indent=0)
# --- build model --- #
# preparation: data placeholders and model parameters
Z, X, T = model.create_placeholders(batch_size, seq_length, latent_dim,
num_variables)
discriminator_vars = [
'hidden_units_d', 'seq_length', 'batch_size', 'batch_mean'
]
discriminator_settings = dict((k, settings[k]) for k in discriminator_vars)
generator_vars = ['hidden_units_g', 'seq_length', 'batch_size', 'learn_scale']
generator_settings = dict((k, settings[k]) for k in generator_vars)
generator_settings['num_signals'] = num_variables
# model: GAN losses
D_loss, G_loss = model.GAN_loss(Z, X, generator_settings,
discriminator_settings)
D_solver, G_solver, priv_accountant = model.GAN_solvers(
D_loss,
G_loss,
learning_rate,
batch_size,
total_examples=samples.shape[0],
l2norm_bound=l2norm_bound,
batches_per_lot=batches_per_lot,
sigma=dp_sigma,
dp=dp)
# model: generate samples for visualization
G_sample = model.generator(Z, **generator_settings, reuse=True)
# # --- evaluation settings--- #
#
'hidden_units_d', 'seq_length', 'cond_dim', 'batch_size', 'batch_mean'
]
discriminator_settings = dict((k, settings[k]) for k in discriminator_vars)
generator_vars = [
'hidden_units_g', 'seq_length', 'batch_size', 'num_generated_features',
'cond_dim', 'learn_scale'
]
generator_settings = dict((k, settings[k]) for k in generator_vars)
CGAN = (cond_dim > 0)
print(CGAN)
D_loss, G_loss, accuracy = model.GAN_loss(Z,
X,
generator_settings,
discriminator_settings,
CGAN,
CG,
CD,
CS,
wrong_labels=wrong_labels)
D_solver, G_solver = model.GAN_solvers(
D_loss,
G_loss,
learning_rate,
batch_size,
total_examples=samples['train'].shape[0],
l2norm_bound=0,
batches_per_lot=0,
sigma=0,
dp=False)
G_sample = model.generator(Z, **generator_settings, reuse=True, c=CG)
'hidden_units_g', 'seq_length', 'batch_size', 'num_generated_features',
'cond_dim', 'learn_scale'
]
generator_settings = dict((k, settings[k]) for k in generator_vars)
CGAN = (cond_dim > 0)
if CGAN: assert not predict_labels
if info:
info_loss, D_loss, G_loss = model.GAN_loss(info,
latent_C,
latent_C_dim,
Z,
X,
generator_settings,
discriminator_settings,
kappa,
CGAN,
CG,
CD,
CS,
cond_sine,
wrong_labels=wrong_labels)
info_solver, D_solver, G_solver, priv_accountant = model.GAN_solvers(
info,
info_loss,
D_loss,
G_loss,
learning_rate,
batch_size,
total_examples=samples['train'].shape[0],
l2norm_bound=l2norm_bound,
