How to get index of non-zero values for every row in a matrix pytorch

1

I want to return a dense tensor of the non-zero indices for each row. For example, given the tensors:

[0,1,1]
[1,0,0]
[0,0,1]
[0,1,0]

Should return

[1,2]
[0]
[2]
[1]
2

What do you want to do accomplish ?
The output you gave as an example is not a tensor, as it does not have the same number of elements in each “row”.

You can try torch.nonzero(x), or x.ne(0) (meaning ‘not equal’).
One of the two solutions will probably suit your needs.

1 Like
3

One way to accomplish this is to do:

# x is a tensor
x.nonzero(as_tuple=True)[0]

Docs for reference

4

Hello, is there a way to obtain indices of nonzero values such that they are differentiable?