def _init(self):
super(GRN1D, self)._init()
if self.aux_discriminators is None:
self.aux_discriminators = [
Uniform1D(low=-5, high=-1),
Uniform1D(low=1, high=5)
]
self.num_aux = len(
self.aux_discriminators) # number of auxiliary distributions
if self.aux_coeffs is None:
self.aux_coeffs = [0.1] * self.num_aux
def test_cgan_mnist(show_figure=False, block_figure_on_end=False):
print("========== Test CGAN on MNIST data ==========")
np.random.seed(random_seed())
(x_train, y_train), (x_test, y_test) = demo.load_mnist()
x_train = x_train.astype(np.float32).reshape([-1, 28, 28, 1]) / 0.5 - 1.
x_test = x_test.astype(np.float32).reshape([-1, 28, 28, 1]) / 0.5 - 1.
loss_display = Display(layout=(1, 1),
dpi='auto',
show=show_figure,
block_on_end=block_figure_on_end,
monitor=[
{
'metrics': ['d_loss', 'g_loss'],
'type':
'line',
'labels':
["discriminator loss", "generator loss"],
'title':
"Losses",
'xlabel':
"epoch",
'ylabel':
"loss",
},
])
sample_display = Display(layout=(1, 1),
dpi='auto',
figsize=(10, 10),
freq=1,
show=show_figure,
block_on_end=block_figure_on_end,
monitor=[
{
'metrics': ['x_samples'],
'title': "Generated data",
'type': 'img',
'num_samples': 100,
'tile_shape': (10, 10),
},
])
model = CGAN(
model_name="CGAN_MNIST",
num_z=10, # set to 100 for a full run
z_prior=Uniform1D(low=-1.0, high=1.0),
img_size=(28, 28, 1),
batch_size=64, # set to 64 for a full run
num_conv_layers=3, # set to 3 for a full run
num_gen_feature_maps=2, # set to 32 for a full run
num_dis_feature_maps=2, # set to 32 for a full run
metrics=['d_loss', 'g_loss'],
callbacks=[loss_display, sample_display],
num_epochs=1, # set to 100 for a full run
random_state=random_seed(),
verbose=1)
model.fit(x_train, y_train)
def test_wgan_cifar10(show_figure=False, block_figure_on_end=False):
print("========== Test WGAN on CIFAR10 data ==========")
np.random.seed(random_seed())
num_data = 128
(x_train, y_train), (x_test, y_test) = demo.load_cifar10()
x_train = x_train[:num_data].astype(np.float32).reshape([-1, 32, 32, 3]) / 0.5 - 1.
x_test = x_test.astype(np.float32).reshape([-1, 32, 32, 3]) / 0.5 - 1.
root_dir = os.path.join(model_dir(), "male/WGAN/CIFAR10")
loss_display = Display(layout=(1, 1),
dpi='auto',
show=show_figure,
block_on_end=block_figure_on_end,
filepath=[os.path.join(root_dir, "loss/loss_{epoch:04d}.png"),
os.path.join(root_dir, "loss/loss_{epoch:04d}.pdf")],
monitor=[{'metrics': ['d_loss', 'g_loss'],
'type': 'line',
'labels': ["discriminator loss", "generator loss"],
'title': "Losses",
'xlabel': "epoch",
'ylabel': "loss",
},
])
sample_display = Display(layout=(1, 1),
dpi='auto',
figsize=(10, 10),
freq=1,
show=show_figure,
block_on_end=block_figure_on_end,
filepath=os.path.join(root_dir, "samples/samples_{epoch:04d}.png"),
monitor=[{'metrics': ['x_samples'],
'title': "Generated data",
'type': 'img',
'num_samples': 100,
'tile_shape': (10, 10),
},
])
model = WGAN(model_name="WGAN_CIFAR10",
num_z=10, # set to 100 for a full run
z_prior=Uniform1D(low=-1.0, high=1.0),
img_size=(32, 32, 3),
batch_size=16, # set to 64 for a full run
num_conv_layers=3, # set to 3 for a full run
num_gen_feature_maps=4, # set to 32 for a full run
num_dis_feature_maps=4, # set to 32 for a full run
metrics=['d_loss', 'g_loss'],
callbacks=[loss_display, sample_display],
num_epochs=4, # set to 100 for a full run
log_path=os.path.join(root_dir, "logs"),
random_state=random_seed(),
verbose=1)
model.fit(x_train)
def test_wgan_gp_mnist(show_figure=False, block_figure_on_end=False):
print("========== Test WGAN-GP on MNIST data ==========")
np.random.seed(random_seed())
(x_train, y_train), (x_test, y_test) = demo.load_mnist()
x_train = x_train.astype(np.float32).reshape([-1, 28, 28, 1]) / 0.5 - 1.
