def test_rrf_cv_gridsearch():
print(
"========== Tune parameters for RRF including cross-validation =========="
)
np.random.seed(random_seed())
(x_train, y_train), (x_test, y_test) = demo.load_iris()
print("Number of training samples = {}".format(x_train.shape[0]))
print("Number of testing samples = {}".format(x_test.shape[0]))
x = np.vstack([x_train, x_test, x_test])
y = np.concatenate([y_train, y_test, y_test])
params = {'gamma': [0.5, 1.0], 'learning_rate': [0.01, 0.05, 0.1]}
ps = PredefinedSplit(test_fold=[-1] * x_train.shape[0] +
[-1] * x_test.shape[0] + [1] * x_test.shape[0])
early_stopping = EarlyStopping(monitor='val_err', patience=2)
filepath = os.path.join(
model_dir(), "male/RRF/search/mnist_{epoch:04d}_{val_err:.6f}.pkl")
checkpoint = ModelCheckpoint(filepath,
mode='min',
monitor='val_err',
verbose=0,
save_best_only=True)
clf = RRF(model_name="RRF_hinge",
D=100,
lbd=0.01,
gamma=0.125,
mode='batch',
loss='hinge',
num_epochs=10,
learning_rate=0.001,
learning_rate_gamma=0.001,
metrics=['loss', 'err'],
callbacks=[early_stopping, checkpoint],
cv=[-1] * x_train.shape[0] + [0] * x_test.shape[0],
catch_exception=True,
random_state=random_seed())
gs = GridSearchCV(clf, params, cv=ps, n_jobs=-1, refit=False, verbose=True)
gs.fit(x, y)
print("Best error {} @ params {}".format(1 - gs.best_score_,
gs.best_params_))
best_clf = clone(clf).set_params(**gs.best_params_)
best_clf.fit(np.vstack([x_train, x_test]),
np.concatenate([y_train, y_test]))
train_err = 1.0 - best_clf.score(x_train, y_train)
test_err = 1.0 - best_clf.score(x_test, y_test)
print("Training error = %.4f" % train_err)
print("Testing error = %.4f" % test_err)
assert abs(test_err - (1.0 - gs.best_score_)) < 1e-4
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_image_saver_callback():
np.random.seed(random_seed())
(x_train, y_train), (_, _) = demo.load_mnist()
(cifar10_train, _), (_, _) = demo.load_cifar10()
imgsaver1 = ImageSaver(freq=1,
filepath=os.path.join(
model_dir(), "male/callbacks/imagesaver/"
"mnist/mnist_{epoch:04d}.png"),
monitor={
'metrics': 'x_data',
'img_size': (28, 28, 1),
'tile_shape': (10, 10),
'images': x_train[:100].reshape([-1, 28, 28, 1])
})
imgsaver2 = ImageSaver(freq=1,
filepath=os.path.join(
model_dir(), "male/callbacks/imagesaver/"
"cifar10/cifar10_{epoch:04d}.png"),
monitor={
'metrics':
'x_data',
'img_size': (32, 32, 3),
'tile_shape': (10, 10),
'images':
cifar10_train[:100].reshape([-1, 32, 32, 3])
})
optz = SGD(learning_rate=0.001)
clf = GLM(model_name="imagesaver_callback",
link='softmax',
loss='softmax',
optimizer=optz,
num_epochs=4,
batch_size=100,
task='classification',
callbacks=[imgsaver1, imgsaver2],
random_state=random_seed(),
verbose=1)
clf.fit(x_train, y_train)
def test_kmm_cv():
print("========== Test cross-validation for KMM ==========")
np.random.seed(random_seed())
(x_train, y_train), (x_test, y_test) = demo.load_iris()
print("Number of training samples = {}".format(x_train.shape[0]))
print("Number of testing samples = {}".format(x_test.shape[0]))
x = np.vstack([x_train, x_test])
y = np.concatenate([y_train, y_test])
early_stopping = EarlyStopping(monitor='val_err', patience=2, verbose=1)
filepath = os.path.join(model_dir(),
"male/KMM/iris_{epoch:04d}_{val_err:.6f}.pkl")
checkpoint = ModelCheckpoint(filepath,
mode='min',
monitor='val_err',
verbose=0,
save_best_only=True)
clf = KMM(model_name="KMM_hinge",
D=20,
lbd=0.0,
gamma=0.1,
mode='batch',
loss='hinge',
num_kernels=3,
batch_size=100,
temperature=1.0,
num_epochs=10,
num_nested_epochs=1,
learning_rate=0.1,
learning_rate_mu=0.0,
learning_rate_gamma=0.1,
learning_rate_alpha=0.1,
metrics=['loss', 'err'],
callbacks=[early_stopping, checkpoint],
cv=[-1] * x_train.shape[0] + [0] * x_test.shape[0],
random_state=random_seed(),
verbose=1)
clf.fit(x, y)
train_err = 1.0 - clf.score(x_train, y_train)
test_err = 1.0 - clf.score(x_test, y_test)
print("Training error = %.4f" % train_err)
print("Testing error = %.4f" % test_err)
def test_checkpoint():
np.random.seed(random_seed())
(x_train, y_train), (x_test, y_test) = demo.load_iris()
print("Number of training samples = {}".format(x_train.shape[0]))
print("Number of testing samples = {}".format(x_test.shape[0]))
x = np.vstack([x_train, x_test])
y = np.concatenate([y_train, y_test])
filepath = os.path.join(
model_dir(), "male/glm/checkpoint_{epoch:04d}_{val_loss:.6f}.pkl")
checkpoint = ModelCheckpoint(filepath,
mode='min',
monitor='val_loss',
verbose=1,
save_best_only=True)
optz = SGD(learning_rate=0.01)
clf = GLM(model_name="checkpoint_callback",
link='softmax',
loss='softmax',
optimizer=optz,
num_epochs=5,
batch_size=10,
task='classification',
metrics=['loss', 'err'],
callbacks=[checkpoint],
cv=[-1] * x_train.shape[0] + [0] * x_test.shape[0],
random_state=random_seed(),
verbose=1)
clf.fit(x, y)
train_err = 1.0 - clf.score(x_train, y_train)
test_err = 1.0 - clf.score(x_test, y_test)
print("Training error = %.4f" % train_err)
print("Testing error = %.4f" % test_err)
model_filepath = filepath.format(epoch=5, val_loss=0.968786)
print("Load model at checkpoint: ", model_filepath, ", and predict:")
clf1 = Model.load_model(model_filepath)
train_err = 1.0 - clf1.score(x_train, y_train)
test_err = 1.0 - clf1.score(x_test, y_test)
print("Training error = %.4f" % train_err)
print("Testing error = %.4f" % test_err)
def test_wgan_gp_cifar10_fid(show_figure=False, block_figure_on_end=False):
print("========== Test WGAN-GP with FID 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-GP/CIFAR10")
checkpoints = ModelCheckpoint(
os.path.join(root_dir, "checkpoints/{epoch:04d}_{FID:.6f}.ckpt"),
mode='min',
monitor='FID',
verbose=1,
save_best_only=True)
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",
},
])
fid_display = Display(layout=(1, 1),
dpi='auto',
show=show_figure,
