def main():
SECTION = 'vae'
RUN_ID = '0002'
DATA_NAME = 'digits'
RUN_FOLDER = 'run/{}/'.format(SECTION)
RUN_FOLDER += '_'.join([RUN_ID, DATA_NAME])
print('RUN_FOLDER: ', RUN_FOLDER)
if not os.path.exists(RUN_FOLDER):
os.makedirs(RUN_FOLDER)
os.makedirs(os.path.join(RUN_FOLDER, 'viz'))
os.makedirs(os.path.join(RUN_FOLDER, 'images'))
os.makedirs(os.path.join(RUN_FOLDER, 'weights'))
mode = 'build' #'load' #
(x_train, y_train), (x_test, y_test) = load_mnist()
vae = VariationalAutoencoder(input_dim=(28, 28, 1),
encoder_conv_filters=[32, 64, 64, 64],
encoder_conv_kernel_size=[3, 3, 3, 3],
encoder_conv_strides=[1, 2, 2, 1],
decoder_conv_t_filters=[64, 64, 32, 1],
decoder_conv_t_kernel_size=[3, 3, 3, 3],
decoder_conv_t_strides=[1, 2, 2, 1],
z_dim=2)
if mode == 'build':
vae.save(RUN_FOLDER)
else:
vae.load_weights(os.path.join(RUN_FOLDER, 'weights/weights.h5'))
vae.encoder.summary()
vae.decoder.summary()
LEARNING_RATE = 0.0005
R_LOSS_FACTOR = 1000
vae.compile(LEARNING_RATE, R_LOSS_FACTOR)
BATCH_SIZE = 32
EPOCHS = 200
PRINT_EVERY_N_BATCHES = 100
INITIAL_EPOCH = 0
vae.train(x_train,
batch_size=BATCH_SIZE,
epochs=EPOCHS,
run_folder=RUN_FOLDER,
print_every_n_batches=PRINT_EVERY_N_BATCHES,
initial_epoch=INITIAL_EPOCH)
print('Done!')
def main():
SECTION = 'vae'
RUN_ID = '0001'
DATA_NAME = 'digits'
RUN_FOLDER = 'run/{}/'.format(SECTION)
RUN_FOLDER += '_'.join([RUN_ID, DATA_NAME])
print('RUN_FOLDER: ', RUN_FOLDER)
(x_train, y_train), (x_test, y_test) = load_mnist()
AE = load_model(Autoencoder, RUN_FOLDER)
reconstruct(AE, x_test)
coder_wall(AE, x_test, y_test)
def main():
SECTION = 'vae'
RUN_ID = '0001'
DATA_NAME = 'digits'
RUN_FOLDER = './run/{}/'.format(SECTION)
RUN_FOLDER += '_'.join([RUN_ID, DATA_NAME])
if not os.path.exists(RUN_FOLDER):
os.makedirs(RUN_FOLDER)
os.makedirs(os.path.join(RUN_FOLDER, 'viz'))
os.makedirs(os.path.join(RUN_FOLDER, 'images'))
os.makedirs(os.path.join(RUN_FOLDER, 'weights'))
MODE = 'build' # 'load' #
(x_train, y_train), (x_test, y_test) = load_mnist()
AE = Autoencoder(input_dim=(28, 28, 1),
encoder_conv_filters=[32, 64, 64, 64],
encoder_conv_kernel_size=[3, 3, 3, 3],
encoder_conv_strides=[1, 2, 2, 1],
decoder_conv_t_filters=[64, 64, 32, 1],
decoder_conv_t_kernel_size=[3, 3, 3, 3],
decoder_conv_t_strides=[1, 2, 2, 1],
z_dim=2)
if MODE == 'build':
AE.save(RUN_FOLDER)
else:
AE.load_weights(os.path.join(RUN_FOLDER, 'weights/weights.h5'))
AE.encoder.summary()
AE.decoder.summary()
LEARNING_RATE = 0.0005
BATCH_SIZE = 32
INITIAL_EPOCH = 0
AE.compile(LEARNING_RATE)
AE.train(x_train[:1000],
batch_size=BATCH_SIZE,
epochs=200,
run_folder=RUN_FOLDER,
initial_epoch=INITIAL_EPOCH)
print("Done")
import numpy as np
import os
from scipy.stats import norm
from models.AE import Autoencoder
from utils.loaders import load_mnist, load_model
# %%
SECTION = 'vae'
RUN_ID = '0001'
DATA_NAME = 'digits'
RUN_FOLDER = f'run/{SECTION}/'
RUN_FOLDER += '_'.join([RUN_ID, DATA_NAME])
# %%
(x_train, y_train), (x_test, y_test) = load_mnist()
# %%
# Load the model architecture
AE = load_model(Autoencoder, RUN_FOLDER)
# reconstructing original paintings
n_to_show = 10
example_idx = np.random.choice(range(len(x_test)), n_to_show)
example_images = x_test[example_idx]
z_points = AE.encoder.predict(example_images)
reconst_images = AE.decoder.predict(z_points)
fig = plt.figure(figsize=(15, 3))