Tests with OpenML and internal benchmarking suites suggest that the DataTransformer un-biasing procedure may hinder balanced accuracy. It is a useful behavior, but it might be better to have it disabled by default.
How
Setting the equal_accuracy_for_all_output_categories to False by default.
Why
Tests with OpenML and internal benchmarking suites suggest that the
DataTransformerun-biasing procedure may hinder balanced accuracy. It is a useful behavior, but it might be better to have it disabled by default.How
Setting the
equal_accuracy_for_all_output_categoriestoFalseby default.