Loss.SoftCrossEntropyLoss() not work!!!

[YanqingWu]

Loss.SoftCrossEntropyLoss() not work.

[maksimovkonstantin]

@YanqingWu without full case description it is useless issue.

[BloodAxe]

Would you please share code to illustrate the problem you experience? пн, 24 февр. 2020 г. в 2:56 PM, Yanqing Wu <notifications@github.com>:

…

Loss.SoftCrossEntropyLoss() not work. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#38?email_source=notifications&email_token=AAEB6YF63YVQHOYEGYLG2NDREO7XDA5CNFSM4K2HVJU2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IPXBMJA>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAEB6YDHJQKHENPKYHZ3PLTREO7XDANCNFSM4K2HVJUQ> .

[YanqingWu]

@BloodAxe @konstantin Maksimov

import torch
import pytorch_toolbelt.losses as loss
sfce = loss.SoftCrossEntropyLoss()
outputs = torch.rand(10, 2).float()
targets = torch.randint(0, 2, (10, 1)).view(-1).long()
targets2 = torch.randint(0, 2, (10, 1)).long()
print('test 1: ', sfce(outputs, targets))
print('test 2: ', sfce(outputs, targets2))

two tests I get the same error: It should not be a dimension not match or some other common problem.
May be I have the wrong torch version?
print(torch.version)
I get : 1.3.1

Traceback (most recent call last):
File "/usr/local/anaconda3/lib/python3.7/site-packages/IPython/core/interactiveshell.py", line 3326, in run_code
exec(code_obj, self.user_global_ns, self.user_ns)
File "", line 7, in
print('test 1: ', sfce(outputs, targets))
File "/usr/local/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 541, in call
result = self.forward(*input, **kwargs)
File "/usr/local/anaconda3/lib/python3.7/site-packages/pytorch_toolbelt/losses/soft_ce.py", line 15, in forward
not_ignored = label_target != self.ignore_index
TypeError: ne() received an invalid combination of arguments - got (NoneType), but expected one of:

• (Tensor other)
  didn't match because some of the arguments have invalid types: (!NoneType!)
• (Number other)
  didn't match because some of the arguments have invalid types: (!NoneType!)

[BloodAxe]

Thanks for the reproduction code. As for now you can use quick workaround loss.SoftCrossEntropyLoss(ignore_index=-100), Correct default value will be fixed in 0.3.1 release.