Closed lorenzozanisi closed 1 year ago
Hi @lorenzozanisi, you are right, the strategy of saving the model with the best validation performance will not have a real effect. Thanks for reporting, will fix it soon.
@all-contributors please add @lorenzozanisi for bug
@xuyxu
I've put up a pull request to add @lorenzozanisi! :tada:
Great, thanks for the fix!
I was having a look at bagging models in
bagging.py
. Specifically, at this line it looks like the ensemble state at the end of the training is saved, thus overwriting the decisions made in the validation loop. Can someone confirm if I'm reading this correctly?