How can I implement the cross validation for a GAN model composed of 3 networks (generator + discriminator + Encoder)