def main(argv):
input_fns, charset = make_wikitext_char_input_fn(
data_dir=tf.flags.FLAGS.data_dir,
batch_length=tf.flags.FLAGS.batch_length,
batch_size=tf.flags.FLAGS.batch_size
)
layers = 6
kernel_size = 5
model_mode = 'cnn'
config = TextConfig(
model_fn=make_model_gan_fn(
charset=charset,
decoder_fn=make_decoder_cnn_fn(
bn=True,
layers=layers,
kernel_size=kernel_size,
padding='valid'),
gan_discriminator_fn=make_discriminator_gan_cnn_ml_fn(
layers=6,
kernel_size=5,
padding='valid'),
gan_loss_fn=wgan_losses,
model_mode=model_mode,
dis_opt=tf.train.AdamOptimizer(1e-5, name='dis_opt'),
gen_opt=tf.train.AdamOptimizer(1e-5, name='gen_opt'),
combined=True,
padding_size=layers * (kernel_size - 1)
),
input_fns=input_fns,
mode=model_mode
)
text_aae.trainer.main(argv, config=config)
def main(argv):
input_fns, charset = make_wikitext_char_input_fn(
data_dir=tf.flags.FLAGS.data_dir,
batch_length=tf.flags.FLAGS.batch_length,
batch_size=tf.flags.FLAGS.batch_size)
model_mode = 'cnn'
config = TextConfig(model_fn=make_model_vae_fn(charset=charset,
encoder_fn=encoder_cnn_fn,
decoder_fn=decoder_cnn_fn,
model_mode=model_mode),
input_fns=input_fns,
mode=model_mode)
text_aae.trainer.main(argv, config=config)
def main(argv):
input_fns, charset = make_wikitext_char_input_fn(
data_dir=tf.flags.FLAGS.data_dir,
batch_length=tf.flags.FLAGS.batch_length,
batch_size=tf.flags.FLAGS.batch_size
)
layers = 9
kernel_size = 3
model_mode = 'cnn'
config = TextConfig(
model_fn=make_model_gan_fn(
charset=charset,
decoder_fn=make_decoder_resnet_cnn_fn(
bn=True,
layers=layers,
kernel_size=kernel_size,
activation_fn=tf.nn.leaky_relu,
padding='valid'),
gan_discriminator_fn=make_discriminator_gan_cnn_ml_fn(
bn_fn=make_batch_norm(eps=1e-6, clip_var=1.),
layers=9,
kernel_size=5,
padding='valid',
emedding_scale=10,
activation_fn=tf.nn.leaky_relu
),
gan_loss_fn=wgan_losses,
model_mode=model_mode,
dis_opt=tf.train.AdamOptimizer(1e-5, name='dis_opt'),
gen_opt=tf.train.AdamOptimizer(1e-5, name='gen_opt'),
combined=True,
padding_size=layers * (kernel_size - 1)*2
),
input_fns=input_fns,
mode=model_mode
)
text_aae.trainer.main(argv, config=config)