new connections between input and output nodes. By default, these weights are initialized by sampling from the same distribution as the init function. Optionally, these weights can be set at zero (with init=False argument).",
"_____no_output_____"
]
],
[
[
"sprs.evolve_connections()\n\nplt.hist(np.array(sprs.weight.data), bins=40)\nplt.title('Weights distribution after evolving connections')\nplt.show()\n\nplt.title('Connections to inputs')\nplt.bar(values, counts)\nplt.xlabel('Input vector index')\nplt.ylabel('Number of connections')\nplt.show()\nprint(\"done\")",
"_____no_output_____"
]
],
[
[
"We can see these weight values have been re-distributed; the new connections conform to the same uniform distribution as before. (We see in the SET paper, and here later on, that the adaptive algorithm learns to allocate these connections to more important input values.)\n\n## A Simple SET Model\n\nThe following is a simple sparsely-connected model using SETLayers with default hyperparameters.",
"_____no_output_____"
]
],
[
[
"class SparseNet(nn.Module):\n def __init__(self):\n super(SparseNet, self).__init__()\n self.set_layers = []\n self.set1 = SETLayer(784, 512)\n self.set_layers.append(self.set1)\n #self.set2 = SETLayer(512, 512)\n #self.set_layers.append(self.set2)\n self.set2 = SETLayer(512, 128)\n self.set_layers.append(self.set2)\n #Use a dense layer for output because of low output dimensionality\n self.fc1 = nn.Linear(128, 10)\n \n def zero_connections(self):\n \"\"\"Sets connections to zero for inferences.\"\"\"\n for layer in self.set_layers:\n layer.zero_connections()\n \n def evolve_connections(self):\n \"\"\"Evolves connections.\"\"\"\n for layer in self.set_layers:\n layer.evolve_connections()\n \n def forward(self, x):\n x = x.reshape(-1, 784)\n x = F.relu(self.set1(x))\n x = F.relu(self.set2(x))\n #x = F.relu(self.set3(x))\n x = self.fc1(x)\n return F.log_softmax(x, dim=1)\n \ndef count_params(model):\n\n prms = 0\n for parameter in model.parameters():\n n_params = 1\n for prm in parameter.shape:\n n_params *= prm\n prms += n_params\n\n return prms\n \ndevice = \"cpu\"\nsparse_net = SparseNet().to(device)\nprint('number of params: ', count_params(sparse_net))",
"number of params: 23226\n"
]
],
[
[
"Consider a fully-connected model with the same architecture: It would contain more than 20 times the number of parameters!<br>\n## Training on MNIST\nThis code was adapted directly from the [pytorch mnist tutorial](https://github.com/pytorch/examples/blob/master/mnist/main.py).",
"_____no_output_____"
]
],
[
[
"class History(object):\n \"\"\"Tracks and plots training history\"\"\"\n def __init__(self):\n self.train_loss = []\n self.val_loss = []\n self.train_acc = []\n self.val_acc = []\n \n def plot(self):\n clear_output()\n plt.plot(self.train_loss, label='train loss')\n plt.plot(self.train_acc, label='train acc')\n plt.plot(self.val_loss, label='val loss')\n plt.plot(self.val_acc, label='val acc')\n plt.legend()\n plt.show()\n \ndef train(log_interval, model, device, train_loader, optimizer, epoch, history):\n model.train()\n correct = 0\n loss_ = []\n for batch_idx, (data, target) in enumerate(train_loader):\n data, target = data.to(device), target.to(device)\n optimizer.zero_grad()\n output = model(data)\n pred = output.max(1, keepdim=True)[1] # get the index of the max log-probability\n correct += pred.eq(target.view_as(pred)).sum().item()\n loss = F.nll_loss(output, target)\n loss.backward()\n loss_.append(loss.item())\n optimizer.step()\n if batch_idx % log_interval == 0:\n print('Train Epoch: {} [{}/{} ({:.0f}%)]\\tLoss: {:.6f}'.format(\n epoch, batch_idx * len(data), len(train_loader.dataset),\n 100. * batch_idx / len(train_loader), loss.item()))\n history.train_loss.append(np.array(loss_).mean())\n history.train_acc.append(correct/len(train_loader.dataset))\n return history\n\ndef test(model, device, test_loader, history):\n model.eval()\n test_loss = 0\n correct = 0\n with torch.no_grad():\n for data, target in test_loader:\n data, target = data.to(device), target.to(device)\n output = model(data)\n test_loss += F.nll_loss(output, target, reduction='sum').item() # sum up batch loss\n pred = output.max(1, keepdim=True)[1] # get the index of the max log-probability\n correct += pred.eq(target.view_as(pred)).sum().item()\n acc = correct / len(test_loader.dataset)\n test_loss /= len(test_loader.dataset)\n print('Test set: Average loss: {:.4f}, Accuracy: {}/{} ({:.2f}%)'.format(\n test_loss, correct, len(test_loader.dataset), 100. * acc))\n history.val_loss.append(test_loss)\n history.val_acc.append(acc)\n return history\n\nprint(\"done\")",
"done\n"
],
[
"torch.manual_seed(0)\n#Optimizer settings\nlr = .01\nmomentum = .5\nepochs = 50\nbatch_size=128\n\nlog_interval = 64\n\ntest_batch_size=128\n\ntrain_loader = torch.utils.data.DataLoader(\n datasets.MNIST('../data', train=True, download=True,\n transform=transforms.Compose([\n transforms.ToTensor(),\n transforms.Normalize((0.1307,), (0.3081,))\n ])),\n batch_size=batch_size, shuffle=True)\ntest_loader = torch.utils.data.DataLoader(\n datasets.MNIST('../data', train=False, transform=transforms.Compose([\n transforms.ToTensor(),\n transforms.Normalize((0.1307,), (0.3081,))\n ])),\n batch_size=test_batch_size, shuffle=True)\n\nprint(\"done\")",
"done\n"
]
],
[
[
"## Dealing with Optimizer Buffers\nSynapses recycles parameters. When connections are broken and reassigned, its parameter gets set to zero.<br><br>\nThis system is designed to be computationally efficient, but it comes with a nasty side-effect. Often, we use optimizers with some sort of buffer; the simplest example is momentum in SGD. When we reset a parameter, the information about the overwritten parameter in the optimizer buffer is not useful. We need to overwrite specific values in the buffer also. To do this in pytorch, we need to pass the optimizer to each SETLayer to let synapses do this for us. <br><br>\n<b>Notice: I'm still working out the best way to initialize adaptive optimizers (current version makes a naive attempt to pick good values); SGD with momentum works fine</b>",
"_____no_output_____"
]
],
[
[
"optimizer = optim.SGD(sparse_net.parameters(), lr=lr, momentum=momentum, weight_decay=1e-2)\nfor layer in sparse_net.set_layers:\n #here we tell our set layers about \n layer.optimizer = optimizer\n\n#This guy will keep track of optimization metrics.\nset_history = History()\nprint(\"done\")",
"done\n"
],
[
"def show_MNIST_connections(model):\n vec = model.set1.connections[:, 0]\n vec = np.array(vec)\n _, counts = np.unique(vec, return_counts=True)\n t = counts.reshape(28, 28)\n sns.heatmap(t, cmap='viridis', xticklabels=[], yticklabels=[], square=True);\n plt.title('Connections per input pixel');\n plt.show();\n v = [t[13-i:15+i,13-i:15+i].mean() for i in range(14)]\n plt.plot(v)\n plt.show()\n \nprint(\"done\")",
"done\n"
],
[
"import time\nepochs = 1000\nfor epoch in range(1, epochs + 1):\n #In the paper, evolutions occur on each epoch\n if epoch != 1:\n set_history.plot()\n show_MNIST_connections(sparse_net)\n if epoch != 1:\n print('Train set: Average loss: {:.4f}, Accuracy: {:.2f}%'.format(\n set_history.train_loss[epoch-2], 100. * set_history.train_acc[epoch-2]))\n print('Test set: Average loss: {:.4f}, Accuracy: {:.2f}%'.format(\n set_history.val_loss[epoch-2], 100. * set_history.val_acc[epoch-2]))\n sparse_net.evolve_connections()\n show_MNIST_connections(sparse_net)\n set_history = train(log_interval, sparse_net, device, train_loader, optimizer, epoch, set_history)\n #And smallest connections are removed during inference.\n sparse_net.zero_connections()\n set_history = test(sparse_net, device, test_loader, set_history)\n time.sleep(10)",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
