fklearn - Browse /4.2.0 at SourceForge.net

The interactive file manager requires Javascript. Please enable it or use sftp or scp.
You may still browse the files here.

Name	Modified	Size	InfoDownloads / Week
Parent folder
4.2.0 source code.tar.gz	2026-04-22	2.1 MB	1
4.2.0 source code.zip	2026-04-22	2.2 MB	0
README.md	2026-04-22	1.5 kB	0
Totals: 3 Items		4.3 MB	1

Enhancement
Support numpy>=1.26,<3 (adds numpy 2.x support).
Bump scikit-learn range to >=1.5,<1.8 (numpy 2.x support starts in 1.5; this raises the minimum supported version, which is a consumer-visible change).
Bump shap range to >=0.43,<0.49.
Bump lightgbm range to >=4,<5.
Bug Fix
xgb_classification_learner / lgbm_classification_learner: handle SHAP >=0.45 returning a single 3D ndarray for multiclass TreeExplainer (instead of the legacy list of per-class 2D arrays).
lgbm_classification_learner: handle SHAP >=0.45 returning a single 2D ndarray with a scalar expected_value for binary TreeExplainer (instead of the legacy [neg_class, pos_class] pair).
lgbm_classification_learner: pass feature_name and categorical_feature to lightgbm.Dataset instead of lightgbm.train (required by lightgbm >=4). Note: a user-supplied feature_name is now honored at the Dataset level — previously the Dataset always used string-cast features regardless of the feature_name argument.
cumulative_effect_curve (and the gain / relative-gain / effect_curves functions that build on it): use a stable sort when ordering rows by prediction score. Previous default (kind="quicksort") produced non-deterministic tie-breaking across numpy versions, which changed the rows included in each cumulative prefix when prediction had ties.