Spaces:
Configuration error
Configuration error
celpri commited on
Commit ·
29fec18
1
Parent(s): db7ff99
Update monitoring dashboard
Browse files- .github/workflows/ci.yml +2 -2
- .gitignore +3 -1
- .vscode/settings.json +3 -0
- README.md +208 -44
- api_logs.jsonl +0 -0
- data_drift_analysis.ipynb +159 -0
- evidently_report.html → data_drift_report.html +0 -0
- evidently_report.py +0 -27
- export_model.py +15 -0
- analyze_logs.py → monitoring/analyze_logs.py +0 -0
- monitoring/drift_analysis.py +66 -0
- scripts/export_model.py +20 -0
- src/api/main.py +6 -1
- src/model/model.py +1 -34
- streamlit_app.py +102 -7
.github/workflows/ci.yml
CHANGED
|
@@ -52,9 +52,9 @@ jobs:
|
|
| 52 |
run: |
|
| 53 |
git config user.name "github-actions"
|
| 54 |
git config user.email "github-actions@github.com"
|
| 55 |
-
#
|
| 56 |
git remote add hf https://PCelia:${HF_TOKEN}@huggingface.co/spaces/PCelia/Pret-a-depenser || git remote set-url hf https://PCelia:${HF_TOKEN}@huggingface.co/spaces/PCelia/Pret-a-depenser
|
| 57 |
-
#
|
| 58 |
git fetch hf main
|
| 59 |
# On pousse les changements de code
|
| 60 |
git push hf main --force
|
|
|
|
| 52 |
run: |
|
| 53 |
git config user.name "github-actions"
|
| 54 |
git config user.email "github-actions@github.com"
|
| 55 |
+
# Ajoute le remote HF s'il n'existe pas déjà
|
| 56 |
git remote add hf https://PCelia:${HF_TOKEN}@huggingface.co/spaces/PCelia/Pret-a-depenser || git remote set-url hf https://PCelia:${HF_TOKEN}@huggingface.co/spaces/PCelia/Pret-a-depenser
|
| 57 |
+
# On fetch d'abord pour ne pas écraser le modèle uploadé manuellement
|
| 58 |
git fetch hf main
|
| 59 |
# On pousse les changements de code
|
| 60 |
git push hf main --force
|
.gitignore
CHANGED
|
@@ -6,4 +6,6 @@ __pycache__/
|
|
| 6 |
model.joblib
|
| 7 |
projet8
|
| 8 |
app
|
| 9 |
-
|
|
|
|
|
|
|
|
|
| 6 |
model.joblib
|
| 7 |
projet8
|
| 8 |
app
|
| 9 |
+
expo
|
| 10 |
+
debug_model.py
|
| 11 |
+
api_logs.jsonl
|
.vscode/settings.json
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"python-envs.pythonProjects": []
|
| 3 |
+
}
|
README.md
CHANGED
|
@@ -82,15 +82,23 @@ projet8/
|
|
| 82 |
│ └── application_train_fused.csv # Données fusionnées
|
| 83 |
├── app/
|
| 84 |
│ └── model.joblib # Modèle sérialisé
|
| 85 |
-
├── tests/ # Tests
|
| 86 |
-
|
| 87 |
-
|
| 88 |
-
├──
|
| 89 |
-
│ └──
|
| 90 |
-
├──
|
| 91 |
-
├──
|
| 92 |
-
|
| 93 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 94 |
```
|
| 95 |
|
| 96 |
---
|
|
@@ -101,11 +109,12 @@ projet8/
|
|
| 101 |
|-----------|-------------|
|
| 102 |
| **ML/Data Science** | LightGBM, scikit-learn, pandas, numpy |
|
| 103 |
| **Web Backend** | FastAPI, Uvicorn |
|
| 104 |
-
| **Monitoring** | Streamlit, MLflow |
|
|
|
|
| 105 |
| **Versioning Modèle** | MLflow, Hugging Face Hub |
|
| 106 |
| **Testing** | pytest, httpx |
|
| 107 |
| **Containerisation** | Docker |
|
| 108 |
-
| **Python Version** | 3.
|
| 109 |
|
| 110 |
---
|
| 111 |
|
|
@@ -342,11 +351,52 @@ asyncio.run(get_prediction())
|
|
| 342 |
|
| 343 |
### Tableau de bord Streamlit
|
| 344 |
|
| 345 |
-
|
| 346 |
|
| 347 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 348 |
- 📉 **Distribution des scores** : Analyse des décisions de crédit
|
| 349 |
-
- 💾 **Historique complet** : Tous les appels enregistrés
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 350 |
|
| 351 |
### Format des logs
|
| 352 |
|
|
@@ -367,6 +417,17 @@ tail -10 api_logs.jsonl
|
|
| 367 |
python analyze_logs.py
|
| 368 |
```
|
| 369 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 370 |
---
|
| 371 |
|
| 372 |
## 🐳 Déploiement
|
|
@@ -382,7 +443,16 @@ docker build -t credit-scoring:latest .
