In successive halving, must the resource always be the number of training samples?
Classic MLLevel: IntermediateSuccessive halving resource
In successive halving, must the resource always be the number of training samples?
Hints
- Resource and hyperparameter candidate are different roles.
- Low-budget rankings need predictive value.
Answer
No. It can be another supported positive integer resource such as n_estimators. The resource must make low-budget scores meaningful enough to screen candidates.
Source and licence: scikit-learn 1.7.2 — grid search · BSD-3-Clause · answer: Quantoros editorial