← ...
ml pipelines preparation
focus on feature engineering, feature stores, offline/online sync, model training and serving, and drift detection.
key concepts
- feature engineering and feature store (feast)
- offline/online sync for ml pipelines
- retraining workflow
- drift detection and monitoring
- data validation for ml features
explanation practice
- feature store flow diagram
- retraining dag diagram
- drift detection illustration
projects
1. feature pipeline
- extract from postgres → clean → feast store
2. retraining dag in airflow
- automated scheduled model retrain
3. feature drift detection tool
- compare distributions and flag anomalies