|
|
| 382 |
#### 2. Exécuter le conteneur
|
| 383 |
|
| 384 |
```bash
|
|
|
|
| 385 |
docker run -p 8000:7860 \
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 386 |
-e HF_TOKEN=hf_votre_token \
|
| 387 |
credit-scoring:latest
|
| 388 |
```
|
|
@@ -393,6 +463,16 @@ docker run -p 8000:7860 \
|
|
| 393 |
http://localhost:8000
|
| 394 |
```
|
| 395 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 396 |
### Docker Compose
|
| 397 |
|
| 398 |
```yaml
|
|
@@ -431,6 +511,23 @@ sdk: docker
|
|
| 431 |
|
| 432 |
## 🧪 Tests
|
| 433 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 434 |
### Lancer les tests
|
| 435 |
|
| 436 |
```bash
|
|
@@ -440,37 +537,53 @@ pytest
|
|
| 440 |
# Avec verbose
|
| 441 |
pytest -v
|
| 442 |
|
| 443 |
-
# Coverage
|
| 444 |
pytest --cov=src
|
| 445 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 446 |
# Test spécifique
|
| 447 |
-
pytest tests/
|
| 448 |
```
|
| 449 |
|
| 450 |
-
###
|
| 451 |
|
| 452 |
-
```
|
| 453 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
| 454 |
|
| 455 |
-
|
| 456 |
-
|
| 457 |
-
|
| 458 |
-
|
|
|
|
| 459 |
```
|
| 460 |
|
| 461 |
### Ajouter vos propres tests
|
| 462 |
|
| 463 |
```python
|
| 464 |
-
# tests/
|
|
|
|
|
|
|
|
|
|
| 465 |
|
| 466 |
-
def
|
| 467 |
-
"""
|
| 468 |
-
|
| 469 |
-
|
| 470 |
-
|
| 471 |
-
|
| 472 |
-
|
| 473 |
-
|
|
|
|
|
|
|
| 474 |
```
|
| 475 |
|
| 476 |
---
|
|
@@ -483,19 +596,46 @@ def test_mon_test():
|
|
| 483 |
|----------|-------------|
|
| 484 |
| [01_eda.ipynb](notebooks/01_eda.ipynb) | Analyse exploratoire des données (EDA) |
|
| 485 |
| [02_fusion.ipynb](notebooks/02_fusion.ipynb) | Fusion de sources et préparation |
|
| 486 |
-
| [03_modelisation.ipynb](notebooks/03_modelisation.ipynb) | Entraînement et validation du modèle |
|
| 487 |
-
|
| 488 |
### Scripts utilitaires
|
| 489 |
|
| 490 |
```bash
|
| 491 |
-
# Exporter le modèle
|
| 492 |
python scripts/export_model.py
|
| 493 |
|
| 494 |
-
# Déboguer le modèle
|
| 495 |
python debug_model.py
|
| 496 |
|
| 497 |
-
# Analyser les logs
|
| 498 |
-
python analyze_logs.py
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 499 |
```
|
| 500 |
|
| 501 |
### Ressources externes
|
|
@@ -628,13 +768,37 @@ Pour toute question ou problème :
|
|
| 628 |
|
| 629 |
## ✨ Roadmap future
|
| 630 |
|
| 631 |
-
- [ ]
|
| 632 |
-
- [ ]
|
| 633 |
-
- [ ]
|
|
|
|
|
|
|
|
|
|
|
|
|
| 634 |
- [ ] Déploiement Kubernetes
|
| 635 |
-
- [ ] Tests de performance
|
| 636 |
-
- [ ] CI/CD pipeline
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 637 |
|
| 638 |
---
|
| 639 |
|
| 640 |
-
**Dernière mise à jour** : Février 2026
|
|
|
|
| 82 |
│ └── application_train_fused.csv # Données fusionnées
|
| 83 |
├── app/
|
| 84 |
│ └── model.joblib # Modèle sérialisé
|
| 85 |
+
├── tests/ # Tests automatisés
|
| 86 |
+
├── monitoring/ # Monitoring et analyse
|
| 87 |
+
│ └── (contient les scripts de monitoring)
|
| 88 |
+
├── mlruns/ # Artefacts MLflow
|
| 89 |
+
│ └── (versioning du modèle)
|
| 90 |
+
├── docker/ # Configuration Docker
|
| 91 |
+
├── scripts/ # Scripts utilitaires
|
| 92 |
+
│ └── export_model.py # Export du modèle
|
| 93 |
+
├── streamlit_app.py # Dashboard de monitoring Streamlit
|
| 94 |
+
├── drift_analysis.py # Analyse de data drift
|
| 95 |
+
├── analyze_logs.py # Analyse des logs API
|
| 96 |
+
├── debug_model.py # Script de débogage
|
| 97 |
+
├── Dockerfile # Configuration Docker
|
| 98 |
+
├── mlflow.db # Base de données MLflow
|
| 99 |
+
├── api_logs.jsonl # Logs des prédictions API
|
| 100 |
+
├── data_drift_report.html # Rapport de drift Evidently
|
| 101 |
+
└── requirements.txt # Dépendances Python
|
| 102 |
```
|
| 103 |
|
| 104 |
---
|
|
|
|
| 109 |
|-----------|-------------|
|
| 110 |
| **ML/Data Science** | LightGBM, scikit-learn, pandas, numpy |
|
| 111 |
| **Web Backend** | FastAPI, Uvicorn |
|
| 112 |
+
| **Monitoring** | Streamlit, MLflow, Evidently.ai |
|
| 113 |
+
| **Data Drift Detection** | Evidently.ai (rapports HTML) |
|
| 114 |
| **Versioning Modèle** | MLflow, Hugging Face Hub |
|
| 115 |
| **Testing** | pytest, httpx |
|
| 116 |
| **Containerisation** | Docker |
|
| 117 |
+
| **Python Version** | 3.12 (compatible 3.9+) |
|
| 118 |
|
| 119 |
---
|
| 120 |
|
|
|
|
| 351 |
|
| 352 |
### Tableau de bord Streamlit
|
| 353 |
|
| 354 |
+
Lancez le tableau de bord de monitoring :
|
| 355 |
|
| 356 |
+
```bash
|
| 357 |
+
streamlit run streamlit_app.py
|
| 358 |
+
```
|
| 359 |
+
|
| 360 |
+
L'application `streamlit_app.py` fournit en temps réel :
|
| 361 |
+
|
| 362 |
+
- 📊 **Latence API** : Métrique et graphique des temps de réponse
|
| 363 |
- 📉 **Distribution des scores** : Analyse des décisions de crédit
|
| 364 |
+
- 💾 **Historique complet** : Tous les appels enregistrés en temps réel
|
| 365 |
+
- 🔍 **Data Drift** : Surveillance de la dérive des données avec Evidently
|
| 366 |
+
- 🎯 **Statut du système** : CPU et mémoire en temps réel
|
| 367 |
+
|
| 368 |
+
Accessible sur http://localhost:8501
|
| 369 |
+
|
| 370 |
+
### Analyse du Data Drift
|
| 371 |
+
|
| 372 |
+
