Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -3,12 +3,12 @@ import pandas as pd
|
|
| 3 |
import numpy as np
|
| 4 |
import joblib
|
| 5 |
import __main__
|
| 6 |
-
from fastapi import FastAPI
|
| 7 |
from fastapi.responses import RedirectResponse
|
| 8 |
from pydantic import BaseModel, Field
|
| 9 |
from sklearn.base import BaseEstimator, TransformerMixin
|
| 10 |
|
| 11 |
-
# --- 1. CLASSE PERSONNALISÉE (
|
| 12 |
class DFToHashedTokens(BaseEstimator, TransformerMixin):
|
| 13 |
def __init__(self, columns=None):
|
| 14 |
self.columns = columns
|
|
@@ -17,16 +17,20 @@ class DFToHashedTokens(BaseEstimator, TransformerMixin):
|
|
| 17 |
return self
|
| 18 |
|
| 19 |
def transform(self, X):
|
|
|
|
|
|
|
|
|
|
|
|
|
| 20 |
X = X.copy()
|
| 21 |
-
# Sécurité : on récupère 'columns' dynamiquement pour éviter l'AttributeError
|
| 22 |
cols = getattr(self, 'columns', None)
|
|
|
|
| 23 |
if cols is not None:
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
return X
|
| 28 |
|
| 29 |
-
# Injection
|
| 30 |
__main__.DFToHashedTokens = DFToHashedTokens
|
| 31 |
|
| 32 |
# --- 2. CHARGEMENT DU MODÈLE ---
|
|
@@ -35,16 +39,12 @@ try:
|
|
| 35 |
print("✅ Modèle chargé avec succès")
|
| 36 |
except Exception as e:
|
| 37 |
model = None
|
| 38 |
-
print(f"❌ Erreur critique
|
| 39 |
|
| 40 |
-
# --- 3. CONFIGURATION API
|
| 41 |
-
app = FastAPI(
|
| 42 |
-
title="Fraud Detection API",
|
| 43 |
-
description="API de prédiction de fraude bancaire basée sur un modèle de Hashing."
|
| 44 |
-
)
|
| 45 |
|
| 46 |
class Transaction(BaseModel):
|
| 47 |
-
# Field(..., example=...) permet de remplir automatiquement la doc Swagger
|
| 48 |
amt: float = Field(..., example=85.20)
|
| 49 |
trans_date_trans_time: str = Field(..., example="2024-02-18 14:30:00")
|
| 50 |
dob: str = Field(..., example="1985-05-20")
|
|
@@ -79,22 +79,19 @@ def prepare_input(data: dict):
|
|
| 79 |
df["day_of_week"] = dt.dt.dayofweek
|
| 80 |
df["day"] = dt.dt.day
|
| 81 |
df["month"] = dt.dt.month
|
| 82 |
-
|
| 83 |
-
# Âge
|
| 84 |
df["age"] = ((dt - dob).dt.days / 365.25).astype("float32")
|
| 85 |
|
| 86 |
-
# Distance
|
| 87 |
lat1, lon1 = np.radians(df["lat"]), np.radians(df["long"])
|
| 88 |
lat2, lon2 = np.radians(df["merch_lat"]), np.radians(df["merch_long"])
|
| 89 |
d = np.sin((lat2-lat1)/2)**2 + np.cos(lat1)*np.cos(lat2)*np.sin((lon2-lon1)/2)**2
|
| 90 |
df["distance"] = (6371 * 2 * np.arcsin(np.sqrt(d))).astype("float32")
|
| 91 |
|
| 92 |
-
#
|
| 93 |
df["avg_amt"] = df["amt"]
|
| 94 |
df["std_amt"] = 0.0
|
| 95 |
df["nb_trans"] = 1.0
|
| 96 |
|
| 97 |
-
# Sélection stricte des 18 colonnes attendues par le modèle
|
| 98 |
expected_cols = [
|
| 99 |
'amt', 'hour', 'day_of_week', 'day', 'month', 'age', 'lat', 'long',
|
| 100 |
'city_pop', 'distance', 'avg_amt', 'std_amt', 'nb_trans',
|
|
@@ -106,9 +103,10 @@ def prepare_input(data: dict):
|
|
| 106 |
@app.post("/predict")
|
| 107 |
def predict(data: Transaction):
|
| 108 |
if model is None:
|
| 109 |
-
return {"status": "error", "message": "Modèle non chargé
|
| 110 |
try:
|
| 111 |
X_processed = prepare_input(data.dict())
|
|
|
|
| 112 |
prediction = model.predict(X_processed)
|
| 113 |
return {
|
| 114 |
"prediction": int(prediction[0]),
|
|
@@ -120,5 +118,4 @@ def predict(data: Transaction):
|
|
| 120 |
|
| 121 |
if __name__ == "__main__":
|
| 122 |
import uvicorn
