Why can native Booster.predict disagree with sklearn predictions immediately after early stopping?
Classic MLLevel: IntermediateXGBoost · Prediction
Why can native Booster.predict disagree with sklearn predictions immediately after early stopping?
Hints
- Training may retain rounds after the best score.
- Check the exclusive endpoint of iteration_range.
Answer
The native interface may use all retained rounds, while sklearn uses best_iteration. Use the explicit half-open range (0, best_iteration + 1) or a correctly saved best model when comparing them.
Source and licence: XGBoost · Prediction · Apache-2.0 · answer: Quantoros editorial