v0.3.0

What's Changed

• Create GitHub Releases on PyPI publish by @alxhslm in #3
• Add retraining-after-load tests by @alxhslm in #1
• Add columnar tensor serialization for LightGBM and XGBoost by @alxhslm in #2
• Rewrite README positioning by @alxhslm in #4
• Use hatch-vcs for automatic version management by @alxhslm in #5

New Contributors

• @alxhslm made their first contribution in #3

Full Changelog: v0.2.1...v0.3.0

v0.2.1

• Fix: RobustScaler center_ was missing from serialized arrays
• Add parametrized scaler tests for all three scaler types

v0.2.0

• Add serialize() / deserialize() for in-memory round-trips (no file I/O required)
• save() / load() are now thin wrappers around these

v0.1.1

Fix logo not rendering on PyPI (use absolute URL for README image).

v0.1.0

Initial release.

• save() / load() for fitted sklearn estimators
• Supported: linear models, MLP, DecisionTree, RandomForest, GradientBoosting, HistGradientBoosting, GaussianProcess, TransformedTargetRegressor, scalers, pipelines
• Optional: LightGBM, XGBoost
• Output: safetensors (arrays) + JSON (hyperparameters, state)
• Version tracking: warns when loading with a different sklearn/lightgbm/xgboost version
• CI: Python 3.11-3.13, sklearn 1.5/1.6/latest, nightly