**Evidently.ai** est intégré pour détecter la dérive des données en temps réel :
|
| 373 |
+
|
| 374 |
+
#### Génération de rapports
|
| 375 |
+
|
| 376 |
+
```bash
|
| 377 |
+
# Générer un rapport de drift
|
| 378 |
+
python monitoring/drift_analysis.py
|
| 379 |
+
```
|
| 380 |
+
|
| 381 |
+
Cela génère `data_drift_report.html` avec :
|
| 382 |
+
- ✅ Détection automatique des dérives
|
| 383 |
+
- ✅ Comparaison des distributions (référence vs. données actuelles)
|
| 384 |
+
- ✅ Alertes sur les changements significatifs
|
| 385 |
+
- ✅ Graphiques détaillés par feature
|
| 386 |
+
|
| 387 |
+
#### Analyse interactive
|
| 388 |
+
|
| 389 |
+
Vous pouvez aussi utiliser le notebook interactif :
|
| 390 |
+
|
| 391 |
+
```bash
|
| 392 |
+
jupyter notebook data_drift_analysis.ipynb
|
| 393 |
+
```
|
| 394 |
+
|
| 395 |
+
Ce notebook permet de :
|
| 396 |
+
- Explorer les dérives en temps réel
|
| 397 |
+
- Configurer les seuils d'alerte personnalisés
|
| 398 |
+
- Générer des rapports HTML automatiques
|
| 399 |
+
- Visualiser les changements de distribution
|
| 400 |
|
| 401 |
### Format des logs
|
| 402 |
|
|
|
|
| 417 |
python analyze_logs.py
|
| 418 |
```
|
| 419 |
|
| 420 |
+
### MLflow Tracking
|
| 421 |
+
|
| 422 |
+
Les expériences de modélisation sont tracées avec MLflow :
|
| 423 |
+
|
| 424 |
+
```bash
|
| 425 |
+
# Consulter l'historique des modèles
|
| 426 |
+
mlflow ui
|
| 427 |
+
|
| 428 |
+
# Accéder à http://localhost:5000
|
| 429 |
+
```
|
| 430 |
+
|
| 431 |
---
|
| 432 |
|
| 433 |
## 🐳 Déploiement
|
|
|
|
| 443 |
#### 2. Exécuter le conteneur
|
| 444 |
|
| 445 |
```bash
|
| 446 |
+
# Mode développement avec volumes
|
| 447 |
docker run -p 8000:7860 \
|
| 448 |
+
-e HF_TOKEN=hf_votre_token \
|
| 449 |
+
-v $(pwd)/Data:/app/Data \
|
| 450 |
+
-v $(pwd)/api_logs.jsonl:/app/api_logs.jsonl \
|
| 451 |
+
credit-scoring:latest
|
| 452 |
+
|
| 453 |
+
# Mode production
|
| 454 |
+
docker run -d -p 8000:7860 \
|
| 455 |
+
--name credit-api \
|
| 456 |
-e HF_TOKEN=hf_votre_token \
|
| 457 |
credit-scoring:latest
|
| 458 |
```
|
|
|
|
| 463 |
http://localhost:8000
|
| 464 |
```
|
| 465 |
|
| 466 |
+
#### 4. Monitorer le conteneur
|
| 467 |
+
|
| 468 |
+
```bash
|
| 469 |
+
# Voir les logs
|
| 470 |
+
docker logs credit-api
|
| 471 |
+
|
| 472 |
+
# Accéder à Streamlit (dans le conteneur)
|
| 473 |
+
docker exec credit-api streamlit run streamlit_app.py --server.port 8501
|
| 474 |
+
```
|
| 475 |
+
|
| 476 |
### Docker Compose
|
| 477 |
|
| 478 |
```yaml
|
|
|
|
| 511 |
|
| 512 |
## 🧪 Tests
|
| 513 |
|
| 514 |
+
### Structure des tests
|
| 515 |
+
|
| 516 |
+
```
|
| 517 |
+
tests/
|
| 518 |
+
├── unit/ # Tests unitaires
|
| 519 |
+
│ ├── test_model_unit.py # Tests du modèle
|
| 520 |
+
│ ├── test_preprocessing.py # Tests du prétraitement
|
| 521 |
+
│ ├── test_input_validation.py # Validation des entrées
|
| 522 |
+
│ └── test_model_loading.py # Chargement du modèle
|
| 523 |
+
├── fonctionnal/ # Tests fonctionnels/intégration
|
| 524 |
+
│ ├── test_api.py # Tests de l'API REST
|
| 525 |
+
│ ├── test_response_schema.py # Schéma des réponses
|
| 526 |
+
│ ├── test_error_handling.py # Gestion des erreurs
|
| 527 |
+
│ └── test_latency.py # Latence des réponses
|
| 528 |
+
└── conftest.py # Configurations pytest
|
| 529 |
+
```
|
| 530 |
+
|
| 531 |
### Lancer les tests
|
| 532 |
|
| 533 |
```bash
|
|
|
|
| 537 |
# Avec verbose
|
| 538 |
pytest -v
|
| 539 |
|
| 540 |
+
# Coverage (couverture de code)
|
| 541 |
pytest --cov=src
|
| 542 |
|
| 543 |
+
# Seulement tests unitaires
|
| 544 |
+
pytest tests/unit/
|
| 545 |
+
|
| 546 |
+
# Seulement tests fonctionnels
|
| 547 |
+
pytest tests/fonctionnal/
|
| 548 |
+
|
| 549 |
# Test spécifique
|
| 550 |
+
pytest tests/unit/test_model_unit.py::test_model_prediction -v
|
| 551 |
```
|
| 552 |
|
| 553 |
+
### Exemples de tests
|
| 554 |
|
| 555 |
+
```bash
|
| 556 |
+
# Tests unitaires du modèle
|
| 557 |
+
pytest tests/unit/test_model_unit.py -v
|
| 558 |
+
|
| 559 |
+
# Tests API
|
| 560 |
+
pytest tests/fonctionnal/test_api.py -v
|
| 561 |
|
| 562 |
+
# Tests de latence
|
| 563 |
+
pytest tests/fonctionnal/test_latency.py -v
|
| 564 |
+
|
| 565 |
+
# Rapport coverage détaillé
|
| 566 |
+
pytest --cov=src --cov-report=html
|
| 567 |
```
|
| 568 |
|
| 569 |
### Ajouter vos propres tests
|
| 570 |
|
| 571 |
```python
|
| 572 |
+
# tests/unit/test_mon_test.py
|
| 573 |
+
|
| 574 |
+
import pytest
|
| 575 |
+
from src.model.model import load_model
|
| 576 |
|
| 577 |
+
def test_model_loading():
|
| 578 |
+
"""Teste le chargement du modèle"""
|
| 579 |
+
model = load_model()
|
| 580 |
+
assert model is not None
|
| 581 |
+
|
| 582 |
+
def test_prediction_shape():
|
| 583 |
+
"""Teste que la prédiction a la bonne forme"""
|
| 584 |
+
model = load_model()
|
| 585 |
+
predictions = model.predict([[1, 2, 3, 4, 5]])
|
| 586 |
+
assert predictions.shape[0] == 1
|
| 587 |
```
|
| 588 |
|
| 589 |
---
|
|
|
|
| 596 |
|----------|-------------|
|
| 597 |
| [01_eda.ipynb](notebooks/01_eda.ipynb) | Analyse exploratoire des données (EDA) |
|
| 598 |
