DummyCodeEncoder#

DummyCodeEncoder is the scikit-learn-style fitted encoder. It supports fit, transform, inverse_transform and get_feature_names_out and can produce sparse output.

Unknown-category handling should be chosen deliberately. ignore can keep inference running but may map an unseen category to an all-zero representation; error is safer when schema drift must be surfaced immediately.

Warning

The current maintenance review records two correctness findings in this encoder: cross-feature category collapse and an infrequent-category grouping path that is not fully applied. Do not assume those advanced behaviors are release-ready until the findings are closed and regression-tested.