How to stack over for loop?

1

I have a loop, and I am getting a 10x10 tensor for each iteration of that loop. Lets assume that I am running that loop five times, and the output after the loop completes should be the concatenation of these tensors, i.e., a size of 10x10x5. How to concatenate this?

outx = []
for i in range(5):
    tmp = net(x) # this will return a 10x10 tensor
    outx = # need to cat tmp with outx in dim=2

outx - should have a dimension of 10x10x5
3 Likes
2

Building the list and then using stack at the end is reasonable:

outx = []
for i in range(5):
    tmp = net(x) # this will return a 10x10 tensor
    outx.append(tmp)

outx = torch.stack(outx, 2)

Best regards

Thomas

5 Likes
3

Hi @tom ,
I had a question, if the outputs that I want to append to a list are my model outputs, Will appending them to a list and then applying torch.stack break the computation graph?
Thank you.

4

No. If any of the inputs in the list (which torch.stack will iterate over) requires gradient, so does the output and the graph is connected accordingly.

Best regards

Thomas

1 Like