| [02_fusion.ipynb](notebooks/02_fusion.ipynb) | Fusion de sources et préparation |
|
| 599 |
+
| [03_modelisation.ipynb](notebooks/03_modelisation.ipynb) | Entraînement et validation du modèle || [data_drift_analysis.ipynb](data_drift_analysis.ipynb) | **NOUVEAU** : Analyse interactive du data drift avec Evidently |
|
|
|
|
| 600 |
### Scripts utilitaires
|
| 601 |
|
| 602 |
```bash
|
| 603 |
+
# Exporter le modèle depuis MLflow
|
| 604 |
python scripts/export_model.py
|
| 605 |
|
| 606 |
+
# Déboguer et tester le modèle
|
| 607 |
python debug_model.py
|
| 608 |
|
| 609 |
+
# Analyser les logs API en détail
|
| 610 |
+
python monitoring/analyze_logs.py
|
| 611 |
+
|
| 612 |
+
# Analyser la dérive des données (Data Drift)
|
| 613 |
+
python monitoring/drift_analysis.py
|
| 614 |
+
```
|
| 615 |
+
|
| 616 |
+
### Chaining des outils
|
| 617 |
+
|
| 618 |
+
Pipeline complet de monitoring :
|
| 619 |
+
|
| 620 |
+
```bash
|
| 621 |
+
# 1. Lancer l'API
|
| 622 |
+
uvicorn src.api.main:app --reload &
|
| 623 |
+
|
| 624 |
+
# 2. Générer quelques prédictions
|
| 625 |
+
for i in {1..10}; do
|
| 626 |
+
curl -X POST "http://localhost:8000/predict" \
|
| 627 |
+
-H "Content-Type: application/json" \
|
| 628 |
+
-d "{\"sk_id_curr\": $((100000 + i))}"
|
| 629 |
+
done
|
| 630 |
+
|
| 631 |
+
# 3. Analyser les logs
|
| 632 |
+
python monitoring/analyze_logs.py
|
| 633 |
+
|
| 634 |
+
# 4. Générer le rapport de drift
|
| 635 |
+
python monitoring/drift_analysis.py
|
| 636 |
+
|
| 637 |
+
# 5. Consulter le tableau de bord
|
| 638 |
+
streamlit run streamlit_app.py
|
| 639 |
```
|
| 640 |
|
| 641 |
### Ressources externes
|
|
|
|
| 768 |
|
| 769 |
## ✨ Roadmap future
|
| 770 |
|
| 771 |
+
- [ ] ✅ **Data Drift Detection** (Evidently.ai) - COMPLÉTÉ
|
| 772 |
+
- [ ] ✅ **Monitoring Dashboard** (Streamlit) - COMPLÉTÉ
|
| 773 |
+
- [ ] ✅ **API Logging & Analytics** - COMPLÉTÉ
|
| 774 |
+
- [ ] Ajouter explication des prédictions (SHAP/LIME)
|
| 775 |
+
- [ ] Interface web avancée (React/Next.js)
|
| 776 |
+
- [ ] Alertes email sur data drift
|
| 777 |
+
- [ ] Améliorer le monitoring (Prometheus + Grafana)
|
| 778 |
- [ ] Déploiement Kubernetes
|
| 779 |
+
- [ ] Tests de performance E2E
|
| 780 |
+
- [ ] CI/CD pipeline GitHub Actions
|
| 781 |
+
|
| 782 |
+
---
|
| 783 |
+
|
| 784 |
+
---
|
| 785 |
+
|
| 786 |
+
## 📝 Nouveautés récentes
|
| 787 |
+
|
| 788 |
+
### v1.1.0 (Février 2026)
|
| 789 |
+
|
| 790 |
+
✨ **Nouvelles fonctionnalités** :
|
| 791 |
+
- 🔍 Détection automatique du **Data Drift** avec Evidently.ai
|
| 792 |
+
- 📊 Tableau de bord **Streamlit** pour le monitoring en temps réel
|
| 793 |
+
- 📈 Analyse des logs API avec **psutil** (CPU, mémoire)
|
| 794 |
+
- 📓 Notebook interactif pour l'analyse du drift
|
| 795 |
+
- 🚀 Support Docker amélioré avec volumes persistants
|
| 796 |
+
|
| 797 |
+
🐛 **Corrections** :
|
| 798 |
+
- Amélioration du chargement du modèle (fallback multi-sources)
|
| 799 |
+
- Meilleure gestion des erreurs API
|
| 800 |
+
- Optimisation des performances
|
| 801 |
|
| 802 |
---
|
| 803 |
|
| 804 |
+
**Dernière mise à jour** : 15 Février 2026
|
api_logs.jsonl
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
data_drift_analysis.ipynb
ADDED
|
@@ -0,0 +1,159 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"cells": [
|
| 3 |
+
{
|
| 4 |
+
"cell_type": "markdown",
|
| 5 |
+
"id": "57a500a1",
|
| 6 |
+
"metadata": {},
|
| 7 |
+
"source": [
|
| 8 |
+
"Charger le dataset initial"
|
| 9 |
+
]
|
| 10 |
+
},
|
| 11 |
+
{
|
| 12 |
+
"cell_type": "code",
|
| 13 |
+
"execution_count": 1,
|
| 14 |
+
"id": "265ff33b",
|
| 15 |
+
"metadata": {},
|
| 16 |
+
"outputs": [],
|
| 17 |
+
"source": [
|
| 18 |
+
"import pandas as pd\n",
|
| 19 |
+
"from sklearn.model_selection import train_test_split\n",
|
| 20 |
+
"\n",
|
| 21 |
+
"df = pd.read_csv(\"Data/features_clients.csv\")\n",
|
| 22 |
+
"df = df.drop(columns=[\"SK_ID_CURR\"])"
|
| 23 |
+
]
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"cell_type": "markdown",
|
| 27 |
+
"id": "55f5c7f9",
|
| 28 |
+
"metadata": {},
|
| 29 |
+
"source": [
|
| 30 |
+
"Train/Test Split"
|
| 31 |
+
]
|
| 32 |
+
},
|
| 33 |
+
{
|
| 34 |
+
"cell_type": "code",
|
| 35 |
+
"execution_count": 2,
|
| 36 |
+
"id": "33025b1c",
|
| 37 |
+
"metadata": {},
|
| 38 |
+
"outputs": [],
|
| 39 |
+
"source": [
|
| 40 |
+
"df_train, df_test = train_test_split(\n",
|
| 41 |
+
" df,\n",
|
| 42 |
+
" test_size=0.3,\n",
|
| 43 |
+
" random_state=42\n",
|
| 44 |
+
")"
|
| 45 |
+
]
|
| 46 |
+
},
|
| 47 |
+
{
|
| 48 |
+
"cell_type": "code",
|
| 49 |
+
"execution_count": 8,
|
| 50 |
+
"id": "ee84412a",
|
| 51 |
+
"metadata": {},
|
| 52 |
+
"outputs": [
|
| 53 |
+
{
|
| 54 |
+
"name": "stdout",
|
| 55 |
+
"output_type": "stream",
|
| 56 |
+
"text": [
|
| 57 |
+
"0.7.20\n",
|
| 58 |
+
"c:\\Users\\User\\Desktop\\Formation IA\\projet8\\projet8\\Lib\\site-packages\\evidently\\__init__.py\n"
|
| 59 |
+
]
|
| 60 |
+
}
|
| 61 |
+
],
|
| 62 |
+
"source": [
|
| 63 |
+
"import evidently\n",
|
| 64 |
+
"print(evidently.__version__)\n",
|
| 65 |
+
"print(evidently.__file__)"
|
| 66 |
+
]
|
