How pooria0v/bike-retrain-mlops works
Bike Demand Auto-Retraining is an MLOps application: monthly bike-demand data is checked for drift, retrained into a challenger model when needed, evaluated against the champion, and conditionally promoted; a FastAPI service serves the champion and records operational signals. Edges follow the documented Prefect orchestration and sampled call sites. Monitoring server configuration and Grafana capabilities are documented, but their runtime wiring is not shown in source excerpts.
Main entry points: main.py Prediction API · retrain_flow.py Retrain Flow · train.py Model Training · evaluate.py Challenger Evaluation
Components 14
Data Preparation
- Monthly Data
src/data/data_prep.py
Retraining Pipeline
- Retrain Flow
src/flows/retrain_flow.py - Drift Detection
src/monitoring/drift_check.py - Model Training
src/training/train.py - Challenger Evaluation
src/training/evaluate.py - Model Promotion
src/registry/promote.py
Prediction Service
- Prediction API
src/api/main.py - Request Schemas
src/api/schemas.py
Operational Monitoring
- Prediction CSV Log
src/monitoring/prediction_logger.py - Prometheus Metrics
src/monitoring/metrics.py - Prometheus
monitoring/prometheus.yml - Grafana Dashboards
monitoring/grafana/dashboards/bike-demand-dashboard.json
Other components
- MLflow Registry
- Prediction Client
Connections 15
- Monthly Data → Retrain Flow: provides monthly inputs
README.md - Retrain Flow → Drift Detection: checks drift
src/flows/retrain_flow.py - Retrain Flow → Model Training: trains on drift
src/flows/retrain_flow.py - Retrain Flow → Challenger Evaluation: evaluates challenger
src/flows/retrain_flow.py - Retrain Flow → Model Promotion: routes decision
src/flows/retrain_flow.py - Model Training → MLflow Registry: logs model
src/training/train.py - Challenger Evaluation → MLflow Registry: loads models
src/training/evaluate.py - Model Promotion → MLflow Registry: updates aliases
src/registry/promote.py - Prediction API → MLflow Registry: loads champion
src/api/main.py - Prediction Client → Prediction API: submits features
README.md - Prediction API → Request Schemas: validates and returns
src/api/main.py - Prediction API → Prediction CSV Log: logs prediction
src/api/main.py - Prediction API → Prometheus Metrics: updates and exposes
src/api/main.py - Prometheus → Prediction API: scrapes metrics
README.md - Grafana Dashboards → Prometheus: visualizes metrics
README.md