Open
Description
Problem:
BasePytorchForecaster.tune
(src code) currently only support numpy array tuple as input, so that batch size can be tuned.
Solution:
We could also support train_dataloaders as input. Just in this case, batch size cannot be tuned option.