Does CatBoost usebestmodel necessarily stop training as soon as validation stops improving?
Classic MLLevel: IntermediateCatBoost · Training parameters
Does CatBoost use_best_model necessarily stop training as soon as validation stops improving?
Hints
- Model selection and stopping computation are different operations.
- Inspect the retained prefix separately from the training log.
Answer
No. It selects the retained tree prefix using the validation metric; stopping computation is a separate overfitting-detector concern. Check both training iterations and saved tree_count_ when diagnosing runtime.
Source and licence: CatBoost · Training parameters · Apache-2.0 · answer: Quantoros editorial