block_on_end=block_figure_on_end,
filepath=[os.path.join(root_dir, "FID/"
"FID_{epoch:04d}.png"),
os.path.join(root_dir, "FID/"
"FID_{epoch:04d}.pdf")],
monitor=[{'metrics': ['FID'],
'type': 'line',
'labels': ["FID"],
'title': "Scores",
'xlabel': "epoch",
'ylabel': "score",
},
],
)
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_GP(model_name="WGAN_GP_CIFAR10",
num_z=10, # set to 100 for a full run
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', 'FID', 'FID_100points'],
callbacks=[loss_display, fid_display,
sample_display, checkpoints],
num_epochs=4, # set to 100 for a full run
inception_metrics=[FID(data="cifar10"),
FID(name="FID_100points", data=x_train[:100])],
inception_metrics_freq=1,
# summary_freq=1, # uncomment this 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_pytorch_mlp_v1(show=False, block_figure_on_end=False):
print("========== Test PytorchMLPv1 ==========")
np.random.seed(random_seed())
(x_train, y_train), (x_test, y_test) = demo.load_mnist()
idx_train = np.random.permutation(x_train.shape[0])
x_train = x_train[idx_train].astype(np.float32)
y_train = y_train[idx_train].astype(np.uint8)
print("Number of training samples = {}".format(x_train.shape[0]))
idx_test = np.random.permutation(x_test.shape[0])
x_test = x_test[idx_test].astype(np.float32)
y_test = y_test[idx_test].astype(np.uint8)
print("Number of testing samples = {}".format(x_test.shape[0]))
x = np.vstack([x_train, x_test])
y = np.concatenate([y_train, y_test])
err_display = Display(title="Error curves",
dpi='auto',
layout=(1, 1),
freq=1,
show=show,
block_on_end=block_figure_on_end,
monitor=[{
'metrics': ['err', 'val_err'],
'type': 'line',
'title': "Learning errors",
'xlabel': "epoch",
'ylabel': "error",
}])
loss_display = Display(
title="Learning curves",
dpi='auto',
layout=(3, 1),
freq=1,
show=show,
block_on_end=block_figure_on_end,
filepath=[
os.path.join(model_dir(), "male/PyTorchMLP/"
"loss/loss_{epoch:04d}.png"),
os.path.join(model_dir(), "male/PyTorchMLP/"
"loss/loss_{epoch:04d}.pdf")
],
monitor=[
{
'metrics': ['loss', 'val_loss'],
'type': 'line',
'labels': ["training loss", "validation loss"],
'title': "Learning losses",
'xlabel': "epoch",
'xlabel_params': {
'fontsize': 50
},
'ylabel': "loss",
},
{
'metrics': ['err', 'val_err'],
'type': 'line',
'title': "Learning errors",
'xlabel': "epoch",
'ylabel': "error",
},
{
'metrics': ['err'],
'type': 'line',
'labels': ["training error"],
'title': "Learning errors",
'xlabel': "epoch",
'ylabel': "error",
},
])
weight_display = Display(title="Filters",
dpi='auto',
layout=(1, 1),
figsize=(6, 15),
freq=1,
show=show,
block_on_end=block_figure_on_end,
filepath=os.path.join(
model_dir(), "male/PyTorchMLP/"
"weights/weights_{epoch:04d}.png"),
monitor=[
{
'metrics': ['weights'],
'title': "Learned weights",
'type': 'img',
'tile_shape': (5, 2),
},
])
clf = PyTorchMLP(model_name='PyTorchMLP',
arch='MLPv1',
num_epochs=4,
batch_size=100,
metrics=['loss', 'err'],
callbacks=[loss_display, err_display, weight_display],
cv=[-1] * x_train.shape[0] + [0] * x_test.shape[0],
random_state=random_seed(),
verbose=1)
clf.fit(x, y)
print("Training error = %.4f" % (1.0 - clf.score(x_train, y_train)))
print("Testing error = %.4f" % (1.0 - clf.score(x_test, y_test)))
def test_glm_save_load(show=False, block_figure_on_end=False):
print("========== Test Save and Load functions for GLM ==========")
np.random.seed(random_seed())
(x_train, y_train), (x_test, y_test) = demo.load_iris()
print("Number of training samples = {}".format(x_train.shape[0]))
print("Number of testing samples = {}".format(x_test.shape[0]))
x = np.vstack([x_train, x_test])
y = np.concatenate([y_train, y_test])
early_stopping = EarlyStopping(monitor='val_err', patience=5, verbose=1)
filepath = os.path.join(model_dir(),
"male/GLM/iris_{epoch:04d}_{val_err:.6f}.pkl")
checkpoint = ModelCheckpoint(filepath,
mode='min',
monitor='val_err',
verbose=0,
save_best_only=True)
loss_display = Display(title="Learning curves",
dpi='auto',
layout=(3, 1),
freq=1,
show=show,
block_on_end=block_figure_on_end,
monitor=[
{
'metrics': ['loss', 'val_loss'],
'type': 'line',
'labels':
["training loss", "validation loss"],
'title': "Learning losses",
'xlabel': "epoch",
'ylabel': "loss",
},
{
'metrics': ['err', 'val_err'],
'type': 'line',
'title': "Learning errors",
'xlabel': "epoch",
'ylabel': "error",
},
{
'metrics': ['err'],
'type': 'line',
'labels': ["training error"],
'title': "Learning errors",
'xlabel': "epoch",
'ylabel': "error",
},
])
weight_display = Display(title="Filters",
dpi='auto',
layout=(1, 1),
figsize=(6, 15),
freq=1,
show=show,
block_on_end=block_figure_on_end,
monitor=[
{
'metrics': ['weights'],
'title': "Learned weights",
'type': 'img',
'disp_dim': (2, 2),
'tile_shape': (3, 1),
},
])
clf = GLM(
model_name="GLM_softmax_cv",
link='softmax',
loss='softmax',
optimizer='sgd',
num_epochs=4,
batch_size=10,
task='classification',
metrics=['loss', 'err'],
callbacks=[early_stopping, checkpoint, loss_display, weight_display],
cv=[-1] * x_train.shape[0] + [0] * x_test.shape[0],
random_state=random_seed(),
verbose=1)
clf.fit(x, y)
train_err = 1.0 - clf.score(x_train, y_train)
test_err = 1.0 - clf.score(x_test, y_test)
print("Training error = %.4f" % train_err)
print("Testing error = %.4f" % test_err)
save_file_path = os.path.join(model_dir(), "male/GLM/saved_model.pkl")
clf.save(file_path=save_file_path)
clf1 = Model.load_model(save_file_path)
clf1.num_epochs = 10
clf1.fit(x, y)
train_err = 1.0 - clf1.score(x_train, y_train)
test_err = 1.0 - clf1.score(x_test, y_test)
print("Training error = %.4f" % train_err)
print("Testing error = %.4f" % test_err)
def test_gank_image_saver():
print("========== Test GANK-Logit with Image Saver ==========")
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.
