How should I implement cross-entropy loss with continuous target outputs?

18

I meant I don’t get good results (of course there might be a different cause for this)
Anyway I would feel more comfortable if the model could just output a softmax and this way I could make sure there is no problem there.