Pytorch implementation of adversarial autoencoders