Hands-On Generative Adversarial Networks with Keras

The training loop

Finally, let's look at the training loop in the following steps:

1. Freeze the generator weights and the sample z using the following code:

    for i in range(n_iterations):         # freeze generator weights and sample z        D.trainable = True        G.trainable = False        z = np.random.normal(0, 1, size=(batch_size, z_dim))

2. Load the training data with images and embedded text:

        real_batch = next(data_iterator)        images_batch = images_from_bytes(real_batch[0])        emb_text_batch = real_batch[1]

3. Compute the discriminator's loss:

            loss_d = D_model.train_on_batch(                [images_batch, emb_text_batch, z],                [ones, zeros, zeros])

4. Freeze the discriminator weights and sample z :

        D.trainable = False        G.trainable = True z = np.random.normal(0, 1, size=(batch_size, ...