x_test = x_test.astype(np.float32).reshape([-1, 28, 28, 1]) / 0.5 - 1.
root_dir = os.path.join(model_dir(), "male/WGAN-GP/MNIST")
loss_display = Display(layout=(1, 1),
dpi='auto',
show=show_figure,
block_on_end=block_figure_on_end,
filepath=[os.path.join(root_dir, "loss/loss_{epoch:04d}.png"),
os.path.join(root_dir, "loss/loss_{epoch:04d}.pdf")],
monitor=[{'metrics': ['d_loss', 'g_loss'],
'type': 'line',
'labels': ["discriminator loss", "generator loss"],
'title': "Losses",
'xlabel': "epoch",
'ylabel': "loss",
},
])
sample_display = Display(layout=(1, 1),
dpi='auto',
figsize=(10, 10),
freq=1,
show=show_figure,
block_on_end=block_figure_on_end,
monitor=[{'metrics': ['x_samples'],
'title': "Generated data",
'type': 'img',
'num_samples': 100,
'tile_shape': (10, 10),
},
])
model = WGAN_GP(model_name="WGAN_GP_MNIST_z_uniform",
num_z=10, # set to 100 for a full run
z_prior=Uniform1D(low=-1.0, high=1.0),
img_size=(28, 28, 1),
batch_size=16, # set to 64 for a full run
num_conv_layers=3, # set to 3 for a full run
num_gen_feature_maps=4, # set to 32 for a full run
num_dis_feature_maps=4, # set to 32 for a full run
metrics=['d_loss', 'g_loss'],
callbacks=[loss_display, sample_display],
num_epochs=4, # set to 100 for a full run
# summary_freq=1, # uncomment this for a full run
random_state=random_seed(),
log_path=os.path.join(root_dir, "logs"),
verbose=1)
model.fit(x_train)
model = WGAN_GP(model_name="WGAN_GP_MNIST_z_Gaussian",
num_z=10, # set to 100 for a full run
z_prior=Gaussian1D(mu=0.0, sigma=1.0),
img_size=(28, 28, 1),
batch_size=32, # set to 64 for a full run
num_conv_layers=3, # set to 3 for a full run
num_gen_feature_maps=4, # set to 32 for a full run
num_dis_feature_maps=4, # set to 32 for a full run
metrics=['d_loss', 'g_loss'],
callbacks=[loss_display, sample_display],
num_epochs=4, # set to 100 for a full run
# summary_freq=1, # uncomment this for a full run
random_state=random_seed(),
log_path=os.path.join(root_dir, "logs"),
verbose=1)
model.fit(x_train)
def test_d2gan1d_gaussian1d(show_figure=False, block_figure_on_end=False):
loss_display = Display(layout=(1, 3),
dpi='auto',
title='Loss',
show=show_figure,
block_on_end=block_figure_on_end,
monitor=[
{
'metrics': ['d_loss', 'g_loss'],
'type':
'line',
'labels':
["discriminator loss", "generator loss"],
'title':
"Losses",
'xlabel':
"epoch",
'ylabel':
"loss",
},
{
'metrics': ['loglik'],
'type': 'line',
'labels': ["Log-likelihood"],
'title': "Evaluation",
'xlabel': "epoch",
'ylabel': "loglik",
},
{
'metrics': ['d1x', 'd2x'],
'type': 'line',
'labels': ["d1x_mean", "d2x_mean"],
'title': "Discriminative scores",
'xlabel': "epoch",
'ylabel': "score",
},
])
distribution_display = Display(layout=(1, 1),
dpi='auto',
freq=10,
title='Histogram',
show=show_figure,
block_on_end=block_figure_on_end,
monitor=[
{
'metrics': ['distribution'],
'type': 'hist',
'title': "Histogram of D2GAN1D",
'xlabel': "Data values",
'ylabel': "Probability density",
},
])
avg_distribution_display = Display(layout=(1, 1),
dpi='auto',
freq=10,