imgsaver = ImageSaver(freq=1,
filepath=os.path.join(
model_dir(), "male/GANK/imagesaver/"
"mnist/mnist_{epoch:04d}.png"),
monitor={
'metrics': 'x_samples',
'num_samples': 100,
'tile_shape': (10, 10),
})
model = GANK(
model_name="GANK_MNIST",
num_random_features=50, # set to 1000 for a full run
gamma_init=0.01,
loss='logit',
num_z=10, # set to 100 for a full run
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=[imgsaver],
num_epochs=4, # set to 100 for a full run
random_state=random_seed(),
verbose=1)
model.fit(x_train)
np.random.seed(random_seed())
(x_train, y_train), (x_test, y_test) = demo.load_cifar10()
x_train = x_train.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.
imgsaver = ImageSaver(freq=1,
filepath=os.path.join(
model_dir(), "male/GANK/imagesaver/"
"cifar10/cifar10_{epoch:04d}.png"),
monitor={
'metrics': 'x_samples',
'num_samples': 100,
'tile_shape': (10, 10),
})
model = GANK(
model_name="GANK_CIFAR10",
num_random_features=50, # set 1000 for a full run
gamma_init=0.01,
loss='logit',
num_z=10, # set to 100 for a full run
img_size=(32, 32, 3),
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=[imgsaver],
num_epochs=4, # set to 500 for a full run
random_state=random_seed(),
verbose=1)
model.fit(x_train)
def test_gank_logit_cifar10_inception_score(show_figure=False,
block_figure_on_end=False):
print(
"========== Test GANK-Logit with Inception Score on CIFAR10 data =========="
)
np.random.seed(random_seed())
(x_train, y_train), (x_test, y_test) = demo.load_cifar10()
x_train = x_train.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.
filepath = os.path.join(
model_dir(), "male/GANK/Logit/cifar10/checkpoints/"
"{epoch:04d}_{inception_score:.6f}.ckpt")
checkpoint = ModelCheckpoint(filepath,
mode='max',
monitor='inception_score',
verbose=0,
save_best_only=True)
loss_display = Display(layout=(1, 1),
dpi='auto',
show=show_figure,
block_on_end=block_figure_on_end,
filepath=[
os.path.join(
model_dir(), "male/GANK/Logit/cifar10/"
"loss/loss_{epoch:04d}.png"),
os.path.join(
model_dir(), "male/GANK/Logit/cifar10/"
"loss/loss_{epoch:04d}.pdf")
],
monitor=[
{
'metrics': ['d_loss', 'g_loss'],
'type':
'line',
'labels':
["discriminator loss", "generator loss"],
'title':
"Losses",
'xlabel':
"epoch",
'ylabel':
"loss",
},
])
inception_score_display = Display(
layout=(1, 1),
dpi='auto',
show=show_figure,
block_on_end=block_figure_on_end,
filepath=[
os.path.join(
model_dir(), "male/GANK/Logit/cifar10/inception_score/"
"inception_score_{epoch:04d}.png"),
os.path.join(
model_dir(), "male/GANK/Logit/cifar10/inception_score/"
"inception_score_{epoch:04d}.pdf")
],
monitor=[
{
'metrics': ['inception_score'],
'type': 'line',
'labels': ["Inception Score"],
'title': "Scores",
'xlabel': "epoch",
'ylabel': "score",
},
],
)
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(
model_dir(),
"male/GANK/Logit/cifar10/samples/"
"samples_{epoch:04d}.png"),
monitor=[
{
'metrics': ['x_samples'],
'title': "Generated data",
'type': 'img',
'num_samples': 100,
'tile_shape': (10, 10),
},
])
model = GANK(
model_name="GANK-Logit_CIFAR10",
num_random_features=50, # set 1000 for a full run
gamma_init=0.01,
loss='logit',
num_z=10, # set to 100 for a full run
img_size=(32, 32, 3),
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', 'inception_score'],
# callbacks=[loss_display, inception_score_display, sample_display, checkpoint],
callbacks=[checkpoint],
num_epochs=4, # set to 500 for a full run
inception_score_freq=1,
random_state=random_seed(),
verbose=1)
model.fit(x_train)
def test_srbm_regression(show_figure=False, block_figure_on_end=False):
print("========== Test Supervised RBM for Regression ==========")
from sklearn.metrics import mean_squared_error
from sklearn.linear_model import LinearRegression
np.random.seed(random_seed())
(x_train, y_train), (x_test, y_test) = demo.load_mnist()
x = np.vstack([x_train, x_test])
y = np.concatenate([y_train, y_test])
learning_display = Display(
title="Learning curves",
dpi='auto',
layout=(3, 1),
freq=1,
show=show_figure,
block_on_end=block_figure_on_end,
monitor=[
{
'metrics': ['recon_err', 'val_recon_err'],
'type': 'line',
'labels': ["training recon error", "validation recon error"],
'title': "Reconstruction Errors",
'xlabel': "epoch",
'ylabel': "error",
},
{
'metrics': ['loss', 'val_loss'],
'type': 'line',
'labels': ["training loss", "validation loss"],
'title': "Learning Losses",
'xlabel': "epoch",
'ylabel': "loss",
},
{
'metrics': ['err', 'val_err'],
'type': 'line',
'labels': ["training error", "validation error"],
'title': "Prediction Errors",
'xlabel': "epoch",
'ylabel': "error",
},
# {'metrics': ['loglik_csl', 'val_loglik_csl'],
# 'type': 'line',
# 'labels': ["training loglik (CSL)", "validation loglik (CSL)"],
# 'title': "Loglikelihoods using CSL",
# 'xlabel': "epoch",
# 'ylabel': "loglik",
# },
])
filter_display = Display(title="Receptive Fields",
dpi='auto',
layout=(1, 1),
figsize=(8, 8),
freq=1,
show=show_figure,
