Spaces:
Running
Running
Upload folder using huggingface_hub
Browse files- README.md +295 -77
- README_HF.md +1 -1
- api.py +424 -0
- app.py +8 -416
- src/gradio_ui.py +28 -24
README.md
CHANGED
|
@@ -1,106 +1,324 @@
|
|
| 1 |
-
|
| 2 |
-
title: Employee Turnover Prediction API
|
| 3 |
-
emoji: 👔
|
| 4 |
-
colorFrom: blue
|
| 5 |
-
colorTo: purple
|
| 6 |
-
sdk: docker
|
| 7 |
-
pinned: true
|
| 8 |
-
license: mit
|
| 9 |
-
app_port: 7860
|
| 10 |
-
---
|
| 11 |
|
|
|
|
| 12 |
|
| 13 |
-
|
| 14 |
|
| 15 |
-
API de prédiction du turnover des employés (XGBoost + SMOTE) avec endpoints batch, validation stricte et documentation à jour.
|
| 16 |
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
- ✅ Prédiction de turnover (0 = reste, 1 = part)
|
| 20 |
- 📦 Endpoint batch CSV (3 fichiers bruts)
|
| 21 |
-
-
|
| 22 |
-
-
|
| 23 |
-
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 27 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 28 |
|
| 29 |
-
|
|
|
|
|
|
|
| 30 |
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
| `/docs` | Documentation interactive Swagger |
|
| 34 |
-
| `/health` | Status de l'API |
|
| 35 |
-
| `/ui` | Interface Gradio interactive |
|
| 36 |
-
| `/predict` | Prédiction unitaire (JSON, contraintes réelles) |
|
| 37 |
-
| `/predict/batch` | Prédiction batch (3 fichiers CSV bruts) |
|
| 38 |
|
|
|
|
|
|
|
|
|
|
| 39 |
|
| 40 |
-
##
|
| 41 |
|
| 42 |
-
### Prédiction unitaire (toutes contraintes appliquées)
|
| 43 |
```bash
|
| 44 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 45 |
-H "Content-Type: application/json" \
|
| 46 |
-
-
|
| 47 |
-
-d '{
|
| 48 |
-
"nombre_participation_pee": 0,
|
| 49 |
-
"nb_formations_suivies": 2,
|
| 50 |
-
"nombre_employee_sous_responsabilite": 1,
|
| 51 |
-
"distance_domicile_travail": 15,
|
| 52 |
-
"niveau_education": 3,
|
| 53 |
-
"domaine_etude": "Infra & Cloud",
|
| 54 |
-
"ayant_enfants": "Y",
|
| 55 |
-
"frequence_deplacement": "Occasionnel",
|
| 56 |
-
"annees_depuis_la_derniere_promotion": 2,
|
| 57 |
-
"annes_sous_responsable_actuel": 5,
|
| 58 |
-
"satisfaction_employee_environnement": 3,
|
| 59 |
-
"note_evaluation_precedente": 4,
|
| 60 |
-
"niveau_hierarchique_poste": 2,
|
| 61 |
-
"satisfaction_employee_nature_travail": 3,
|
| 62 |
-
"satisfaction_employee_equipe": 3,
|
| 63 |
-
"satisfaction_employee_equilibre_pro_perso": 2,
|
| 64 |
-
"note_evaluation_actuelle": 4,
|
| 65 |
-
"heure_supplementaires": "Non",
|
| 66 |
-
"augementation_salaire_precedente": 5.5,
|
| 67 |
-
"age": 35,
|
| 68 |
-
"genre": "M",
|
| 69 |
-
"revenu_mensuel": 4500.0,
|
| 70 |
-
"statut_marital": "Marié(e)",
|
| 71 |
-
"departement": "Commercial",
|
| 72 |
-
"poste": "Manager",
|
| 73 |
-
"nombre_experiences_precedentes": 3,
|
| 74 |
-
"nombre_heures_travailless": 80,
|
| 75 |
-
"annee_experience_totale": 10,
|
| 76 |
-
"annees_dans_l_entreprise": 5,
|
| 77 |
-
"annees_dans_le_poste_actuel": 2
|
| 78 |
-
}'
|
| 79 |
```
|
| 80 |
|
| 81 |
-
|
|
|
|
|
|
|
|
|
|
| 82 |
```bash
|
| 83 |
-
|
| 84 |
-
|
| 85 |
-
|
| 86 |
-
|
| 87 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 88 |
```
|
| 89 |
|
| 90 |
-
|
| 91 |
-
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 92 |
{
|
| 93 |
"total_employees": 1470,
|
| 94 |
-
"predictions": [
|
|
|
|
|
|
|
|
|
|
| 95 |
"summary": {
|
| 96 |
"total_stay": 1169,
|
| 97 |
"total_leave": 301,
|
| 98 |
-
"high_risk_count": 222
|
|
|
|
|
|
|
| 99 |
}
|
| 100 |
}
|
| 101 |
```
|
| 102 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 103 |
|
| 104 |
-
##
|
| 105 |
|
| 106 |
-
|
|
|
|
|
|
| 1 |
+
# 🚀 Employee Turnover Prediction API - v3.2.1
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
|
| 3 |
+
## 📊 Vue d'ensemble
|
| 4 |
|
| 5 |
+
API REST de prédiction du turnover des employés basée sur un modèle XGBoost avec SMOTE.
|
| 6 |
|
|
|
|
| 7 |
|
| 8 |
+
**✨ Nouveautés v3.2.1** :
|
| 9 |
+
- 🎛️ Sliders Gradio et schémas Pydantic alignés sur les min/max réels des données d'entraînement
|
|
|
|
| 10 |
- 📦 Endpoint batch CSV (3 fichiers bruts)
|
| 11 |
+
- 🔑 Authentification API Key (prod)
|
| 12 |
+
- 🔧 Correction preprocessing (scaling, ordre des colonnes)
|
| 13 |
+
- 📝 Documentation et exemples mis à jour
|
| 14 |
+
|
| 15 |
+
## 🏗️ Architecture
|
| 16 |
+
|
| 17 |
+
```
|
| 18 |
+
OC_P5/
|
| 19 |
+
├── app.py # Point d'entrée FastAPI
|
| 20 |
+
├── src/
|
| 21 |
+
│ ├── auth.py # Authentification API Key
|
| 22 |
+
│ ├── config.py # Configuration centralisée
|
| 23 |
+
│ ├── logger.py # Logging structuré (NOUVEAU)
|
| 24 |
+
│ ├── models.py # Chargement modèle HF Hub
|
| 25 |
+
│ ├── preprocessing.py # Pipeline preprocessing
|
| 26 |
+
│ ├── rate_limit.py # Rate limiting (NOUVEAU)
|
| 27 |
+
│ └── schemas.py # Validation Pydantic
|
| 28 |
+
├── tests/ # Suite pytest (33 tests, 88% couverture)
|
| 29 |
+
├── logs/ # Logs JSON (NOUVEAU)
|
| 30 |
+
│ ├── api.log # Tous les logs
|
| 31 |
+
│ └── error.log # Erreurs uniquement
|
| 32 |
+
├── docs/ # Documentation
|
| 33 |
+
├── ml_model/ # Scripts training
|
| 34 |
+
└── data/ # Données sources
|
| 35 |
+
## 🗄️ Schéma de la Base de Données (PostgreSQL)
|
| 36 |
+
|
| 37 |
+
Schéma UML pour traçabilité ML (basé sur P5 prédiction turnover employé) :
|
| 38 |
+

|
| 39 |
+
|
| 40 |
+
- **dataset** : Dataset original (référence pour tests/retraining). Colonnes adaptées au modèle de prédiction turnover.
|
| 41 |
+
- **ml_logs** : Logs inputs/outputs (JSON pour flexibilité, timestamp pour audits).
|
| 42 |
+
|
| 43 |
+
Choix : Structure relationnelle pour efficacité volume data ; sécurité via user dédié (ml_user).
|
| 44 |
+
Instructions : Voir create_db.py pour création.