| 67 |
+
},
|
| 68 |
+
{
|
| 69 |
+
"cell_type": "code",
|
| 70 |
+
"execution_count": 14,
|
| 71 |
+
"id": "dc5d67c4",
|
| 72 |
+
"metadata": {},
|
| 73 |
+
"outputs": [
|
| 74 |
+
{
|
| 75 |
+
"name": "stdout",
|
| 76 |
+
"output_type": "stream",
|
| 77 |
+
"text": [
|
| 78 |
+
"['AbsMaxError', 'Accuracy', 'AlmostConstantColumnsCount', 'AlmostDuplicatedColumnsCount', 'CategoryCount', 'ColumnCorrelationMatrix', 'ColumnCorrelations', 'ColumnCount', 'ConstantColumnsCount', 'CorrelationMatrix', 'DatasetCorrelations', 'DatasetMissingValueCount', 'Diversity', 'DriftedColumnsCount', 'DummyAccuracy', 'DummyF1Score', 'DummyFNR', 'DummyFPR', 'DummyLogLoss', 'DummyMAE', 'DummyMAPE', 'DummyPrecision', 'DummyRMSE', 'DummyRecall', 'DummyRocAuc', 'DummyTNR', 'DummyTPR', 'DuplicatedColumnsCount', 'DuplicatedRowCount', 'EmptyColumnsCount', 'EmptyRowsCount', 'F1ByLabel', 'F1Score', 'FBetaTopK', 'FNR', 'FPR', 'GroupBy', 'HitRate', 'InListValueCount', 'InRangeValueCount', 'ItemBias', 'LogLoss', 'MAE', 'MAP', 'MAPE', 'MRR', 'MaxValue', 'MeanError', 'MeanValue', 'MedianValue', 'MinValue', 'MissingValueCount', 'NDCG', 'Novelty', 'OutListValueCount', 'OutRangeValueCount', 'Personalization', 'PopularityBiasMetric', 'Precision', 'PrecisionByLabel', 'PrecisionTopK', 'QuantileValue', 'R2Score', 'RMSE', 'RecCasesTable', 'Recall', 'RecallByLabel', 'RecallTopK', 'RocAuc', 'RocAucByLabel', 'RowCount', 'RowTestSummary', 'ScoreDistribution', 'Serendipity', 'StdValue', 'SumValue', 'TNR', 'TPR', 'UniqueValueCount', 'UserBias', 'ValueDrift', '__all__', '__builtins__', '__cached__', '__doc__', '__file__', '__loader__', '__name__', '__package__', '__path__', '__spec__', '_legacy', 'classification', 'column_statistics', 'data_quality', 'dataset_statistics', 'group_by', 'recsys', 'regression', 'row_test_summary']\n"
|
| 79 |
+
]
|
| 80 |
+
}
|
| 81 |
+
],
|
| 82 |
+
"source": [
|
| 83 |
+
"import evidently.metrics\n",
|
| 84 |
+
"print(dir(evidently.metrics))"
|
| 85 |
+
]
|
| 86 |
+
},
|
| 87 |
+
{
|
| 88 |
+
"cell_type": "markdown",
|
| 89 |
+
"id": "4a4bba44",
|
| 90 |
+
"metadata": {},
|
| 91 |
+
"source": [
|
| 92 |
+
"Lancer Evidently"
|
| 93 |
+
]
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"cell_type": "code",
|
| 97 |
+
"execution_count": null,
|
| 98 |
+
"id": "ba976620",
|
| 99 |
+
"metadata": {},
|
| 100 |
+
"outputs": [
|
| 101 |
+
{
|
| 102 |
+
"ename": "AttributeError",
|
| 103 |
+
"evalue": "'Snapshot' object has no attribute 'save'",
|
| 104 |
+
"output_type": "error",
|
| 105 |
+
"traceback": [
|
| 106 |
+
"\u001b[31m---------------------------------------------------------------------------\u001b[39m",
|
| 107 |
+
"\u001b[31mAttributeError\u001b[39m Traceback (most recent call last)",
|
| 108 |
+
"\u001b[36mCell\u001b[39m\u001b[36m \u001b[39m\u001b[32mIn[16]\u001b[39m\u001b[32m, line 21\u001b[39m\n\u001b[32m 14\u001b[39m report.run(reference_data=reference, current_data=current)\n\u001b[32m 16\u001b[39m snapshot = report.run(\n\u001b[32m 17\u001b[39m reference_data=reference,\n\u001b[32m 18\u001b[39m current_data=current\n\u001b[32m 19\u001b[39m )\n\u001b[32m---> \u001b[39m\u001b[32m21\u001b[39m \u001b[43msnapshot\u001b[49m\u001b[43m.\u001b[49m\u001b[43msave\u001b[49m(\u001b[33m\"\u001b[39m\u001b[33mdata_drift_report.html\u001b[39m\u001b[33m\"\u001b[39m)\n",
|
| 109 |
+
"\u001b[31mAttributeError\u001b[39m: 'Snapshot' object has no attribute 'save'"
|
| 110 |
+
]
|
| 111 |
+
}
|
| 112 |
+
],
|
| 113 |
+
"source": [
|
| 114 |
+
"from evidently import Report, Dataset\n",
|
| 115 |
+
"from evidently.metrics import ValueDrift\n",
|
| 116 |
+
"\n",
|
| 117 |
+
"reference = Dataset.from_pandas(df_train)\n",
|
| 118 |
+
"current = Dataset.from_pandas(df_test)\n",
|
| 119 |
+
"\n",
|
| 120 |
+
"metrics = []\n",
|
| 121 |
+
"\n",
|
| 122 |
+
"for col in df_train.columns:\n",
|
| 123 |
+
" metrics.append(ValueDrift(column=col))\n",
|
| 124 |
+
"\n",
|
| 125 |
+
"report = Report(metrics)\n",
|
| 126 |
+
"\n",
|
| 127 |
+
"report.run(reference_data=reference, current_data=current)\n",
|
| 128 |
+
"\n",
|
| 129 |
+
"snapshot = report.run(\n",
|
| 130 |
+
" reference_data=reference,\n",
|
| 131 |
+
" current_data=current\n",
|
| 132 |
+
")\n",
|
| 133 |
+
"\n",
|
| 134 |
+
"snapshot.save_html(\"data_drift_report.html\")"
|
| 135 |
+
]
|
| 136 |
+
}
|
| 137 |
+
],
|
| 138 |
+
"metadata": {
|
| 139 |
+
"kernelspec": {
|
| 140 |
+
"display_name": "projet8 (3.12.10)",
|
| 141 |
+
"language": "python",
|
| 142 |
+
"name": "python3"
|
| 143 |
+
},
|
| 144 |
+
"language_info": {
|
| 145 |
+
"codemirror_mode": {
|
| 146 |
+
"name": "ipython",
|
| 147 |
+
"version": 3
|
| 148 |
+
},
|
| 149 |
+
"file_extension": ".py",
|
| 150 |
+
"mimetype": "text/x-python",
|
| 151 |
+
"name": "python",
|
| 152 |
+
"nbconvert_exporter": "python",
|
| 153 |
+
"pygments_lexer": "ipython3",
|
| 154 |
+
"version": "3.12.10"
|
| 155 |
+
}
|
| 156 |
+
},
|
| 157 |
+
"nbformat": 4,
|
| 158 |
+
"nbformat_minor": 5
|
| 159 |
+
}
|
evidently_report.html → data_drift_report.html
RENAMED
|
The diff for this file is too large to render.
