Confusion with dataloader

Morning all,

I am reading in a 2048x420000 numpy array, this array is 420000 samples each 2048 long.
I convert this np array into a tensor using torch.Tensor and then set up a single tensor 1x2048 long for labels.
Using TensorDataset(X_train, y_train) to combine (I think as i understand???) this is then passed into DataLoader with a batch sie of 32.

This then gives me the error

RuntimeError: Expected 3-dimensional input for 3-dimensional weight 16 1, but got 2-dimensional input of size [32, 42000] instead

I’m confused, why isn’t the tensor generated by dataload [32 , 2048]? it is using the size of the row (42000) and not the columns (2048).


The error indicates that the supplied input dimension and weight dimension doe not match.
It is expecting a 3 dimensional input.

Further to this. Looking at the input vector, it has shape 32x42000, and not 32x3048 as intended, how do I change this?