pdfcoffee

[ 195 ]
Chapter 6
Next, we combine the generator and discriminator together to form a GAN. In the
GAN we ensure that the discriminator weights are fixed by setting the trainable
argument to False:
discriminator.trainable = False
ganInput = Input(shape=(randomDim,))
x = generator(ganInput)
ganOutput = discriminator(x)
gan = Model(inputs=ganInput, outputs=ganOutput)
The trick to train the two is that we first train the discriminator separately; we
use binary cross entropy loss for the discriminator. Later we freeze the weights of
the discriminator and train the combined GAN; this results in the training of the
generator. The loss this time is also binary cross entropy:
discriminator.compile(loss='binary_crossentropy', optimizer='adam')
gan.compile(loss='binary_crossentropy', optimizer='adam')
Let us now perform the training. For each epoch we take a sample of random noise
first, feed it to the generator, and the generator produces a fake image. We combine
the generated fake images and the actual training images in a batch with their
specific labels and use them to train the discriminator first on the given batch:
def train(epochs=1, batchSize=128):
batchCount = int(X_train.shape[0] / batchSize)
print ('Epochs:', epochs)
print ('Batch size:', batchSize)
print ('Batches per epoch:', batchCount)
for e in range(1, epochs+1):
print ('-'*15, 'Epoch %d' % e, '-'*15)
for _ in range(batchCount):
# Get a random set of input noise and images
noise = np.random.normal(0, 1, size=[batchSize, randomDim])
imageBatch = X_train[np.random.randint(0, X_train.
shape[0], size=batchSize)]
# Generate fake MNIST images
generatedImages = generator.predict(noise)
# print np.shape(imageBatch), np.shape(generatedImages)
X = np.concatenate([imageBatch, generatedImages])
# Labels for generated and real data
yDis = np.zeros(2*batchSize)
# One-sided label smoothing
yDis[:batchSize] = 0.9