Release notes
2.0.0 (2025-11-24)
Suspended support fro UARFF files
Significant improvement of the backbone decision tree generation mechanism by moving away from object-oriented structure and focusing purely on Python-optimized structures (dicts)
Parsing dataframe is up to 10 times faster (usually more like 5x)
Fitting model using SHAP values, but without oversampling is up to 4 times faster
Oversampling alone is up 30 times faster
Fixes in dependencies related to upgraded numpy and shap packages.
1.3.2 (2025-02-05)
Fixes in dependencies
Fixes in numpy float representation of dataframes read from UARFF
Added more examples to documentation
1.3.0 (2024-09-06)
Fixed catBoost problem and minor visualization bugs
Fixed class balancing code
Added labelling of phantom branches in the leaves
Fixed bug with additional columns added when parity==’local’
Improved SHAP-guided sampling
Bugfixes in importance samplers
Added medoid CF for categorical variables
Fixed invalid requirements in build file
gower replaced with gower-multiprocessing
Fixed visualization issue when no instance nor counterfactual is passed.
Fixed issue with categorical variables.
Fixed with gower installed without multiprocessing
Added balancing to BOTH sampling strategy, to overcome heavy imbalance generated by SHAP-sampler
1.2.0 (2024-04-30)
Fixes in SHAP guided sampler
Fixed in export to HMR
Added documentation section on visualization
1.1.0 (2024-03-08)
Changed modularization (internal, so it will not affect usage). Samplers were moved to separate module
1.0.3 (2024-03-01)
Fixes in tree visualization
Improved documentation
1.0.1 (2024-02-23)
Added pypl installation
Added sphinx documentation