Need to train Discriminator adversarially

[liujiahao11]

In your code, D was jointly trained based on sum of 3 loss function. As a variant of GAN, you should train G+E and D based on opposite loss function.