block_on_end=block_figure_on_end,
monitor=[
{
'metrics': ['filters'],
'title': "Receptive Fields",
'type': 'img',
'num_filters': 15,
'disp_dim': (28, 28),
'tile_shape': (3, 5),
},
])
hidden_display = Display(title="Hidden Activations",
dpi='auto',
layout=(1, 1),
figsize=(8, 8),
freq=1,
show=show_figure,
block_on_end=block_figure_on_end,
monitor=[
{
'metrics': ['hidden_activations'],
'title': "Hidden Activations",
'type': 'img',
'data': x_train[:100],
},
])
early_stopping = EarlyStopping(monitor='val_loss', patience=2, verbose=1)
filepath = os.path.join(model_dir(),
"male/sRBM/mnist_{epoch:04d}_{val_loss:.6f}.pkl")
checkpoint = ModelCheckpoint(filepath,
mode='min',
monitor='val_loss',
verbose=0,
save_best_only=True)
model = SupervisedRBM(task='regression',
num_hidden=15,
num_visible=784,
batch_size=100,
num_epochs=4,
w_init=0.01,
learning_rate=0.01,
momentum_method='sudden',
weight_cost=0.0,
inference_engine='variational_inference',
approx_method='first_order',
metrics=['recon_err', 'loss', 'err'],
callbacks=[
filter_display, learning_display, hidden_display,
early_stopping, checkpoint
],
cv=[-1] * x_train.shape[0] + [0] * x_test.shape[0],
random_state=random_seed(),
verbose=1)
model.fit(x, y)
print("Test reconstruction error = %.4f" %
model.get_reconstruction_error(x_test).mean())
print("Test loss = %.4f" % model.get_loss(x_test, y_test))
print("=========== Predicted by sRBM ============")
print("Train MSE = {0:>1.4f}\tTest MSE = {1:>1.4f}".format(
-model.score(x_train, y_train), -model.score(x_test, y_test)))
# fit a Linear Regressor
lr = LinearRegression()
lr.fit(x_train, y_train)
print("=========== Predicted by Linear Regressor ============")
print("Train MSE = {0:>1.4f}\tTest MSE = {1:>1.4f}".format(
mean_squared_error(y_train, lr.predict(x_train)),
mean_squared_error(y_test, lr.predict(x_test))))
def test_display_callbacks(show=False, block_figure_on_end=False):
np.random.seed(random_seed())
(x_train, y_train), (x_test, y_test) = demo.load_mnist()
idx_train = np.random.permutation(x_train.shape[0])
x_train = x_train[idx_train]
y_train = y_train[idx_train]
print("Number of training samples = {}".format(x_train.shape[0]))
idx_test = np.random.permutation(x_test.shape[0])
x_test = x_test[idx_test]
y_test = y_test[idx_test]
print("Number of testing samples = {}".format(x_test.shape[0]))
x = np.vstack([x_train, x_test])
y = np.concatenate([y_train, y_test])
err_display = Display(title="Error curves",
dpi='auto',
layout=(1, 1),
freq=1,
show=show,
block_on_end=block_figure_on_end,
monitor=[{
'metrics': ['err', 'val_err'],
'type': 'line',
'title': "Learning errors",
'xlabel': "epoch",
'ylabel': "error",
}])
loss_display = Display(
title="Learning curves",
dpi='auto',
layout=(3, 1),
freq=1,
show=show,
block_on_end=block_figure_on_end,
filepath=[
os.path.join(model_dir(), "male/callbacks/"
"display/loss/loss_{epoch:04d}.png"),
os.path.join(model_dir(), "male/callbacks/"
"display/loss/loss_{epoch:04d}.pdf")
],
monitor=[
{
'metrics': ['loss', 'val_loss'],
'type': 'line',
'labels': ["training loss", "validation loss"],
'title': "Learning losses",
'xlabel': "epoch",
'xlabel_params': {
'fontsize': 50
},
'ylabel': "loss",
},
{
'metrics': ['err', 'val_err'],
'type': 'line',
'title': "Learning errors",
'xlabel': "epoch",
'ylabel': "error",
},
{
'metrics': ['err'],
'type': 'line',
'labels': ["training error"],
'title': "Learning errors",
'xlabel': "epoch",
'ylabel': "error",
},
])
weight_display = Display(title="Filters",
dpi='auto',
layout=(1, 1),
figsize=(6, 15),
freq=1,
show=show,
block_on_end=block_figure_on_end,
filepath=os.path.join(
model_dir(), "male/callbacks/display/"
"weights/weights_{epoch:04d}.png"),
monitor=[
{
'metrics': ['weights'],
'title': "Learned weights",
'type': 'img',
'tile_shape': (5, 2),
},
])
optz = SGD(learning_rate=0.001)
clf = GLM(model_name="display_callbacks",
link='softmax',
loss='softmax',
optimizer=optz,
num_epochs=20,
batch_size=100,
task='classification',
metrics=['loss', 'err'],
callbacks=[loss_display, weight_display, err_display],
cv=[-1] * x_train.shape[0] + [0] * x_test.shape[0],
random_state=random_seed(),
verbose=1)
clf.fit(x, y)
print("Training error = %.4f" % (1.0 - clf.score(x_train, y_train)))
print("Testing error = %.4f" % (1.0 - clf.score(x_test, y_test)))
def test_dcgan_image_saver():
print("========== Test DCGAN with Image Saver ==========")
np.random.seed(random_seed())
num_data = 128
(x_train, y_train), (x_test, y_test) = demo.load_mnist()
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/DCGAN/imagesaver/mnist")
imgsaver = ImageSaver(freq=1,
filepath=os.path.join(root_dir,
"mnist_{epoch:04d}.png"),
monitor={
'metrics': 'x_samples',
'num_samples': 100,
'tile_shape': (10, 10),
})
model = DCGAN(
model_name="DCGAN_MNIST",
num_z=10, # set to 100 for a full run
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=[imgsaver],
num_epochs=4, # set to 100 for a full run
random_state=random_seed(),
log_path=os.path.join(root_dir, "logs"),
