Hands-On Generative Adversarial Networks with Keras

Training

Our training method is divided into two parts; let's take a look at the function signature first:

def train(data_filepath='data/flowers.hdf5', ndf=64, ngf=128, z_dim=128,          emb_dim=128, lr_d=2e-4, lr_g=2e-4, n_iterations=int(1e6),          batch_size=64, iters_per_checkpoint=500, n_checkpoint_samples=16,          out_dir='gan'):

Let's look at each argument of our train method:

Argument	Description
data_filepath	File path to the dataset
ndf, ngf	Number of discriminator and generator filters
z_dim, emb_dim	Dimensionality of the noise vector and the text-embedding projection
lr_d, lr_g	Learning rate of the discriminator and generator
n_iterations, batch_size, iters_per_checkpoint	Self-descriptive
n_checkpoint_samples	Samples to log on each ...