Difference between reshape and view?

[massonix]

Hi! Thanks for this wonderful course!

I have a question related with the following code

tensor = torch.rand(4)
a = tensor.reshape(2, 2)
b = tensor.view(2, 2)
a[0][2] = 4
b[0][1] = 3

From the video, I understood that the main difference between view and reshape is that reshape allocates the tensor a to new memory, while view shares the same memory as tensor. I thought this would translate in the fact that if you update an element of b, then tensor would also be updated, while this would not be the case for a. Instead however, both of them are updated

a
tensor([[4.0000, 3.0000],
        [0.5108, 0.4083]])
b
tensor([[4.0000, 3.0000],
        [0.5108, 0.4083]])
tensor
tensor([4.0000, 3.0000, 0.5108, 0.4083])

Could you please explain me why that is? Thank you!

[37Lime]

Hi massonix,

In your case, both a and b are updated because they are views of the original tensor. That is, they point to the same underlying memory location.

The tensor.reshape() method does not always return a copy of the tensor. In the documentation they say:

When possible, the returned tensor will be a view of input. Otherwise, it will be a copy.

I hope this helps.

Best wishes.