verbose=1)
model.fit(x_train)
np.random.seed(random_seed())
(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/DCGAN/imagesaver/cifar10")
imgsaver = ImageSaver(freq=1,
filepath=os.path.join(root_dir,
"cifar10_{epoch:04d}.png"),
monitor={
'metrics': 'x_samples',
'num_samples': 100,
'tile_shape': (10, 10),
})
model = DCGAN(
model_name="DCGAN_CIFAR10",
num_z=10, # set to 100 for a full run
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=[imgsaver],
num_epochs=4, # set to 100 for a full run
random_state=random_seed(),
log_path=os.path.join(root_dir, "logs"),
verbose=1)
model.fit(x_train)
def test_ssrbm_classification(show_figure=False, block_figure_on_end=False):
print("========== Test Semi-Supervised RBM for Classification ==========")
num_labeled_data = 1000
from sklearn.metrics import accuracy_score
from sklearn.neighbors import KNeighborsClassifier
np.random.seed(random_seed())
(x_train, y_train), (x_test, y_test) = demo.load_mnist()
# remove some labels
idx_train, idx_test = next(
iter(
StratifiedShuffleSplit(n_splits=1,
test_size=num_labeled_data,
random_state=random_seed()).split(
x_train, y_train)))
y_train[idx_train] = 10**8
x = np.vstack([x_train, x_test])
y = np.concatenate([y_train, y_test])
learning_display = Display(
title="Learning curves",
dpi='auto',
layout=(3, 1),
freq=1,
show=show_figure,
block_on_end=block_figure_on_end,
monitor=[
{
'metrics': ['recon_err', 'val_recon_err'],
'type': 'line',
'labels': ["training recon error", "validation recon error"],
'title': "Reconstruction Errors",
'xlabel': "epoch",
'ylabel': "error",
},
{
'metrics': ['loss', 'val_loss'],
'type': 'line',
'labels': ["training loss", "validation loss"],
'title': "Learning Losses",
'xlabel': "epoch",
'ylabel': "loss",
},
{
'metrics': ['err', 'val_err'],
'type': 'line',
'labels': ["training error", "validation error"],
'title': "Prediction Errors",
'xlabel': "epoch",
'ylabel': "error",
},
# {'metrics': ['loglik_csl', 'val_loglik_csl'],
# 'type': 'line',
# 'labels': ["training loglik (CSL)", "validation loglik (CSL)"],
# 'title': "Loglikelihoods using CSL",
# 'xlabel': "epoch",
# 'ylabel': "loglik",
# },
])
filter_display = Display(title="Receptive Fields",
dpi='auto',
layout=(1, 1),
figsize=(8, 8),
freq=1,
show=show_figure,
block_on_end=block_figure_on_end,
monitor=[
{
'metrics': ['filters'],
'title': "Receptive Fields",
'type': 'img',
'num_filters': 15,
'disp_dim': (28, 28),
'tile_shape': (3, 5),
},
])
hidden_display = Display(title="Hidden Activations",
dpi='auto',
layout=(1, 1),
figsize=(8, 8),
freq=1,
show=show_figure,
block_on_end=block_figure_on_end,
monitor=[
{
'metrics': ['hidden_activations'],
'title': "Hidden Activations",
'type': 'img',
'data': x_train[:100],
},
])
early_stopping = EarlyStopping(monitor='val_loss', patience=2, verbose=1)
filepath = os.path.join(model_dir(),
"male/ssRBM/mnist_{epoch:04d}_{val_loss:.6f}.pkl")
checkpoint = ModelCheckpoint(filepath,
mode='min',
monitor='val_loss',
verbose=0,
save_best_only=True)
model = SemiSupervisedRBM(num_hidden=15,
num_visible=784,
batch_size=100,
num_epochs=4,
learning_rate=0.1,
w_init=0.1,
momentum_method='sudden',
weight_cost=0.0,
inference_engine='variational_inference',
approx_method='first_order',
metrics=['recon_err', 'loss', 'err'],
callbacks=[
filter_display, learning_display,
hidden_display, early_stopping, checkpoint
],
cv=[-1] * x_train.shape[0] +
[0] * x_test.shape[0],
random_state=random_seed(),
verbose=1)
model.fit(x, y)
print("Test reconstruction error = %.4f" %
model.get_reconstruction_error(x_test).mean())
print("=========== Predicted by Semi-Supervised RBM ============")
print("Train accuracy = {0:>1.4f}\tTest accuracy = {1:>1.4f}".format(
accuracy_score(y_train, model.predict(x_train)),
accuracy_score(y_test, model.predict(x_test))))
x_train1 = model.transform(x_train)
x_test1 = model.transform(x_test)
clf = KNeighborsClassifier(n_neighbors=4)
clf.fit(x_train1, y_train)
print("=========== Predicted by kNN ============")
print("Train accuracy = {0:>1.4f}\tTest accuracy = {1:>1.4f}".format(
accuracy_score(y_train, clf.predict(x_train1)),
accuracy_score(y_test, clf.predict(x_test1))))
def test_mggan2d_gmm2d(show_figure=False, block_figure_on_end=False):
loss_display = Display(
layout=(4, 1),
dpi='auto',
title='Loss',
freq=1, # set to 1000 for a full run
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': ['d_bin_loss', 'd_mul_loss'],
'type': 'line',
'labels': ["d_bin_loss", "d_mul_loss"],
'title': "Discriminator Losses",
'xlabel': "epoch",
'ylabel': "loss",
},
{
'metrics': ['g_bin_loss', 'g_mul_loss'],
'type': 'line',
'labels': ["g_bin_loss", "g_mul_loss"],
'title': "Generator Losses",
'xlabel': "epoch",
'ylabel': "loss",
},
],
)
scatter_display = Display(
layout=(1, 1),
figsize=(12, 12),
dpi='auto',
filepath=[
os.path.join(model_dir(), "male/MGGAN2D/samples/"
"scatter_{epoch:04d}.png"),
os.path.join(model_dir(), "male/MGGAN2D/samples/"
"scatter_{epoch:04d}.pdf")
],
