What's the difference between passing numpy to Dataloader and converting to tensors and passing to TensorDataset then Dataloader?

1

As above. I tried passing numpy to Dataloader directly, it worked. How does this contrast with converting numpy to tensor and passing it to TensorDataset then pass the dataset to Dataloader?

2

How did you pass the numpy arrays to the DataLoader?
Did you get the expected batches for both, the data and target?

I would personally see it as a bit of a hack, if that’s working.
This simple example won’t throw an error, but doesn’t yield the expected results:

data, target = np.zeros((100, 1)), np.ones((100, 1))
loader_np = DataLoader((data, target), batch_size=1)
for x in loader_np:
    print(x.shape)
    print(type(x))
    print(x)