See raw diff
|
|
|
evidently_report.py
DELETED
|
@@ -1,27 +0,0 @@
|
|
| 1 |
-
import json
|
| 2 |
-
import pandas as pd
|
| 3 |
-
import warnings
|
| 4 |
-
from evidently import Report
|
| 5 |
-
from evidently.presets import DataDriftPreset
|
| 6 |
-
|
| 7 |
-
warnings.filterwarnings("ignore") # enlève le spam de warnings
|
| 8 |
-
|
| 9 |
-
# Charger logs
|
| 10 |
-
records = []
|
| 11 |
-
with open("api_logs.jsonl") as f:
|
| 12 |
-
for line in f:
|
| 13 |
-
records.append(json.loads(line))
|
| 14 |
-
|
| 15 |
-
df = pd.DataFrame(records)
|
| 16 |
-
|
| 17 |
-
# Split référence / courant
|
| 18 |
-
reference_df = df.iloc[: len(df)//2][["score", "total_time"]]
|
| 19 |
-
current_df = df.iloc[len(df)//2 :][["score", "total_time"]]
|
| 20 |
-
|
| 21 |
-
report = Report([DataDriftPreset()])
|
| 22 |
-
|
| 23 |
-
# IMPORTANT: le résultat (snapshot) porte save_html()
|
| 24 |
-
snapshot = report.run(reference_data=reference_df, current_data=current_df)
|
| 25 |
-
snapshot.save_html("evidently_report.html")
|
| 26 |
-
|
| 27 |
-
print("OK: evidently_report.html généré")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
export_model.py
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import mlflow
|
| 2 |
+
import mlflow.sklearn
|
| 3 |
+
import joblib
|
| 4 |
+
import os
|
| 5 |
+
|
| 6 |
+
mlflow.set_tracking_uri("sqlite:///mlflow.db")
|
| 7 |
+
|
| 8 |
+
model = mlflow.sklearn.load_model(
|
| 9 |
+
"models:/CreditScoring_LightGBM/Production"
|
| 10 |
+
)
|
| 11 |
+
|
| 12 |
+
os.makedirs("app", exist_ok=True)
|
| 13 |
+
joblib.dump(model, "app/model.joblib")
|
| 14 |
+
|
| 15 |
+
print("OK")
|
analyze_logs.py → monitoring/analyze_logs.py
RENAMED
|
File without changes
|
monitoring/drift_analysis.py
ADDED
|
@@ -0,0 +1,66 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import json
|
| 2 |
+
import pandas as pd
|
| 3 |
+
from evidently import Report
|
| 4 |
+
from evidently.presets import DataDriftPreset
|
| 5 |
+
|
| 6 |
+
print("=== Reconstruction dataset production ===")
|
| 7 |
+
|
| 8 |
+
# --- 1. Charger données production depuis logs ---
|
| 9 |
+
inputs = []
|
| 10 |
+
|
| 11 |
+
with open("api_logs.jsonl") as f:
|
| 12 |
+
for line in f:
|
| 13 |
+
log = json.loads(line)
|
| 14 |
+
if "inputs" in log:
|
| 15 |
+
inputs.append(log["inputs"])
|
| 16 |
+
|
| 17 |
+
df_current = pd.DataFrame(inputs)
|
| 18 |
+
|
| 19 |
+
print("df_current shape:", df_current.shape)
|
| 20 |
+
|
| 21 |
+
if df_current.empty:
|
| 22 |
+
raise ValueError("Aucune donnée production trouvée dans les logs.")