freq=1, # set to 1000 for a full run
title='Scatter',
show=show_figure,
block_on_end=block_figure_on_end,
monitor=[
{
'metrics': ['scatter'],
'type': 'scatter',
'title': "Data scatter generated by MGGAN2D",
'num_samples': 512,
'xlim': (-3, 3),
'ylim': (-3, 3),
},
])
num_mixtures = 8
radius = 2.0
std = 0.02
import numpy as np
thetas = np.linspace(0, 2 * np.pi, num_mixtures + 1)[:num_mixtures]
xs, ys = radius * np.sin(thetas), radius * np.cos(thetas)
model = MGGAN2D(
data=GMM(mix_coeffs=[1 / num_mixtures] * num_mixtures,
mean=list(zip(xs, ys)),
cov=[[std, std]] * num_mixtures),
beta=0.05, # weight for g_bin_loss
num_z=4, # set to 128 for a full run
generator=Gaussian1D(mu=0.0, sigma=1.0),
num_gens=4,
d_batch_size=4, # set to 512 for full run
g_batch_size=4, # set to 128 for full run
num_epochs=4, # set to 25000 for a full run
d_hidden_size=4, # set to 128 for a full run
g_hidden_size=4, # set to 128 for a full run
loglik_freq=1,
generator_learning_rate=0.0002,
discriminator_learning_rate=0.0002,
summary_freq=250,
metrics=[
'd_bin_loss', 'd_mul_loss', 'd_loss', 'g_bin_loss', 'g_mul_loss',
'g_loss', 'loglik'
],
callbacks=[loss_display, scatter_display],
random_state=random_seed(),
verbose=1)
model.fit()
def test_wgan_gp_resnet_cifar10(show_figure=False, block_figure_on_end=False):
print("========== Test WGAN-GP-ResNet 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-GP-ResNet/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,
monitor=[
{
'metrics': ['x_samples'],
'title': "Generated data",
'type': 'img',
'num_samples': 100,
'tile_shape': (10, 10),
},
])
model = WGAN_GP_ResNet(
model_name="WGAN_GP_ResNet_CIFAR10",
num_z=10, # set to 100 for a full run
img_size=(32, 32, 3),
batch_size=16, # set to 64 for a full run
g_blocks=('up', 'up', 'up'),
d_blocks=('down', 'down', None, None),
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=2, # set to 100 for a full run
random_state=random_seed(),
log_path=os.path.join(root_dir, "logs"),
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_wgan_gp_resnet_cifar10_inception_metric(show_figure=False,
block_figure_on_end=False):
print(
"========== Test WGAN-GP-ResNet with Inception Score and FID 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.
# uncomment for full run
'''
import pickle
from male.configs import data_dir
tmp = pickle.load(open(os.path.join(data_dir(), "cifar10/cifar10_train.pkl"), "rb"))
x_train = tmp['data'].astype(np.float32).reshape(
[-1, 32, 32, 3]) / 127.5 - 1.
'''
root_dir = os.path.join(model_dir(), "male/WGAN-GP-ResNet/CIFAR10")
checkpoints_is = ModelCheckpoint(os.path.join(
root_dir, "checkpoints_is/the_best_is.ckpt"),
mode='max',
monitor='inception_score',
verbose=1,
save_best_only=True)
checkpoints_fid = ModelCheckpoint(os.path.join(
root_dir, "checkpoints_fid/the_best_fid.ckpt"),
mode='min',
monitor='FID',
verbose=1,
save_best_only=True)
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",
},
])
inception_score_display = Display(
layout=(1, 1),
dpi='auto',
show=show_figure,
block_on_end=block_figure_on_end,
filepath=[
os.path.join(root_dir, "inception_score/"
"inception_score_{epoch:04d}.png"),
os.path.join(root_dir, "inception_score/"
"inception_score_{epoch:04d}.pdf")
],
monitor=[
{
'metrics': ['inception_score'],
'type': 'line',
'labels': ["Inception Score"],
'title': "Scores",
'xlabel': "epoch",
'ylabel': "score",
},
],
)
fid_display = Display(
layout=(1, 1),
dpi='auto',
show=show_figure,
block_on_end=block_figure_on_end,
filepath=[
os.path.join(root_dir, "FID/"
"FID_{epoch:04d}.png"),
os.path.join(root_dir, "FID/"
"FID_{epoch:04d}.pdf")
],
monitor=[
{
'metrics': ['FID'],
'type': 'line',
'labels': ["FID"],
'title': "Scores",
'xlabel': "epoch",
'ylabel': "score",
},
],
)
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_GP_ResNet(
model_name="WGAN_GP_ResNet_CIFAR10",
num_z=8, # set to 128 for a full run
img_size=(32, 32, 3),
batch_size=64, # set to 64 for a full run
g_blocks=('up', 'up', 'up'),
d_blocks=('down', 'down', None, None),
num_gen_feature_maps=8, # set to 128 for a full run
num_dis_feature_maps=8, # set to 128 for a full run
metrics=[
'd_loss', 'g_loss', 'inception_score', 'inception_score_std', 'FID'
],
callbacks=[
loss_display, inception_score_display, fid_display, sample_display,
checkpoints_is, checkpoints_fid
],
num_epochs=2, # set to 500 for a full run
inception_metrics=[InceptionScore(),
FID(data='cifar10')],
inception_metrics_freq=1,
num_inception_samples=100, # set to 50000 for a full run
# summary_freq=1, # uncomment this for a full run
log_path=os.path.join(root_dir, 'logs'),
random_state=random_seed(),
verbose=1)
model.fit(x_train)
filepath = os.path.join(root_dir, 'checkpoints_fid/the_best_fid.ckpt')
print('Reloading the latest model at: {}'.format(filepath))
model1 = TensorFlowModel.load_model(filepath)
model1.inception_metrics = InceptionMetricList(
[InceptionScore(), FID(data='cifar10')])
model1.num_epochs = 4
model1.fit(x_train)
print('Done!')