|
| 45 |
+
|
| 46 |
+
📖 **Guide complet pour débutants** : [docs/database_guide.md](docs/database_guide.md)
|
| 47 |
+
|
| 48 |
+
### 💾 Insertion du Dataset
|
| 49 |
+
```bash
|
| 50 |
+
# Insérer le dataset complet (1470 employés)
|
| 51 |
+
poetry run python scripts/insert_dataset.py
|
| 52 |
+
|
| 53 |
+
# Vérifier l'insertion
|
| 54 |
+
psql -h localhost -U ml_user -d oc_p5_db -c "SELECT COUNT(*) FROM dataset;"
|
| 55 |
+
```
|
| 56 |
+
|
| 57 |
+
### Prérequis
|
| 58 |
+
- Python 3.12+
|
| 59 |
+
- Poetry 1.7+
|
| 60 |
+
- Git
|
| 61 |
+
|
| 62 |
+
### Setup rapide
|
| 63 |
|
| 64 |
+
```bash
|
| 65 |
+
# 1. Cloner le repo
|
| 66 |
+
git clone https://github.com/chaton59/OC_P5.git
|
| 67 |
+
cd OC_P5
|
| 68 |
+
|
| 69 |
+
# 2. Installer les dépendances
|
| 70 |
+
poetry install
|
| 71 |
|
| 72 |
+
# 3. Configurer l'environnement
|
| 73 |
+
cp .env.example .env
|
| 74 |
+
# Éditer .env avec vos valeurs
|
| 75 |
|
| 76 |
+
# 4. Lancer l'API
|
| 77 |
+
poetry run uvicorn app:app --reload
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 78 |
|
| 79 |
+
# 5. Accéder à la documentation
|
| 80 |
+
# http://localhost:8000/docs
|
| 81 |
+
```
|
| 82 |
|
| 83 |
+
## 📝 Configuration (.env)
|
| 84 |
|
|
|
|
| 85 |
```bash
|
| 86 |
+
# Mode développement (désactive auth + active logs détaillés)
|
| 87 |
+
DEBUG=true
|
| 88 |
+
|
| 89 |
+
# API Key (requis en production)
|
| 90 |
+
API_KEY=your-secret-key-here
|
| 91 |
+
|
| 92 |
+
# Logging (DEBUG, INFO, WARNING, ERROR, CRITICAL)
|
| 93 |
+
LOG_LEVEL=INFO
|
| 94 |
+
|
| 95 |
+
# HuggingFace Model
|
| 96 |
+
HF_MODEL_REPO=ASI-Engineer/employee-turnover-model
|
| 97 |
+
MODEL_FILENAME=model/model.pkl
|
| 98 |
+
```
|
| 99 |
+
|
| 100 |
+
## 🔒 Authentification
|
| 101 |
+
|
| 102 |
+
### Mode DEBUG (développement)
|
| 103 |
+
```bash
|
| 104 |
+
# L'API Key n'est PAS requise
|
| 105 |
+
curl http://localhost:8000/predict -H "Content-Type: application/json" -d '{...}'
|
| 106 |
+
```
|
| 107 |
+
|
| 108 |
+
### Mode PRODUCTION
|
| 109 |
+
```bash
|
| 110 |
+
# L'API Key est REQUISE
|
| 111 |
+
curl http://localhost:8000/predict \
|
| 112 |
+
-H "X-API-Key: your-secret-key" \
|
| 113 |
-H "Content-Type: application/json" \
|
| 114 |
+
-d '{...}'
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 115 |
```
|
| 116 |
|
| 117 |
+
|
| 118 |
+
## 📡 Endpoints
|
| 119 |
+
|
| 120 |
+
### 🏥 Health Check
|
| 121 |
```bash
|
| 122 |
+
GET /health
|
| 123 |
+
|
| 124 |
+
# Réponse
|
| 125 |
+
{
|
| 126 |
+
"status": "healthy",
|
| 127 |
+
"model_loaded": true,
|
| 128 |
+
"model_type": "Pipeline",
|
| 129 |
+
"version": "3.2.1"
|
| 130 |
+
}
|
| 131 |
```
|
| 132 |
|
| 133 |
+
### 🔮 Prédiction unitaire
|
| 134 |
+
```bash
|
| 135 |
+
POST /predict
|
| 136 |
+
Content-Type: application/json
|
| 137 |
+
X-API-Key: your-key (en production)
|
| 138 |
+
|
| 139 |
+
# Payload (exemple, contraintes réelles appliquées)
|
| 140 |
+
{
|
| 141 |
+
"nombre_participation_pee": 0,
|
| 142 |
+
"nb_formations_suivies": 2,
|
| 143 |
+
"nombre_employee_sous_responsabilite": 1,
|
| 144 |
+
"distance_domicile_travail": 15,
|
| 145 |
+
"niveau_education": 3,
|
| 146 |
+
"domaine_etude": "Infra & Cloud",
|
| 147 |
+
"ayant_enfants": "Y",
|
| 148 |
+
"frequence_deplacement": "Occasionnel",
|
| 149 |
+
"annees_depuis_la_derniere_promotion": 2,
|
| 150 |
+
"annes_sous_responsable_actuel": 5,
|
| 151 |
+
"satisfaction_employee_environnement": 3,
|
| 152 |
+
"note_evaluation_precedente": 4,
|
| 153 |
+
"niveau_hierarchique_poste": 2,
|
| 154 |
+
"satisfaction_employee_nature_travail": 3,
|
| 155 |
+
"satisfaction_employee_equipe": 3,
|
| 156 |
+
"satisfaction_employee_equilibre_pro_perso": 2,
|
| 157 |
+
"note_evaluation_actuelle": 4,
|
| 158 |
+
"heure_supplementaires": "Non",
|
| 159 |
+
"augementation_salaire_precedente": 5.5,
|
| 160 |
+
"age": 35,
|
| 161 |
+
"genre": "M",
|
| 162 |
+
"revenu_mensuel": 4500.0,
|
| 163 |
+
"statut_marital": "Marié(e)",
|
| 164 |
+
"departement": "Commercial",
|
| 165 |
+
"poste": "Manager",
|
| 166 |
+
"nombre_experiences_precedentes": 3,
|
| 167 |
+
"nombre_heures_travailless": 80,
|
| 168 |
+
"annee_experience_totale": 10,
|
| 169 |
+
"annees_dans_l_entreprise": 5,
|
| 170 |
+
"annees_dans_le_poste_actuel": 2
|
| 171 |
+
}
|
| 172 |
+
|
| 173 |
+
# Réponse
|
| 174 |
+
{
|
| 175 |
+
"prediction": 0, # 0 = reste, 1 = part
|
| 176 |
+
"probability_0": 0.85, # Probabilité de rester
|
| 177 |
+
"probability_1": 0.15, # Probabilité de partir
|
| 178 |
+
"risk_level": "Low" # Low, Medium, High
|
| 179 |
+
}
|
| 180 |
+
```
|
| 181 |
+
|
| 182 |
+
### 📦 Prédiction batch (CSV)
|
| 183 |
+
```bash
|
| 184 |
+
POST /predict/batch
|
| 185 |
+
X-API-Key: your-key (en production)
|
| 186 |
+
|
| 187 |
+
# Envoi des 3 fichiers CSV bruts
|
| 188 |
+
curl -X POST "http://localhost:8000/predict/batch" \
|
| 189 |
+
-H "X-API-Key: your-key" \
|
| 190 |
+
-F "sondage_file=@data/extrait_sondage.csv" \
|
| 191 |
+
-F "eval_file=@data/extrait_eval.csv" \
|
| 192 |
+
-F "sirh_file=@data/extrait_sirh.csv"
|
| 193 |
+
|
| 194 |
+
# Réponse
|
| 195 |
{
|
| 196 |
"total_employees": 1470,
|
| 197 |
+
"predictions": [
|
| 198 |
+
{"employee_id": 1, "prediction": 1, "probability_leave": 0.84, "risk_level": "High"},
|
| 199 |
+
{"employee_id": 2, "prediction": 0, "probability_leave": 0.11, "risk_level": "Low"}
|
| 200 |
+
],
|
| 201 |
"summary": {
|
| 202 |
"total_stay": 1169,
|
| 203 |
"total_leave": 301,
|
| 204 |
+
"high_risk_count": 222,
|
| 205 |
+
"medium_risk_count": 233,
|
| 206 |
+
"low_risk_count": 1015
|
| 207 |
}
|
| 208 |
}
|
| 209 |
```
|
| 210 |
|
| 211 |
+
## 📊 Logging
|
| 212 |
+
|
| 213 |
+
### Logs structurés JSON
|
| 214 |
+
|
| 215 |
+
**Fichiers** :
|
| 216 |
+
- `logs/api.log` : Tous les logs
|
| 217 |
+
- `logs/error.log` : Erreurs uniquement
|
| 218 |
+
|
| 219 |
+
**Format** :
|
| 220 |
+
```json
|
| 221 |
+
{
|
| 222 |
+
"timestamp": "2025-12-26T10:30:45",
|
| 223 |
+
"level": "INFO",
|
| 224 |
+
"logger": "employee_turnover_api",
|
| 225 |
+
"message": "Request POST /predict",
