Torch.sum() has a problem while adding byte tensor

1

I work in python 2.7 , I noticed when summing a logical values as a byte tensor sometimes Torch.sum() gives wrong values which resolves when the logical is type casted to int etc . Can Somebody give me a reason for this . Thank You

2

This is a byte overflow. A common way is to avoid it is to cast: count = flag.long().sum().

Best regards

Thomas

3

Oh, Thank you very much, I never encountered this in numpy . Maybe torch.nonzero() is a an efficient option ?

4

If you don’t need the count, you can use .any(). (There is .all(), too.)

Best regards

Thomas

5

Thank you for your kind clarification