Masking the columns of pytorch matrix

Question

I have a matrix of shape (batch_size, N, N) and a masking tensor of shape (batch_size, N).

I want to put -infinity values only for the columns (and not rows) of the matrix, according to the given mask.

Answer 1

Alternatively, you can use torch.Tensor.expand_as:

>>> x[m[...,None].expand_as(x)]

Answer 2

The solution is via repeating the mask in the correct dimension:

repeated_mask[mask.unsqueeze(-1).repeat(1,1,mask.shape[-1])!=1] = 0