Why can two numeric features use different missing-value modes despite one global CatBoost nanmode setting?
Classic MLLevel: IntermediateCatBoost · Missing values
Why can two numeric features use different missing-value modes despite one global CatBoost nan_mode setting?
Hints
- There can be more than one configuration layer.
- Look for a feature-level override.
Answer
A custom quantization-borders file can specify per-feature missing modes that override the global default. Inspect that file along with training parameters when reproducing a model.
Source and licence: CatBoost · Missing values · Apache-2.0 · answer: Quantoros editorial