|
| 23 |
+
|
| 24 |
+
# --- 2. Charger référence ---
|
| 25 |
+
df_reference = pd.read_csv("Data/features_clients.csv")
|
| 26 |
+
|
| 27 |
+
if "SK_ID_CURR" in df_reference.columns:
|
| 28 |
+
df_reference = df_reference.drop(columns=["SK_ID_CURR"])
|
| 29 |
+
|
| 30 |
+
print("df_reference shape:", df_reference.shape)
|
| 31 |
+
|
| 32 |
+
# --- 3. Aligner colonnes ---
|
| 33 |
+
common_cols = df_current.columns.intersection(df_reference.columns)
|
| 34 |
+
|
| 35 |
+
df_current = df_current[common_cols]
|
| 36 |
+
df_reference = df_reference[common_cols]
|
| 37 |
+
|
| 38 |
+
# Supprimer colonnes entièrement vides dans current
|
| 39 |
+
non_empty_cols = df_current.columns[df_current.notna().any()]
|
| 40 |
+
|
| 41 |
+
df_current = df_current[non_empty_cols]
|
| 42 |
+
df_reference = df_reference[non_empty_cols]
|
| 43 |
+
|
| 44 |
+
print("Colonnes finales utilisées :", len(non_empty_cols))
|
| 45 |
+
|
| 46 |
+
# --- 4. Échantillonner référence pour éviter biais taille ---
|
| 47 |
+
df_reference = df_reference.sample(n=len(df_current), random_state=42)
|
| 48 |
+
|
| 49 |
+
# --- 5. Simulation drift volontaire ---
|
| 50 |
+
df_current["AMT_INCOME_TOTAL"] *= 3
|
| 51 |
+
df_current["AMT_CREDIT"] *= 2
|
| 52 |
+
df_current["AMT_ANNUITY"] *= 2
|
| 53 |
+
|
| 54 |
+
# --- 6. Lancer Data Drift ---
|
| 55 |
+
print("=== Lancement Evidently ===")
|
| 56 |
+
|
| 57 |
+
report = Report(metrics=[DataDriftPreset()])
|
| 58 |
+
|
| 59 |
+
snapshot = report.run(
|
| 60 |
+
reference_data=df_reference,
|
| 61 |
+
current_data=df_current
|
| 62 |
+
)
|
| 63 |
+
|
| 64 |
+
snapshot.save_html("data_drift_report.html")
|
| 65 |
+
|
| 66 |
+
print("Rapport généré : data_drift_report.html")
|
scripts/export_model.py
ADDED
|
@@ -0,0 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import mlflow
|
| 3 |
+
import mlflow.sklearn
|
| 4 |
+
import joblib
|
| 5 |
+
|
| 6 |
+
|
| 7 |
+
BASE_DIR = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
|
| 8 |
+
DB_PATH = os.path.join(BASE_DIR, "mlflow.db")
|
| 9 |
+
|
| 10 |
+
mlflow.set_tracking_uri(f"sqlite:///{DB_PATH}")
|
| 11 |
+
|
| 12 |
+
MODEL_URI = "models:/CreditScoring_LightGBM/Production"
|
| 13 |
+
OUT_PATH = os.path.join(BASE_DIR, "app", "model.joblib")
|
| 14 |
+
|
| 15 |
+
os.makedirs(os.path.dirname(OUT_PATH), exist_ok=True)
|
| 16 |
+
|
| 17 |
+
model = mlflow.sklearn.load_model(MODEL_URI)
|
| 18 |
+
joblib.dump(model, OUT_PATH)
|
| 19 |
+
|
| 20 |
+
print("Export OK ->", OUT_PATH)
|
src/api/main.py
CHANGED
|
@@ -98,6 +98,8 @@ def predict_by_id(payload: ClientID):
|
|
| 98 |
f"infer={infer_time:.3f}s | "
|
| 99 |
f"total={total_time:.3f}s"
|
| 100 |
)
|
|
|
|
|
|
|
| 101 |
log_entry = {
|
| 102 |
"timestamp": datetime.utcnow().isoformat(),
|
| 103 |
"endpoint": "/predict_by_id",
|
|
@@ -105,9 +107,12 @@ def predict_by_id(payload: ClientID):
|
|
| 105 |
"score": score,
|
| 106 |
"features_time": features_time,
|
| 107 |
"inference_time": infer_time,
|
| 108 |
-
"total_time": total_time
|
|
|
|
| 109 |
}
|
| 110 |
|
|
|
|
|
|
|
| 111 |
with open("api_logs.jsonl", "a") as f:
|
| 112 |
f.write(json.dumps(log_entry) + "\n")
|
| 113 |
|
|
|
|
| 98 |
f"infer={infer_time:.3f}s | "
|
| 99 |
f"total={total_time:.3f}s"
|
| 100 |
)
|
| 101 |
+
inputs_dict = X.to_dict(orient="records")[0]
|
| 102 |
+
|
| 103 |
log_entry = {
|
| 104 |
"timestamp": datetime.utcnow().isoformat(),
|
| 105 |
"endpoint": "/predict_by_id",
|
|
|
|
| 107 |
"score": score,
|
| 108 |
"features_time": features_time,
|
| 109 |
"inference_time": infer_time,
|
| 110 |
+
"total_time": total_time,
|
| 111 |
+
"inputs": inputs_dict
|
| 112 |
}
|
| 113 |
|
| 114 |
+
import os
|
| 115 |
+
print("LOG PATH:", os.path.abspath("api_logs.jsonl"))
|
| 116 |
with open("api_logs.jsonl", "a") as f:
|
| 117 |
f.write(json.dumps(log_entry) + "\n")
|
| 118 |
|
src/model/model.py
CHANGED
|
@@ -27,7 +27,7 @@ def load_model():
|
|
| 27 |
# Local
|
| 28 |
try:
|
| 29 |
import mlflow.sklearn
|
| 30 |
-
# On définit le chemin de
|
| 31 |
current_dir = os.path.dirname(os.path.abspath(__file__))
|
| 32 |
db_path = os.path.join(current_dir, "..", "..", "mlflow.db")
|
| 33 |
mlflow.set_tracking_uri(f"sqlite:///{db_path}")
|
|
@@ -39,36 +39,3 @@ def load_model():
|
|
| 39 |
|
| 40 |
raise FileNotFoundError("Impossible de charger le modèle (ni HF Hub, ni MLflow)")
|
| 41 |
|
| 42 |
-
# import joblib
|
| 43 |
-
# from pathlib import Path
|
| 44 |
-
|
| 45 |
-
# def load_model():
|
| 46 |
-
# # HF Space
|
| 47 |
-
# hf_path = Path("model.joblib")
|
| 48 |
-
# if hf_path.exists():
|
| 49 |
-
# return joblib.load(hf_path)
|
| 50 |
-
|
| 51 |
-
# # Local
|
| 52 |
-
# local_path = Path(__file__).resolve().parents[2] / "app" / "model.joblib"
|
| 53 |
-
# if local_path.exists():
|
| 54 |
-
# return joblib.load(local_path)
|
| 55 |
-
|