|
| 226 |
+
"method": "POST",
|
| 227 |
+
"path": "/predict",
|
| 228 |
+
"status_code": 200,
|
| 229 |
+
"duration_ms": 23.45,
|
| 230 |
+
"client_host": "127.0.0.1"
|
| 231 |
+
}
|
| 232 |
+
```
|
| 233 |
+
|
| 234 |
+
## 🛡️ Rate Limiting
|
| 235 |
+
|
| 236 |
+
**Configuration** :
|
| 237 |
+
- **Développement** : Désactivé (DEBUG=true)
|
| 238 |
+
- **Production** : 20 requêtes/minute par IP ou API Key
|
| 239 |
+
|
| 240 |
+
**En cas de dépassement** :
|
| 241 |
+
```json
|
| 242 |
+
{
|
| 243 |
+
"error": "Rate limit exceeded",
|
| 244 |
+
"message": "20 per 1 minute"
|
| 245 |
+
}
|
| 246 |
+
```
|
| 247 |
+
|
| 248 |
+
## ✅ Tests
|
| 249 |
+
|
| 250 |
+
```bash
|
| 251 |
+
# Tous les tests
|
| 252 |
+
poetry run pytest tests/ -v
|
| 253 |
+
|
| 254 |
+
# Avec couverture
|
| 255 |
+
poetry run pytest tests/ --cov --cov-report=html
|
| 256 |
+
|
| 257 |
+
# Voir rapport HTML
|
| 258 |
+
open htmlcov/index.html
|
| 259 |
+
```
|
| 260 |
+
|
| 261 |
+
**Résultats** :
|
| 262 |
+
- ✅ 33 tests passés
|
| 263 |
+
- 📊 88% de couverture globale
|
| 264 |
+
|
| 265 |
+
## 🚀 Déploiement
|
| 266 |
+
|
| 267 |
+
### Variables d'environnement requises
|
| 268 |
+
```bash
|
| 269 |
+
DEBUG=false
|
| 270 |
+
API_KEY=<votre-clé-sécurisée>
|
| 271 |
+
LOG_LEVEL=INFO
|
| 272 |
+
```
|
| 273 |
+
|
| 274 |
+
### HuggingFace Spaces
|
| 275 |
+
Prêt pour déploiement avec `app.py` et `requirements.txt`
|
| 276 |
+
|
| 277 |
+
## 📚 Documentation
|
| 278 |
+
|
| 279 |
+
- **API Interactive** : http://localhost:8000/docs
|
| 280 |
+
- **ReDoc** : http://localhost:8000/redoc
|
| 281 |
+
- **Guide complet** : [docs/API_GUIDE.md](docs/API_GUIDE.md)
|
| 282 |
+
- **Standards** : [docs/standards.md](docs/standards.md)
|
| 283 |
+
- **Couverture tests** : [docs/TEST_COVERAGE.md](docs/TEST_COVERAGE.md)
|
| 284 |
+
|
| 285 |
+
## 📦 Dépendances principales
|
| 286 |
+
|
| 287 |
+
- **FastAPI** 0.115.14 : Framework web
|
| 288 |
+
- **Pydantic** 2.12.5 : Validation données
|
| 289 |
+
- **XGBoost** 2.1.3 : Modèle ML
|
| 290 |
+
- **SlowAPI** 0.1.9 : Rate limiting
|
| 291 |
+
- **python-json-logger** 4.0.0 : Logs structurés
|
| 292 |
+
- **pytest** 9.0.2 : Tests
|
| 293 |
+
|
| 294 |
+
|
| 295 |
+
## 🔄 Changelog
|
| 296 |
+
|
| 297 |
+
### v3.2.1 (janvier 2026)
|
| 298 |
+
- 🎛️ Sliders Gradio et schémas Pydantic alignés sur les min/max réels des données d'entraînement
|
| 299 |
+
- 📦 Endpoint batch CSV (3 fichiers bruts)
|
| 300 |
+
- 🔑 Authentification API Key (prod)
|
| 301 |
+
- 🔧 Correction preprocessing (scaling, ordre des colonnes)
|
| 302 |
+
- 📝 Documentation et exemples mis à jour
|
| 303 |
+
|
| 304 |
+
### v2.2.0 (27 décembre 2025)
|
| 305 |
+
- 📦 Nouvel endpoint `/predict/batch` pour traitement CSV direct
|
| 306 |
+
- 🔧 Fix preprocessing : ajout du scaling des features
|
| 307 |
+
- 🔧 Fix preprocessing : correction de l'ordre des colonnes
|
| 308 |
+
- 📊 Amélioration précision des prédictions (~90%)
|
| 309 |
+
|
| 310 |
+
### v2.1.0 (26 décembre 2025)
|
| 311 |
+
- ✨ Système de logging structuré JSON
|
| 312 |
+
- 🛡️ Rate limiting avec SlowAPI
|
| 313 |
+
- ⚡ Amélioration gestion d'erreurs
|
| 314 |
+
- 📊 Monitoring des performances
|
| 315 |
+
|
| 316 |
+
### v2.0.0 (26 décembre 2025)
|
| 317 |
+
- ✅ Suite de tests complète (36 tests)
|
| 318 |
+
- 🔐 Authentification API Key
|
| 319 |
+
- 📊 88% de couverture de code
|
| 320 |
|
| 321 |
+
## 👥 Auteurs
|
| 322 |
|
| 323 |
+
- **Projet** : OpenClassrooms P5
|
| 324 |
+
- **Repo** : [github.com/chaton59/OC_P5](https://github.com/chaton59/OC_P5)
|
README_HF.md
CHANGED
|
@@ -3,7 +3,7 @@ title: Employee Turnover Prediction API
|
|
| 3 |
emoji: 👔
|
| 4 |
colorFrom: blue
|
| 5 |
colorTo: purple
|
| 6 |
-
sdk:
|
| 7 |
pinned: true
|
| 8 |
license: mit
|
| 9 |
app_port: 7860
|
|
|
|
| 3 |
emoji: 👔
|
| 4 |
colorFrom: blue
|
| 5 |
colorTo: purple
|
| 6 |
+
sdk: gradio
|
| 7 |
pinned: true
|
| 8 |
license: mit
|
| 9 |
app_port: 7860
|
api.py
ADDED
|
@@ -0,0 +1,424 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python3
|
| 2 |
+
"""
|
| 3 |
+
API FastAPI pour le modèle Employee Turnover.
|
| 4 |
+
|
| 5 |
+
Cette API expose le modèle de prédiction de départ des employés avec :
|
| 6 |
+
- Validation stricte des inputs via Pydantic
|
| 7 |
+
- Preprocessing automatique
|
| 8 |
+
- Health check pour monitoring
|
| 9 |
+
- Documentation OpenAPI/Swagger automatique
|
| 10 |
+
- Interface Gradio pour utilisation interactive
|
| 11 |
+
- Endpoint batch pour traitement de fichiers CSV
|
| 12 |
+
"""
|
| 13 |
+
import io
|
| 14 |
+
import time
|
| 15 |
+
from contextlib import asynccontextmanager
|
| 16 |
+
|
| 17 |
+
import gradio as gr
|
| 18 |
+
import pandas as pd
|
| 19 |
+
from fastapi import Depends, FastAPI, File, HTTPException, Request, UploadFile
|
| 20 |
+
from fastapi.middleware.cors import CORSMiddleware
|
| 21 |
+
from slowapi import _rate_limit_exceeded_handler
|
| 22 |
+
from slowapi.errors import RateLimitExceeded
|
| 23 |
+
|
| 24 |
+
from src.auth import verify_api_key
|
| 25 |
+
from src.config import get_settings
|
| 26 |
+
from src.gradio_ui import create_gradio_interface
|
| 27 |
+
from src.logger import logger, log_model_load, log_request
|
| 28 |
+
from src.models import get_model_info, load_model
|
| 29 |
+
from src.preprocessing import (
|
| 30 |
+
merge_csv_dataframes,
|
| 31 |
+
preprocess_dataframe_for_prediction,
|
| 32 |
+
preprocess_for_prediction,
|
| 33 |
+
)
|
| 34 |
+
from src.rate_limit import limiter
|
| 35 |
+
from src.schemas import (
|
| 36 |
+
BatchPredictionOutput,
|
| 37 |
+
EmployeeInput,
|
| 38 |
+
EmployeePrediction,
|
| 39 |
+
HealthCheck,
|
| 40 |
+
PredictionOutput,
|
| 41 |
+
)
|
| 42 |
+
|
| 43 |
+
# Charger la configuration
|
| 44 |
+
settings = get_settings()
|
| 45 |
+
API_VERSION = settings.API_VERSION
|
| 46 |
+
|
| 47 |
+
|
| 48 |
+
@asynccontextmanager
|
| 49 |
+
async def lifespan(app: FastAPI):
|
| 50 |
+
"""
|
| 51 |
+
Gestion du cycle de vie de l'application.
|
| 52 |
+
|
| 53 |
+
Charge le modèle au démarrage et le garde en cache.
