File size: 9,745 Bytes
554d184
23bce77
 
 
7d55365
554d184
0e52835
7d55365
0e52835
 
 
 
 
 
 
 
7d55365
 
0e52835
7d55365
 
 
 
 
 
 
0e52835
 
 
7d55365
0e52835
 
 
 
 
 
 
 
 
80c3833
0e52835
 
 
 
 
 
 
 
 
7d55365
0e52835
 
 
7d55365
 
80c3833
0e52835
80c3833
0e52835
 
 
7d55365
0e52835
80c3833
 
7d55365
23bce77
7d55365
23bce77
0d73c4d
23bce77
92944d2
80c3833
 
7d55365
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
80c3833
 
 
 
 
 
 
7d55365
0e52835
7d55365
 
 
 
 
 
 
 
 
80c3833
23bce77
7d55365
0e52835
 
23bce77
 
0e52835
80c3833
0e52835
 
 
 
 
 
 
 
 
 
7d55365
92944d2
0e52835
 
 
92944d2
7d55365
554d184
 
23bce77
80c3833
23bce77
80c3833
23bce77
 
 
 
5b272d7
b11e6cc
 
b907c28
 
 
5b272d7
b907c28
23bce77
 
 
7d55365
 
23bce77
 
7d55365
23bce77
0d73c4d
7d55365
 
 
 
 
 
 
 
 
0d73c4d
7d55365
23bce77
80c3833
23bce77
 
 
 
 
 
 
 
 
7d55365
23bce77
7d55365
 
23bce77
 
 
80c3833
23bce77
 
 
 
7d55365
 
 
 
 
 
 
 
 
23bce77
80c3833
7d55365
 
0e52835
80c3833
7d55365
0e52835
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
import gradio as gr
import json
import os
import random
import datasets

from dipromats_evaluation_v2 import evaluate_results

# CONSTANTS

# Hugging Face datasets

DATASET_GOLD = "NLP-UNED/dipromats2024-t2_leaderboard-gold"
FILE_GOLD = 'gold_test.json'

DATASET_RESULTS = "NLP-UNED/dipromats2024-t2_leaderboard-results"
SPLIT_EN = 'results_en'
SPLIT_ES = 'results_es'
FEATURES_RESULTS = datasets.Features({
        "team_name": datasets.Value("string"),
        "run_id": datasets.Value("string"),
        "description": datasets.Value("string"),
        "lenient_f1": datasets.Value("float64"),
        "strict_f1": datasets.Value("float64"),
        "average_f1": datasets.Value("float64") })

EMPTY_RESULT={"team_name": [], "run_id": [], "description": [], "lenient_f1": [], "strict_f1": [], "average_f1": []}

# Before, you must create the Tokens in HF User Settings to give read and write access only to the datasets
try:
    from google.colab import userdata
    # Token must be copied and activated in Colab Secrets
    HF_TOKEN_GOLD = userdata.get('HF_DIPROMATS2024_T2_GOLD_TOKEN')
    HF_TOKEN_RESULTS = userdata.get('HF_DIPROMATS2024_T2_RESULTS_TOKEN')
except:
    # Assume running in HF Space
    # Tokens must be copied in Secrets under Space Settings
    HF_TOKEN_GOLD = os.getenv('HF_DIPROMATS2024_T2_GOLD_TOKEN')
    HF_TOKEN_RESULTS = os.getenv('HF_DIPROMATS2024_T2_RESULTS_TOKEN')

# LOAD DATASETS

# Load the Gold Standard data
# FILE_GOLD was uploaded directly through HF web, and the default split is train
dataset_gold = datasets.load_dataset(DATASET_GOLD, split='train', data_files=FILE_GOLD, token=HF_TOKEN_GOLD)
        
# Load the English dataset or create an empty one instead
try:
    dataset_en = datasets.load_dataset(DATASET_RESULTS, split=SPLIT_EN)
except Exception as e:
    print(f"Error loading English dataset: {e}. Creating it...")
    dataset_en = datasets.Dataset.from_dict(EMPTY_RESULT, features=FEATURES_RESULTS, split=SPLIT_EN)
    dataset_en.push_to_hub(DATASET_RESULTS, split=SPLIT_EN, token=HF_TOKEN_RESULTS)

# Load the Spanish dataset or create an empty one instead
try:
    dataset_es = datasets.load_dataset(DATASET_RESULTS, split=SPLIT_ES)
except Exception as e:
    print(f"Error loading Spanish dataset: {e}. Creating it...")
    dataset_es = datasets.Dataset.from_dict(EMPTY_RESULT, features=FEATURES_RESULTS, split=SPLIT_ES)
    dataset_es.push_to_hub(DATASET_RESULTS, split=SPLIT_ES, token=HF_TOKEN_RESULTS)

# AUX FUNCTIONS

# Funci贸n para convertir el dataset en tabla
def data_to_table(dataset):
    table_data = []
    for item in dataset:
        table_data.append([item.get("team_name", ""), item.get("run_id", ""),
                           item.get("lenient_f1", ""), item.get("strict_f1", ""), item.get("average_f1", ""), item.get("description", "")])
    return table_data


