Bug in Cross-Validation estimation about flaml

[sonichi]

Hi David, thanks for the comment. It doesn't produce unwanted result but I understand why it looks overcomplicated in the current form. With n_repeats=1 in kf, val_loss_list only contains one element which is the average of loss from valid_fold_num folds. So we don't really need the list as of now. Using the list, however, allows setting n_repeats larger than 1.

Please let us know if you find other ways of cross validation are needed in your application.

[davidwaroquiers]

Hi,

Thank you for your answer. Ok indeed, I see. What I wanted is to have access to the different validation scores of each fold and it was difficult (actually not possible without changing a little bit the code of course) to do that with the current implementation.

Best,

David

[sonichi]

"to have access to the different validation scores of each fold", do you need the scores per fold to be logged? Otherwise they are not going to be accessible from outside.

[sonichi]

@davidwaroquiers Since this is not really a bug I will close this issue now. If you need the scores per fold to be logged please create a separate issue for that feature request.