|
| 54 |
+
"""
|
| 55 |
+
logger.info(
|
| 56 |
+
"🚀 Démarrage de l'API Employee Turnover...", extra={"version": API_VERSION}
|
| 57 |
+
)
|
| 58 |
+
|
| 59 |
+
start_time = time.time()
|
| 60 |
+
try:
|
| 61 |
+
# Pré-charger le modèle au démarrage
|
| 62 |
+
model = load_model()
|
| 63 |
+
duration_ms = (time.time() - start_time) * 1000
|
| 64 |
+
|
| 65 |
+
model_type = type(model).__name__
|
| 66 |
+
log_model_load(model_type, duration_ms, True)
|
| 67 |
+
logger.info("✅ Modèle chargé avec succès")
|
| 68 |
+
except Exception as e:
|
| 69 |
+
duration_ms = (time.time() - start_time) * 1000
|
| 70 |
+
log_model_load("Unknown", duration_ms, False)
|
| 71 |
+
logger.error("Le modèle n'a pas pu être chargé", extra={"error": str(e)})
|
| 72 |
+
|
| 73 |
+
yield # L'application tourne
|
| 74 |
+
|
| 75 |
+
logger.info("🛑 Arrêt de l'API")
|
| 76 |
+
|
| 77 |
+
|
| 78 |
+
# Créer l'application FastAPI
|
| 79 |
+
app = FastAPI(
|
| 80 |
+
title="Employee Turnover Prediction API",
|
| 81 |
+
description="API de prédiction du turnover des employés avec XGBoost + SMOTE",
|
| 82 |
+
version=API_VERSION,
|
| 83 |
+
lifespan=lifespan,
|
| 84 |
+
docs_url="/docs",
|
| 85 |
+
redoc_url="/redoc",
|
| 86 |
+
)
|
| 87 |
+
|
| 88 |
+
# Ajouter rate limiting
|
| 89 |
+
app.state.limiter = limiter
|
| 90 |
+
app.add_exception_handler(RateLimitExceeded, _rate_limit_exceeded_handler)
|
| 91 |
+
|
| 92 |
+
# Configurer CORS (autoriser tous les domaines en dev)
|
| 93 |
+
app.add_middleware(
|
| 94 |
+
CORSMiddleware,
|
| 95 |
+
allow_origins=["*"],
|
| 96 |
+
allow_credentials=True,
|
| 97 |
+
allow_methods=["*"],
|
| 98 |
+
allow_headers=["*"],
|
| 99 |
+
)
|
| 100 |
+
|
| 101 |
+
|
| 102 |
+
# Middleware de logging des requêtes
|
| 103 |
+
@app.middleware("http")
|
| 104 |
+
async def log_requests(request: Request, call_next):
|
| 105 |
+
"""
|
| 106 |
+
Middleware pour logger toutes les requêtes HTTP.
|
| 107 |
+
"""
|
| 108 |
+
start_time = time.time()
|
| 109 |
+
|
| 110 |
+
# Traiter la requête
|
| 111 |
+
response = await call_next(request)
|
| 112 |
+
|
| 113 |
+
# Calculer la durée
|
| 114 |
+
duration_ms = (time.time() - start_time) * 1000
|
| 115 |
+
|
| 116 |
+
# Logger
|
| 117 |
+
log_request(
|
| 118 |
+
method=request.method,
|
| 119 |
+
path=request.url.path,
|
| 120 |
+
status_code=response.status_code,
|
| 121 |
+
duration_ms=duration_ms,
|
| 122 |
+
client_host=request.client.host if request.client else None,
|
| 123 |
+
)
|
| 124 |
+
|
| 125 |
+
return response
|
| 126 |
+
|
| 127 |
+
|
| 128 |
+
@app.get("/health", response_model=HealthCheck, tags=["Monitoring"])
|
| 129 |
+
async def health_check():
|
| 130 |
+
"""
|
| 131 |
+
Health check endpoint pour monitoring.
|
| 132 |
+
|
| 133 |
+
Vérifie que l'API est opérationnelle et que le modèle est chargé.
|
| 134 |
+
|
| 135 |
+
Returns:
|
| 136 |
+
HealthCheck: Status de l'API et du modèle.
|
| 137 |
+
|
| 138 |
+
Raises:
|
| 139 |
+
HTTPException: 503 si le modèle n'est pas disponible.
|
| 140 |
+
"""
|
| 141 |
+
try:
|
| 142 |
+
model_info = get_model_info()
|
| 143 |
+
|
| 144 |
+
return HealthCheck(
|
| 145 |
+
status="healthy",
|
| 146 |
+
model_loaded=model_info.get("cached", False),
|
| 147 |
+
model_type=model_info.get("model_type", "Unknown"),
|
| 148 |
+
version=API_VERSION,
|
| 149 |
+
)
|
| 150 |
+
except Exception as e:
|
| 151 |
+
raise HTTPException(
|
| 152 |
+
status_code=503,
|
| 153 |
+
detail={
|
| 154 |
+
"status": "unhealthy",
|
| 155 |
+
"error": "Model not available",
|
| 156 |
+
"message": str(e),
|
| 157 |
+
},
|
| 158 |
+
)
|
| 159 |
+
|
| 160 |
+
|
| 161 |
+
@app.post(
|
| 162 |
+
"/predict",
|
| 163 |
+
response_model=PredictionOutput,
|
| 164 |
+
tags=["Prediction"],
|
| 165 |
+
dependencies=[Depends(verify_api_key)] if settings.is_api_key_required else [],
|
| 166 |
+
)
|
| 167 |
+
@limiter.limit("20/minute")
|
| 168 |
+
async def predict(request: Request, employee: EmployeeInput):
|
| 169 |
+
"""
|
| 170 |
+
Endpoint de prédiction du turnover d'un employé.
|
| 171 |
+
|
| 172 |
+
**PROTÉGÉ PAR API KEY** : Requiert le header `X-API-Key` en production.
|
| 173 |
+
|
| 174 |
+
Prend en entrée les données d'un employé, applique le preprocessing
|
| 175 |
+
et retourne la prédiction avec les probabilités.
|
| 176 |
+
|
| 177 |
+
Args:
|
| 178 |
+
employee: Données de l'employé validées par Pydantic.
|
| 179 |
+
|
| 180 |
+
Returns:
|
| 181 |
+
PredictionOutput: Prédiction et probabilités.
|
| 182 |
+
|
| 183 |
+
Raises:
|
| 184 |
+
HTTPException: 401 si API key invalide ou manquante.
|
| 185 |
+
HTTPException: 500 si erreur lors de la prédiction.
|
| 186 |
+
|
| 187 |
+
Examples:
|
| 188 |
+
```bash
|
| 189 |
+
# Avec authentification
|
| 190 |
+
curl -X POST http://localhost:8000/predict \\
|
| 191 |
+
-H "X-API-Key: your-secret-key" \\
|
| 192 |
+
-H "Content-Type: application/json" \\
|
| 193 |
+
-d '{...}'
|
| 194 |
+
```
|
| 195 |
+
"""
|
| 196 |
+
try:
|
| 197 |
+
# 1. Charger le modèle
|
| 198 |
+
model = load_model()
|
| 199 |
+
|
| 200 |
+
# 2. Préprocessing
|
| 201 |
+
X = preprocess_for_prediction(employee)
|
| 202 |
+
|
| 203 |
+
# 3. Prédiction
|
| 204 |
+
prediction = int(model.predict(X)[0])
|
| 205 |
+
|
| 206 |
+
# 4. Probabilités (si le modèle supporte predict_proba)
|
| 207 |
+
try:
|
| 208 |
+
probabilities = model.predict_proba(X)[0]
|
| 209 |
+
prob_0 = float(probabilities[0])
|
| 210 |
+
prob_1 = float(probabilities[1])
|
| 211 |
+
except AttributeError:
|
| 212 |
+
# Si le modèle ne supporte pas predict_proba
|
| 213 |
+
prob_0 = 1.0 if prediction == 0 else 0.0
|
| 214 |
+
prob_1 = 1.0 if prediction == 1 else 0.0
|
| 215 |
+
|
| 216 |
+
# 5. Niveau de risque
|
| 217 |
+
if prob_1 < 0.3:
|
| 218 |
+
risk_level = "Low"
|
| 219 |
+
elif prob_1 < 0.7:
|
| 220 |
+
risk_level = "Medium"
|
| 221 |
+
else:
|
| 222 |
+
risk_level = "High"
|
| 223 |
+
|
| 224 |
+
# 6. Enregistrer dans la base de données
|
| 225 |
+
try:
|
| 226 |
+
from sqlalchemy import create_engine
|
| 227 |
+
from sqlalchemy.orm import sessionmaker
|
| 228 |
+
from db_models import MLLog
|
| 229 |
+
|
| 230 |
+
engine = create_engine(settings.DATABASE_URL)
|
| 231 |
+
Session = sessionmaker(bind=engine)
|
| 232 |
+
session = Session()
|
| 233 |
+
|
| 234 |
+
log_entry = MLLog(
|
| 235 |
+
input_json=employee.dict(),
|
| 236 |
+
prediction="Oui" if prediction == 1 else "Non",
|
| 237 |
+
)
|
| 238 |
+
session.add(log_entry)
|
| 239 |
+
session.commit()
|
| 240 |
+
session.close()
|
| 241 |
+
|
| 242 |
+
logger.info(f"Prediction logged to database: {prediction}")
|
| 243 |
+
except Exception as db_error:
|
| 244 |
+
logger.warning(f"Failed to log prediction to database: {db_error}")
|
| 245 |
+
|
| 246 |
+
return PredictionOutput(
|
| 247 |
+
prediction=prediction,
|
| 248 |
+
probability_0=prob_0,
|
| 249 |
+
probability_1=prob_1,
|
| 250 |
+
risk_level=risk_level,
|
| 251 |
+
)
|
| 252 |
+
|
| 253 |
+
except Exception:
|
| 254 |
+
logger.exception("Unexpected error during prediction")
|
| 255 |
+
raise HTTPException(
|
| 256 |
+
status_code=500,
|
| 257 |
+
detail={
|
| 258 |
+
"error": "Prediction failed",
|
| 259 |
+
"message": "An unexpected error occurred. Please contact support.",
|
| 260 |
+
},
|
| 261 |
+
)
|
| 262 |
+
|
| 263 |
+
|
| 264 |
+
@app.post(
|
| 265 |
+
"/predict/batch",
|
| 266 |
+
response_model=BatchPredictionOutput,
|
| 267 |
+
tags=["Prediction"],
|
| 268 |
+
dependencies=[Depends(verify_api_key)] if settings.is_api_key_required else [],
|
| 269 |
+
)
|
| 270 |
+
@limiter.limit("5/minute")
|
| 271 |
+
async def predict_batch(
|
| 272 |
+
request: Request,
|
| 273 |
+
sondage_file: UploadFile = File(..., description="Fichier CSV du sondage"),
|
| 274 |
+
eval_file: UploadFile = File(..., description="Fichier CSV des évaluations"),
|
| 275 |
+
sirh_file: UploadFile = File(..., description="Fichier CSV SIRH"),
|
| 276 |
+
):
|
| 277 |
+
"""
|
| 278 |
+
Endpoint de prédiction batch à partir de fichiers CSV.
