What is the difference between permutationa and reshape?

[Endtroducing1993]

Hi Team,

Suppose we have a tensor called tensor_A of shape (3,4).

What is the difference between tensor.reshape(4,3) and tensor.permute(1,0)

I have seen by creating a random tensor that we obtain different results.

Can you please suggest, which one to use and where?

original tensor
tensor_original=torch.rand(size=(3,4))
tensor([[0.9829, 0.2585, 0.1642, 0.6212],
[0.6378, 0.7740, 0.8801, 0.7784],
[0.0042, 0.5443, 0.8029, 0.4538]])

rehshaped_tensor
tensor_reshape=tensor_original.reshape(4,3)
tensor([[0.9829, 0.2585, 0.1642],
[0.6212, 0.6378, 0.7740],
[0.8801, 0.7784, 0.0042],
[0.5443, 0.8029, 0.4538]])

permuted tensor
tensor_permuted=tensor_original.permute(1,0)
tensor([[0.9829, 0.6378, 0.0042],
[0.2585, 0.7740, 0.5443],
[0.1642, 0.8801, 0.8029],
[0.6212, 0.7784, 0.4538]])

[Perian-Yan]

Hi,
reshape changes the shape of the tensor arbitrarily, as long as the number of entries is the same. It's like assigning an index for each entry:
In the shape (12,1), indices are [0], [1], ..., [11];
in the shape (4, 3), indices are [0,0], [0,1], [0,2], [1,0], [1,1], ...

permute changes the order of axes. For 2d tensor, it's transpose (you swap x and y axis).