Found some strange behavior of masked_scatter_. which confuses the C type indices and Pascal type indices.
Following is a minimum example:
a = autograd.Variable(torch.zeros(3,2))
Variable containing:
0 0
0 0
0 0
[torch.FloatTensor of size 3x2]
m = autograd.Variable(torch.ByteTensor([1,0]))
Variable containing:
1
0
[torch.ByteTensor of size 2]
c = autograd.Variable(torch.arange(0, 6).view(3,2))
Variable containing:
0 1
2 3
4 5
[torch.FloatTensor of size 3x2]
a.masked_scatter_(m.expand(3,-1), c)
I am planning to obtain
Variable containing:
0 0
2 0
4 0
[torch.FloatTensor of size 3x2]
But got
Variable containing:
0 0
1 0
2 0
[torch.FloatTensor of size 3x2]
Any idea that can help solve this?
Thanks a lot!