]
] |
cbdc275b5fafdd5f9d30cc956f9e1d27fb616f77
| 18,944 |
ipynb
|
Jupyter Notebook
|
notebooks/circles/train_circles_pairs.ipynb
|
xenomarz/deep-signature
|
f831f05971727c5d00cf3b5c556b6a8b658048df
|
[
"MIT"
] | null | null | null |
notebooks/circles/train_circles_pairs.ipynb
|
xenomarz/deep-signature
|
f831f05971727c5d00cf3b5c556b6a8b658048df
|
[
"MIT"
] | null | null | null |
notebooks/circles/train_circles_pairs.ipynb
|
xenomarz/deep-signature
|
f831f05971727c5d00cf3b5c556b6a8b658048df
|
[
"MIT"
] | null | null | null | 31.00491 | 162 | 0.555743 |
[
[
[
"# ** IMPORT PACKAGES: **",
"_____no_output_____"
]
],
[
[
"# Python peripherals\nimport os\nimport random\n\n# Scipy\nimport scipy.io\nimport scipy.stats as ss\n\n# Numpy\nimport numpy\n\n# Matplotlib\nimport matplotlib.pyplot as plt\nimport matplotlib.collections as mcoll\nimport matplotlib.ticker as ticker\n\n# PyTorch\nimport torch\nfrom torch.utils.data.sampler import SubsetRandomSampler\nfrom torch.utils.data.sampler import SequentialSampler\nfrom torch.utils.data import DataLoader\n\n# IPython\nfrom IPython.display import display, HTML\n\n# Deep signature\nimport deep_signature.utils\nfrom deep_signature.data_generation import SimpleCurveDatasetGenerator\nfrom deep_signature.data_generation import SimpleCurveManager\nfrom deep_signature.training import DeepSignatureNet\nfrom deep_signature.training import ContrastiveLoss\nfrom deep_signature.training import ModelTrainer\nfrom deep_signature.training import DeepSignaturePairsDataset\nfrom deep_signature import curve_processing",
"_____no_output_____"
]
],
[
[
"# ** HELPER FUNCTIONS: **",
"_____no_output_____"
]
],
[
[
"def chunker(seq, size):\n return (seq[pos:pos + size] for pos in range(0, len(seq), size))\n\n# https://stackoverflow.com/questions/36074455/python-matplotlib-with-a-line-color-gradient-and-colorbar\ndef colorline(ax, x, y, z=None, cmap='copper', norm=plt.Normalize(0.0, 1.0), linewidth=3, alpha=1.0):\n \"\"\"\n http://nbviewer.ipython.org/github/dpsanders/matplotlib-examples/blob/master/colorline.ipynb\n http://matplotlib.org/examples/pylab_examples/multicolored_line.html\n Plot a colored line with coordinates x and y\n Optionally specify colors in the array z\n Optionally specify a colormap, a norm function and a line width\n \"\"\"\n\n # Default colors equally spaced on [0,1]:\n if z is None:\n z = numpy.linspace(0.0, 1.0, len(x))\n\n # Special case if a single number:\n # to check for numerical input -- this is a hack\n if not hasattr(z, \"__iter__\"):\n z = numpy.array([z])\n\n z = numpy.asarray(z)\n\n segments = make_segments(x, y)\n lc = mcoll.LineCollection(segments, array=z, cmap=cmap, norm=norm,\n linewidth=linewidth, alpha=alpha)\n\n # ax = plt.gca()\n ax.add_collection(lc)\n\n return lc\n\ndef make_segments(x, y):\n \"\"\"\n Create list of line segments from x and y coordinates, in the correct format\n for LineCollection: an array of the form numlines x (points per line) x 2 (x\n and y) array\n \"\"\"\n\n points = numpy.array([x, y]).T.reshape(-1, 1, 2)\n segments = numpy.concatenate([points[:-1], points[1:]], axis=1)\n return segments\n\ndef plot_dist(ax, dist):\n x = numpy.array(range(dist.shape[0]))\n y = dist\n ax.set_xlim(x.min(), x.max())\n ax.set_ylim(y.min(), y.max())\n colorline(ax=ax, x=x, y=y, cmap='hsv')\n\ndef plot_curve_sample(ax, curve, curve_sample, indices, zorder, point_size=10, alpha=1, cmap='hsv'):\n x = curve_sample[:, 0]\n y = curve_sample[:, 1]\n c = numpy.linspace(0.0, 1.0, curve.shape[0])\n\n ax.scatter(\n x=x,\n y=y,\n c=c[indices],\n s=point_size,\n cmap=cmap,\n alpha=alpha,\n norm=plt.Normalize(0.0, 1.0),\n zorder=zorder)\n\ndef plot_curve_section_center_point(ax, x, y, zorder, radius=1, color='white'):\n circle = plt.Circle((x, y), radius=radius, color=color, zorder=zorder)\n ax.add_artist(circle)\n\ndef plot_curve(ax, curve, linewidth=2, color='red', alpha=1):\n x = curve[:, 0]\n y = curve[:, 1]\n ax.plot(x, y, linewidth=linewidth, color=color, alpha=alpha)\n\ndef plot_curvature(ax, curvature, color='red', linewidth=2):\n x = range(curvature.shape[0])\n y = curvature\n ax.plot(x, y, color=color, linewidth=linewidth)\n\ndef plot_sample(ax, sample, color, zorder, point_size=10, alpha=1):\n x = sample[:, 0]\n y = sample[:, 1]\n\n ax.scatter(\n x=x,\n y=y,\n s=point_size,\n color=color,\n alpha=alpha,\n zorder=zorder)\n\ndef all_subdirs_of(b='.'):\n result = []\n for d in os.listdir(b):\n bd = os.path.join(b, d)\n if os.path.isdir(bd): result.append(bd)\n return result",
"_____no_output_____"
]
],
[
[
"# ** GLOBAL SETTINGS: **",
"_____no_output_____"
]
],
[
[
"curves_dir_path_train = 'C:/deep-signature-data/circles/curves/pairs/train'\ncurves_dir_path_test = 'C:/deep-signature-data/circles/curves/pairs/test'\nnegative_pairs_dir_path = 'C:/deep-signature-data/circles/datasets/pairs/negative-pairs'\npositive_pairs_dir_path = 'C:/deep-signature-data/circles/datasets/pairs/positive-pairs'\nresults_base_dir_path = 'C:/deep-signature-data/circles/results/pairs'\n\nepochs = 100\nbatch_size = 256\nvalidation_split = .05\nlearning_rate = 1e-4\nmu = 1\n\nrotation_factor=1\nsampling_factor=1\nmultimodality_factor=15\nsupporting_points_count=3\nsampling_points_count=None\nsampling_points_ratio=0.15\nsectioning_points_count=None\nsectioning_points_ratio=0.1\nsample_points=3\n\nplt.style.use(\"dark_background\")",
"_____no_output_____"
]
],
[
[
"# ** SANITY CHECK - CURVES: **",
"_____no_output_____"
]
],
[
[
"curves = SimpleCurveDatasetGenerator.load_curves(dir_path=curves_dir_path_train)\n\nfig, ax = plt.subplots(1, 1, figsize=(80,40))\nfor label in (ax.get_xticklabels() + ax.get_yticklabels()):\n label.set_fontsize(30)\nax.axis('equal')\n\nlimit = 200\ncolor_map = plt.get_cmap('rainbow', limit)\n\nfor i, curve in enumerate(curves[:limit]): \n plot_curve(ax=ax, curve=curve, linewidth=5, color=color_map(i))\n\nplt.show()",
"_____no_output_____"
]
],
[
[
"# ** SANITY CHECK - NEGATIVE PAIRS **",
"_____no_output_____"
]
],
[
[
"\nnegative_pairs = SimpleCurveDatasetGenerator.load_negative_pairs(dir_path=negative_pairs_dir_path)\n\nrows = 6\ncols = 6\ncells = rows * cols\nfig, ax = plt.subplots(rows, cols, figsize=(40,100))\n\naxes = []\nfor i in range(rows):\n for j in range(cols):\n for label in (ax[i,j].get_xticklabels() + ax[i,j].get_yticklabels()):\n label.set_fontsize(10)\n # ax[i,j].axis('equal')\n axes.append(ax[i,j])\n\nnumpy.random.shuffle(negative_pairs)\nfor negative_pair_index, negative_pair in enumerate(negative_pairs[:cells]):\n ax = axes[negative_pair_index]\n plot_sample(ax, negative_pair[0], point_size=50, alpha=1, color='red', zorder=50)\n plot_sample(ax, negative_pair[1], point_size=50, alpha=1, color='green', zorder=50)\n plot_sample(ax, numpy.array([[0,0]]), point_size=50, alpha=1, color='white', zorder=100)\n\nplt.show()",
"_____no_output_____"
]
],
[
[
"# ** SANITY CHECK - POSITIVE PAIRS **",
"_____no_output_____"
]
],
[
[