def test_dcgan_save_and_load(show_figure=False, block_figure_on_end=False):
print(
"========== Test Save and Load functions of DCGAN 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/DCGAN/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 = DCGAN(
model_name="DCGAN_MNIST_SaveLoad",
num_z=10, # set to 100 for a full run
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=2, # 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)
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_kmm_syn2d(show=False, block_figure_on_end=False):
print("========== Test KMM on 2D data ==========")
np.random.seed(random_seed())
(x_train, y_train), (x_test, y_test) = demo.load_synthetic_2d()
idx_train, idx_test = next(
iter(
StratifiedShuffleSplit(n_splits=1,
test_size=40,
random_state=random_seed()).split(
x_train, y_train)))
x0 = x_train[idx_train]
y0 = y_train[idx_train]
x1 = x_train[idx_test]
y1 = y_train[idx_test]
x = np.vstack([x0, x1])
y = np.concatenate([y0, y1])
early_stopping = EarlyStopping(monitor='val_loss', patience=2, verbose=1)
filepath = os.path.join(
model_dir(), "male/KMM/syn2d_data_{epoch:04d}_{val_err:.6f}.pkl")
checkpoint = ModelCheckpoint(filepath,
mode='min',
monitor='val_err',
verbose=0,
save_best_only=True)
display = Display(layout=(3, 1),
dpi='auto',
show=show,
block_on_end=block_figure_on_end,
monitor=[
{
'metrics': ['loss', 'val_loss'],
'type': 'line',
'labels': ["training loss", "validation loss"],
'title': "Learning losses",
'xlabel': "epoch",
'ylabel': "loss",
},
{
'metrics': ['err', 'val_err'],
'type': 'line',
'title': "Learning errors",
'xlabel': "epoch",
'ylabel': "error",
},
{
'metrics': ['err'],
'type': 'line',
'labels': ["training error"],
'title': "Learning errors",
'xlabel': "epoch",
'ylabel': "error",
},
])
clf = KMM(model_name="KMM_hinge",
D=10,
lbd=0.0,
gamma=0.5,
mode='batch',
loss='hinge',
num_kernels=4,
batch_size=4,
temperature=0.1,
num_epochs=10,
num_nested_epochs=0,
learning_rate=0.001,
learning_rate_mu=0.0,
learning_rate_gamma=0.001,
learning_rate_alpha=0.001,
metrics=['loss', 'err'],
callbacks=[display, early_stopping, checkpoint],
cv=[-1] * x0.shape[0] + [0] * x1.shape[0],
random_state=random_seed())
clf.fit(x, y)
train_err = 1.0 - clf.score(x_train, y_train)
print("Training error = %.4f" % train_err)
if block_figure_on_end:
# save predictions
y_test_pred = clf.predict(x_test)
x_test[x_test == 0] = 1e-4
dump_svmlight_file(x_test,
y_test_pred,
os.path.join(
data_dir(),
"demo/synthetic_2D_data_test_predict.libsvm"),
zero_based=False)
def test_dcgan_cifar10_inception_metric(show_figure=False,
block_figure_on_end=False):
print(
"========== Test DCGAN with Inception Score and FID 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/DCGAN/cifar10")
checkpoints_is = ModelCheckpoint(os.path.join(
root_dir, "checkpoints_is/the_best_is.ckpt"),
mode='max',
monitor='inception_score',
verbose=1,
save_best_only=True)
checkpoints_fid = ModelCheckpoint(os.path.join(
root_dir, "checkpoints_fid/the_best_fid.ckpt"),
mode='min',
monitor='FID',
verbose=1,
save_best_only=True)
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",
},
])
inception_score_display = Display(
layout=(1, 1),
dpi='auto',
show=show_figure,
block_on_end=block_figure_on_end,
filepath=[
os.path.join(root_dir, "inception_score/"
"inception_score_{epoch:04d}.png"),
os.path.join(root_dir, "inception_score/"
"inception_score_{epoch:04d}.pdf")
],
monitor=[
{
'metrics': ['inception_score'],
'type': 'line',
'labels': ["Inception Score"],
'title': "Scores",
'xlabel': "epoch",
'ylabel': "score",
},
],
)
fid_display = Display(
layout=(1, 1),
dpi='auto',
show=show_figure,
block_on_end=block_figure_on_end,
filepath=[
os.path.join(root_dir, "FID/"
"FID_{epoch:04d}.png"),
os.path.join(root_dir, "FID/"
"FID_{epoch:04d}.pdf")
],
monitor=[
{
'metrics': ['FID'],
'type': 'line',
'labels': ["FID"],
'title': "Scores",
'xlabel': "epoch",
'ylabel': "score",
},
],
)
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 = DCGAN(
model_name="DCGAN_CIFAR10",
num_z=10, # set to 100 for a full run
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', 'inception_score', 'inception_score_std', 'FID'
],
callbacks=[
loss_display, inception_score_display, fid_display, sample_display,
checkpoints_is, checkpoints_fid
],
num_epochs=4, # set to 100 for a full run
inception_metrics=[InceptionScore(),
FID(data="cifar10")],
inception_metrics_freq=1,
# summary_freq=1, # uncomment this for a full run
log_path=os.path.join(root_dir, "logs"),
random_state=random_seed(),
verbose=1)
model.fit(x_train)
filepath = os.path.join(root_dir, "checkpoints_fid/the_best_fid.ckpt")
print("Reloading the latest model at: {}".format(filepath))
model1 = TensorFlowModel.load_model(filepath)
model1.inception_metrics = InceptionMetricList(
[InceptionScore(), FID(data="cifar10")])
model1.num_epochs = 6
model1.fit(x_train)
print("Done!")