title='Average Histogram',
show=show_figure,
block_on_end=block_figure_on_end,
monitor=[
{
'metrics': ['avg_distribution'],
'type': 'hist',
'title':
"Averaged Histogram of D2GAN1D",
'xlabel': "Data values",
'ylabel': "Probability density",
},
])
model = D2GAN1D(
data=Gaussian1D(mu=4.0, sigma=0.5),
generator=Uniform1D(low=-8.0, high=8.0),
alpha=1.0,
beta=1.0,
num_z=10, # increase this to 100 for a full run
num_epochs=4, # increase this to 1000 for a full run
hidden_size=16, # increase this to 128 for a full run
batch_size=16,
minibatch_discriminator=False,
loglik_freq=1,
generator_learning_rate=0.0001,
discriminator_learning_rate=0.0001,
metrics=[
'd1x', 'd2x', 'd1_loss', 'd2_loss', 'd_loss', 'g_loss', 'loglik'
],
callbacks=[
loss_display, distribution_display, avg_distribution_display
],
random_state=random_seed(),
verbose=1)
model.fit()
def test_gan1d_gaussian1d(show_figure=False, block_figure_on_end=False):
print(
"========== Test GAN on 1D data generated from a Gaussian distribution =========="
)
loss_display = Display(layout=(2, 1),
dpi='auto',
title='Loss',
show=show_figure,
block_on_end=block_figure_on_end,
monitor=[
{
'metrics': ['d_loss', 'g_loss'],
'type':
'line',
'labels':
["discriminator loss", "generator loss"],
'title':
"Losses",
'xlabel':
"epoch",
'ylabel':
"loss",
},
{
'metrics': ['loglik'],
'type': 'line',
'labels': ["Log-likelihood"],
'title': "Evaluation",
'xlabel': "epoch",
'ylabel': "loglik",
},
])
distribution_display = Display(
layout=(1, 1),
dpi='auto',
freq=1,
title='Histogram',
show=show_figure,
block_on_end=block_figure_on_end,
# filepath=[os.path.join(model_dir(),
# "GAN1D/samples/hist_{epoch:04d}.png")],
monitor=[
{
'metrics': ['distribution'],
'type': 'hist',
'title': "Histogram of GAN1D",
'xlabel': "Data values",
'ylabel': "Probability density",
},
])
avg_distribution_display = Display(layout=(1, 1),
dpi='auto',
freq=1,
title='Average Histogram',
show=show_figure,
block_on_end=block_figure_on_end,
monitor=[
{
'metrics': ['avg_distribution'],
'type': 'hist',
'title':
"Averaged Histogram of GAN1D",
'xlabel': "Data values",
'ylabel': "Probability density",
},
])
model = GAN1D(
data=Gaussian1D(mu=4.0, sigma=0.5),
generator=Uniform1D(low=-8.0, high=8.0),
num_z=8, # set to 100 for a full run
num_epochs=4, # set to 355 for a full run
hidden_size=4, # set to 128 for a full run
batch_size=8, # set to 128 for a full run
minibatch_discriminator=False,
loglik_freq=1,
generator_learning_rate=0.0001,
discriminator_learning_rate=0.0001,
metrics=['d_loss', 'g_loss', 'loglik'],
callbacks=[
loss_display, distribution_display, avg_distribution_display
],
random_state=random_seed(),
verbose=1)
model.fit()
def test_wgan_save_and_load(show_figure=False, block_figure_on_end=False):
print("========== Test Save and Load functions of WGAN on MNIST data ==========")
np.random.seed(random_seed())
(x_train, y_train), (x_test, y_test) = demo.load_mnist()
num_data = 128
x_train = x_train[:num_data].astype(np.float32).reshape([-1, 28, 28, 1]) / 0.5 - 1.