"\npositive_pairs = SimpleCurveDatasetGenerator.load_positive_pairs(dir_path=positive_pairs_dir_path)\n\nrows = 6\ncols = 8\ncells = rows * cols\nfig, ax = plt.subplots(rows, cols, figsize=(40,100))\n\naxes = []\nfor i in range(rows):\n for j in range(cols):\n for label in (ax[i,j].get_xticklabels() + ax[i,j].get_yticklabels()):\n label.set_fontsize(10)\n ax[i,j].axis('equal')\n axes.append(ax[i,j])\n\nnumpy.random.shuffle(positive_pairs)\nfor positive_pair_index, positive_pair in enumerate(positive_pairs[:cells]):\n ax = axes[positive_pair_index]\n plot_sample(ax, positive_pair[0], point_size=50, alpha=1, color='red', zorder=50)\n plot_sample(ax, positive_pair[1], point_size=50, alpha=1, color='green', zorder=50)\n plot_sample(ax, numpy.array([[0,0]]), point_size=50, alpha=1, color='white', zorder=100)\n\nplt.show()",
"_____no_output_____"
]
],
[
[
"# ** SANITY CHECK - DATASET PAIRS **",
"_____no_output_____"
]
],
[
[
"dataset = DeepSignaturePairsDataset()\ndataset.load_dataset(\n negative_pairs_dir_path=negative_pairs_dir_path, \n positive_pairs_dir_path=positive_pairs_dir_path)\ndataset_size = len(dataset)\nindices = list(range(dataset_size))\n# numpy.random.shuffle(indices)\nsampler = SubsetRandomSampler(indices)\ndata_loader = DataLoader(dataset, batch_size=1, sampler=sampler)\n\ndisplay(HTML('<h3>Random samples of positive and negative examples:</h3>'))\nfor pair_index, data in enumerate(data_loader, 0):\n if pair_index == 10:\n break\n\n curve1 = torch.squeeze(torch.squeeze(data['input'])[0])\n curve2 = torch.squeeze(torch.squeeze(data['input'])[1])\n label = int(torch.squeeze(data['labels']))\n\n if label == 1:\n pair_type = 'Positive'\n else:\n pair_type = 'Negative'\n\n display(HTML(f'<h3>{pair_type} sample #{pair_index}:</h3>'))\n\n curve1 = curve1.cpu().numpy()\n curve2 = curve2.cpu().numpy()\n\n fig, ax = plt.subplots(1, 1, figsize=(5,5))\n ax.axis('equal')\n\n plot_sample(\n ax=ax, \n sample=curve1, \n point_size=50,\n color='lightcoral',\n zorder=50)\n\n plot_sample(\n ax=ax, \n sample=curve2, \n point_size=50,\n color='skyblue',\n zorder=50)\n\n plot_sample(ax, numpy.array([[0,0]]), point_size=50, alpha=1, color='white', zorder=100)\n\n for label in (ax.get_xticklabels() + ax.get_yticklabels()):\n label.set_fontsize(10)\n \n plt.show()",
"_____no_output_____"
]
],
[
[
"# ** TRAINING **",
"_____no_output_____"
]
],
[
[
"torch.set_default_dtype(torch.float64)\ndataset = SimpleDeepSignatureDataset()\ndataset.load_dataset(\n negative_pairs_dir_path=negative_pairs_dir_path, \n positive_pairs_dir_path=positive_pairs_dir_path)\nmodel = SimpleDeepSignatureNet(layers=20, sample_points=sample_points).cuda()\noptimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)\nloss_fn = ContrastiveLoss(mu)\nmodel_trainer = ModelTrainer(model=model, loss_fn=loss_fn, optimizer=optimizer)\n\nprint(model)\n\ndef epoch_handler(epoch_index):\n return\n\nresults = model_trainer.fit(dataset=dataset, epochs=epochs, batch_size=batch_size, results_base_dir_path=results_base_dir_path, epoch_handler=epoch_handler)",
"_____no_output_____"
]
],
[
[
"# ** TRAIN/VALIDATION LOSS **",
"_____no_output_____"
]
],
[
[
"# results_file_path = os.path.normpath(os.path.join(results_base_dir_path, 'results.npy'))\nall_subdirs = all_subdirs_of(results_base_dir_path)\nlatest_subdir = os.path.normpath(max(all_subdirs, key=os.path.getmtime))\nresults = numpy.load(f\"{latest_subdir}/results.npy\", allow_pickle=True).item()\n\nepochs = results['epochs']\nbatch_size = results['batch_size']\ntrain_loss_array = results['train_loss_array']\nvalidation_loss_array = results['validation_loss_array']\nepochs_list = numpy.array(range(len(train_loss_array)))\n\nfig, ax = plt.subplots(1, 1, figsize=(10,10))\nax.xaxis.set_major_locator(ticker.MaxNLocator(integer=True))\n\nfor label in (ax.get_xticklabels() + ax.get_yticklabels()):\n label.set_fontsize(20)\n\nax.plot(epochs_list, train_loss_array, label='Train Loss', linewidth=7.0)\nax.plot(epochs_list, validation_loss_array, label='Validation Loss', linewidth=7.0)\nplt.legend(fontsize=20, title_fontsize=20)\n\n# print(train_loss_array)\n# print(validation_loss_array)\nplt.show()",
"_____no_output_____"
]
],
[
[
"# ** TEST MODEL **",
"_____no_output_____"
]
],
[
[
"torch.set_default_dtype(torch.float64)\ndevice = torch.device('cuda')\nmodel = DeepSignatureNet(layers=2, sample_points=sample_points).cuda()\nmodel.load_state_dict(torch.load(results['model_file_path'], map_location=device))\nmodel.eval()\n\nlimit = 50\ncurves = SimpleCurveDatasetGenerator.load_curves(dir_path=curves_dir_path_test)\nnumpy.random.seed(50)\nnumpy.random.shuffle(curves)\ncurves = curves[:limit]\n\ncolor_map = plt.get_cmap('rainbow', limit)\nfig, ax = plt.subplots(2, 1, figsize=(80,100))\nax[0].axis('equal')\nfor label in (ax[0].get_xticklabels() + ax[0].get_yticklabels()):\n label.set_fontsize(30)\n\nfor label in (ax[1].get_xticklabels() + ax[1].get_yticklabels()):\n label.set_fontsize(30)\n\nlow = 0.1\nhigh = 0.4\ndelta = numpy.random.uniform(low=low, high=high, size=[4000, 2])\nfor curve_index, curve in enumerate(curves): \n plot_curve(ax=ax[0], curve=curve, color=color_map(curve_index), linewidth=5)\n predicted_curvature = numpy.zeros(curve.shape[0])\n center_index = 1\n for i in range(curve.shape[0]):\n current_delta = delta[i, :] * curve.shape[0]\n indices = numpy.array([i - int(current_delta[0]), i, i + int(current_delta[1])])\n indices = numpy.mod(indices, curve.shape[0])\n sample = curve[indices]\n center_point = sample[center_index]\n sample = sample - center_point\n\n if curve_processing.is_ccw(curve_sample=sample) is False:\n sample = numpy.flip(sample, axis=0)\n\n radians = curve_processing.calculate_tangent_angle(curve_sample=sample)\n sample = curve_processing.rotate_curve(curve=sample, radians=radians)\n batch_data = torch.unsqueeze(torch.unsqueeze(torch.from_numpy(sample).double(), dim=0), dim=0).cuda()\n\n with torch.no_grad():\n predicted_curvature[i] = torch.squeeze(model(batch_data), dim=0).cpu().detach().numpy()\n\n plot_curvature(ax=ax[1], curvature=predicted_curvature, color=color_map(curve_index), linewidth=5)\nplt.show()",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
]
] |
cbdc3256999095c79be69ab2b56f2e6ac4b58a11
| 6,958 |
ipynb
|
Jupyter Notebook
|
sorting_searching/merge_sort/merge_sort_challenge.ipynb
|
hanbf/interactive-coding-challenges
|
1676ac16c987e35eeb4be6ab57a3c10ed9b71b8b
|
[
"Apache-2.0"
] | null | null | null |
sorting_searching/merge_sort/merge_sort_challenge.ipynb
|
hanbf/interactive-coding-challenges
|
1676ac16c987e35eeb4be6ab57a3c10ed9b71b8b
|
[
"Apache-2.0"
] | null | null | null |
sorting_searching/merge_sort/merge_sort_challenge.ipynb
|
hanbf/interactive-coding-challenges
|
1676ac16c987e35eeb4be6ab57a3c10ed9b71b8b
|
[
"Apache-2.0"
] | null | null | null | 26.761538 | 296 | 0.45904 |
[
[
[
"This notebook was prepared by [Donne Martin](http://donnemartin.com). Source and license info is on [GitHub](https://github.com/donnemartin/interactive-coding-challenges).",
"_____no_output_____"
],
[
"# Challenge Notebook",
"_____no_output_____"
],
[
"## Problem: Implement merge sort.\n\n* [Constraints](#Constraints)\n* [Test Cases](#Test-Cases)\n* [Algorithm](#Algorithm)\n* [Code](#Code)\n* [Unit Test](#Unit-Test)\n* [Solution Notebook](#Solution-Notebook)",
"_____no_output_____"
],
[
"## Constraints\n\n* Is a naive solution sufficient?\n * Yes\n* Are duplicates allowed?\n * Yes\n* Can we assume the input is valid?\n * No\n* Can we assume this fits memory?\n * Yes",
"_____no_output_____"
],
[
"## Test Cases\n\n* None -> Exception\n* Empty input -> []\n* One element -> [element]\n* Two or more elements\n* Left and right subarrays of different lengths",
"_____no_output_____"
],
[