def test_gan_save_and_load(show_figure=False, block_figure_on_end=False):
print(
"========== Test Save and Load functions of GAN 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/GAN/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 = GAN(model_name="GAN_MNIST_SaveLoad",
num_x=784,
num_discriminator_hiddens=(16, ),
discriminator_batchnorm=False,
discriminator_act_funcs=('lrelu', ),
discriminator_learning_rate=0.001,
num_z=8,
generator_distribution=Uniform(low=(-1.0, ) * 8,
high=(1.0, ) * 8),
generator_batchnorm=False,
num_generator_hiddens=(16, ),
generator_act_funcs=('lrelu', ),
generator_out_func='sigmoid',
generator_learning_rate=0.001,
batch_size=32,
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_pytorch_convnet_v1(show=False, block_figure_on_end=False):
print("========== Test PytorchConvNetv1 ==========")
np.random.seed(random_seed())
err_display = Display(title="Error curves",
dpi='auto',
layout=(1, 1),
freq=1,
show=show,
block_on_end=block_figure_on_end,
monitor=[{
'metrics': ['err'],
'type': 'line',
'title': "Learning errors",
'xlabel': "epoch",
'ylabel': "error",
}])
loss_display = Display(
title="Learning curves",
dpi='auto',
layout=(3, 1),
freq=1,
show=show,
block_on_end=block_figure_on_end,
filepath=[
os.path.join(model_dir(), "male/PyTorchConvNet/"
"loss/loss_{epoch:04d}.png"),
os.path.join(model_dir(), "male/PyTorchConvNet/"
"loss/loss_{epoch:04d}.pdf")
],
monitor=[
{
'metrics': ['loss'],
'type': 'line',
'labels': ["training loss"],
'title': "Learning losses",
'xlabel': "epoch",
'xlabel_params': {
'fontsize': 50
},
'ylabel': "loss",
},
{
'metrics': ['err'],
'type': 'line',
'title': "Learning errors",
'xlabel': "epoch",
'ylabel': "error",
},
{
'metrics': ['err'],
'type': 'line',
'labels': ["training error"],
'title': "Learning errors",
'xlabel': "epoch",
'ylabel': "error",
},
])
weight_display = Display(title="Filters",
dpi='auto',
layout=(1, 1),
figsize=(6, 15),
freq=1,
show=show,
block_on_end=block_figure_on_end,
filepath=os.path.join(
model_dir(), "male/PyTorchConvNet/"
"weights/weights_{epoch:04d}.png"),
monitor=[
{
'metrics': ['weights'],
'title': "Learned weights",
'type': 'img',
'tile_shape': (5, 2),
},
])
# Construct the model.
clf = PyTorchConvNet(model_name='PyTorchConvNet',
num_epochs=1,
batch_size=64,
metrics=['loss', 'err'],
callbacks=[loss_display, err_display],
random_state=random_seed(),
verbose=1)
print('Show some of the training images.')
if False:
clf.num_epochs = 0
# Build the network as well as data loaders.
clf.fit()
import torchvision
def imshow(img):
img = img / 2 + 0.5 # unnormalize
npimg = img.numpy()
plt.imshow(np.transpose(npimg, (1, 2, 0)))
plt.show(block=True)
# get some random training images
dataiter = iter(clf.trainloader)
images, labels = dataiter.next()
# show images
imshow(torchvision.utils.make_grid(images))
# print labels
print(' '.join('%5s' % clf.classes[labels[j]] for j in range(4)))
clf.num_epochs = 1
clf.fit()
print('Testing accuracy = %.4f' % (100 * clf.acc_test()))
save_file_path = clf.save()
clf = PyTorchModel.load_model(save_file_path)
clf.num_epochs = 4
clf.fit()
print('Testing accuracy = %.4f' % (100 * clf.acc_test()))
def test_kmm_cv_disp(show=False, block_figure_on_end=False):
print("========== Test cross-validation for KMM with Display ==========")
np.random.seed(random_seed())
(x_train, y_train), (x_test, y_test) = demo.load_iris()
print("Number of training samples = {}".format(x_train.shape[0]))
print("Number of testing samples = {}".format(x_test.shape[0]))
x = np.vstack([x_train, x_test])
y = np.concatenate([y_train, y_test])
early_stopping = EarlyStopping(monitor='val_err', patience=2, verbose=1)
filepath = os.path.join(model_dir(),
"male/KMM/iris_{epoch:04d}_{val_err:.6f}.pkl")
checkpoint = ModelCheckpoint(filepath,
mode='min',
monitor='val_err',
verbose=0,
save_best_only=True)
display = Display(layout=(3, 1),
dpi='auto',
show=show,
block_on_end=block_figure_on_end,
monitor=[
{
'metrics': ['loss', 'val_loss'],
'type': 'line',
'labels': ["training loss", "validation loss"],
'title': "Learning losses",
'xlabel': "epoch",
'ylabel': "loss",
},
{
'metrics': ['err', 'val_err'],
'type': 'line',
'title': "Learning errors",
'xlabel': "epoch",
'ylabel': "error",
},
{
'metrics': ['err'],
'type': 'line',
'labels': ["training error"],
'title': "Learning errors",
'xlabel': "epoch",
'ylabel': "error",
},
])
clf = KMM(model_name="KMM_hinge",
D=20,
lbd=0.0,
gamma=0.1,
mode='batch',
loss='hinge',
num_kernels=3,
batch_size=100,
temperature=1.0,
num_epochs=10,
num_nested_epochs=1,
learning_rate=0.1,
learning_rate_mu=0.0,
learning_rate_gamma=0.1,
learning_rate_alpha=0.1,
metrics=['loss', 'err'],
callbacks=[display, early_stopping, checkpoint],
cv=[-1] * x_train.shape[0] + [0] * x_test.shape[0],
random_state=random_seed(),
verbose=1)
clf.fit(x, y)
train_err = 1.0 - clf.score(x_train, y_train)
test_err = 1.0 - clf.score(x_test, y_test)
print("Training error = %.4f" % train_err)
print("Testing error = %.4f" % test_err)
