Errata in the readme?

[lezcano]

In the MNIST example in the readme, the functions tng_dataloader, val_dataloader, and test_dataloader return the same object. I think that at least for test, the train flag used to create the MNIST object should be False. I am not aware of how to extract automatically a validation dataset from the MNIST dataloader in torch.

[williamFalcon]

nah, it’s just for illustration and simplicity only. mnist has no val set. in the real world you’d split the train portion into a train/val split by subclassing the mnist dataset.

the readme is to illustrate the main ideas