def main():
# inputs
input_shape = (1, 28, 20)
filters = 32
latent_channels = 1
kernel_size = 2
beta = 1.0 # entangled latent space
epochs = 10
batch_size = 1
# log directory
experiment = 'experiment_optimal_network_convolutional_latent_frey'
run = 'cvae_frey_entangled_with_fully_connected_filters_same_borders_15_May_13_50_05_batch_size_1_beta_1.0_epochs_20_filters_32_kernel_size_2_latent_channels_64_loss_vae_loss_optimizer_adam'
log_dir = './summaries/' + experiment + '/' + run + '/'
# define model
vae = FreyOptimalConvolutionalLatentExperimentFullyConnectedFilterVAE(
input_shape,
log_dir,
filters=filters,
latent_channels=latent_channels,
kernel_size=kernel_size,
beta=beta)
# load weights
vae.load_model()
# extract models
model = vae.get_model()
decoder = vae.get_decoder()
encoder = vae.get_encoder()
# get dataset
(X_train, _), (X_test, _) = utils.load_frey()
train_generator = utils.make_generator(X_train, batch_size=batch_size)
test_generator = utils.make_generator(X_test, batch_size=batch_size)
train_size = len(X_train)
test_size = len(X_test)
# show original and reconstruction
# sampling.encode_decode_sample(X_test, model)
# plot filters
# sampling.show_convolutional_layers(X_test, encoder, 8, 8)
# sample from prior
# sampling.decode_prior_samples(5, decoder, latent_shape=(1, 64, 1, 1))
# sample from posterior
num_iter = 1000
sampling.sample_posterior(X_test, model, num_iter, show_every=5)
def main():
# inputs
input_shape = (1, 28, 20)
filters = 32
latent_filters = 1
kernel_size = 2
pool_size = 2
lr = 1e-4
beta = 1.0
batch_size = 1
# log directory
run = 'cvae_frey_20_May_17_07_04_batch_size_1_beta_1_epochs_20_filters_32_kernel_size_2_latent_filters_1_loss_vae_loss_lr_0.0001_optimizer_adam'
log_dir = './summaries/' + experiment + '/' + run + '/'
# define model
vae = FreyConvolutionalLatentSpaceNoBatchNormVAE(
input_shape,
log_dir,
filters=filters,
latent_filters=latent_filters,
kernel_size=kernel_size,
pool_size=pool_size,
beta=beta)
# load weights
vae.load_model()
# extract models
model = vae.get_model()
decoder = vae.get_decoder()
encoder = vae.get_encoder()
# get dataset
(X_train, _), (X_test, _) = utils.load_frey()
train_generator = utils.make_generator(X_train, batch_size=batch_size)
test_generator = utils.make_generator(X_test, batch_size=batch_size)
train_size = len(X_train)
test_size = len(X_test)
# show original and reconstruction
sampling.encode_decode_sample(X_test, model)
vae = FreyVAE((1, 28, 20), log_dir)
# load architecture and weights
encoder = vae.load_model()
# extract models
model = vae.get_model()
encoder = vae.get_encoder()
decoder = vae.get_decoder()
# print summaries
vae.print_model_summaries()
'''
Load data
'''
(_, _), (X_test, _) = utils.load_frey()
'''
Sampling functions
'''
def __decode_prior_samples(num_samples, latent_shape=(1, 2)):
# take num_sample samples
for i in range(num_samples):
# sample from prior
prior_sample = np.random.normal(
size=latent_shape, loc=0.0,
scale=1.0) # sample from standard normal
# decode sample
sample_decoded = decoder.predict(prior_sample)
# plot decoded sample
def train_reconstruction_only_frey_network_with_image_latent_space(latent_filters):
# inputs
input_shape = (1, 28, 20)
filters = 32
kernel_size = 2
pool_size = 2
beta = 1.0
lr = 1e-4
epochs = 20
batch_size = 1
# define filename
name = 'cvae_frey'
# builder hyperparameter dictionary
hp_dictionary = {
'epochs': epochs,
'batch_size': batch_size,
'filters': filters,
'latent_filters': latent_filters,
'kernel_size': kernel_size,
'beta': beta,
'lr': lr,
'loss': 'vae_loss',
'optimizer': 'adam'
}
# define log directory
log_dir = './summaries/' + experiment + '/' + utils.build_hyperparameter_string(name, hp_dictionary) + '/'
# make VAE
vae = FreyConvolutionalLatentSpaceNoBatchNormVAE(input_shape,
log_dir,
filters=filters,
latent_filters=latent_filters,
kernel_size=kernel_size,
pool_size=pool_size,
beta=beta)
# compile VAE
from keras import optimizers
optimizer = optimizers.Adam(lr=lr)
vae.compile(optimizer=optimizer)
# get dataset
(X_train, _), (X_test, _) = utils.load_frey()
train_generator = utils.make_generator(X_train, batch_size=batch_size)
test_generator = utils.make_generator(X_test, batch_size=batch_size)
train_size = len(X_train)
test_size = len(X_test)
# print summaries
vae.print_model_summaries()