"## Algorithm\n\nRefer to the [Solution Notebook](http://nbviewer.ipython.org/github/donnemartin/interactive-coding-challenges/blob/master/sorting_searching/merge_sort/merge_sort_solution.ipynb). If you are stuck and need a hint, the solution notebook's algorithm discussion might be a good place to start.",
"_____no_output_____"
],
[
"## Code",
"_____no_output_____"
]
],
[
[
"class MergeSort(object):\n \n # merge sorted subarray left, right\n def _merge(self, left, right):\n result = []\n i = 0\n j = 0\n while i < len(left) or j < len(right):\n if i >= len(left):\n result.append(right[j])\n j += 1\n elif j >= len(right):\n result.append(left[i])\n i += 1\n elif left[i] < right[j]:\n result.append(left[i])\n i += 1\n else: # left[i] >= right[j]\n result.append(right[j])\n j += 1\n \n return result \n \n def _sort(self, data):\n if (len(data) < 2):\n return data\n\n # divide data to 2 halves\n mid = len(data) / 2\n # copy [0 ~ mid)\n left = data[:mid]\n # copy [mid, len)\n right = data[mid:]\n return self._merge(self._sort(left), self._sort(right))\n \n def _sort2(self, data):\n if (len(data) < 2):\n return data\n \n i = 1\n \n # i is size of each piece of subarray\n while 1:\n # j is index of the 'left' subarray of each merging pair.\n j = 0\n result = []\n while j < len(data):\n left = data[j:(j + i)]\n right = data[(j + i):(j + i + i)]\n result += self._merge(left, right)\n j += i + i\n \n data = result\n i += i\n if i >= len(data):\n break\n\n \n return data\n \n def sort(self, data):\n if data is None:\n raise\n \n return self._sort2(data)",
"_____no_output_____"
]
],
[
[
"## Unit Test\n\n\n\n**The following unit test is expected to fail until you solve the challenge.**",
"_____no_output_____"
]
],
[
[
"# %load test_merge_sort.py\nfrom nose.tools import assert_equal, assert_raises\n\n\nclass TestMergeSort(object):\n\n def test_merge_sort(self):\n merge_sort = MergeSort()\n\n print('None input')\n assert_raises(TypeError, merge_sort.sort, None)\n\n print('Empty input')\n assert_equal(merge_sort.sort([]), [])\n\n print('One element')\n assert_equal(merge_sort.sort([5]), [5])\n\n print('Two or more elements, odd len')\n data = [5, 1, 7, 2, 6, -3, 5, 7]\n assert_equal(merge_sort.sort(data), sorted(data))\n\n print('Two or more elements')\n data = [5, 1, 7, 2, 6, -3, 5, 7, -1]\n assert_equal(merge_sort.sort(data), sorted(data))\n\n print('Success: test_merge_sort')\n\n\ndef main():\n test = TestMergeSort()\n test.test_merge_sort()\n\n\nif __name__ == '__main__':\n main()",
"None input\nEmpty input\nOne element\nTwo or more elements, odd len\nTwo or more elements\nSuccess: test_merge_sort\n"
]
],
[
[
"## Solution Notebook\n\nReview the [Solution Notebook](http://nbviewer.ipython.org/github/donnemartin/interactive-coding-challenges/blob/master/sorting_searching/merge_sort/merge_sort_solution.ipynb) for a discussion on algorithms and code solutions.",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown"
] |
[
[
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown",
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
]
] |
cbdc47c4799ab066bf4c7eb112c485ecb1423b7e
| 3,074 |
ipynb
|
Jupyter Notebook
|
DAY 001 ~ 100/DAY017_[Programmers] 정렬 H-index (Python).ipynb
|
SOMJANG/CODINGTEST_PRACTICE
|
1a7304e9063579441b8a67765175c82b0ad93ac9
|
[
"MIT"
] | 15 |
2020-03-17T01:18:33.000Z
|
2021-12-24T06:31:06.000Z
|
DAY 001 ~ 100/DAY017_[Programmers] 정렬 H-index (Python).ipynb
|
SOMJANG/CODINGTEST_PRACTICE
|
1a7304e9063579441b8a67765175c82b0ad93ac9
|
[
"MIT"
] | null | null | null |
DAY 001 ~ 100/DAY017_[Programmers] 정렬 H-index (Python).ipynb
|
SOMJANG/CODINGTEST_PRACTICE
|
1a7304e9063579441b8a67765175c82b0ad93ac9
|
[
"MIT"
] | 10 |
2020-03-17T01:18:34.000Z
|
2022-03-30T10:53:07.000Z
| 19.333333 | 95 | 0.450553 |
[
[
[
"## 2020년 2월 24일 일요일\n### 프로그래머스 정렬 : H-index\n### 문제 : https://programmers.co.kr/learn/courses/30/lessons/42747#fn1\n### 블로그 : https://somjang.tistory.com/entry/Programmers-%EC%A0%95%EB%A0%AC-H-Index-Python",
"_____no_output_____"
],
[
"### 첫번째 시도",
"_____no_output_____"
]
],
[
[
"def solution(citations):\n answer = 0\n \n reversed_citations = citations\n reversed_citations.sort(reverse=True)\n \n for i in range(len(reversed_citations)):\n if i+1 >= reversed_citations[i]:\n answer = reversed_citations[i]\n break\n \n return answer",
"_____no_output_____"
],
[
"solution([3, 0, 6, 1, 5])",
"_____no_output_____"
]
],
[
[
"---",
"_____no_output_____"
],
[
"### 두번째 시도",
"_____no_output_____"
]
],
[
[
"def solution(citations):\n answer=[]\n citations.sort(reverse=True)\n for i in range(len(citations),0,-1):\n count=0\n for j in citations:\n if j >= i:\n count+=1\n if count <= i:\n answer.append(count)\n else:\n count=0\n answer.append(count)\n return max(answer)",
"_____no_output_____"
],
[
"solution([3, 0, 6, 1, 5])",
"_____no_output_____"
],
[
"solution([20, 19, 18 , 1])",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown",
"markdown"
],
[
"code",
"code"
],
[
"markdown",
"markdown"
],
[
"code",
"code",
"code"
]
] |
cbdc586688bcb305b2d01efb02888aff4ab77f3c
| 2,581 |
ipynb
|
Jupyter Notebook
|
notebooks/test_quadrature_partial_integration_of_domain.ipynb
|
saullocastro/panels
|
5387daab1ec656b065f7a9be820e2c234e53764e
|
[
"BSD-3-Clause"
] | 1 |
2020-03-12T14:39:13.000Z
|
2020-03-12T14:39:13.000Z
|
notebooks/test_quadrature_partial_integration_of_domain.ipynb
|
saullocastro/panels
|
5387daab1ec656b065f7a9be820e2c234e53764e
|
[
"BSD-3-Clause"
] | null | null | null |
notebooks/test_quadrature_partial_integration_of_domain.ipynb
|
saullocastro/panels
|
5387daab1ec656b065f7a9be820e2c234e53764e
|
[
"BSD-3-Clause"
] | null | null | null | 26.608247 | 73 | 0.462611 |
[
[
[
"import numpy as np\nfrom panels.legendre_gauss_quadrature import get_points_weights\n\nn = 3\npts = np.zeros(n, dtype=float)\nws = np.zeros(n, dtype=float)\nget_points_weights(n, pts, ws)\n\ndef func(x):\n return 4 + 2*x - 4*x**2 + 2*x**3 - 10*x**4\n\ndef funcxi(xi, xinf, xsup):\n x = (xi + 1)*(xsup - xinf)/2 + xinf\n return 4 + 2*x - 4*x**2 + 2*x**3 - 10*x**4\n\ndef intref(x1, x2):\n res1 = 4*x1 + 2*x1**2/2 - 4*x1**3/3 + 2*x1**4/4 - 10*x1**5/5\n res2 = 4*x2 + 2*x2**2/2 - 4*x2**3/3 + 2*x2**4/4 - 10*x2**5/5\n return res2 - res1\n\ndef intquad(x1, x2):\n ans = 0\n for xi, w in zip(pts, ws):\n x = (xi + 1)*(x2-x1)/2 + x1\n #dx = (x2-x1)/2*(dxi)\n ans += w * func(x) * ((x2-x1)/2)\n return ans\n\ndef intquadxi(x1, x2, xinf, xsup):\n xi1 = (x1 - xinf)/(xsup - xinf)*2 - 1\n xi2 = (x2 - xinf)/(xsup - xinf)*2 - 1\n print('DEBUG xi1, xi2', xi1, xi2)\n ans = 0\n for xi, w in zip(pts, ws):\n xip = (xi - (-1))/2 * (xi2 - xi1) + xi1\n ans += w * funcxi(xip, xinf, xsup) * (x2 - x1)/2\n return ans\n\nprint(intref(0, 10))\nprint(intref(4, 6))\n\nprint(intquad(0, 10))\nprint(intquad(4, 6))\n\nxinf = 0\nxsup = 10\nprint(intquadxi(0, 10, xinf, xsup))\nprint(intquadxi(4, 6, xinf, xsup))",
"-196193.33333333334\n-13158.666666666666\n-196193.33333333328\n-13158.666666666668\nDEBUG xi1, xi2 -1.0 1.0\n-196193.33333333328\nDEBUG xi1, xi2 -0.19999999999999996 0.19999999999999996\n-13158.666666666664\n"
]
]
] |
[
"code"
] |
[
[
"code"
]
] |
cbdc5bc53adbf7c1727e10aac469876314e5282e
| 908 |
ipynb
|
Jupyter Notebook
|
Untitled0.ipynb
|
keyula/Nudity-Detector
|
0f7ecdf962306e7e5b06de306abfca0ee3f48817
|
[
"MIT"
] | null | null | null |
Untitled0.ipynb
|
keyula/Nudity-Detector
|
0f7ecdf962306e7e5b06de306abfca0ee3f48817
|
[
"MIT"
] | null | null | null |
Untitled0.ipynb
|
keyula/Nudity-Detector
|
0f7ecdf962306e7e5b06de306abfca0ee3f48817
