Bool value of Variable objects containing non-empty torch.ByteTensor is ambiguous runtime error

1

I am getting ‘bool value of Variable objects containing non-empty torch.ByteTensor is ambiguous’ runtime error when I run the below code.

elif mask[i] ==1 and var[i] - mu[i] > 5:

Not sure what this means.

2

mask[i], var[i] or mu[i] will most likely return a non-scalar tensor, which cannot be converted to bool.
Have a look at this dummy example:

x = torch.randn(1, 2)
if x[0]==1:
    print('passed')

As x[0] returns a tensor of shape torch.Size([2]), the comparison will also have the same shape, e.g. tensor([0, 0], dtype=torch.uint8).

If you want to check, if all values are equal to 1, you can use:

if (x[0]==1).all():
    print('passed')

, but obviously this depends on your use case.

1 Like
3

Thanks. But what if I do something like mask[0][1] ==1? I still get the same error where mask is a Variable

4

Could you print mask[0][1]? Is it a 0-dim tensor?

5

Ah. Ok. I figured it out. No it is a 1-dim tensor. But the if statement gives weird results if mask is a Variable. If its a tensor, it works fine.

m=Variable(torch.rand(3,4))
if m[0][1] > 0:
print(‘passed’)

vs

m=torch.rand(3,4)
if m[0][1] > 0:
print(‘passed’)

6

Not sure how to make it work with variables.

7

I’m not sure, but you should update to the latest release.
Variables and tensors were already merged in 0.4.0.
You can find the install instructions on the website.