def __init__(self, z_dim, batch_size):
"""."""
logger.info("MNIST GAN CNN")
logger.info("z_dim: {}".format(z_dim))
mnist_data_dir = get_project_directory("mnist", "dataset")
logger.info("mnist_data_dir: {}".format(mnist_data_dir))
self.mnist_data = input_data.read_data_sets(mnist_data_dir,
one_hot=True)
self.z_dim = z_dim
self.z_mean = np.zeros(self.z_dim)
self.z_cov = np.diag(np.ones(self.z_dim))
self.x_dim = 784
self.y_dim = 10
self.batch_size = batch_size
self.G_input_dim = self.y_dim
self.D_input_dim = self.x_dim + self.y_dim
self.real_prob_val = 0.9
self.fake_prob_val = 0.1
self.plot_batch_x, self.plot_batch_y = self.get_data_batches(
10 * 10, normalize="no")
self.plot_batch_z = self.get_noise_batches(10 * 10, random_seed=1)
self.std_test_data = False
self.__build_model()
self.__start_session()
def __init__(self):
"""."""
mnist_data_dir = get_project_directory("mnist", "dataset")
self.mnist_data = input_data.read_data_sets(mnist_data_dir,
one_hot=True)
self.__build_model()
self.__build_accuracy_computation()
self.__start_session()
def plot_fake_data(self, epoch, batch_fx, fig_plots=(10, 10), figsize=(10, 10)):
"""."""
batch_fx = batch_fx.reshape(-1, 28, 28)
plt.figure(figsize=figsize)
for i in range(batch_fx.shape[0]):
plt.subplot(fig_plots[0], fig_plots[1], i + 1)
plt.imshow(batch_fx[i], interpolation='nearest', cmap='gray_r')
plt.axis('off')
plt.tight_layout()
img_dir = get_project_directory("mnist", "results", self.start_timestamp)
plt.savefig(os.path.join(img_dir, "mnist_gan_mlp_{}.png".format(epoch)))
plt.close()
self.std_batch_z = self.get_noise_batches(batch_size=10 *
10)
self.std_test_data = True
loss_G, loss_D, D1, D2, batch_fx = self.sess.run(
[self.loss_G, self.loss_D, self.D1, self.D2, self.G],
feed_dict={
self.x: self.std_batch_x,
self.y: self.std_batch_y,
self.z: self.std_batch_z,
self.keep_prob: 1.0
})
self.plot_losses(epoch=i, loss_G=loss_G, loss_D=loss_D)
self.plot_fake_data(epoch=i, batch_fx=batch_fx)
time_diff = time.time() - start_time
start_time = time.time()
logger.info(
"Epoch: {:3d} - L_G: {:0.3f} - L_D: {:0.3f} - D1: {:0.3f} - D2: {:0.3f} - Time: {:0.1f}"
.format(i, loss_G, loss_D, D1[0][0], D2[0][0], time_diff))
if __name__ == "__main__":
setup_logger(log_directory=get_project_directory("mnist_cnn", "logs"),
file_handler_type=HandlerType.TIME_ROTATING_FILE_HANDLER,
allow_console_logging=True,
allow_file_logging=True,
max_file_size_bytes=10000,
change_log_level=None)
mnist_gan_mlp = GAN_CNN(z_dim=10, batch_size=100)
mnist_gan_mlp.run(epochs=1000, batch_size=100, summary_epochs=1)