Update app.py

app.py

@@ -4,15 +4,25 @@ import mlflow.pyfunc
 import pandas as pd
 import config
 import logging
-logger = logging.getLogger(__name__)
-
 
-#logging.basicConfig(filename='myapp.log', level=logging.INFO)
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger(__name__)
 
 
 # 1) URL du Space HF qui héberge MLflow (tracking server)
-SPACE_URL = "https://pradelf-getaround-mlflow.hf.space"  # <-- remplace ici
-mlflow.set_tracking_uri(SPACE_URL)  # ou via env MLFLOW_TRACKING_URI
+MLFLOW_TRACKING_URI = os.getenv("MLFLOW_TRACKING_URI", "https://pradelf-getaround-mlflow.hf.space")
+MODEL_URI = os.getenv("MODEL_URI", "models:/getaround-price-prediction-model@certification")
+mlflow.set_tracking_uri(MLFLOW_TRACKING_URI)  # ou via env MLFLOW_TRACKING_URI
+BOOL_COLS = [
+    "private_parking_available",
+    "has_gps",
+    "has_air_conditioning",
+    "automatic_car",
+    "has_getaround_connect",
+    "has_speed_regulator",
+    "winter_tires",
+]
+
 
 description = """
 Bienvenue sur l'API de Getaround pour prédire le prix journalier de location d'une voiture en fonction de son année d'expérience!
@@ -42,11 +52,11 @@ La documentation est ci-dessous 👇 pour chaque point de terminaison (endpoints
 
 tags_metadata = [
     {
-        "name": "Introduction Endpoints",
+        "name": "Point de terminaison d'introduction",
         "description": "Terminaison simple de présentation de l'API et de sa version pour vérifier que tout fonctionne correctement.",
     },
     {
-        "name": "Health",
+        "name": "Point de terminaison de ligne de vie",
         "description": "Point de terminaison de ligne de vie pour surveiller le statut du serveur de l'API. Retourne 'ok' si le serveur fonctionne sans problèmes.",
     },
     {
@@ -58,6 +68,20 @@ tags_metadata = [
         "description": "Point de terminaison de prediction du prix de location journalier d'un véhicule en fonction de ses caractéristiques.",
     },
 ]
+
+@app.on_event("startup")
+def load_model_on_startup():
+    global model
+    try:
+        logger.info(f"Chargement du modèle depuis {MLFLOW_TRACKING_URI}")
+        logger.info(f"MLFLOW_TRACKING_URI={os.getenv('MLFLOW_TRACKING_URI')}")
+        model = mlflow.pyfunc.load_model(MODEL_URI)
+        logger.info("Modèle chargé avec succès")
+    except Exception:
+        logger.exception("Impossible de charger le modèle au démarrage")
+        model = None
+
+
 app = FastAPI(
     title="Getaround API pour le prix journalier de location d'une voiture.",
     description=description,
@@ -71,7 +95,6 @@ app = FastAPI(
 
 
 class RentalFeatures(BaseModel):
-    model_key: str
     model_key: str
     mileage: int
     engine_power: int
@@ -92,16 +115,14 @@ def greet_json():
     """
     Bonjour
     """
-    logger.info('Hello')
     return {"Hello": "World!", "version": config.__version__}
 
 
-@app.get("/health", tags=["Health"])
+@app.get("/health")
 def health():
     """
     Health line to monitor the server status of the API. Returns "ok" if the server is running without issues.
     """
-    logger.info('Health')
     return {"status": "ok"}
 
 
@@ -111,39 +132,47 @@ async def predict(predictionFeatures: RentalFeatures):
     Prediction of car daily rental cost for a given properties of a car !
     """
     # Read data
-    pf = [
-         predictionFeatures.model_key or "Peugeot",   
-         predictionFeatures.mileage or 0,
-         predictionFeatures.engine_power or 0,
-         predictionFeatures.fuel or "petrol",
-         predictionFeatures.car_type or "sedan",
-         predictionFeatures.private_parking_available or 1,
-         predictionFeatures.has_gps or 0,
-         predictionFeatures.has_air_conditioning or 0,
-         predictionFeatures.automatic_car or 0,
-         predictionFeatures.has_getaround_connect or 0,
-         predictionFeatures.has_speed_regulator or 0,
-         predictionFeatures.winter_tires or 0,
-     ]
-
-    car_caracteristic = pd.DataFrame([pf])
-    logger.info('Started')
-
-    logger.info(car_caracteristic)
-    #car_caracteristic = pd.DataFrame({"Car": pf})
-    # Log model from mlflow
-    logged_model = "models:/getaround-price-prediction-model@champion"  # REPLACE WITH YOUR OWN RUN ID
-    #logged_model = "runs:/d3856ef1a2ee42db989496949f47ac74/getaround-price-prediction-model"
-    # Load model as a PyFuncModel.
-    loaded_model = mlflow.pyfunc.load_model(logged_model)
-    prediction = loaded_model.predict(car_caracteristic)
-
-    # Format response
-    response = {
-        "prediction": prediction.tolist()[0],
-        "detail": "Prédiction du tarif journalier (nul si aucun modèle : model.pkl).",
-    }
-    logger.info(response)
-
-    logger.info('Finished')
-    return response
+    # pf = [
+    #     predictionFeatures.model_key or "Peugeot",   
+    #     predictionFeatures.mileage or 0,
+    #     predictionFeatures.engine_power or 0,
+    #     predictionFeatures.fuel or "petrol",
+    #     predictionFeatures.car_type or "sedan",
+    #     predictionFeatures.private_parking_available or 1,
+    #     predictionFeatures.has_gps or 0,
+    #     predictionFeatures.has_air_conditioning or 0,
+    #     predictionFeatures.automatic_car or 0,
+    #     predictionFeatures.has_getaround_connect or 0,
+    #     predictionFeatures.has_speed_regulator or 0,
+    #     predictionFeatures.winter_tires or 0,
+    # ]
+    try:
+        car_caracteristic = pd.DataFrame([predictionFeatures.model_dump()])
+        #car_caracteristic = pd.DataFrame({"Car": pf})
+        # Log model from mlflow
+   
+        BOOL_COLS = [
+            "private_parking_available",
+            "has_gps",
+            "has_air_conditioning",
+            "automatic_car",
+            "has_getaround_connect",
+            "has_speed_regulator",
+            "winter_tires",
+        ]
+        # Conversion explicite des colonnes booléennes
+        for col in BOOL_COLS:
+            car_caracteristic[col] = car_caracteristic[col].astype(bool)
+        # Prediction from previously loaded model as a PyFuncModel.
+        prediction = model.predict(car_caracteristic)
+
+        # Format response
+        response = {
+            "prediction": prediction.tolist()[0],
+            "detail": "Prédiction du tarif journalier (nul si aucun modèle : model.pkl).",
+        }
+        return response
+    except Exception as e:
+        logger.exception("Erreur dans /predict")
+        raise Exception(status_code=500, detail=str(e))
+