Release History
Track breaking changes, new features, and important updates to the system.February 2026
New Features
- Multi-objective Optuna training — Optimize for both RMSE and MAE simultaneously
- Ensemble model stacking — CatBoost + XGBoost + LightGBM + Random Forest ensembles
- A/B testing framework — Route traffic between production and challenger models
- Evidently drift monitoring — Per-feature KS-test drift detection with configurable thresholds
- Security middleware stack — OWASP headers, rate limiting, trusted hosts, CORS
Breaking Changes
weekdayparameter key renamed today_of_weekinparameters.yml- Demand lags changed from
[1, 2, 24]to[1, 2, 22, 23] RetrainingSchedulerclass removed — replaced by Airflow DAG atdags/automated_retraining_dag.py- Retraining schedule changed from daily to weekly (
0 2 * * 0)
January 2026
New Features
- Kedro pipeline framework — Reproducible, modular ML pipelines
- Airflow orchestration — DAG-based automation for training and monitoring
- MLflow experiment tracking — Full model registry with versioning and aliases
- MinIO object storage — S3-compatible storage for all artifacts
- Docker Compose stack — 20-service infrastructure with health checks
Infrastructure
- PostgreSQL backend for Airflow + Optuna metadata
- Prometheus + Grafana monitoring with custom dashboards
- Nginx reverse proxy with SSL-ready configuration
- ClearML integration for experiment management