How to use multiple train dataloaders with different lengths

LightningModule
1

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

2

This is possible by using a custom dataset:
https://pytorch-lightning.readthedocs.io/en/stable/multiple_loaders.html

class ConcatDataset(torch.utils.data.Dataset):
    def __init__(self, *datasets):
        self.datasets = datasets

    def __getitem__(self, i):
        return tuple(d[i] for d in self.datasets)

    def __len__(self):
        return min(len(d) for d in self.datasets)

def train_dataloader(self):
    concat_dataset = ConcatDataset(
        datasets.ImageFolder(traindir_A),
        datasets.ImageFolder(traindir_B)
    )

    loader = torch.utils.data.DataLoader(
        concat_dataset,
        batch_size=args.batch_size,
        shuffle=True,
        num_workers=args.workers,
        pin_memory=True
    )
    return loader

def val_dataloader(self):
    # SAME
    ...

def test_dataloader(self):
    # SAME
1 Like