# fit VAE
steps_per_epoch = int(train_size / batch_size)
validation_steps = int(test_size / batch_size)
vae.fit_generator(train_generator,
epochs=epochs,
steps_per_epoch=steps_per_epoch,
validation_data=test_generator,
validation_steps=validation_steps)
def run_convolutional_latent_space_with_same_number_of_parameters():
'''
Latent shape: (128, 5, 3)
Number of parameters in network: 119,521
Number of parameters in latent space: 32,896
'''
# inputs
input_shape = (1, 28, 20)
epochs = 1
batch_size = 1
beta = 1.0
filters = 32
kernel_size = 2
latent_channels = 128
pool_size = 2
# define filename
name = 'cvae_frey_convolutional_latent'
# builder hyperparameter dictionary
hp_dictionary = {
'epochs': epochs,
'batch_size': batch_size,
'beta': beta,
'filters': filters,
'kernel_size': kernel_size,
'latent_channels': latent_channels,
'pool_size': pool_size,
'loss': 'vae_loss',
'optimizer': 'adam'
}
# define log directory
log_dir = './summaries/' + utils.build_hyperparameter_string(
name, hp_dictionary) + '/'
# make VAE
vae = FreyConvolutionalLatentVAE(input_shape,
log_dir,
filters=filters,
kernel_size=kernel_size,
latent_channels=latent_channels,
pool_size=pool_size)
# compile VAE
from keras import optimizers
optimizer = optimizers.Adam(lr=1e-3)
vae.compile(optimizer=optimizer)
# print summaries
vae.print_model_summaries()
# get dataset
(X_train, _), (X_test, _) = utils.load_frey()
train_generator = utils.make_generator(X_train, batch_size=batch_size)
test_generator = utils.make_generator(X_test, batch_size=batch_size)
train_size = len(X_train)
test_size = len(X_test)
# fit VAE
steps_per_epoch = int(train_size / batch_size)
validation_steps = int(test_size / batch_size)
vae.fit_generator(train_generator,
epochs=epochs,
steps_per_epoch=steps_per_epoch,
validation_data=test_generator,
validation_steps=validation_steps)
def train_entangled_frey_network_with_fully_connected_filters_same_borders_no_pooling_less_filters(
):
# inputs
input_shape = (1, 28, 20)
filters = 8
latent_channels = 16
kernel_size = 2
beta = 1.0 # entangled latent space
epochs = 20
batch_size = 1
# define filename
name = 'cvae_frey_entangled_with_fully_connected_filters_same_borders_no_pooling_less_filters'
# builder hyperparameter dictionary
hp_dictionary = {
'epochs': epochs,
'batch_size': batch_size,
'beta': beta,
'filters': filters,
'latent_channels': latent_channels,
'kernel_size': kernel_size,
'loss': 'vae_loss',
'optimizer': 'adam'
}
# define log directory
log_dir = './summaries/experiment_optimal_network_convolutional_latent_frey/' + utils.build_hyperparameter_string(
name, hp_dictionary) + '/'
# make VAE
vae = FreyOptimalConvolutionalLatentExperimentFullyConnectedFilterVAE(
input_shape,
log_dir,
filters=filters,
latent_channels=latent_channels,
kernel_size=kernel_size,
beta=beta)
# compile VAE
from keras import optimizers
optimizer = optimizers.Adam(lr=1e-3)
vae.compile(optimizer=optimizer)
# get dataset
(X_train, _), (X_test, _) = utils.load_frey()
train_generator = utils.make_generator(X_train, batch_size=batch_size)
test_generator = utils.make_generator(X_test, batch_size=batch_size)
train_size = len(X_train)
test_size = len(X_test)
# print summaries
vae.print_model_summaries()
# fit VAE
steps_per_epoch = int(train_size / batch_size)
validation_steps = int(test_size / batch_size)
vae.fit_generator(train_generator,
epochs=epochs,
steps_per_epoch=steps_per_epoch,
validation_data=test_generator,
validation_steps=validation_steps)
# make VAE
vae = FreyVAE(input_shape,
log_dir,
filters=filters,
kernel_size=kernel_size,
pre_latent_size=pre_latent_size,
latent_size=latent_size)
# compile VAE
from keras import optimizers
optimizer = optimizers.Adam(lr=1e-3)
vae.compile(optimizer=optimizer)
# print summaries
vae.print_model_summaries()
# get dataset
(X_train, _), (X_test, _) = utils.load_frey()
train_generator = utils.make_generator(X_train, batch_size=batch_size)
test_generator = utils.make_generator(X_test, batch_size=batch_size)
train_size = len(X_train)
test_size = len(X_test)
# fit VAE
steps_per_epoch = int(train_size / batch_size)
validation_steps = int(test_size / batch_size)
vae.fit_generator(train_generator,
epochs=epochs,
steps_per_epoch=steps_per_epoch,
validation_data=test_generator,
validation_steps=validation_steps)