- Increased the interpretability of interactions with the new _get_unique_term_affiliation_shape_ method. - Removed the dependency on numpy, which was previously used for type hints. - Updated examples and documentation.
10.3.0
- Added support for MacOS. - Added a convenience method, get_base_predictors_in_each_unique_term_affiliation(), which makes it easier to merge local feature contribution with relevant predictor values.
10.2.1
Bugfix - Fixed a bug in the get_main_effect_shape method. - Removed unnecessary complexity in the formation of interaction terms.
10.2.0
Improved interpretability and updated the documentation. Feature importance and local feature contribution are now broken down into main effects and interactions. Local feature contribution can for example be used to interpret interactions by plotting the local feature contribution for a particular two-way interaction against relevant predictor values in a 3D scatter plot.
10.1.0
Improved the handling of interactions. Improved the documentation.
10.0.0
Replaced get_coefficient_shape_function() with get_main_effect_shape(). The former was difficult to compute correctly. The new method also makes it easier to interpret main effects.