Hands-On Generative Adversarial Networks with Keras

Initializations

Let's implement the first part of our train method using the following steps:

1. Define a global batch for the Random Weighted Average calculation:

    global BATCH_SIZE    BATCH_SIZE = batch_size

2. Instantiate the logger to save logs in out_dir and log learning rates of G and D:

    logger = SummaryWriter(out_dir)    logger.add_scalar('d_lr', lr_d, 0)    logger.add_scalar('g_lr', lr_g, 0)

3. Load the data and set up the mini-batch iterator:

 data, token_to_id, id_to_token = get_data(        GLOB_STR, vocabulary_size, sos, eos, unk_token, symbol_based)    data_iterator = iterate_minibatches(        data, token_to_id, vocabulary_size,                              max_sentence_length, batch_size)

4. Sample real data, report the shape, and log the text for reference:

 real_sample = ...