Several datamodules ignoring batch_size

[hecoding]

🐛 Bug

Realized MNISTDataModule was ignoring batch_size parameter. I found a closed issue (#171) referring to that without a fix.

While fixing it myself (PR #331), I found more datamodules had this problem too - MNISTDataModule, BinaryMNISTDataModule, FashionMNISTDataModule, SklearnDataModule, SSLImagenetDataModule.

I could take care of that. My question is, is this signature from MNIST in use anymore?:

def train_dataloader(self, batch_size=64, transforms=None):

That is basically where the bug comes from. Other datamodules working fine, like CIFAR10DataModule, are simply:

def train_dataloader(self):

Same thing goes for val_dataloader and test_dataloader.

To Reproduce

Steps to reproduce the behavior:

1. Set up MNISTDataModule to use a batch size anything else than 32.
2. Run an experiment
3. Check the batch sizes are 32 regardless

Code sample

from pytorch_lightning.core.lightning import LightningModule
from pytorch_lightning import Trainer
from pl_bolts.datamodules.mnist_datamodule import MNISTDataModule

batch_size = 64


class SampleModel(LightningModule):
    def configure_optimizers(self):
        pass

    def training_step(self, batch, batch_idx):
        x, y = batch
        assert x.shape[0] == batch_size
        print('fine')


model = SampleModel()
dm = MNISTDataModule(data_dir='~/Datasets/', batch_size=batch_size)

trainer = Trainer()
trainer.fit(model, dm)

Expected behavior

Batch sizes of 64.
Printing fine instead of assert error.

Environment

• PyTorch Version (e.g., 1.0): 1.7
• OS (e.g., Linux): Ubuntu 18.04
• How you installed PyTorch (conda, pip, source): conda
• Build command you used (if compiling from source):
• Python version: 3.8
• CUDA/cuDNN version:
• GPU models and configuration:
• Any other relevant information: pytorch-lightning-bolts 0.2.5

Additional context

[github-actions]

Hi! thanks for your contribution!, great first issue!

[ananyahjha93]

@hecoding send in the PR.

the signature is def train_dataloader(self):. The datamodule class init should take the batch size params and not the train_dataloader.

[hecoding]

Cool, I fixed the signatures too. Please @ananyahjha93 have a look at the PR here #331
Should I assign somebody for review myself? I'm a bit lost on that.

[hecoding]

I can fix the rest of the data modules listed too. Let me know.