parameter-aware scoring function on GridSearch sklearn

user_4156713 0
Viewed 43

Is there a way to pass the current candidate model as an input to a custom scoring function?

I am looking for something similar to thecross_val_score method, where clf is one of the inputs.

Passing the model as an extra argument does not work since the copy of the candidate is model is not carried in the scoring procedure. What I get instead is an unfitted model with the default parameters.

gridsearchcv scikit-learn python
1 Answers

Yes, that's how scoring works in GridSearchCV: the scoring parameter can be a callable, with signature (estimator, X, y). The search fits a copy of the estimator on training folds, and passes that into your scorer together with the test fold data. See the User Guide.

user_104958930
Related