How does reshape moves entries?

#1

Hello, I have a neural network with two input tensors:

a shape (batch_size=128, 3)
b shape (60, batch_size=128, 1)

I need to augment a to have the same shape as b:

a.repeat(60, 1, 1) -> a shape (60, 128, 3)

Then I want to put a and b in the same neural network that accepts two dimensions. I need to reshape so that the 60 can be considered the same as the batch size:

result = neural_network(a.reshape(-1, 3), b.reshape(-1, 1))

This result now has the same ((60 x 128), 1)

Now I want to do the reshape so that the result has the shape (60, 128, 1) and it has the same order than a and b had initially. If I do:

result_reshaped = result.reshape(60, 128, 1) I will get that, however, how can I be sure that the order is correct?

Thanks!

#2

One way to double check the reshaping operations is to initialize the tensor via torch.arange, reshape it, and print the tensor to see where the values landed. To do so I would reduce the shapes a bit, as the currently used number of elements is quite large.