|
[
"MIT"
] | null | null | null | 23.282051 | 228 | 0.506608 |
[
[
[
"<a href=\"https://colab.research.google.com/github/keyula/Nudity-Detector/blob/main/Untitled0.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>",
"_____no_output_____"
]
],
[
[
"",
"_____no_output_____"
]
]
] |
[
"markdown",
"code"
] |
[
[
"markdown"
],
[
"code"
]
] |
cbdc5be707858d57463fb9d923f77dfd88ba8bde
| 20,590 |
ipynb
|
Jupyter Notebook
|
src/datacleaning/Chapter 9/4_wide_to_long.py.ipynb
|
vidyabhandary/DataScienceNotebooks
|
1a9a423ffe6bdcd803ed195da97a751e44567813
|
[
"MIT"
] | null | null | null |
src/datacleaning/Chapter 9/4_wide_to_long.py.ipynb
|
vidyabhandary/DataScienceNotebooks
|
1a9a423ffe6bdcd803ed195da97a751e44567813
|
[
"MIT"
] | null | null | null |
src/datacleaning/Chapter 9/4_wide_to_long.py.ipynb
|
vidyabhandary/DataScienceNotebooks
|
1a9a423ffe6bdcd803ed195da97a751e44567813
|
[
"MIT"
] | null | null | null | 28.051771 | 773 | 0.390675 |
[
[
[
"<h1>Table of Contents<span class=\"tocSkip\"></span></h1>\n<div class=\"toc\"><ul class=\"toc-item\"><li><span><a href=\"#Import-pandas-and-load-the-NLS-data\" data-toc-modified-id=\"Import-pandas-and-load-the-NLS-data-1\"><span class=\"toc-item-num\">1 </span>Import pandas and load the NLS data</a></span></li><li><span><a href=\"#View-some-of-the-weeks-worked-and-college-enrollment-data\" data-toc-modified-id=\"View-some-of-the-weeks-worked-and-college-enrollment-data-2\"><span class=\"toc-item-num\">2 </span>View some of the weeks worked and college enrollment data</a></span></li><li><span><a href=\"#Run-the-wide_to_long-function\" data-toc-modified-id=\"Run-the-wide_to_long-function-3\"><span class=\"toc-item-num\">3 </span>Run the wide_to_long function</a></span></li></ul></div>",
"_____no_output_____"
],
[
"# Import pandas and load the NLS data",
"_____no_output_____"
]
],
[
[
"import pandas as pd",
"_____no_output_____"
],
[
"# pd.set_option('display.width', 200)\n# pd.set_option('display.max_columns', 30)\n# pd.set_option('display.max_rows', 200)\n# pd.options.display.float_format = '{:,.0f}'.format",
"_____no_output_____"
],
[
"import watermark\n%load_ext watermark\n\n%watermark -n -i -iv",
"pandas : 1.2.1\nwatermark: 2.1.0\njson : 2.0.9\n\n"
],
[
"nls97 = pd.read_csv('data/nls97f.csv')",
"_____no_output_____"
],
[
"nls97.set_index('personid', inplace=True)",
"_____no_output_____"
]
],
[
[
"# View some of the weeks worked and college enrollment data",
"_____no_output_____"
]
],
[
[
"weeksworkedcols = [\n 'weeksworked00', 'weeksworked01', 'weeksworked02', 'weeksworked03',\n 'weeksworked04'\n]",
"_____no_output_____"
],
[
"colenrcols = [\n 'colenroct00', 'colenroct01', 'colenroct02', 'colenroct03', 'colenroct04'\n]",
"_____no_output_____"
],
[
"nls97.loc[nls97['originalid'].isin([1, 2]),\n ['originalid'] + weeksworkedcols + colenrcols].T",
"_____no_output_____"
]
],
[
[
"# Run the wide_to_long function",
"_____no_output_____"
]
],
[
[
"workschool = pd.wide_to_long(nls97[['originalid'] + weeksworkedcols +\n colenrcols],\n stubnames=['weeksworked', 'colenroct'],\n i=['originalid'],\n j='year').reset_index()",
"_____no_output_____"
],
[
"workschool.head(2)",
"_____no_output_____"
],
[
"workschool['year'] = workschool['year'] + 2000",
"_____no_output_____"
],
[
"workschool = workschool.sort_values(['originalid', 'year'])",
"_____no_output_____"
],
[
"workschool.set_index(['originalid'], inplace=True)",
"_____no_output_____"
],
[
"# wide_to_long accomplishes in one step what it took us several steps to accomplish in\n# the previous recipe using melt\n\nworkschool.head(10)",
"_____no_output_____"
],
[
"weeksworkedcols = [\n 'weeksworked00', 'weeksworked01', 'weeksworked02', 'weeksworked04',\n 'weeksworked05'\n]",
"_____no_output_____"
],
[
"workschool = pd.wide_to_long(nls97[['originalid'] + weeksworkedcols +\n colenrcols],\n stubnames=['weeksworked', 'colenroct'],\n i=['originalid'],\n j='year').reset_index()",
"_____no_output_____"
],
[
"workschool['year'] = workschool['year'] + 2000",
"_____no_output_____"
],
[
"workschool = workschool.sort_values(['originalid', 'year'])",
"_____no_output_____"
],
[
"workschool.set_index(['originalid'], inplace=True)",
"_____no_output_____"
],
[
"workschool.head(12)",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown",
"markdown"
],
[
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
]
] |
cbdc6bc7635701a26e4206fdae92f971d7799170
| 4,055 |
ipynb
|
Jupyter Notebook
|
Getting-Started/Core-JavaScript/Introduction.ipynb
|
kodaman2/fab-mentorship
|
f99ea7080f0f8bd9a8544deed4b0e3ee5c8ea2a6
|
[
"MIT"
] | 3 |
2019-10-25T17:40:15.000Z
|
2019-10-28T20:26:19.000Z
|
Getting-Started/Core-JavaScript/Introduction.ipynb
|
kodaman2/fab-mentorship
|
f99ea7080f0f8bd9a8544deed4b0e3ee5c8ea2a6
|
[
"MIT"
] | null | null | null |
Getting-Started/Core-JavaScript/Introduction.ipynb
|
kodaman2/fab-mentorship
|
f99ea7080f0f8bd9a8544deed4b0e3ee5c8ea2a6
|
[
"MIT"
] | null | null | null | 25.503145 | 459 | 0.54402 |
[
[
[
"empty"
]
]
] |
[
"empty"
] |
[
[
"empty"
]
] |
cbdc795e4cee2ff46330e9b270a075bd8a57d036
| 244,512 |
ipynb
|
Jupyter Notebook
|
RNN - binary and multiclass classification.ipynb
|
aduphad/Predictive-Maintenance-of-Aircraft-Engine
|
cc687ee820eb1510e6d0c998a000044977eb85ee
|
[
"MIT"
] | null | null | null |
RNN - binary and multiclass classification.ipynb
|
aduphad/Predictive-Maintenance-of-Aircraft-Engine
|
cc687ee820eb1510e6d0c998a000044977eb85ee
|
[
"MIT"
] | null | null | null |
RNN - binary and multiclass classification.ipynb
|
aduphad/Predictive-Maintenance-of-Aircraft-Engine
|
cc687ee820eb1510e6d0c998a000044977eb85ee
|
[
"MIT"
] | null | null | null | 157.647969 | 59,968 | 0.864031 |
[
[
[
"## RNN Time Series Classification",
"_____no_output_____"
],
[
"### Importing Required Libraries and Data",
"_____no_output_____"
]
],
[
[
"import pandas as pd\nimport numpy as np\n\n#to plot the data\nimport matplotlib.pyplot as plt\nimport seaborn as sns\n%matplotlib inline\nimport matplotlib as mpl\nmpl.rcParams.update(mpl.rcParamsDefault)\n\nimport time\nimport os\nos.chdir(\"C:/Data/aircraft/\")\n\nfrom sklearn.preprocessing import MinMaxScaler #to normalize data\nfrom sklearn.metrics import classification_report, confusion_matrix, roc_curve\nfrom sklearn.metrics import f1_score, roc_auc_score, accuracy_score, precision_score, recall_score\nfrom sklearn.utils import class_weight\n\n#for deep learning\nimport keras\nimport keras.backend as k\nfrom keras.models import Sequential\nfrom keras.layers import Dense, Activation, Masking, Dropout, SimpleRNN\nfrom keras.callbacks import History\nfrom keras import callbacks\nfrom keras.utils import to_categorical",
"_____no_output_____"
],
[
