edit de l'API

api.py

@@ -1,5 +1,7 @@
+import os
+import mlflow
+import mlflow.pyfunc
 import pandas as pd
-import joblib
 import uvicorn
 from pydantic import BaseModel
 from typing import Literal, List, Union
@@ -46,9 +48,13 @@ class Car(BaseModel):
     winter_tires: bool
 
 
-# Chargement du modèle et du preprocessor au démarrage (une seule fois)
-preprocessor = joblib.load("preprocessor.joblib")
-model = joblib.load("model.joblib")
+# Chargement du modèle depuis MLflow au démarrage (une seule fois)
+MLFLOW_TRACKING_URI = os.getenv("MLFLOW_TRACKING_URI", "https://atomik31-mlflow.hf.space")
+mlflow.set_tracking_uri(MLFLOW_TRACKING_URI)
+
+print(f"Chargement du modèle depuis MLflow : {MLFLOW_TRACKING_URI}")
+model = mlflow.pyfunc.load_model("models:/GetAround_price_predictor@production")
+print("Modèle chargé en mémoire.")
 
 
 @app.get("/", include_in_schema=False)
@@ -66,8 +72,7 @@ async def predict(cars: List[Car]):
     **Sortie** : `{"prediction": [prix_en_euros, ...]}`
     """
     car_features = pd.DataFrame(jsonable_encoder(cars))
-    car_features_transformed = preprocessor.transform(car_features)
-    prediction = model.predict(car_features_transformed)
+    prediction = model.predict(car_features)
     return {"prediction": prediction.tolist()}
 
 

model.joblib

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:137a16b3654076082c80396e67638800496ee2eefe913db17a786ba91aac84a7
-size 44960641
+oid sha256:074b922a05eee7516a7aba920a1c2608c3eb5b76c2305b0c44f278eed8cfe399
+size 42983521

preprocessor.joblib

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:132daf149c1c2f7a8e907e9a3988fd461425783c9d9c56037fa6c7f2465aaead
-size 5708
+oid sha256:d0646694ed246c9fb4c3b6619b1b36615dec18badc23784e9d733d6d72653597
+size 4912

request.py

@@ -0,0 +1,65 @@
+"""
+Script de test de l'API GetAround.
+Lancer d'abord l'API avec : uvicorn api:app --port 8000
+"""
+import requests
+
+# Test avec une seule voiture
+response = requests.post("http://localhost:8000/predict", json=[
+    {
+        "model_key": "Citroën",
+        "mileage": 13929,
+        "engine_power": 317,
+        "fuel": "petrol",
+        "paint_color": "grey",
+        "car_type": "convertible",
+        "private_parking_available": True,
+        "has_gps": True,
+        "has_air_conditioning": False,
+        "automatic_car": False,
+        "has_getaround_connect": False,
+        "has_speed_regulator": True,
+        "winter_tires": True
+    }
+])
+
+print("Status :", response.status_code)
+print("Réponse :", response.json())
+
+# Test avec plusieurs voitures
+response2 = requests.post("http://localhost:8000/predict", json=[
+    {
+        "model_key": "BMW",
+        "mileage": 5000,
+        "engine_power": 200,
+        "fuel": "petrol",
+        "paint_color": "black",
+        "car_type": "sedan",
+        "private_parking_available": True,
+        "has_gps": True,
+        "has_air_conditioning": True,
+        "automatic_car": True,
+        "has_getaround_connect": True,
+        "has_speed_regulator": True,
+        "winter_tires": False
+    },
+    {
+        "model_key": "Renault",
+        "mileage": 80000,
+        "engine_power": 90,
+        "fuel": "diesel",
+        "paint_color": "white",
+        "car_type": "hatchback",
+        "private_parking_available": False,
+        "has_gps": False,
+        "has_air_conditioning": True,
+        "automatic_car": False,
+        "has_getaround_connect": False,
+        "has_speed_regulator": False,
+        "winter_tires": True
+    }
+])
+
+print("\nTest multi-voitures :")
+print("Status :", response2.status_code)
+print("Réponse :", response2.json())

requirements.txt

@@ -5,3 +5,5 @@ pandas
 scikit-learn==1.6.1
 numpy
 joblib
+mlflow
+boto3