Each trainer is modified such that code related to the validation dataset and model selection is NOT executed when USE_LAST_EPOCH is set to True.
Main.py and config.py are modified to better indicate to users whether or not the validation dataset matters and to catch any weird behavior a user may encounter if they modify their configuration file incorrectly.
I don't think the config (YAML) files themselves have to be updated, especially since any additional functionality related to using a validation dataset (e.g., preprocessing, loading) should only ever occur if a validation dataset is actually being used for model selection in subsequent testing.
This PR makes two key changes:
I don't think the config (YAML) files themselves have to be updated, especially since any additional functionality related to using a validation dataset (e.g., preprocessing, loading) should only ever occur if a validation dataset is actually being used for model selection in subsequent testing.