|
| 279 |
+
|
| 280 |
+
**PROTÉGÉ PAR API KEY** : Requiert le header `X-API-Key` en production.
|
| 281 |
+
|
| 282 |
+
Prend en entrée les 3 fichiers CSV (sondage, évaluation, SIRH),
|
| 283 |
+
les fusionne, applique le preprocessing et retourne les prédictions
|
| 284 |
+
pour tous les employés.
|
| 285 |
+
|
| 286 |
+
Args:
|
| 287 |
+
sondage_file: Fichier CSV contenant les données de sondage.
|
| 288 |
+
eval_file: Fichier CSV contenant les données d'évaluation.
|
| 289 |
+
sirh_file: Fichier CSV contenant les données SIRH.
|
| 290 |
+
|
| 291 |
+
Returns:
|
| 292 |
+
BatchPredictionOutput: Prédictions pour tous les employés.
|
| 293 |
+
|
| 294 |
+
Raises:
|
| 295 |
+
HTTPException: 400 si les fichiers sont invalides.
|
| 296 |
+
HTTPException: 500 si erreur lors du traitement.
|
| 297 |
+
"""
|
| 298 |
+
try:
|
| 299 |
+
# 1. Lire les fichiers CSV
|
| 300 |
+
sondage_content = await sondage_file.read()
|
| 301 |
+
eval_content = await eval_file.read()
|
| 302 |
+
sirh_content = await sirh_file.read()
|
| 303 |
+
|
| 304 |
+
sondage_df = pd.read_csv(io.BytesIO(sondage_content))
|
| 305 |
+
eval_df = pd.read_csv(io.BytesIO(eval_content))
|
| 306 |
+
sirh_df = pd.read_csv(io.BytesIO(sirh_content))
|
| 307 |
+
|
| 308 |
+
logger.info(
|
| 309 |
+
f"Fichiers CSV chargés: sondage={len(sondage_df)}, "
|
| 310 |
+
f"eval={len(eval_df)}, sirh={len(sirh_df)} lignes"
|
| 311 |
+
)
|
| 312 |
+
|
| 313 |
+
# 2. Fusionner les DataFrames
|
| 314 |
+
merged_df = merge_csv_dataframes(sondage_df, eval_df, sirh_df)
|
| 315 |
+
employee_ids = merged_df["original_employee_id"].tolist()
|
| 316 |
+
merged_df = merged_df.drop(columns=["original_employee_id"])
|
| 317 |
+
|
| 318 |
+
# Supprimer la colonne cible si présente
|
| 319 |
+
if "a_quitte_l_entreprise" in merged_df.columns:
|
| 320 |
+
merged_df = merged_df.drop(columns=["a_quitte_l_entreprise"])
|
| 321 |
+
|
| 322 |
+
logger.info(f"DataFrame fusionné: {len(merged_df)} employés")
|
| 323 |
+
|
| 324 |
+
# 3. Preprocessing
|
| 325 |
+
X = preprocess_dataframe_for_prediction(merged_df)
|
| 326 |
+
|
| 327 |
+
# 4. Charger le modèle et prédire
|
| 328 |
+
model = load_model()
|
| 329 |
+
predictions = model.predict(X.values)
|
| 330 |
+
probabilities = model.predict_proba(X.values)
|
| 331 |
+
|
| 332 |
+
# 5. Construire la réponse
|
| 333 |
+
results = []
|
| 334 |
+
risk_counts = {"Low": 0, "Medium": 0, "High": 0}
|
| 335 |
+
leave_count = 0
|
| 336 |
+
|
| 337 |
+
for i, emp_id in enumerate(employee_ids):
|
| 338 |
+
prob_stay = float(probabilities[i][0])
|
| 339 |
+
prob_leave = float(probabilities[i][1])
|
| 340 |
+
pred = int(predictions[i])
|
| 341 |
+
|
| 342 |
+
if prob_leave < 0.3:
|
| 343 |
+
risk = "Low"
|
| 344 |
+
elif prob_leave < 0.7:
|
| 345 |
+
risk = "Medium"
|
| 346 |
+
else:
|
| 347 |
+
risk = "High"
|
| 348 |
+
|
| 349 |
+
risk_counts[risk] += 1
|
| 350 |
+
if pred == 1:
|
| 351 |
+
leave_count += 1
|
| 352 |
+
|
| 353 |
+
results.append(
|
| 354 |
+
EmployeePrediction(
|
| 355 |
+
employee_id=int(emp_id),
|
| 356 |
+
prediction=pred,
|
| 357 |
+
probability_stay=prob_stay,
|
| 358 |
+
probability_leave=prob_leave,
|
| 359 |
+
risk_level=risk,
|
| 360 |
+
)
|
| 361 |
+
)
|
| 362 |
+
|
| 363 |
+
summary = {
|
| 364 |
+
"total_stay": len(results) - leave_count,
|
| 365 |
+
"total_leave": leave_count,
|
| 366 |
+
"high_risk_count": risk_counts["High"],
|
| 367 |
+
"medium_risk_count": risk_counts["Medium"],
|
| 368 |
+
"low_risk_count": risk_counts["Low"],
|
| 369 |
+
}
|
| 370 |
+
|
| 371 |
+
logger.info(f"Prédictions terminées: {summary}")
|
| 372 |
+
|
| 373 |
+
return BatchPredictionOutput(
|
| 374 |
+
total_employees=len(results),
|
| 375 |
+
predictions=results,
|
| 376 |
+
summary=summary,
|
| 377 |
+
)
|
| 378 |
+
|
| 379 |
+
except pd.errors.EmptyDataError:
|
| 380 |
+
raise HTTPException(
|
| 381 |
+
status_code=400,
|
| 382 |
+
detail={
|
| 383 |
+
"error": "Empty CSV file",
|
| 384 |
+
"message": "Un des fichiers CSV est vide.",
|
| 385 |
+
},
|
| 386 |
+
)
|
| 387 |
+
except KeyError as e:
|
| 388 |
+
raise HTTPException(
|
| 389 |
+
status_code=400,
|
| 390 |
+
detail={
|
| 391 |
+
"error": "Missing column",
|
| 392 |
+
"message": f"Colonne manquante dans les CSV: {e}",
|
| 393 |
+
},
|
| 394 |
+
)
|
| 395 |
+
except Exception as e:
|
| 396 |
+
logger.exception("Unexpected error during batch prediction")
|
| 397 |
+
raise HTTPException(
|
| 398 |
+
status_code=500,
|
| 399 |
+
detail={
|
| 400 |
+
"error": "Batch prediction failed",
|
| 401 |
+
"message": str(e),
|
| 402 |
+
},
|
| 403 |
+
)
|
| 404 |
+
|
| 405 |
+
|
| 406 |
+
# Monter l'interface Gradio sur / (racine pour HuggingFace Spaces)
|
| 407 |
+
gradio_app = create_gradio_interface()
|
| 408 |
+
app = gr.mount_gradio_app(app, gradio_app, path="/")
|
| 409 |
+
|
| 410 |
+
|
| 411 |
+
if __name__ == "__main__":
|
| 412 |
+
import uvicorn
|
| 413 |
+
|
| 414 |
+
print("\U0001f680 Lancement de l'API en mode d\u00e9veloppement...")