"def prepare_data(drop_cols = True):\n dependent_var = ['RUL']\n index_columns_names = [\"UnitNumber\",\"Cycle\"]\n operational_settings_columns_names = [\"OpSet\"+str(i) for i in range(1,4)]\n sensor_measure_columns_names =[\"SensorMeasure\"+str(i) for i in range(1,22)]\n input_file_column_names = index_columns_names + operational_settings_columns_names + sensor_measure_columns_names\n\n cols_to_drop = ['OpSet3', 'SensorMeasure1', 'SensorMeasure5', 'SensorMeasure6', 'SensorMeasure10', 'SensorMeasure14',\n 'SensorMeasure16', 'SensorMeasure18', 'SensorMeasure19']\n\n df_train = pd.read_csv('train_FD001.txt',delim_whitespace=True,names=input_file_column_names)\n\n rul = pd.DataFrame(df_train.groupby('UnitNumber')['Cycle'].max()).reset_index()\n rul.columns = ['UnitNumber', 'max']\n df_train = df_train.merge(rul, on=['UnitNumber'], how='left')\n df_train['RUL'] = df_train['max'] - df_train['Cycle']\n df_train.drop('max', axis=1, inplace=True)\n df_train['failure_lbl_1'] = [1 if i < 50 else 0 for i in df_train.RUL]\n df_train['failure_lbl_2'] = df_train['failure_lbl_1']\n df_train.failure_lbl_2[df_train.RUL < 25] = 2\n\n df_test = pd.read_csv('test_FD001.txt', delim_whitespace=True, names=input_file_column_names)\n \n if(drop_cols == True):\n df_train = df_train.drop(cols_to_drop, axis = 1)\n df_test = df_test.drop(cols_to_drop, axis = 1)\n\n y_true = pd.read_csv('RUL_FD001.txt', delim_whitespace=True,names=[\"RUL\"])\n y_true[\"UnitNumber\"] = y_true.index\n y_true['failure_lbl_1'] = [1 if i < 50 else 0 for i in y_true.RUL]\n y_true['failure_lbl_2'] = y_true['failure_lbl_1']\n y_true.failure_lbl_2[y_true.RUL < 25] = 2\n \n return df_train, df_test, y_true",
"_____no_output_____"
],
[
"df_train, df_test, y_true = prepare_data(drop_cols=True)\ndf_train.shape, df_test.shape, y_true.shape",
"C:\\Users\\archd\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:20: SettingWithCopyWarning: \nA value is trying to be set on a copy of a slice from a DataFrame\n\nSee the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy\n"
],
[
"feats = df_train.columns.drop(['UnitNumber', 'Cycle', 'RUL', 'failure_lbl_1', 'failure_lbl_2'])",
"_____no_output_____"
],
[
"min_max_scaler = MinMaxScaler(feature_range=(-1,1))\n\ndf_train[feats] = min_max_scaler.fit_transform(df_train[feats])\ndf_test[feats] = min_max_scaler.transform(df_test[feats])",
"_____no_output_____"
],
[
"df_train.head()",
"_____no_output_____"
],
[
"df_test.head()",
"_____no_output_____"
],
[
"y_true.head()",
"_____no_output_____"
]
],
[
[
"\nLSTM expects an input in the shape of a numpy array of 3 dimensions and I will need to convert train and test data accordingly.",
"_____no_output_____"
]
],
[
[
"def gen_train(id_df, seq_length, seq_cols):\n \"\"\"\n function to prepare train data into (samples, time steps, features)\n id_df = train dataframe\n seq_length = look back period\n seq_cols = feature columns\n \"\"\"\n \n data_array = id_df[seq_cols].values\n num_elements = data_array.shape[0]\n lstm_array=[]\n \n for start, stop in zip(range(0, num_elements-seq_length+1), range(seq_length, num_elements+1)):\n lstm_array.append(data_array[start:stop, :])\n \n return np.array(lstm_array)",
"_____no_output_____"
],
[
"def gen_target(id_df, seq_length, label):\n data_array = id_df[label].values\n num_elements = data_array.shape[0]\n return data_array[seq_length-1:num_elements+1]",
"_____no_output_____"
],
[
"def gen_test(id_df, seq_length, seq_cols, mask_value):\n \"\"\"\n function to prepare test data into (samples, time steps, features)\n function only returns last sequence of data for every unit\n id_df = test dataframe\n seq_length = look back period\n seq_cols = feature columns\n \"\"\"\n df_mask = pd.DataFrame(np.zeros((seq_length-1,id_df.shape[1])),columns=id_df.columns)\n df_mask[:] = mask_value\n \n id_df = df_mask.append(id_df,ignore_index=True)\n \n data_array = id_df[seq_cols].values\n num_elements = data_array.shape[0]\n lstm_array=[]\n \n start = num_elements-seq_length\n stop = num_elements\n \n lstm_array.append(data_array[start:stop, :])\n \n return np.array(lstm_array)",
"_____no_output_____"
]
],
[
[
"### Function to Print results",
"_____no_output_____"
]
],
[
[
"def print_results(y_test, y_pred, multi_class = False):\n \n #f1-score\n if multi_class == True:\n f1 = f1_score(y_test, y_pred, average=\"macro\")\n else:\n f1 = f1_score(y_test, y_pred)\n print(\"F1 Score: \", f1)\n print(classification_report(y_test, y_pred))\n \n conf_matrix = confusion_matrix(y_test, y_pred)\n plt.figure(figsize=(12,12))\n plt.subplot(221)\n sns.heatmap(conf_matrix, fmt = \"d\",annot=True, cmap='Blues')\n b, t = plt.ylim()\n plt.ylim(b + 0.5, t - 0.5)\n plt.title('Confuion Matrix')\n plt.ylabel('True Values')\n plt.xlabel('Predicted Values')\n\n #roc_auc_score\n if multi_class == False:\n model_roc_auc = roc_auc_score(y_test, y_pred) \n print (\"Area under curve : \",model_roc_auc,\"\\n\")\n fpr,tpr,thresholds = roc_curve(y_test, y_pred)\n gmeans = np.sqrt(tpr * (1-fpr))\n ix = np.argmax(gmeans)\n threshold = np.round(thresholds[ix],3)\n\n plt.subplot(222)\n plt.plot(fpr, tpr, color='darkorange', lw=1, label = \"Auc : %.3f\" %model_roc_auc)\n plt.plot([0, 1], [0, 1], color='navy', lw=2, linestyle='--')\n plt.scatter(fpr[ix], tpr[ix], marker='o', color='black', label='Best Threshold:' + str(threshold))\n plt.xlim([0.0, 1.0])\n plt.ylim([0.0, 1.05])\n plt.xlabel('False Positive Rate')\n plt.ylabel('True Positive Rate')\n plt.title('Receiver operating characteristic')\n plt.legend(loc=\"lower right\")",
"_____no_output_____"
]
],
[
[
"## Binary Classification",
"_____no_output_____"
]
],
[
[
"sequence_length = 50\nmask_value = 0\nlabel = \"failure_lbl_1\"",
"_____no_output_____"
]
],
[
[
"\nLet's prepare data using above functions.",
"_____no_output_____"
]
],
[
[
"#generate train\nx_train=np.concatenate(list(list(gen_train(df_train[df_train['UnitNumber']==unit], sequence_length, feats)) for unit in df_train['UnitNumber'].unique()))\nprint(x_train.shape)",
"(15731, 50, 15)\n"
],
[
"#generate target of train\ny_train = np.concatenate(list(list(gen_target(df_train[df_train['UnitNumber']==unit], sequence_length, label)) for unit in df_train['UnitNumber'].unique()))\ny_train.shape",
"_____no_output_____"
],
[
"#generate test\nx_test=np.concatenate(list(list(gen_test(df_test[df_test['UnitNumber']==unit], sequence_length, feats, mask_value)) for unit in df_test['UnitNumber'].unique()))\nprint(x_test.shape)",
"(100, 50, 15)\n"
],
[
"#true target of test \ny_test = y_true.RUL.values\ny_test.shape",
"_____no_output_____"
],
[
"nb_features = x_train.shape[2]\nnb_out = 1",
"_____no_output_____"
],
[
"nb_features",
"_____no_output_____"
],
[
"cls_wt= class_weight.compute_class_weight('balanced',np.unique(y_train), y_train)\ncls_wt",
"_____no_output_____"
]
],
[
[
"### Model",
"_____no_output_____"
]
],
[
[
"history = History()\n\nmodel = Sequential()\nmodel.add(SimpleRNN(16, input_shape=(sequence_length, nb_features),\n activation = 'relu'))\nmodel.add(Dense(8, activation = 'relu'))\nmodel.add(Dense(1, activation='sigmoid'))\n\nmodel.compile(loss=\"binary_crossentropy\", optimizer=\"adam\", metrics=['accuracy'])\n\nmodel.summary()",
"_________________________________________________________________\nLayer (type) Output Shape Param # \n=================================================================\nsimple_rnn_13 (SimpleRNN) (None, 16) 512 \n_________________________________________________________________\ndense_16 (Dense) (None, 8) 136 \n_________________________________________________________________\ndense_17 (Dense) (None, 1) 9 \n=================================================================\nTotal params: 657\nTrainable params: 657\nNon-trainable params: 0\n_________________________________________________________________\n"
],
[
"%%time\n# fit the model\nmodel.fit(x_train, y_train, epochs=100, batch_size=64, validation_split=0.2, verbose=1, class_weight = cls_wt,\n callbacks = [history,\n keras.callbacks.EarlyStopping(monitor='val_loss', min_delta=0, patience=5, verbose=0, mode='auto')])",