| 56 |
-
# raise FileNotFoundError("model.joblib not found")
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
# # import mlflow
|
| 62 |
-
# # import mlflow.sklearn
|
| 63 |
-
# # import os
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
# # current_dir = os.path.dirname(os.path.abspath(__file__))
|
| 67 |
-
|
| 68 |
-
# # db_path = os.path.join(current_dir, "..", "..", "mlflow.db")
|
| 69 |
-
|
| 70 |
-
# # mlflow.set_tracking_uri(f"sqlite:///{db_path}")
|
| 71 |
-
|
| 72 |
-
# # def load_model():
|
| 73 |
-
# # model_uri = "models:/CreditScoring_LightGBM/Production"
|
| 74 |
-
# # return mlflow.sklearn.load_model(model_uri)
|
|
|
|
| 27 |
# Local
|
| 28 |
try:
|
| 29 |
import mlflow.sklearn
|
| 30 |
+
# On définit le chemin de DB mlflow relative à ce fichier
|
| 31 |
current_dir = os.path.dirname(os.path.abspath(__file__))
|
| 32 |
db_path = os.path.join(current_dir, "..", "..", "mlflow.db")
|
| 33 |
mlflow.set_tracking_uri(f"sqlite:///{db_path}")
|
|
|
|
| 39 |
|
| 40 |
raise FileNotFoundError("Impossible de charger le modèle (ni HF Hub, ni MLflow)")
|
| 41 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
streamlit_app.py
CHANGED
|
@@ -1,10 +1,16 @@
|
|
| 1 |
import json
|
| 2 |
import pandas as pd
|
| 3 |
import streamlit as st
|
|
|
|
|
|
|
|
|
|
| 4 |
|
| 5 |
st.title("Monitoring API – Credit Scoring")
|
| 6 |
|
| 7 |
-
|
|
|
|
|
|
|
|
|
|
| 8 |
records = []
|
| 9 |
with open("api_logs.jsonl") as f:
|
| 10 |
for line in f:
|
|
@@ -12,11 +18,100 @@ with open("api_logs.jsonl") as f:
|
|
| 12 |
|
| 13 |
df = pd.DataFrame(records)
|
| 14 |
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
|
| 20 |
st.subheader("Distribution des scores")
|
| 21 |
-
|
| 22 |
-
st.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import json
|
| 2 |
import pandas as pd
|
| 3 |
import streamlit as st
|
| 4 |
+
import matplotlib.pyplot as plt
|
| 5 |
+
import psutil
|
| 6 |
+
from datetime import datetime
|
| 7 |
|
| 8 |
st.title("Monitoring API – Credit Scoring")
|
| 9 |
|
| 10 |
+
|
| 11 |
+
# Chargement des logs
|
| 12 |
+
|
| 13 |
+
|
| 14 |
records = []
|
| 15 |
with open("api_logs.jsonl") as f:
|
| 16 |
for line in f:
|
|
|
|
| 18 |
|
| 19 |
df = pd.DataFrame(records)
|
| 20 |
|
| 21 |
+
# Conversion timestamp
|
| 22 |
+
df["timestamp"] = pd.to_datetime(df["timestamp"])
|
| 23 |
+
|
| 24 |
+
|
| 25 |
+
# Latence API (temps total)
|
| 26 |
+
|
| 27 |
+
|
| 28 |
+
st.subheader("Latence API (temps total)")
|
| 29 |
+
|
| 30 |
+
st.metric(
|
| 31 |
+
"Latence moyenne (ms)",
|
| 32 |
+
round(df["total_time"].mean() * 1000, 2)
|
| 33 |
+
)
|
| 34 |
+
|
| 35 |
+
st.metric(
|
| 36 |
+
"Latence max (ms)",
|
| 37 |
+
round(df["total_time"].max() * 1000, 2)
|
| 38 |
+
)
|
| 39 |
+
|
| 40 |
+
st.line_chart(df["total_time"] * 1000)
|
| 41 |
+
|
| 42 |
+
|
| 43 |
+
# Temps d'inférence modèle
|
| 44 |
+
|
| 45 |
+
|
| 46 |
+
st.subheader("Temps d'inférence modèle")
|
| 47 |
+
|
| 48 |
+
st.metric(
|
| 49 |
+
"Temps moyen (ms)",
|
| 50 |
+
round(df["inference_time"].mean() * 1000, 2)
|
| 51 |
+
)
|
| 52 |
+
|
| 53 |
+
st.metric(
|
| 54 |
+
"Temps max (ms)",
|
| 55 |
+
round(df["inference_time"].max() * 1000, 2)
|
| 56 |
+
)
|
| 57 |
+
|
| 58 |
+
fig, ax = plt.subplots()
|
| 59 |
+
|
| 60 |
+
ax.plot(df["inference_time"].values * 1000)
|
| 61 |
+
|
| 62 |
+
# Point rouge dernière requête
|
| 63 |
+
ax.scatter(
|
| 64 |
+
len(df) - 1,
|
| 65 |
+
df["inference_time"].iloc[-1] * 1000,
|
| 66 |
+
color="red",
|
| 67 |
+
s=80
|
| 68 |
+
)
|
| 69 |
+
|
| 70 |
+
ax.set_xlabel("Requête")
|
| 71 |
+
ax.set_ylabel("Inference time (ms)")
|
| 72 |
+
|
| 73 |
+
st.pyplot(fig)
|
| 74 |
+
|
| 75 |
+
|
| 76 |
+
# Distribution des scores
|
| 77 |
+
|
| 78 |
|
| 79 |
st.subheader("Distribution des scores")
|
| 80 |
+
|
| 81 |
+
st.metric(
|
| 82 |
+
"Score moyen",
|
| 83 |
+
round(df["score"].mean(), 4)
|
| 84 |
+
)
|
| 85 |
+
|
| 86 |
+
st.bar_chart(df["score"])
|
| 87 |
+
|
| 88 |
+
|
| 89 |
+
# Requêtes par minute
|
| 90 |
+
|
| 91 |
+
|
| 92 |
+
requests_per_min = (
|
| 93 |
+
df.set_index("timestamp")
|
| 94 |
+
.resample("1min")
|
| 95 |
+
.size()
|
| 96 |
+
)
|
| 97 |
+
|
| 98 |
+
st.subheader("Requêtes par minute")
|
| 99 |
+
st.line_chart(requests_per_min)
|
| 100 |
+
|
| 101 |
+
|
| 102 |
+
# Utilisation CPU et RAM
|
| 103 |
+
|
| 104 |
+
st.subheader("Utilisation système")
|
| 105 |
+
|
| 106 |
+
cpu_usage = psutil.cpu_percent(interval=None)
|
| 107 |
+
ram_usage = psutil.virtual_memory().percent
|
| 108 |
+
|
| 109 |
+
col1, col2 = st.columns(2)
|
| 110 |
+
|
| 111 |
+
#col1.metric("CPU usage (%)", cpu_usage)
|
| 112 |
+
col2.metric("RAM usage (%)", ram_usage)
|
| 113 |
+
|
| 114 |
+
# Dernière requête
|
| 115 |
+
|
| 116 |
+
last_request = df.iloc[-1]
|
| 117 |
+
st.write("Dernière requête :", last_request["timestamp"])
|