|
| 415 |
+
print("\U0001f4d6 Documentation : http://localhost:8000/docs")
|
| 416 |
+
print("\U0001f3a8 Interface Gradio : http://localhost:8000/")
|
| 417 |
+
|
| 418 |
+
uvicorn.run(
|
| 419 |
+
"app:app",
|
| 420 |
+
host="0.0.0.0",
|
| 421 |
+
port=8000,
|
| 422 |
+
reload=True,
|
| 423 |
+
log_level="info",
|
| 424 |
+
)
|
app.py
CHANGED
|
@@ -1,424 +1,16 @@
|
|
| 1 |
#!/usr/bin/env python3
|
| 2 |
"""
|
| 3 |
-
|
| 4 |
|
| 5 |
-
|
| 6 |
-
- Validation stricte des inputs via Pydantic
|
| 7 |
-
- Preprocessing automatique
|
| 8 |
-
- Health check pour monitoring
|
| 9 |
-
- Documentation OpenAPI/Swagger automatique
|
| 10 |
-
- Interface Gradio pour utilisation interactive
|
| 11 |
-
- Endpoint batch pour traitement de fichiers CSV
|
| 12 |
"""
|
| 13 |
-
import
|
| 14 |
-
import
|
| 15 |
-
from contextlib import asynccontextmanager
|
| 16 |
|
| 17 |
-
|
| 18 |
-
import pandas as pd
|
| 19 |
-
from fastapi import Depends, FastAPI, File, HTTPException, Request, UploadFile
|
| 20 |
-
from fastapi.middleware.cors import CORSMiddleware
|
| 21 |
-
from slowapi import _rate_limit_exceeded_handler
|
| 22 |
-
from slowapi.errors import RateLimitExceeded
|
| 23 |
-
|
| 24 |
-
from src.auth import verify_api_key
|
| 25 |
-
from src.config import get_settings
|
| 26 |
-
from src.gradio_ui import create_gradio_interface
|
| 27 |
-
from src.logger import logger, log_model_load, log_request
|
| 28 |
-
from src.models import get_model_info, load_model
|
| 29 |
-
from src.preprocessing import (
|
| 30 |
-
merge_csv_dataframes,
|
| 31 |
-
preprocess_dataframe_for_prediction,
|
| 32 |
-
preprocess_for_prediction,
|
| 33 |
-
)
|
| 34 |
-
from src.rate_limit import limiter
|
| 35 |
-
from src.schemas import (
|
| 36 |
-
BatchPredictionOutput,
|
| 37 |
-
EmployeeInput,
|
| 38 |
-
EmployeePrediction,
|
| 39 |
-
HealthCheck,
|
| 40 |
-
PredictionOutput,
|
| 41 |
-
)
|
| 42 |
-
|
| 43 |
-
# Charger la configuration
|
| 44 |
-
settings = get_settings()
|
| 45 |
-
API_VERSION = settings.API_VERSION
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
@asynccontextmanager
|
| 49 |
-
async def lifespan(app: FastAPI):
|
| 50 |
-
"""
|
| 51 |
-
Gestion du cycle de vie de l'application.
|
| 52 |
-
|
| 53 |
-
Charge le modèle au démarrage et le garde en cache.
|
| 54 |
-
"""
|
| 55 |
-
logger.info(
|
| 56 |
-
"🚀 Démarrage de l'API Employee Turnover...", extra={"version": API_VERSION}
|
| 57 |
-
)
|
| 58 |
-
|
| 59 |
-
start_time = time.time()
|
| 60 |
-
try:
|
| 61 |
-
# Pré-charger le modèle au démarrage
|
| 62 |
-
model = load_model()
|
| 63 |
-
duration_ms = (time.time() - start_time) * 1000
|
| 64 |
-
|
| 65 |
-
model_type = type(model).__name__
|
| 66 |
-
log_model_load(model_type, duration_ms, True)
|
| 67 |
-
logger.info("✅ Modèle chargé avec succès")
|
| 68 |
-
except Exception as e:
|
| 69 |
-
duration_ms = (time.time() - start_time) * 1000
|
| 70 |
-
log_model_load("Unknown", duration_ms, False)
|
| 71 |
-
logger.error("Le modèle n'a pas pu être chargé", extra={"error": str(e)})
|
| 72 |
-
|
| 73 |
-
yield # L'application tourne
|
| 74 |
-
|
| 75 |
-
logger.info("🛑 Arrêt de l'API")
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
# Créer l'application FastAPI
|
| 79 |
-
app = FastAPI(
|
| 80 |
-
title="Employee Turnover Prediction API",
|
| 81 |
-
description="API de prédiction du turnover des employés avec XGBoost + SMOTE",
|
| 82 |
-
version=API_VERSION,
|
| 83 |
-
lifespan=lifespan,
|
| 84 |
-
docs_url="/docs",
|
| 85 |
-
redoc_url="/redoc",
|
| 86 |
-
)
|
| 87 |
-
|
| 88 |
-
# Ajouter rate limiting
|
| 89 |
-
app.state.limiter = limiter
|
| 90 |
-
app.add_exception_handler(RateLimitExceeded, _rate_limit_exceeded_handler)
|
| 91 |
-
|
| 92 |
-
# Configurer CORS (autoriser tous les domaines en dev)
|
| 93 |
-
app.add_middleware(
|
| 94 |
-
CORSMiddleware,
|
| 95 |
-
allow_origins=["*"],
|
| 96 |
-
allow_credentials=True,
|
| 97 |
-
allow_methods=["*"],
|
| 98 |
-
allow_headers=["*"],
|
| 99 |
-
)
|
| 100 |
-
|
| 101 |
-
|
| 102 |
-
# Middleware de logging des requêtes
|
| 103 |
-
@app.middleware("http")
|
| 104 |
-
async def log_requests(request: Request, call_next):
|
| 105 |
-
"""
|
| 106 |
-
Middleware pour logger toutes les requêtes HTTP.
|
| 107 |
-
"""
|
| 108 |
-
start_time = time.time()
|
| 109 |
-
|
| 110 |
-
# Traiter la requête
|
| 111 |
-
response = await call_next(request)
|
| 112 |
-
|
| 113 |
-
# Calculer la durée
|
| 114 |
-
duration_ms = (time.time() - start_time) * 1000
|
| 115 |
-
|
| 116 |
-
# Logger
|
| 117 |
-
log_request(
|
| 118 |
-
method=request.method,
|
| 119 |
-
path=request.url.path,
|
| 120 |
-
status_code=response.status_code,
|
| 121 |
-
duration_ms=duration_ms,
|
| 122 |
-
client_host=request.client.host if request.client else None,
|
| 123 |
-
)
|
| 124 |
-
|
| 125 |
-
return response
|
| 126 |
-
|
| 127 |
-
|
| 128 |
-
@app.get("/health", response_model=HealthCheck, tags=["Monitoring"])
|
| 129 |
-
async def health_check():
|
| 130 |
-
"""
|
| 131 |
-
Health check endpoint pour monitoring.
|
| 132 |
-
|
| 133 |
-
Vérifie que l'API est opérationnelle et que le modèle est chargé.
|
| 134 |
-
|
| 135 |
-
Returns:
|
| 136 |
-
HealthCheck: Status de l'API et du modèle.
|
| 137 |
-
|
| 138 |
-
Raises:
|
| 139 |
-
HTTPException: 503 si le modèle n'est pas disponible.
|
| 140 |
-
"""
|
| 141 |
-
try:
|
| 142 |
-
model_info = get_model_info()
|
| 143 |
-
|
| 144 |
-
return HealthCheck(
|
| 145 |
-
status="healthy",
|
| 146 |
-
model_loaded=model_info.get("cached", False),
|
| 147 |
-
model_type=model_info.get("model_type", "Unknown"),
|
| 148 |
-
version=API_VERSION,
|
| 149 |
-
)
|
| 150 |
-
except Exception as e:
|
| 151 |
-
raise HTTPException(
|
| 152 |
-
status_code=503,
|
| 153 |
-
detail={
|
| 154 |
-
"status": "unhealthy",
|
| 155 |
-
"error": "Model not available",
|
| 156 |
-
"message": str(e),
|
| 157 |
-
},
|
| 158 |
-
)
|
| 159 |
-
|
| 160 |
-
|
| 161 |
-
@app.post(
|
| 162 |
-
"/predict",
|
| 163 |
-
response_model=PredictionOutput,
|
| 164 |
-
tags=["Prediction"],
|
| 165 |
-
dependencies=[Depends(verify_api_key)] if settings.is_api_key_required else [],
|
| 166 |
-
)
|
| 167 |
-
@limiter.limit("20/minute")
|
| 168 |
-
async def predict(request: Request, employee: EmployeeInput):
|
| 169 |
-
"""
|
| 170 |
-
Endpoint de prédiction du turnover d'un employé.
|
| 171 |
-
|
| 172 |
-
**PROTÉGÉ PAR API KEY** : Requiert le header `X-API-Key` en production.
|
| 173 |
-
|
| 174 |
-
Prend en entrée les données d'un employé, applique le preprocessing
|
| 175 |
-
et retourne la prédiction avec les probabilités.