"Train on 12584 samples, validate on 3147 samples\nEpoch 1/100\n12584/12584 [==============================] - 6s 478us/step - loss: 0.3399 - acc: 0.8586 - val_loss: 0.2351 - val_acc: 0.8958\nEpoch 2/100\n12584/12584 [==============================] - 3s 249us/step - loss: 0.2144 - acc: 0.9074 - val_loss: 0.2258 - val_acc: 0.9047\nEpoch 3/100\n12584/12584 [==============================] - 3s 246us/step - loss: 0.2030 - acc: 0.9135 - val_loss: 0.2214 - val_acc: 0.9059\nEpoch 4/100\n12584/12584 [==============================] - 3s 249us/step - loss: 0.1943 - acc: 0.9163 - val_loss: 0.2152 - val_acc: 0.9053\nEpoch 5/100\n12584/12584 [==============================] - 3s 247us/step - loss: 0.1856 - acc: 0.9197 - val_loss: 0.2076 - val_acc: 0.9018\nEpoch 6/100\n12584/12584 [==============================] - 3s 242us/step - loss: 0.1724 - acc: 0.9279 - val_loss: 0.1800 - val_acc: 0.9196\nEpoch 7/100\n12584/12584 [==============================] - 3s 249us/step - loss: 0.1585 - acc: 0.9309 - val_loss: 0.1677 - val_acc: 0.9234\nEpoch 8/100\n12584/12584 [==============================] - 3s 248us/step - loss: 0.1465 - acc: 0.9381 - val_loss: 0.1575 - val_acc: 0.9317\nEpoch 9/100\n12584/12584 [==============================] - 3s 246us/step - loss: 0.1435 - acc: 0.9394 - val_loss: 0.1790 - val_acc: 0.9263\nEpoch 10/100\n12584/12584 [==============================] - 3s 251us/step - loss: 0.1358 - acc: 0.9447 - val_loss: 0.1901 - val_acc: 0.9314\nEpoch 11/100\n12584/12584 [==============================] - 3s 249us/step - loss: 0.1322 - acc: 0.9464 - val_loss: 0.1545 - val_acc: 0.9323\nEpoch 12/100\n12584/12584 [==============================] - 3s 248us/step - loss: 0.1282 - acc: 0.9456 - val_loss: 0.1621 - val_acc: 0.9307\nEpoch 13/100\n12584/12584 [==============================] - 3s 275us/step - loss: 0.1300 - acc: 0.9472 - val_loss: 0.1841 - val_acc: 0.9152\nEpoch 14/100\n12584/12584 [==============================] - 3s 247us/step - loss: 0.1245 - acc: 0.9499 - val_loss: 0.2081 - val_acc: 0.9148\nEpoch 15/100\n12584/12584 [==============================] - 3s 265us/step - loss: 0.1232 - acc: 0.9507 - val_loss: 0.1767 - val_acc: 0.9339\nEpoch 16/100\n12584/12584 [==============================] - 3s 248us/step - loss: 0.1217 - acc: 0.9478 - val_loss: 0.1932 - val_acc: 0.9295\nWall time: 54.3 s\n"
],
[
"fig, ax = plt.subplots(nrows = 1, ncols = 2, figsize = (10, 4))\n\n# Accuracy\nax[0].plot(history.history['acc'])\nax[0].plot(history.history['val_acc'])\nax[0].set_ylabel('Accuracy')\nax[0].set_xlabel('# Epoch')\nax[0].legend(['train', 'validation'], loc='upper left')\nax[0].set_title('Accuracy')\n\n# Loss\nax[1].plot(history.history['loss'])\nax[1].plot(history.history['val_loss'])\nax[1].set_ylabel('Loss')\nax[1].set_xlabel('# Epoch')\nax[1].legend(['train', 'validation'], loc='upper left')\nax[1].set_title('Loss')",
"_____no_output_____"
],
[
"y_pred = model.predict_classes(x_test)\n\nprint_results(y_true.failure_lbl_1, y_pred)",
"F1 Score: 0.9354838709677419\n precision recall f1-score support\n\n 0 0.99 0.96 0.97 70\n 1 0.91 0.97 0.94 30\n\n accuracy 0.96 100\n macro avg 0.95 0.96 0.95 100\nweighted avg 0.96 0.96 0.96 100\n\nArea under curve : 0.9619047619047619 \n\n"
],
[
"confusion_matrix(y_train, model.predict_classes(x_train))",
"_____no_output_____"
]
],
[
[
"## Multiclass Classification",
"_____no_output_____"
]
],
[
[
"sequence_length = 50\nmask_value = 0\nlabel = \"failure_lbl_2\"",
"_____no_output_____"
]
],
[
[
"\nLet's prepare data using above functions.",
"_____no_output_____"
]
],
[
[
"#generate train\nx_train=np.concatenate(list(list(gen_train(df_train[df_train['UnitNumber']==unit], sequence_length, feats)) for unit in df_train['UnitNumber'].unique()))\nprint(x_train.shape)",
"(15731, 50, 15)\n"
],
[
"#generate target of train\ny_train = np.concatenate(list(list(gen_target(df_train[df_train['UnitNumber']==unit], sequence_length, label)) for unit in df_train['UnitNumber'].unique()))\ny_train.shape",
"_____no_output_____"
],
[
"y_train2 = to_categorical(y_train)\ny_train2",
"_____no_output_____"
],
[
"#generate test\nx_test=np.concatenate(list(list(gen_test(df_test[df_test['UnitNumber']==unit], sequence_length, feats, mask_value)) for unit in df_test['UnitNumber'].unique()))\nprint(x_test.shape)",
"(100, 50, 15)\n"
],
[
"nb_features = x_train.shape[2]\nnb_out = 1",
"_____no_output_____"
],
[
"nb_features",
"_____no_output_____"
],
[
"cls_wt= class_weight.compute_class_weight('balanced',np.unique(y_train), y_train)\ncls_wt",
"_____no_output_____"
]
],
[
[
"### Model",
"_____no_output_____"
]
],
[
[
"history2 = History()\n\nmodel2 = Sequential()\nmodel2.add(SimpleRNN(16, input_shape=(sequence_length, nb_features),\n activation = 'relu'))\nmodel2.add(Dense(8, activation = 'relu'))\nmodel2.add(Dense(units=3, activation='softmax'))\n\nmodel2.compile(loss=\"categorical_crossentropy\", optimizer=\"adam\", metrics=['accuracy'])\n\nmodel2.summary()",
"_________________________________________________________________\nLayer (type) Output Shape Param # \n=================================================================\nsimple_rnn_14 (SimpleRNN) (None, 16) 512 \n_________________________________________________________________\ndense_18 (Dense) (None, 8) 136 \n_________________________________________________________________\ndense_19 (Dense) (None, 3) 27 \n=================================================================\nTotal params: 675\nTrainable params: 675\nNon-trainable params: 0\n_________________________________________________________________\n"
],
[
"%%time\n# fit the model\nmodel2.fit(x_train, y_train2, epochs=10, batch_size=64, validation_split=0.2, verbose=1, class_weight = cls_wt,\n callbacks = [history2,\n keras.callbacks.EarlyStopping(monitor='val_loss', min_delta=0, patience=3, verbose=0, mode='auto')])",
"Train on 12584 samples, validate on 3147 samples\nEpoch 1/10\n12584/12584 [==============================] - 6s 441us/step - loss: 0.5184 - acc: 0.8041 - val_loss: 0.3455 - val_acc: 0.8494\nEpoch 2/10\n12584/12584 [==============================] - 3s 239us/step - loss: 0.3096 - acc: 0.8685 - val_loss: 0.2812 - val_acc: 0.8837\nEpoch 3/10\n12584/12584 [==============================] - 2s 195us/step - loss: 0.2234 - acc: 0.9078 - val_loss: 0.1940 - val_acc: 0.9171\nEpoch 4/10\n12584/12584 [==============================] - 3s 218us/step - loss: 0.1838 - acc: 0.9225 - val_loss: 0.1748 - val_acc: 0.9304\nEpoch 5/10\n12584/12584 [==============================] - 2s 188us/step - loss: 0.1689 - acc: 0.9282 - val_loss: 0.1633 - val_acc: 0.9355\nEpoch 6/10\n12584/12584 [==============================] - 3s 210us/step - loss: 0.1668 - acc: 0.9270 - val_loss: 0.1803 - val_acc: 0.9218\nEpoch 7/10\n12584/12584 [==============================] - 3s 250us/step - loss: 0.1588 - acc: 0.9296 - val_loss: 0.1747 - val_acc: 0.9304\nEpoch 8/10\n12584/12584 [==============================] - 3s 246us/step - loss: 0.1587 - acc: 0.9332 - val_loss: 0.1874 - val_acc: 0.9282\nWall time: 25.8 s\n"
],
[
"fig, ax = plt.subplots(nrows = 1, ncols = 2, figsize = (10, 4))\n\n# Accuracy\nax[0].plot(history2.history['acc'])\nax[0].plot(history2.history['val_acc'])\nax[0].set_ylabel('Accuracy')\nax[0].set_xlabel('# Epoch')\nax[0].legend(['train', 'validation'], loc='upper left')\nax[0].set_title('Accuracy')\n\n# Loss\nax[1].plot(history2.history['loss'])\nax[1].plot(history2.history['val_loss'])\nax[1].set_ylabel('Loss')\nax[1].set_xlabel('# Epoch')\nax[1].legend(['train', 'validation'], loc='upper left')\nax[1].set_title('Loss')",
"_____no_output_____"
],
[
"y_pred = model2.predict_classes(x_test)\n\nprint_results(y_true.failure_lbl_2, y_pred, multi_class=True)",
