Fixing issue #57 related to cuda

[JaoCR]

When using cuda, the code is trying to take the argmin of an array that's being constructed with torch.Tensor objects, which results in an error that aborts the entire execution. See issue #57 for more info.

[JaoCR]

On the issue thread I mentioned there was another line that needed fixing, but that was a mistake, this change is already sufficient. But let me know if I should change the code so it only checks for the type of one of the error variables, I can do that here before the merge.