|
| 176 |
-
|
| 177 |
-
Args:
|
| 178 |
-
employee: Données de l'employé validées par Pydantic.
|
| 179 |
-
|
| 180 |
-
Returns:
|
| 181 |
-
PredictionOutput: Prédiction et probabilités.
|
| 182 |
-
|
| 183 |
-
Raises:
|
| 184 |
-
HTTPException: 401 si API key invalide ou manquante.
|
| 185 |
-
HTTPException: 500 si erreur lors de la prédiction.
|
| 186 |
-
|
| 187 |
-
Examples:
|
| 188 |
-
```bash
|
| 189 |
-
# Avec authentification
|
| 190 |
-
curl -X POST http://localhost:8000/predict \\
|
| 191 |
-
-H "X-API-Key: your-secret-key" \\
|
| 192 |
-
-H "Content-Type: application/json" \\
|
| 193 |
-
-d '{...}'
|
| 194 |
-
```
|
| 195 |
-
"""
|
| 196 |
-
try:
|
| 197 |
-
# 1. Charger le modèle
|
| 198 |
-
model = load_model()
|
| 199 |
-
|
| 200 |
-
# 2. Préprocessing
|
| 201 |
-
X = preprocess_for_prediction(employee)
|
| 202 |
-
|
| 203 |
-
# 3. Prédiction
|
| 204 |
-
prediction = int(model.predict(X)[0])
|
| 205 |
-
|
| 206 |
-
# 4. Probabilités (si le modèle supporte predict_proba)
|
| 207 |
-
try:
|
| 208 |
-
probabilities = model.predict_proba(X)[0]
|
| 209 |
-
prob_0 = float(probabilities[0])
|
| 210 |
-
prob_1 = float(probabilities[1])
|
| 211 |
-
except AttributeError:
|
| 212 |
-
# Si le modèle ne supporte pas predict_proba
|
| 213 |
-
prob_0 = 1.0 if prediction == 0 else 0.0
|
| 214 |
-
prob_1 = 1.0 if prediction == 1 else 0.0
|
| 215 |
-
|
| 216 |
-
# 5. Niveau de risque
|
| 217 |
-
if prob_1 < 0.3:
|
| 218 |
-
risk_level = "Low"
|
| 219 |
-
elif prob_1 < 0.7:
|
| 220 |
-
risk_level = "Medium"
|
| 221 |
-
else:
|
| 222 |
-
risk_level = "High"
|
| 223 |
-
|
| 224 |
-
# 6. Enregistrer dans la base de données
|
| 225 |
-
try:
|
| 226 |
-
from sqlalchemy import create_engine
|
| 227 |
-
from sqlalchemy.orm import sessionmaker
|
| 228 |
-
from db_models import MLLog
|
| 229 |
-
|
| 230 |
-
engine = create_engine(settings.DATABASE_URL)
|
| 231 |
-
Session = sessionmaker(bind=engine)
|
| 232 |
-
session = Session()
|
| 233 |
-
|
| 234 |
-
log_entry = MLLog(
|
| 235 |
-
input_json=employee.dict(),
|
| 236 |
-
prediction="Oui" if prediction == 1 else "Non",
|
| 237 |
-
)
|
| 238 |
-
session.add(log_entry)
|
| 239 |
-
session.commit()
|
| 240 |
-
session.close()
|
| 241 |
-
|
| 242 |
-
logger.info(f"Prediction logged to database: {prediction}")
|
| 243 |
-
except Exception as db_error:
|
| 244 |
-
logger.warning(f"Failed to log prediction to database: {db_error}")
|
| 245 |
-
|
| 246 |
-
return PredictionOutput(
|
| 247 |
-
prediction=prediction,
|
| 248 |
-
probability_0=prob_0,
|
| 249 |
-
probability_1=prob_1,
|
| 250 |
-
risk_level=risk_level,
|
| 251 |
-
)
|
| 252 |
-
|
| 253 |
-
except Exception:
|
| 254 |
-
logger.exception("Unexpected error during prediction")
|
| 255 |
-
raise HTTPException(
|
| 256 |
-
status_code=500,
|
| 257 |
-
detail={
|
| 258 |
-
"error": "Prediction failed",
|
| 259 |
-
"message": "An unexpected error occurred. Please contact support.",
|
| 260 |
-
},
|
| 261 |
-
)
|
| 262 |
-
|
| 263 |
-
|
| 264 |
-
@app.post(
|
| 265 |
-
"/predict/batch",
|
| 266 |
-
response_model=BatchPredictionOutput,
|
| 267 |
-
tags=["Prediction"],
|
| 268 |
-
dependencies=[Depends(verify_api_key)] if settings.is_api_key_required else [],
|
| 269 |
-
)
|
| 270 |
-
@limiter.limit("5/minute")
|
| 271 |
-
async def predict_batch(
|
| 272 |
-
request: Request,
|
| 273 |
-
sondage_file: UploadFile = File(..., description="Fichier CSV du sondage"),
|
| 274 |
-
eval_file: UploadFile = File(..., description="Fichier CSV des évaluations"),
|
| 275 |
-
sirh_file: UploadFile = File(..., description="Fichier CSV SIRH"),
|
| 276 |
-
):
|
| 277 |
-
"""
|
| 278 |
-
Endpoint de prédiction batch à partir de fichiers CSV.
|
| 279 |
-
|
| 280 |
-
**PROTÉGÉ PAR API KEY** : Requiert le header `X-API-Key` en production.
|
| 281 |
-
|
| 282 |
-
Prend en entrée les 3 fichiers CSV (sondage, évaluation, SIRH),
|
| 283 |
-
les fusionne, applique le preprocessing et retourne les prédictions
|
| 284 |
-
pour tous les employés.
|
| 285 |
-
|
| 286 |
-
Args:
|
| 287 |
-
sondage_file: Fichier CSV contenant les données de sondage.
|
| 288 |
-
eval_file: Fichier CSV contenant les données d'évaluation.
|
| 289 |
-
sirh_file: Fichier CSV contenant les données SIRH.
|
| 290 |
-
|
| 291 |
-
Returns:
|
| 292 |
-
BatchPredictionOutput: Prédictions pour tous les employés.
|
| 293 |
-
|
| 294 |
-
Raises:
|
| 295 |
-
HTTPException: 400 si les fichiers sont invalides.
|
| 296 |
-
HTTPException: 500 si erreur lors du traitement.
