This release adds sample weight support to the binning process and scorecard, makes BinningProcess a scikit-learn transformer, and extends the plotting and significance testing tools. Weighted cases are handled consistently: sample and class weights drive the weighted binning metrics, while min_bin_size and min_prebin_size continue to count raw records.
New features:
- Add
sample_weightsupport toOptimalBinning,ContinuousOptimalBinning,BinningProcessandScorecardfor binary and continuous targets #420 - Add
get_feature_names_outtoBinningProcessfor scikit-learn pipeline compatibility #382 - Add
target_dtypeto override the target type detection inBinningProcess#296 - Allow
ivas aselection_criteriametric for continuous targets #307 - Expose per-bin significance tests through
BinningTable.df_tests#283 - Add a binning process plot grid with an automatic near-square layout, shared metric axes and a shared legend by default #315
Improvements:
- Make
BinningProcessa scikit-learnTransformerMixin, withvariable_namesnow optional #343 - Vectorize
transform()using NumPy and pandas gather operations #388 - Auto-detect the dtype in
OptimalBinningandContinuousOptimalBinning#316
Bugfixes:
- Fix
min_bin_sizeandmin_prebin_sizeto count raw records, not class-reweighted mass. Weighted metrics are still computed from the weighted counts. #385 - Fix the WoE and IV formula in the continuous-target tutorial #370
- Fix scorecard bin lookup when per-variable transform overrides are used #412
- Fix NumPy value serialization and categorical state restoration in JSON round trips; preserve unknown categories and mixed values #317
- Fix streaming binning
transformoverrides and row alignment #438 - Fix 2D binning table plot legends showing NumPy scalar representations such as np.int64(1) instead of plain integers.
Packaging and dependencies:
- Require Matplotlib 3.7 or newer for the shared-legend plotting layout.
- Add Python 3.14 to the continuous integration matrix and package classifiers.
Documentation:
- Add a multi-variable overview to the Telco churn tutorial #418
- Document the plotting API and significance results.
- Add
CONTRIBUTING.mdwith the contribution guidelines.
Contributors:
- @lcrmorin: plotting API and grids, significance tests, scikit-learn transformer compatibility, vectorized
transform, JSON fixes, and the continuous-target tutorial fix. - @josp70:
get_feature_names_outimplementation - @Swapnil-byte-798: fix for streaming binning transform overrides and row alignment