Spaces:
Running
Running
Upload folder using huggingface_hub
Browse files
main.py
ADDED
|
@@ -0,0 +1,175 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python3
|
| 2 |
+
"""
|
| 3 |
+
Pipeline principal d'entraînement du modèle Employee Turnover.
|
| 4 |
+
|
| 5 |
+
Ce script enchaîne:
|
| 6 |
+
1. Chargement et préprocessing des données
|
| 7 |
+
2. Entraînement du modèle XGBoost avec RandomizedSearchCV et SMOTE
|
| 8 |
+
3. Logging des résultats dans MLflow (params, metrics, artifacts, model)
|
| 9 |
+
4. Sauvegarde des encoders et scaler pour utilisation future
|
| 10 |
+
|
| 11 |
+
Usage:
|
| 12 |
+
python main.py
|
| 13 |
+
|
| 14 |
+
Le modèle et les artifacts sont enregistrés dans MLflow pour:
|
| 15 |
+
- Suivi des expérimentations
|
| 16 |
+
- Reproductibilité
|
| 17 |
+
Déploiement via Model Registry
|
| 18 |
+
"""
|
| 19 |
+
from pathlib import Path
|
| 20 |
+
|
| 21 |
+
import joblib
|
| 22 |
+
import mlflow
|
| 23 |
+
import mlflow.sklearn
|
| 24 |
+
from ml_model.preprocess import preprocess_data
|
| 25 |
+
from ml_model.train_model import train_model
|
| 26 |
+
|
| 27 |
+
|
| 28 |
+
def main():
|
| 29 |
+
"""Pipeline principal d'entraînement."""
|
| 30 |
+
print("=" * 80)
|
| 31 |
+
print("🚀 PIPELINE D'ENTRAÎNEMENT - Employee Turnover Prediction")
|
| 32 |
+
print("=" * 80)
|
| 33 |
+
print()
|
| 34 |
+
|
| 35 |
+
# Configuration MLflow
|
| 36 |
+
mlflow.set_tracking_uri("sqlite:///mlflow.db")
|
| 37 |
+
mlflow.set_experiment("Employee_Turnover_Training")
|
| 38 |
+
|
| 39 |
+
print("📊 Configuration MLflow:")
|
| 40 |
+
print(f" Tracking URI: {mlflow.get_tracking_uri()}")
|
| 41 |
+
print(" Experiment: Employee_Turnover_Training")
|
| 42 |
+
print()
|
| 43 |
+
|
| 44 |
+
# Chemins des données
|
| 45 |
+
data_paths = {
|
| 46 |
+
"sondage_path": "data/extrait_sondage.csv",
|
| 47 |
+
"eval_path": "data/extrait_eval.csv",
|
| 48 |
+
"sirh_path": "data/extrait_sirh.csv",
|
| 49 |
+
}
|
| 50 |
+
|
| 51 |
+
# Vérifier que les fichiers existent
|
| 52 |
+
for name, path in data_paths.items():
|
| 53 |
+
if not Path(path).exists():
|
| 54 |
+
raise FileNotFoundError(f"❌ Fichier manquant: {path}")
|
| 55 |
+
|
| 56 |
+
print("✅ Fichiers de données trouvés")
|
| 57 |
+
print()
|
| 58 |
+
|
| 59 |
+
# ========================================================================
|
| 60 |
+
# ÉTAPE 1 : Préprocessing
|
| 61 |
+
# ========================================================================
|
| 62 |
+
print("1️⃣ PRÉPROCESSING")
|
| 63 |
+
print("-" * 80)
|
| 64 |
+
|
| 65 |
+
X, y, scaler, onehot_encoder, ordinal_encoder = preprocess_data(data_paths)
|
| 66 |
+
|
| 67 |
+
print(f" Forme X: {X.shape}")
|
| 68 |
+
print(f" Forme y: {y.shape}")
|
| 69 |
+
print(f" Classes: {y.value_counts().to_dict()}")
|
| 70 |
+
print(f" Ratio déséquilibre: {(y == 0).sum() / (y == 1).sum():.2f}:1")
|
| 71 |
+
print()
|
| 72 |
+
|
| 73 |
+
# ========================================================================
|
| 74 |
+
# ÉTAPE 2 : Entraînement avec MLflow tracking
|
| 75 |
+
# ========================================================================
|
| 76 |
+
print("2️⃣ ENTRAÎNEMENT")
|
| 77 |
+
print("-" * 80)
|
| 78 |
+
|
| 79 |
+
# Entraînement (déjà avec MLflow tracking dans train_model.py)
|
| 80 |
+
model, best_params, cv_f1 = train_model(X, y)
|
| 81 |
+
|
| 82 |
+
print(" ✅ Modèle entraîné")
|
| 83 |
+
print(f" 🏆 Meilleur F1 CV: {cv_f1:.4f}")
|
| 84 |
+
print()
|
| 85 |
+
|
| 86 |
+
# Récupérer le run actif pour sauvegarder les artifacts
|
| 87 |
+
active_run = mlflow.active_run()