|
| 297 |
-
"""
|
| 298 |
-
try:
|
| 299 |
-
# 1. Lire les fichiers CSV
|
| 300 |
-
sondage_content = await sondage_file.read()
|
| 301 |
-
eval_content = await eval_file.read()
|
| 302 |
-
sirh_content = await sirh_file.read()
|
| 303 |
-
|
| 304 |
-
sondage_df = pd.read_csv(io.BytesIO(sondage_content))
|
| 305 |
-
eval_df = pd.read_csv(io.BytesIO(eval_content))
|
| 306 |
-
sirh_df = pd.read_csv(io.BytesIO(sirh_content))
|
| 307 |
-
|
| 308 |
-
logger.info(
|
| 309 |
-
f"Fichiers CSV chargés: sondage={len(sondage_df)}, "
|
| 310 |
-
f"eval={len(eval_df)}, sirh={len(sirh_df)} lignes"
|
| 311 |
-
)
|
| 312 |
-
|
| 313 |
-
# 2. Fusionner les DataFrames
|
| 314 |
-
merged_df = merge_csv_dataframes(sondage_df, eval_df, sirh_df)
|
| 315 |
-
employee_ids = merged_df["original_employee_id"].tolist()
|
| 316 |
-
merged_df = merged_df.drop(columns=["original_employee_id"])
|
| 317 |
-
|
| 318 |
-
# Supprimer la colonne cible si présente
|
| 319 |
-
if "a_quitte_l_entreprise" in merged_df.columns:
|
| 320 |
-
merged_df = merged_df.drop(columns=["a_quitte_l_entreprise"])
|
| 321 |
-
|
| 322 |
-
logger.info(f"DataFrame fusionné: {len(merged_df)} employés")
|
| 323 |
-
|
| 324 |
-
# 3. Preprocessing
|
| 325 |
-
X = preprocess_dataframe_for_prediction(merged_df)
|
| 326 |
-
|
| 327 |
-
# 4. Charger le modèle et prédire
|
| 328 |
-
model = load_model()
|
| 329 |
-
predictions = model.predict(X.values)
|
| 330 |
-
probabilities = model.predict_proba(X.values)
|
| 331 |
-
|
| 332 |
-
# 5. Construire la réponse
|
| 333 |
-
results = []
|
| 334 |
-
risk_counts = {"Low": 0, "Medium": 0, "High": 0}
|
| 335 |
-
leave_count = 0
|
| 336 |
-
|
| 337 |
-
for i, emp_id in enumerate(employee_ids):
|
| 338 |
-
prob_stay = float(probabilities[i][0])
|
| 339 |
-
prob_leave = float(probabilities[i][1])
|
| 340 |
-
pred = int(predictions[i])
|
| 341 |
-
|
| 342 |
-
if prob_leave < 0.3:
|
| 343 |
-
risk = "Low"
|
| 344 |
-
elif prob_leave < 0.7:
|
| 345 |
-
risk = "Medium"
|
| 346 |
-
else:
|
| 347 |
-
risk = "High"
|
| 348 |
-
|
| 349 |
-
risk_counts[risk] += 1
|
| 350 |
-
if pred == 1:
|
| 351 |
-
leave_count += 1
|
| 352 |
-
|
| 353 |
-
results.append(
|
| 354 |
-
EmployeePrediction(
|
| 355 |
-
employee_id=int(emp_id),
|
| 356 |
-
prediction=pred,
|
| 357 |
-
probability_stay=prob_stay,
|
| 358 |
-
probability_leave=prob_leave,
|
| 359 |
-
risk_level=risk,
|
| 360 |
-
)
|
| 361 |
-
)
|
| 362 |
-
|
| 363 |
-
summary = {
|
| 364 |
-
"total_stay": len(results) - leave_count,
|
| 365 |
-
"total_leave": leave_count,
|
| 366 |
-
"high_risk_count": risk_counts["High"],
|
| 367 |
-
"medium_risk_count": risk_counts["Medium"],
|
| 368 |
-
"low_risk_count": risk_counts["Low"],
|
| 369 |
-
}
|
| 370 |
-
|
| 371 |
-
logger.info(f"Prédictions terminées: {summary}")
|
| 372 |
-
|
| 373 |
-
return BatchPredictionOutput(
|
| 374 |
-
total_employees=len(results),
|
| 375 |
-
predictions=results,
|
| 376 |
-
summary=summary,
|
| 377 |
-
)
|
| 378 |
-
|
| 379 |
-
except pd.errors.EmptyDataError:
|
| 380 |
-
raise HTTPException(
|
| 381 |
-
status_code=400,
|
| 382 |
-
detail={
|
| 383 |
-
"error": "Empty CSV file",
|
| 384 |
-
"message": "Un des fichiers CSV est vide.",
|
| 385 |
-
},
|
| 386 |
-
)
|
| 387 |
-
except KeyError as e:
|
| 388 |
-
raise HTTPException(
|
| 389 |
-
status_code=400,
|
| 390 |
-
detail={
|
| 391 |
-
"error": "Missing column",
|
| 392 |
-
"message": f"Colonne manquante dans les CSV: {e}",
|
| 393 |
-
},
|
| 394 |
-
)
|
| 395 |
-
except Exception as e:
|
| 396 |
-
logger.exception("Unexpected error during batch prediction")
|
| 397 |
-
raise HTTPException(
|
| 398 |
-
status_code=500,
|
| 399 |
-
detail={
|
| 400 |
-
"error": "Batch prediction failed",
|
| 401 |
-
"message": str(e),
|
| 402 |
-
},
|
| 403 |
-
)
|
| 404 |
-
|
| 405 |
-
|
| 406 |
-
# Monter l'interface Gradio sur / (racine pour HuggingFace Spaces)
|
| 407 |
-
gradio_app = create_gradio_interface()
|
| 408 |
-
app = gr.mount_gradio_app(app, gradio_app, path="/")
|
| 409 |
|
|
|
|
|
|
|
| 410 |
|
| 411 |
if __name__ == "__main__":
|
| 412 |
-
|
| 413 |
-
|
| 414 |
-
print("\U0001f680 Lancement de l'API en mode d\u00e9veloppement...")
|
| 415 |
-
print("\U0001f4d6 Documentation : http://localhost:8000/docs")
|
| 416 |
-
print("\U0001f3a8 Interface Gradio : http://localhost:8000/")
|
| 417 |
-
|
| 418 |
-
uvicorn.run(
|
| 419 |
-
"app:app",
|
| 420 |
-
host="0.0.0.0",
|
| 421 |
-
port=8000,
|
| 422 |
-
reload=True,
|
| 423 |
-
log_level="info",
|
| 424 |
-
)
|
|
|
|
| 1 |
#!/usr/bin/env python3
|
| 2 |
"""
|
| 3 |
+
App Gradio pour Hugging Face Spaces.
|
| 4 |
|
| 5 |
+
Lance l'interface Gradio pour la prédiction de turnover.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
"""
|
| 7 |
+
import sys
|
| 8 |
+
import os
|
|
|
|
| 9 |
|
| 10 |
+
from src.gradio_ui import launch_standalone
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
|
| 12 |
+
# Ajouter le répertoire src au path
|
| 13 |
+
sys.path.insert(0, os.path.join(os.path.dirname(__file__), "src"))
|
| 14 |
|
| 15 |
if __name__ == "__main__":
|
| 16 |
+
launch_standalone()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
src/gradio_ui.py
CHANGED
|
@@ -8,6 +8,7 @@ Cette interface permet de:
|
|
| 8 |
- Comprendre les champs requis
|
| 9 |
"""
|
| 10 |
import gradio as gr
|
|
|
|
| 11 |
|
| 12 |
from src.models import get_model_info, load_model
|
| 13 |
from src.preprocessing import preprocess_for_prediction
|
|
@@ -123,30 +124,33 @@ def predict_turnover(
|
|
| 123 |
|
| 124 |
confidence = max(prob_0, prob_1) * 100
|
| 125 |
|
| 126 |
-
# Enregistrer dans la base de données (
|
|
|
|
| 127 |
try:
|
| 128 |
-
|
| 129 |
-
|
| 130 |
-
|
| 131 |
-
|
| 132 |
-
|
| 133 |
-
|
| 134 |
-
|
| 135 |
-
|
| 136 |
-
|
| 137 |
-
|
| 138 |
-
|
| 139 |
-
|
| 140 |
-
|
| 141 |
-
|
| 142 |
-
|
| 143 |
-
|
| 144 |
-
|
| 145 |
-
|
| 146 |
-
|
| 147 |
-
|
| 148 |
-
|
| 149 |
-
|
|
|
|
|
|
|
| 150 |
except Exception as db_error:
|
| 151 |
db_status = f"⚠️ Erreur DB: {str(db_error)}"
|
| 152 |
|
|
@@ -597,7 +601,7 @@ def launch_standalone():
|
|
| 597 |
demo.launch(
|
| 598 |
server_name="0.0.0.0",
|
| 599 |
server_port=7860,
|
| 600 |
-
share=False,
|
| 601 |
show_error=True,
|
| 602 |
)
|
| 603 |
|
|
|
|
| 8 |
- Comprendre les champs requis
|
| 9 |
"""
|
| 10 |
import gradio as gr
|
| 11 |
+
import os
|
| 12 |
|
| 13 |
from src.models import get_model_info, load_model
|
| 14 |
from src.preprocessing import preprocess_for_prediction
|
|
|
|
| 124 |
|
| 125 |
confidence = max(prob_0, prob_1) * 100
|
| 126 |
|
| 127 |
+
# Enregistrer dans la base de données (uniquement en local)
|
| 128 |
+
db_status = "ℹ️ DB désactivée sur HF Spaces"
|
| 129 |
try:
|
| 130 |
+
# Vérifier si on est sur HF Spaces (variable d'environnement)
|
| 131 |
+
if os.getenv("SPACE_ID") is None: # Pas sur HF Spaces
|
| 132 |
+
from sqlalchemy import create_engine
|
| 133 |
+
from sqlalchemy.orm import sessionmaker
|
| 134 |
+
from src.config import get_settings
|
| 135 |
+
|
| 136 |
+
settings = get_settings()
|
| 137 |
+
engine = create_engine(settings.DATABASE_URL)
|
| 138 |
+
Session = sessionmaker(bind=engine)
|
| 139 |
+
session = Session()
|
| 140 |
+
|
| 141 |
+
# Importer le modèle MLLog
|
| 142 |
+
from db_models import MLLog
|
| 143 |
+
|
| 144 |
+
# Créer le log
|
| 145 |
+
log_entry = MLLog(
|
| 146 |
+
input_json=employee.dict(), # Convertir Pydantic en dict
|
| 147 |
+
prediction="Oui" if prediction == 1 else "Non",
|
| 148 |
+
)
|
| 149 |
+
session.add(log_entry)
|
| 150 |
+
session.commit()
|
| 151 |
+
session.close()
|
| 152 |
+
|
| 153 |
+
db_status = "✅ Enregistré en DB"
|
| 154 |
except Exception as db_error:
|
| 155 |
db_status = f"⚠️ Erreur DB: {str(db_error)}"
|
| 156 |
|
|
|
|
| 601 |
demo.launch(
|
| 602 |
server_name="0.0.0.0",
|
| 603 |
server_port=7860,
|
| 604 |
+
share=False,
|
| 605 |
show_error=True,
|
| 606 |
)
|
| 607 |
|