CatBoost labels are fraud and legitimate. Why is an explicit class-to-weight dictionary safer than an undocumented…
Classic MLLevel: IntermediateCatBoost · Training parameters
CatBoost labels are fraud and legitimate. Why is an explicit class-to-weight dictionary safer than an undocumented two-element list?
Hints
- Ordering is part of the data contract.
- Independent balancing mechanisms should not be stacked blindly.
Answer
It binds each multiplier to a label rather than an assumed class ordering. Check the model class mapping and avoid combining class_weights with auto_class_weights or scale_pos_weight, as documented.
Source and licence: CatBoost · Training parameters · Apache-2.0 · answer: Quantoros editorial