Keras Deep Learning Cookbook

Train the generator using feedback from a discriminator

In this section, we look at how generator loss is calculated and the discriminator is made smarter by feeding it the loss between real and valid images:

• First, create adversarial ground truths: valid and fake image data holders
• Iterate over epochs:
  • Select a random set of images
  • Using noise, generate images using the generator
  • Use real images and valid images to calculate d_loss against fake and valid images
  • Calculate the total discriminator loss as an average
  • Calculate the combined loss for generator and discriminator stacked on top of each other:

def train(self, epochs, batch_size=128, save_interval=50):    # Load the dataset    (X_train, _), (_, _) = fashion_mnist.load_data()    # Rescale ...