Scikit-learn integration#

Both encoders follow estimator conventions closely enough to participate in pipeline-style workflows. Keep the encoder inside the fitted preprocessing pipeline so category state is learned only from training data.

Use get_feature_names_out when downstream diagnostics or model explainers need a stable mapping from encoded columns back to source features. Sparse output can reduce memory use for high-cardinality categories, but downstream estimators must accept sparse input.

Do not call fit_transform independently on train and test sets.