CrossEntropyLoss는 label을 indice로 줘야한다. one-hot-vector는 안돼.
a = Variable(torch.from_numpy(np.array([[0.7,0.5],[0.4,0.9]])))
label = (torch.from_numpy(np.array([1.0,0.0])))
label = label.type(torch.LongTensor)
label = Variable(label)
RuntimeError: multi-target not supported (newbie)
https://discuss.pytorch.org/t/runtimeerror-multi-target-not-supported-newbie/10216