Originally from: Shreeyak Sajjan
I’m training with a strategy of alternate batches of 2 datasets.
I.e., 1 batch of images from dataset A only, then a batch full of images from dataset B only. The sizes of the datasets are mismatched, but both use same batch size.
Any directions to achieve this with pytorch lightning? Normally, I’d look at the batch_idx and select a datset to draw from based on whether it’d odd or even