x_test = x_test.astype(np.float32).reshape([-1, 28, 28, 1]) / 0.5 - 1.
root_dir = os.path.join(model_dir(), "male/WGAN/MNIST")
loss_display = Display(layout=(1, 1),
dpi='auto',
show=show_figure,
block_on_end=block_figure_on_end,
monitor=[{'metrics': ['d_loss', 'g_loss'],
'type': 'line',
'labels': ["discriminator loss", "generator loss"],
'title': "Losses",
'xlabel': "epoch",
'ylabel': "loss",
},
])
sample_display = Display(layout=(1, 1),
dpi='auto',
figsize=(10, 10),
freq=1,
show=show_figure,
block_on_end=block_figure_on_end,
monitor=[{'metrics': ['x_samples'],
'title': "Generated data",
'type': 'img',
'num_samples': 100,
'tile_shape': (10, 10),
},
])
model = WGAN(model_name="WGAN_MNIST_SaveLoad",
num_z=8,
z_prior=Uniform1D(low=-1.0, high=1.0),
img_size=(28, 28, 1),
batch_size=16,
num_conv_layers=3,
num_gen_feature_maps=4,
num_dis_feature_maps=4,
metrics=['d_loss', 'g_loss'],
callbacks=[loss_display, sample_display],
num_epochs=2,
log_path=os.path.join(root_dir, "logs"),
random_state=random_seed(),
verbose=1)
model.fit(x_train)
print("Saving model...")
save_file_path = model.save(os.path.join(root_dir, "checkpoints/ckpt"))
print("Reloading model...")
model1 = TensorFlowModel.load_model(save_file_path)
model1.num_epochs = 4
model1.fit(x_train)
print("Done!")
def test_grn1d_gaussian1d(block_figure_on_end=False):
print(
"========== Test GRN on 1D data generated from a Gaussian distribution =========="
)
loss_display = Display(
layout=(2, 1),
block_on_end=block_figure_on_end,
monitor=[
{
'metrics': ['d_loss', 'g_loss', 'a_loss_1'],
'type':
'line',
'labels':
["discriminator loss", "generator loss", "auxiliary_1 loss"],
'title':
"Losses",
'xlabel':
"epoch",
'ylabel':
"loss",
},
{
'metrics': ['loglik'],
'type': 'line',
'labels': ["Log-likelihood"],
'title': "Evaluation",
'xlabel': "epoch",
'ylabel': "loglik",
},
])
distribution_display = Display(layout=(1, 1),
freq=1,
block_on_end=block_figure_on_end,
monitor=[
{
'metrics': ['distribution'],
'type': 'hist',
'title': "Histograms of GRN1D",
'xlabel': "Data values",
'ylabel': "Probability density",
},
])
avg_distribution_display = Display(layout=(1, 1),
freq=1,
block_on_end=block_figure_on_end,
monitor=[
{
'metrics': ['avg_distribution'],
'type': 'hist',
'title':
"Average Histograms of GRN1D",
'xlabel': "Data values",
'ylabel': "Probability density",
},
])
model = GRN1D(data=Gaussian1D(mu=4.0, sigma=0.5),
generator=Uniform1D(low=-8, high=8.0),
aux_discriminators=[Gaussian1D(mu=0.0, sigma=0.5)],
aux_coeffs=[0.5],
num_epochs=4,
hidden_size=20,
batch_size=12,
aux_batch_size=10,
discriminator_learning_rate=0.001,
generator_learning_rate=0.001,
aux_learning_rate=0.001,
loglik_freq=1,
metrics=['d_loss', 'g_loss', 'a_loss_1', 'loglik'],
callbacks=[
loss_display, distribution_display,
avg_distribution_display
],
random_state=random_seed(),
verbose=1)
model.fit()