CrossEntropyLoss with LogSoftmax?

[shancun-c]

LeNet-5/lenet.py

Line 37 in b9d7f39

('sig7', nn.LogSoftmax(dim=-1))

I think there is no need to put LogSoftmax layer at the end of network if you are going to use CrossEntropyLoss ? Because CrossEntropyLoss combines nn.LogSoftmax() and nn.NLLLoss() in one single class.

[skyqwe123]

LeNet-5/lenet.py

Line 37 in b9d7f39

('sig7', nn.LogSoftmax(dim=-1))

I think there is no need to put LogSoftmax layer at the end of network if you are going to use CrossEntropyLoss ? Because CrossEntropyLoss combines nn.LogSoftmax() and nn.NLLLoss() in one single class.

@Frusciante @activatedgeek I think so. But I am not understand this:
x = self.c2_1(output) output = self.c2_2(output) output += x
It seems that C3 layer is not realized in this way。