SafeFeat documentation
safefeat is an open source, MIT-licensed library for leakage-safe,
point-in-time feature engineering from event logs. It builds ML features for each
(entity_id, cutoff_time) using only events that occurred at or before
each cutoff — so your model learns what it would actually know in production.
Useful links: PyPI · Source Repository · Issues & Ideas ·
New to safefeat? The getting started guide walks through installation, core concepts, and your first leakage-safe feature set.
To the getting started guide →Understand how point-in-time joins work, what the spine pattern is, and how to avoid silent data leakage in ML pipelines.
To the user guide →Detailed description of build_features, WindowAgg, RecencyBlock, and all configuration options.
Found a bug or want to improve safefeat? The repository is open — contributions, issues, and feedback are welcome.
To the GitHub repository →