Update app.py

app.py

@@ -7,6 +7,7 @@ from sklearn.ensemble import RandomForestRegressor
 from huggingface_hub import hf_hub_download
 from gradio_calendar import Calendar
 from datetime import datetime
+from datasets import load_dataset
 
 REPO_ID = "Koaris/rf_france_07042024"
 FILENAME = "random_forest_france_07_04_2024.pkl"
@@ -19,6 +20,16 @@ def get_lat_lon(address,api_key=API_KEY):
     coordinates = resp_json_payload['results'][0]['geometry']['location']
     return (coordinates['lat'], coordinates['lng'])
 
+def compute_reliability(std_pred: float):
+    scores = load_dataset('Koaris/quantiles_predict/scores_13072024.csv')
+    scores = scores.tolist()
+    ct = 5
+    for std_set in scores:
+        if std_pred < std_set:
+            return ct
+        else:
+            ct-=1
+
 def predict_price(date: datetime ,room_count: int, address:str, surface: float , property_type: str ):
     date = int(datetime.timestamp(date))
     print(date)
@@ -27,8 +38,18 @@ def predict_price(date: datetime ,room_count: int, address:str, surface: float ,
     rf_input = pd.DataFrame([{"date_mutation": date, "nombre_pieces_principales": room_count, "longitude" : longitude,"latitude":latitude, "surface_batie_totale": surface, "type_local_Maison": isHouse}])
     print(rf_input)
     rf_pred = rf.predict(rf_input)[0]
-    print(np.exp(rf_pred))
-    return float(np.exp(rf_pred))
+    predictions_all = np.array([tree.predict(rf_input) for tree in rf.estimators_])
+    std_predict = np.std(predictions_all)
+    reliability_index = compute_reliability(std_predict)
+    q1 = np.exp(np.quantile(predictions_all, 0.25))
+    q2 = np.exp(np.quantile(predictions_all, 0.5))
+    q3 = np.exp(np.quantile(predictions_all, 0.75))
+    if (rf_pred_exp <= q1) | (rf_pred_exp >= q3):
+        print(f"Estimated Price: {np.exp(q2)} /n Low Price: {q1}, High Price: {q3}")
+        return float(q1), float(q2), float(q3), reliability_index
+    else:
+        print(f"Estimated Price: {rf_pred_exp} /n Low Price: {q1}, High Price: {q3}")
+        return float(q1), float(rf_pred_exp), float(q3), reliability_index
 
 
 
@@ -41,8 +62,11 @@ if __name__ == '__main__':
         surface = gr.Number(label='Surface', info='Saisissez la surface du bien',value=30)
         property_type = gr.Dropdown(choices=['Maison','Appartement'], label='Type de bien', info='Choisissez le type de bien')
         estimation_button = gr.Button("Estimez le bien")
-        output = gr.Number(label='Estimation du prix au m2')
-        estimation_button.click(fn=predict_price, inputs=[date, room_count, address, surface, property_type], outputs=output,api_name='Estimation')
+        output1 = gr.Number(label='Estimation basse du prix au m2')
+        output2 = gr.Number(label='Estimation du prix au m2')
+        output3 = gr.Number(label='Estimation haute du prix au m2')
+        output4 = gr.Number(label='Indice de fiabilite')
+        estimation_button.click(fn=predict_price, inputs=[date, room_count, address, surface, property_type], outputs=[output1,output2,output3,output4],api_name='Estimation')
 
 
 demo.launch(share=True)