UDA Splits Not Used in domainbed/scripts/train.py During Training

[YizhenLiao]

I noticed that in the code domainbed/scripts/train.py, the uda_splits are not utilized during training. Could you clarify if there is an updated version of the code that integrates the uda_splits into the training step? Additionally, does the updated version unify the step and the uda_split settings?

For the leave-one-domain-out setting, are the reported results based solely on the test set (i.e., the domain used as the test set) with cross-validation (leaving each domain out as the test set in turn)? I am a bit confused because I couldn’t find a validation set in the code. Could you clarify this part?

Looking forward to your response. Thank you!