Why is early stopping not enabled by default in LightGBM?
Classic MLLevel: IntermediateValidation policy for early stopping
Why is early stopping not enabled by default in LightGBM?
Hints
- State the task, inputs and prediction target.
- Compare with a simple baseline: assumptions, objective and independent validation.
Answer
Early stopping requires a validation set whose quality is checked after boosting iterations. LightGBM cannot choose a universally correct split: IID, grouped and temporal data require different policies. The modeler must supply an appropriate holdout and stopping configuration. Because stopping decisions use that holdout repeatedly, it participates in model selection and should not be treated as an untouched final test set.
Source and licence: LightGBM developers — Official FAQ · MIT · answer: Quantoros editorial