|
| 123 |
-
# Port 7860 est le port standard pour Hugging Face Spaces
|
| 124 |
uvicorn.run(app, host="0.0.0.0", port=7860)
|
|
|
|
| 3 |
import numpy as np
|
| 4 |
import joblib
|
| 5 |
import __main__
|
| 6 |
+
from fastapi import FastAPI
|
| 7 |
from fastapi.responses import RedirectResponse
|
| 8 |
from pydantic import BaseModel, Field
|
| 9 |
from sklearn.base import BaseEstimator, TransformerMixin
|
| 10 |
|
| 11 |
+
# --- 1. CLASSE PERSONNALISÉE (CORRECTION DU HASHING) ---
|
| 12 |
class DFToHashedTokens(BaseEstimator, TransformerMixin):
|
| 13 |
def __init__(self, columns=None):
|
| 14 |
self.columns = columns
|
|
|
|
| 17 |
return self
|
| 18 |
|
| 19 |
def transform(self, X):
|
| 20 |
+
# On s'assure que X est un DataFrame
|
| 21 |
+
if not isinstance(X, pd.DataFrame):
|
| 22 |
+
X = pd.DataFrame(X)
|
| 23 |
+
|
| 24 |
X = X.copy()
|
|
|
|
| 25 |
cols = getattr(self, 'columns', None)
|
| 26 |
+
|
| 27 |
if cols is not None:
|
| 28 |
+
# Pour chaque ligne, on crée une liste de chaînes de caractères
|
| 29 |
+
# C'est ce que "iterable over iterables of strings" signifie
|
| 30 |
+
return X[cols].astype(str).values.tolist()
|
| 31 |
+
return X.astype(str).values.tolist()
|
| 32 |
|
| 33 |
+
# Injection pour que joblib retrouve la classe
|
| 34 |
__main__.DFToHashedTokens = DFToHashedTokens
|
| 35 |
|
| 36 |
# --- 2. CHARGEMENT DU MODÈLE ---
|
|
|
|
| 39 |
print("✅ Modèle chargé avec succès")
|
| 40 |
except Exception as e:
|
| 41 |
model = None
|
| 42 |
+
print(f"❌ Erreur critique : {e}")
|
| 43 |
|
| 44 |
+
# --- 3. CONFIGURATION API ---
|
| 45 |
+
app = FastAPI(title="Fraud Detection API")
|
|
|
|
|
|
|
|
|
|
| 46 |
|
| 47 |
class Transaction(BaseModel):
|
|
|
|
| 48 |
amt: float = Field(..., example=85.20)
|
| 49 |
trans_date_trans_time: str = Field(..., example="2024-02-18 14:30:00")
|
| 50 |
dob: str = Field(..., example="1985-05-20")
|
|
|
|
| 79 |
df["day_of_week"] = dt.dt.dayofweek
|
| 80 |
df["day"] = dt.dt.day
|
| 81 |
df["month"] = dt.dt.month
|
|
|
|
|
|
|
| 82 |
df["age"] = ((dt - dob).dt.days / 365.25).astype("float32")
|
| 83 |
|
| 84 |
+
# Distance
|
| 85 |
lat1, lon1 = np.radians(df["lat"]), np.radians(df["long"])
|
| 86 |
lat2, lon2 = np.radians(df["merch_lat"]), np.radians(df["merch_long"])
|
| 87 |
d = np.sin((lat2-lat1)/2)**2 + np.cos(lat1)*np.cos(lat2)*np.sin((lon2-lon1)/2)**2
|
| 88 |
df["distance"] = (6371 * 2 * np.arcsin(np.sqrt(d))).astype("float32")
|
| 89 |
|
| 90 |
+
# Valeurs par défaut pour les agrégats
|
| 91 |
df["avg_amt"] = df["amt"]
|
| 92 |
df["std_amt"] = 0.0
|
| 93 |
df["nb_trans"] = 1.0
|
| 94 |
|
|
|
|
| 95 |
expected_cols = [
|
| 96 |
'amt', 'hour', 'day_of_week', 'day', 'month', 'age', 'lat', 'long',
|
| 97 |
'city_pop', 'distance', 'avg_amt', 'std_amt', 'nb_trans',
|
|
|
|
| 103 |
@app.post("/predict")
|
| 104 |
def predict(data: Transaction):
|
| 105 |
if model is None:
|
| 106 |
+
return {"status": "error", "message": "Modèle non chargé."}
|
| 107 |
try:
|
| 108 |
X_processed = prepare_input(data.dict())
|
| 109 |
+
# Le modèle Pipeline appellera automatiquement DFToHashedTokens.transform()
|
| 110 |
prediction = model.predict(X_processed)
|
| 111 |
return {
|
| 112 |
"prediction": int(prediction[0]),
|
|
|
|
| 118 |
|
| 119 |
if __name__ == "__main__":
|
| 120 |
import uvicorn
|
|
|
|
| 121 |
uvicorn.run(app, host="0.0.0.0", port=7860)
|