app.py

# %%
import gradio as gr
from sklearn.ensemble import RandomForestRegressor
import numpy as np
import pandas as pd
import pickle
# Define model filename
model_filename = "random_forest_regression_extended.pkl"

try:
    # Try to load the model
    with open(model_filename, 'rb') as f:
        model_data = pickle.load(f)
        if isinstance(model_data, dict) and 'model' in model_data and 'feature_names' in model_data:
            random_forest_model = model_data['model']
            feature_names = model_data['feature_names']
            # Check scikit-learn version and handle feature information
            if hasattr(random_forest_model, 'n_features_in_'):
                print('Number of features: ', random_forest_model.n_features_in_)
            else:
                print('Number of features: ', len(feature_names))
            print('Features are: ', feature_names)
        else:
            print("Error: Model file does not contain expected dictionary structure")
            print("Expected keys: 'model' and 'feature_names'")
            print(f"Found keys: {model_data.keys() if isinstance(model_data, dict) else 'not a dictionary'}")
            exit(1)
except FileNotFoundError:
    print(f"Error: Could not find model file '{model_filename}'")
    print("Please run save_model.py first to create the model file.")
    exit(1)
except Exception as e:
    print(f"Error loading model: {str(e)}")
    print(f"scikit-learn version: {sklearn.__version__}")
    exit(1)

# Load and prepare BFS data
df_bfs_data = pd.read_csv('bfs_municipality_and_tax_data.csv', sep=',', encoding='utf-8')
df_bfs_data['tax_income'] = df_bfs_data['tax_income'].str.replace("'", "").astype(float)
df_bfs_data['proximity_to_public_transportation'] = 500  # Default value in meters


# %%
locations = {
    "Zürich": 261,
    "Kloten": 62,
    "Uster": 198,
    "Illnau-Effretikon": 296,
    "Feuerthalen": 27,
    "Pfäffikon": 177,
    "Ottenbach": 11,
    "Dübendorf": 191,
    "Richterswil": 138,
    "Maur": 195,
    "Embrach": 56,
    "Bülach": 53,
    "Winterthur": 230,
    "Oetwil am See": 157,
    "Russikon": 178,
    "Obfelden": 10,
    "Wald (ZH)": 120,
    "Niederweningen": 91,
    "Dällikon": 84,
    "Buchs (ZH)": 83,
    "Rüti (ZH)": 118,
    "Hittnau": 173,
    "Bassersdorf": 52,
    "Glattfelden": 58,
    "Opfikon": 66,
    "Hinwil": 117,
    "Regensberg": 95,
    "Langnau am Albis": 136,
    "Dietikon": 243,
    "Erlenbach (ZH)": 151,
    "Kappel am Albis": 6,
    "Stäfa": 158,
    "Zell (ZH)": 231,
    "Turbenthal": 228,
    "Oberglatt": 92,
    "Winkel": 72,
    "Volketswil": 199,
    "Kilchberg (ZH)": 135,
    "Wetzikon (ZH)": 121,
    "Zumikon": 160,
    "Weisslingen": 180,
    "Elsau": 219,
    "Hettlingen": 221,
    "Rüschlikon": 139,
    "Stallikon": 13,
    "Dielsdorf": 86,
    "Wallisellen": 69,
    "Dietlikon": 54,
    "Meilen": 156,
    "Wangen-Brüttisellen": 200,
    "Flaach": 28,
    "Regensdorf": 96,
    "Niederhasli": 90,
    "Bauma": 297,
    "Aesch (ZH)": 241,
    "Schlieren": 247,
    "Dürnten": 113,
    "Unterengstringen": 249,
    "Gossau (ZH)": 115,
    "Oberengstringen": 245,
    "Schleinikon": 98,
    "Aeugst am Albis": 1,
    "Rheinau": 38,
    "Höri": 60,
    "Rickenbach (ZH)": 225,
    "Rafz": 67,
    "Adliswil": 131,
    "Zollikon": 161,
    "Urdorf": 250,
    "Hombrechtikon": 153,
    "Birmensdorf (ZH)": 242,
    "Fehraltorf": 172,
    "Weiach": 102,
    "Männedorf": 155,
    "Küsnacht (ZH)": 154,
    "Hausen am Albis": 4,
    "Hochfelden": 59,
    "Fällanden": 193,
    "Greifensee": 194,
    "Mönchaltorf": 196,
    "Dägerlen": 214,
    "Thalheim an der Thur": 39,
    "Uetikon am See": 159,
    "Seuzach": 227,
    "Uitikon": 248,
    "Affoltern am Albis": 2,
    "Geroldswil": 244,
    "Niederglatt": 89,
    "Thalwil": 141,
    "Rorbas": 68,
    "Pfungen": 224,
    "Weiningen (ZH)": 251,
    "Bubikon": 112,
    "Neftenbach": 223,
    "Mettmenstetten": 9,
    "Otelfingen": 94,
    "Flurlingen": 29,
    "Stadel": 100,
    "Grüningen": 116,
    "Henggart": 31,
    "Dachsen": 25,
    "Bonstetten": 3,
    "Bachenbülach": 51,
    "Horgen": 295
}
def predict_apartment(rooms, area, proximity, town):
    bfs_number = locations[town]
    df = df_bfs_data[df_bfs_data['bfs_number']==bfs_number].copy()
    df.reset_index(inplace=True)
    df.loc[0, 'rooms'] = rooms
    df.loc[0, 'area'] = area
    df.loc[0, 'proximity_to_public_transportation'] = proximity  # Use user input instead of default
    
    if len(df) != 1:
        return -1
        
    features = ['rooms', 'area', 'pop', 'pop_dens', 'frg_pct', 'emp', 'tax_income', 'proximity_to_public_transportation']
    X = df[features].values
    prediction = random_forest_model.predict(X)
    return np.round(prediction[0], 0)

# Create the Gradio interface
iface = gr.Interface(
    fn=predict_apartment,
    inputs=[
        gr.Number(label="Number of Rooms"),
        gr.Number(label="Area"),
        gr.Slider(minimum=0, maximum=2000, value=500, step=50, 
                 label="Distance to Public Transportation (meters)"),
        gr.Dropdown(choices=locations.keys(), label="Town", type="value")
    ],
    outputs=[gr.Number(label="Predicted Price (CHF)")],
    examples=[
        [4.5, 120, 500, "Dietlikon"], 
        [3.5, 60, 250, "Winterthur"]
    ],
    description="Predict apartment prices in Zürich based on rooms, area, proximity to public transportation, and location."
)
iface.launch()