exps.mnist package

exps.mnist.gan module

exps.mnist.gan.extended_input_fn(noise_dims, mnist_feeder, gan_type)

Make MNIST input fit for input to unconditional and conditional GAN

The TFGAN module feeds the features to generator and discriminator, but the labels are passed only to the discriminator. The labels are supposed to be the real input. For the unconditional GAN, the features merely contain the latent noise, and they are not used by the discriminator. In the case of the conditional GAN, the features include the class label in addition to the latent noise.

exps.mnist.gan.main(model_dir, save_summary_steps, save_checkpoints_steps, log_step_count_steps, gen_lr, crit_lr, max_train_steps, gan_type, mnist_feeder)

Run MNIST GAN Training

Example

$ python -m exps.mnist.gan with mnist_feeder.batch_size=16