Why there is nn.softmax before nn.crossentroyloss in infogan?

[Lucksong]

In 111 line of the implementaion of infogan, nn.softmax is added to aux_layer. But the returned data of aux_layer is used to compute crossentropy loss in 277 line. Doesn`t nn.crossentropyloss module integrated functions of log_softmax and nllloss?

[Lucksong]

One more thing, sample_image() function is for verification, shouldn`t we set the model to test mode firstly?

[mo666666]

In 111 line of the implementaion of infogan, nn.softmax is added to aux_layer. But the returned data of aux_layer is used to compute crossentropy loss in 277 line. Doesn`t nn.crossentropyloss module integrated functions of log_softmax and nllloss?

I also think nn.softmax is not needed here.