- Added stratify sampling option – use for imbalanced datasets. - Added feature bagging option – use for very high-dimensional datasets, e.g., >3000 features, together with ensembling.
1.0.0
In this new release we: - Improved compatibility with the scikit-learn API. - Set ``cat_features`` when initializing the model, not in ``fit()``. - Allow categorical features encoded as strings. - Fixed some minor bugs. - Improved user experience: faster model loading (from cache) when re-initializing; model path can be specified in ``custom_path``.