This should stop after PATIENCE epochs beyond the current best performing epoch (I set this to 25 arbitrarily). So if we peak at epoch 60, it would run until 85 and then stop early if none of 61-85 performed better than 60 did.
It also now saves the state_dict of the best performing epoch and saves that instead of the last one.
This should stop after
PATIENCEepochs beyond the current best performing epoch (I set this to 25 arbitrarily). So if we peak at epoch 60, it would run until 85 and then stop early if none of 61-85 performed better than 60 did.It also now saves the state_dict of the best performing epoch and saves that instead of the last one.