# Funci贸n para subir los resultados al leaderboard
def update_leaderboard(lang, file_path, email, team_input, run_id, description, lenient_f1, strict_f1, average_f1):
    global dataset_en
    global dataset_es
    if lang == "en":
        dataset = dataset_en
    else:
        dataset = dataset_es

    warn = False
    if not email:
        gr.Warning("Email cannot be blank")
        warn=True
    if not team_input:
        gr.Warning("Team name cannot be blank")
        warn=True
    if not run_id:
        gr.Warning("Run ID cannot be blank")
        warn=True
    if not file_path:
        gr.Warning("File cannot be blank")
        warn=True
    if not description:
        gr.Warning("Description cannot be blank")
        warn=True

    if warn:
        return data_to_table(dataset_en), data_to_table(dataset_es), gr.Tabs(selected=1), gr.Button(visible=False), gr.Column(visible=True), team_input, run_id, description, email, file_path, lenient_f1, strict_f1, average_f1

    dataset = dataset.add_item({
        "team_name": team_input,
        "run_id": run_id,
        "description": description,
        "lenient_f1": lenient_f1,
        "strict_f1": strict_f1,
        "average_f1": average_f1
    })
    # Save change in database
    dataset.push_to_hub(DATASET_RESULTS, token=HF_TOKEN_RESULTS)

    # Update dataset in memory
    if lang == "en":
        dataset_en = dataset
    else:
        dataset_es = dataset

    #output: leaderboard_table, tabs, evaluate_button, submission_col, team_input, run_id, description_input, email_input, file_input, lenient_f1, strict_f1, average_f1
    return data_to_table(dataset_en), data_to_table(dataset_es), gr.Tabs(selected=0), gr.Button(visible=True), gr.Column(visible=False), "", "", "", "", None, None, None, None

# Funci贸n para procesar el archivo de resultados
def process_file(lang, file_path):
    global dataset_gold

    if not file_path:
        gr.Warning("File cannot be blank")
        return gr.Button(visible=True), gr.Row(visible=False), None, None, None

    with open(file_path, 'r') as f:
        test = json.load(f)

    try:
        results = evaluate_results(lang, dataset_gold, test)
        #print(results)
    except Exception as e:
        gr.Warning("Invalid JSON file or Incorrect Language")
        print(f"Error in function evaluate_results: {e}.")
        print(dataset_gold)
        return gr.Button(visible=True), gr.Row(visible=False), None, None, None

    lenient_f1 = results['lenient']['micro']['scores']['f1-score']
    strict_f1 = results['strict']['micro']['scores']['f1-score']
    average_f1 = (lenient_f1 + strict_f1) / 2

    return gr.Button(visible=False), gr.Row(visible=True), lenient_f1, strict_f1, average_f1


# Main

with gr.Blocks() as leaderboard:

    gr.Markdown(
        """
        # Dipromats 2024 Task 2 Leaderboard
        # Automatic Detection of Narratives from Diplomats of Major Powers
        These are the leaderboards for DIPROMATS 2024 Task 2 described in <a href=https://nlp.uned.es/dipromats2024>nlp.uned.es/dipromats2024</a>.   
        The Gold Standard is not publicly available so LLMs cannot be contamined with them. However, the datasets are available at https://nlp.uned.es/hamison-project/results.html      
        You can submit your results here and get your system automatically evaluated. Then you will have the choice to submit your results to the leaderboard.   
        The HuggingFace datasets of results displayed here are:   
        - https://huggingface.co/datasets/NLP-UNED/dipromats2024-t2_leaderboard-results/viewer/default/results_en   
        - https://huggingface.co/datasets/NLP-UNED/dipromats2024-t2_leaderboard-results/viewer/default/results_es   
           
              
        """)
    with gr.Tabs() as tabs:

        # Tab English Leaderboard
        with gr.TabItem("English Leaderboard", id=0):
            gr.Markdown(
                """
                # English Leaderboard
                """)
            leaderboard_table_en = gr.Dataframe(headers=["Team", "Run ID", "Lenient F1", "Strict F1", "Average F1", "System Description"],
                        value=data_to_table(dataset_en),
                        interactive=False)

        # Tab Spanish Leaderboard
        with gr.TabItem("Spanish Leaderboard", id=2):
            gr.Markdown(
                """
                # Spanish Leaderboard
                """)
            leaderboard_table_es = gr.Dataframe(headers=["Team", "Run ID", "Lenient F1", "Strict F1", "Average F1", "System Description"],
                        value=data_to_table(dataset_es),
                        interactive=False)

        # Tab Evaluate
        with gr.TabItem("Evaluate your results", id=1):
            gr.Markdown(
                """
                # Upload your results and get evaluated
                Then you can decide to submit your results to the leaderboard or not.
                Make sure that you upload a file with the json format described in...
                """)
            with gr.Row():
                file_input = gr.File(label="Upload a JSON file", file_types=[".json"], type="filepath", file_count="single")
                with gr.Column():
                    lang = gr.Dropdown(label="Language", choices=["en", "es"], interactive=True)
                    evaluate_button = gr.Button("Evaluate")

            # System results table
            with gr.Row(visible=True):
                lenient_f1 = gr.Number(label="Lenient F1", interactive=False)
                strict_f1 = gr.Number(label="Strict F1", interactive=False)
                average_f1 = gr.Number(label="Average F1", interactive=False)

            # Submit to leaderboard
            with gr.Column(visible=False) as submission_col:
                with gr.Row():
                    with gr.Column():
                        with gr.Row():
                            team_input = gr.Textbox(label="Team Name")
                            run_id = gr.Textbox(label="Run ID")
                        email_input = gr.Textbox(label="Email (only for submission verification, it won't be shown)")
                    description_input = gr.Textbox(label="System description", lines=6)
                submit_button = gr.Button("Submit to leaderboard")

    evaluate_button.click(process_file,
                        inputs=[lang, file_input],
                        outputs=[evaluate_button, submission_col,lenient_f1, strict_f1, average_f1])

    submit_button.click(update_leaderboard,
                        inputs=[lang, file_input, email_input, team_input, run_id, description_input, lenient_f1, strict_f1, average_f1],
                        outputs=[leaderboard_table_en, leaderboard_table_es, tabs, evaluate_button, submission_col, team_input, run_id, description_input, email_input, file_input, lenient_f1, strict_f1, average_f1])

leaderboard.launch()