Why can a CatBoost pipeline lose some categorical transformations when target data needed by a CTR type is unavailable?
Classic MLLevel: IntermediateCatBoost · Categorical features
Why can a CatBoost pipeline lose some categorical transformations when target data needed by a CTR type is unavailable?
Hints
- Each statistic has required inputs.
- Device-specific fallbacks are not interchangeable.
Answer
Target-dependent statistics cannot be computed without their required labels. The documented fallback differs between CPU and GPU; audit effective categorical handling rather than assuming that declaring cat_features guarantees all CTR types.
Source and licence: CatBoost · Categorical features · Apache-2.0 · answer: Quantoros editorial