On the issue of model selection

[PIPIKAI]

CIRL/train.py

Line 192 in 1b50cd6

if self.results['test'][self.current_epoch] >= self.best_acc:

Choose the best model in your training code based on the correctness of the test set rather than the best correctness of the validation set.

[SakurajimaMaiii]

I have the same question.
Previous work, such as FACT uses the training domain validation.
It seems that this paper chooses the best model on the (out of distribution) test set.

[JeffLee1874]

I have the same question.

[eadstry]

I have the same question. In Search of Lost Domain Generalization， pointed out that the optimal model on test should not be selected, which leads to overfitting. The appropriate choice is to select the model on the last epoch

[Linzsd]

can you reproduce the experiments?

[undercutspiky]

Same question.