|
| 88 |
+
if active_run is None:
|
| 89 |
+
# Si train_model a fermé le run, on en ouvre un nouveau
|
| 90 |
+
active_run = mlflow.start_run()
|
| 91 |
+
run_id = active_run.info.run_id
|
| 92 |
+
should_end_run = True
|
| 93 |
+
else:
|
| 94 |
+
run_id = active_run.info.run_id
|
| 95 |
+
should_end_run = False
|
| 96 |
+
|
| 97 |
+
# Log des infos dataset
|
| 98 |
+
mlflow.log_param("n_samples", len(X))
|
| 99 |
+
mlflow.log_param("n_features", X.shape[1])
|
| 100 |
+
mlflow.log_param("class_ratio", f"{(y == 0).sum()}:{(y == 1).sum()}")
|
| 101 |
+
|
| 102 |
+
# ========================================================================
|
| 103 |
+
# ÉTAPE 3 : Sauvegarde des artifacts (encoders, scaler)
|
| 104 |
+
# ========================================================================
|
| 105 |
+
print("3️⃣ SAUVEGARDE DES ARTIFACTS")
|
| 106 |
+
print("-" * 80)
|
| 107 |
+
|
| 108 |
+
# Créer dossier temporaire pour artifacts
|
| 109 |
+
artifacts_dir = Path("artifacts_temp")
|
| 110 |
+
artifacts_dir.mkdir(exist_ok=True)
|
| 111 |
+
|
| 112 |
+
# Sauvegarder scaler
|
| 113 |
+
scaler_path = artifacts_dir / "scaler.joblib"
|
| 114 |
+
joblib.dump(scaler, scaler_path)
|
| 115 |
+
mlflow.log_artifact(str(scaler_path), artifact_path="preprocessing")
|
| 116 |
+
print(" ✅ Scaler sauvegardé")
|
| 117 |
+
|
| 118 |
+
# Sauvegarder encoders (onehot et ordinal)
|
| 119 |
+
onehot_path = artifacts_dir / "onehot_encoder.joblib"
|
| 120 |
+
joblib.dump(onehot_encoder, onehot_path)
|
| 121 |
+
mlflow.log_artifact(str(onehot_path), artifact_path="preprocessing")
|
| 122 |
+
|
| 123 |
+
ordinal_path = artifacts_dir / "ordinal_encoder.joblib"
|
| 124 |
+
joblib.dump(ordinal_encoder, ordinal_path)
|
| 125 |
+
mlflow.log_artifact(str(ordinal_path), artifact_path="preprocessing")
|
| 126 |
+
print(" ✅ Encoders sauvegardés (OneHot + Ordinal)")
|
| 127 |
+
|
| 128 |
+
# Log git commit si disponible
|
| 129 |
+
try:
|
| 130 |
+
import subprocess
|
| 131 |
+
|
| 132 |
+
git_commit = (
|
| 133 |
+
subprocess.check_output(["git", "rev-parse", "HEAD"])
|
| 134 |
+
.strip()
|
| 135 |
+
.decode("utf-8")
|
| 136 |
+
)
|
| 137 |
+
mlflow.set_tag("git_commit", git_commit[:8])
|
| 138 |
+
print(f" ✅ Git commit: {git_commit[:8]}")
|
| 139 |
+
except Exception:
|
| 140 |
+
pass
|
| 141 |
+
|
| 142 |
+
# Nettoyer artifacts temporaires
|
| 143 |
+
scaler_path.unlink()
|
| 144 |
+
onehot_path.unlink()
|
| 145 |
+
ordinal_path.unlink()
|
| 146 |
+
artifacts_dir.rmdir()
|
| 147 |
+
|
| 148 |
+
print()
|
| 149 |
+
|
| 150 |
+
# Fermer le run si on l'a ouvert
|
| 151 |
+
if should_end_run:
|
| 152 |
+
mlflow.end_run()
|
| 153 |
+
|
| 154 |
+
# ========================================================================
|
| 155 |
+
# RÉSUMÉ
|
| 156 |
+
# ========================================================================
|
| 157 |
+
print("=" * 80)
|
| 158 |
+
print("✅ ENTRAÎNEMENT TERMINÉ")
|
| 159 |
+
print("=" * 80)
|
| 160 |
+
print()
|
| 161 |
+
print(f"📊 Run ID: {run_id}")
|
| 162 |
+
print(f"🎯 F1 Score (CV): {cv_f1:.4f}")
|
| 163 |
+
print("📦 Artifacts sauvegardés dans MLflow")
|
| 164 |
+
print()
|
| 165 |
+
print("🌐 Pour visualiser les résultats:")
|
| 166 |
+
print(" ./scripts/start_mlflow.sh")
|
| 167 |
+
print(" ou: mlflow ui --backend-store-uri sqlite:///mlflow.db")
|
| 168 |
+
print()
|
| 169 |
+
print("📝 Pour charger le modèle:")
|
| 170 |
+
print(f" model = mlflow.sklearn.load_model('runs:/{run_id}/model')")
|
| 171 |
+
print()
|
| 172 |
+
|
| 173 |
+
|
| 174 |
+
if __name__ == "__main__":
|
| 175 |
+
main()
|