"F1 Score: 0.8692412850307587\n precision recall f1-score support\n\n 0 0.99 0.99 0.99 70\n 1 0.73 0.73 0.73 11\n 2 0.89 0.89 0.89 19\n\n accuracy 0.94 100\n macro avg 0.87 0.87 0.87 100\nweighted avg 0.94 0.94 0.94 100\n\n"
],
[
"confusion_matrix(y_train, model2.predict_classes(x_train))",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code"
] |
[
[
"markdown",
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code",
"code",
"code"
],
[
"markdown"
],
[
"code",
"code",
"code",
"code",
"code"
]
] |
cbdc94c81e806f924896146aa8684800d52b4ed9
| 15,678 |
ipynb
|
Jupyter Notebook
|
sphinx/datascience/source/regression-errors.ipynb
|
oneoffcoder/books
|
84619477294a3e37e0d7538adf819113c9e8dcb8
|
[
"CC-BY-4.0"
] | 26 |
2020-05-05T08:07:43.000Z
|
2022-02-12T03:28:15.000Z
|
sphinx/datascience/source/regression-errors.ipynb
|
oneoffcoder/books
|
84619477294a3e37e0d7538adf819113c9e8dcb8
|
[
"CC-BY-4.0"
] | 19 |
2021-03-10T00:33:51.000Z
|
2022-03-02T13:04:32.000Z
|
sphinx/datascience/source/regression-errors.ipynb
|
oneoffcoder/books
|
84619477294a3e37e0d7538adf819113c9e8dcb8
|
[
"CC-BY-4.0"
] | 2 |
2022-01-09T16:48:21.000Z
|
2022-02-19T17:06:50.000Z
| 31.736842 | 511 | 0.473785 |
[
[
[
"# Regression Errors\n\nLet's talk about errors in regression problems. Typically, in regression, we have a variable $y$ for which we want to learn a model to predict. The prediction from the model is usually denoted as $\\hat{y}$. The error $e$ is thus defined as follows\n\n- $e = y - \\hat{y}$\n\nSince we have many pairs of the truth, $y$ and $\\hat{y}$, we want to average over the differences. I will denote this error as the Mean Error `ME`.\n\n- $\\mathrm{ME} = \\frac{1}{n} \\sum{y - \\hat{y}}$\n\nThe problem with ME is that averaging over the differences may result in something close to zero. The reason is because the positive and negative differences will have a cancelling effect. No one really computes the error of a regression model in this way.\n\nA better way is to consider the Mean Absolute Error `MAE`, where we take the average of the absolute differences.\n\n- $\\mathrm{MAE} = \\frac{1}{n} \\sum |y - \\hat{y}|$\n\nIn MAE, since there are only positive differences resulting from $|y - \\hat{y}|$, we avoid the cancelling effect of positive and negative values when averaging. Many times, data scientists want to punish models that predict values further from the truth. In that case, the Root Mean Squared Error `RMSE` is used.\n\n- $\\mathrm{RMSE} = \\sqrt{\\frac{1}{n} \\sum (y - \\hat{y})^2}$\n\nIn RMSE, we do not take the difference as in ME or the absolute difference as in MAE, rather, we square the difference. The idea is that when a model's prediction is off from the truth, we should exaggerate the consequences as it reflects the reality that being further away from the truth is orders of magnitude worse. However, the squaring of the difference results in something that is no longer in the unit of $y$, as such, we take the square root to bring the scalar value back into unit with $y$.\n\nFor all these measures of performance, the closer the value is to zero, the better.\n\nLet's look at the following made-up example where a hypothetical model has made some prediction $\\hat{y}$ or `y_pred` and for each of these prediction, we have the ground truth $y$ or `y_true`.",
"_____no_output_____"
]
],
[
[
"import pandas as pd\n\ndf = pd.DataFrame({\n 'y_true': [10, 8, 7, 9, 4],\n 'y_pred': [11, 7, 6, 15, 1]\n})\n\ndf = pd.DataFrame({\n 'y_true': [10, 8, 7, 9, 4],\n 'y_pred': [11, 7, 5, 11, 1]\n})\n\ndf",
"_____no_output_____"
]
],
[
[
"We will now compute the error `E`, absolute error `AE` and squared errors `SE` for each pair.",
"_____no_output_____"
]
],
[
[
"import numpy as np\n\ndf['E'] = df.y_true - df.y_pred\ndf['AE'] = np.abs(df.y_true - df.y_pred)\ndf['SE'] = np.power(df.y_true - df.y_pred, 2.0)\n\ndf",
"_____no_output_____"
]
],
[
[
"From E, AE and SE, we can compute the average or mean errors, ME, MAE, RMSE, respectively, as follows.",
"_____no_output_____"
]
],
[
[
"errors = df[['E', 'AE', 'SE']].mean()\nerrors.se = np.sqrt(errors.SE)\n\nerrors.index = ['ME', 'MAE', 'RMSE']\nerrors",
"_____no_output_____"
]
],
[
[
"As you can see, these judgement of errors are saying different things and might lead you to draw contradictory and/or conflicting conclusions. We know ME is defective, and so we will ignore interpreting ME. MAE says we can expect to be `2.4` off from the truth while RMSE says we can expect to be `9.6` off from the truth. The values `2.4` and `9.6` are very different; while `2.4` may seem to be tolerably `good`, on the other hand, `9.6` seems `bad`.\n\nOne thing we can try to do is to `normalize` these values. Let's just look at RMSE. Here are some ways we can normalize RMSE.\n\n- using the `mean` of y, denoted as $\\bar{y}$\n- using the `standard deviation` of y, denoted as $\\sigma_y$\n- using the range of y, denoted as $y_{\\mathrm{max}} - y_{\\mathrm{min}}$\n- using the interquartile range of y, denoted as $Q_y^1 - Q_y^3$\n\nThe code to compute these is as follows.\n\n- $\\bar{y}$ is `me_y`\n- $\\sigma_y$ is `sd_y`\n- $y_{\\mathrm{max}} - y_{\\mathrm{min}}$ is `ra_y`\n- $Q_y^1 - Q_y^3$ is `iq_y`\n\nSince these are used to divide RMSE, let's group them under a series as `denominators`.",
"_____no_output_____"
]
],
[
[
"from scipy.stats import iqr\n\nme_y = df.y_true.mean()\nsd_y = df.y_true.std()\nra_y = df.y_true.max() - df.y_true.min()\niq_y = iqr(df.y_true)\n\ndenominators = pd.Series([me_y, sd_y, ra_y, iq_y], index=['me_y', 'sd_y', 'ra_y', 'iq_y'])\ndenominators",
"_____no_output_____"
]
],
[
[
"Here's the results of normalizing RMSE with the mean `me`, standard deviation `sd`, range `ra` and interquartile range `iq`.",
"_____no_output_____"
]
],
[
[
"pd.DataFrame([{\n r'$\\mathrm{RMSE}_{\\mathrm{me}}$': errors.RMSE / denominators.me_y,\n r'$\\mathrm{RMSE}_{\\mathrm{sd}}$': errors.RMSE / denominators.sd_y,\n r'$\\mathrm{RMSE}_{\\mathrm{ra}}$': errors.RMSE / denominators.ra_y,\n r'$\\mathrm{RMSE}_{\\mathrm{iq}}$': errors.RMSE / denominators.iq_y,\n}]).T.rename(columns={0: 'values'})",
"_____no_output_____"
]
],
[
[
"That we have normalized RMSE, we can make a little bit better interpretation.\n\n- $\\mathrm{RMSE}_{\\mathrm{me}}$ is saying we can expect to be 126% away from the truth.\n- $\\mathrm{RMSE}_{\\mathrm{sd}}$ is saying we can expect to be over 4.2 standard deviation from the truth.\n- $\\mathrm{RMSE}_{\\mathrm{ra}}$ is saying we can expect to be 1.6 \n- $\\mathrm{RMSE}_{\\mathrm{iq}}$",
"_____no_output_____"
]
]
] |
[
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown",
"code",
"markdown"
] |
[
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
],
[
"code"
],
[
"markdown"
]
] |
cbdcba2aa092922727dfadd14f47262720419f00
| 1,191 |
ipynb
|
Jupyter Notebook
|
rec_gee_ui.ipynb
|
sepal-contrib/reclassify
|
a2fd260ff9ea3496d456621b6c9b234d23aa903e
|
[
"MIT"
] | null | null | null |
rec_gee_ui.ipynb
|
sepal-contrib/reclassify
|
a2fd260ff9ea3496d456621b6c9b234d23aa903e
|
[
"MIT"
] | null | null | null |
rec_gee_ui.ipynb
|
sepal-contrib/reclassify
|
a2fd260ff9ea3496d456621b6c9b234d23aa903e
|
[
"MIT"
] | null | null | null | 18.609375 | 48 | 0.539043 |
[
[
[
"from component import tile",
"_____no_output_____"
],
[
"rec_gee = tile.RecTile(gee=True)",
"_____no_output_____"
],
[
"rec_gee",
"_____no_output_____"
]
]
] |
[
"code"
] |
[
[
"code",
"code",
"code"
]
] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.