def train_stacked_autoencoder(X_train, X_test):
model = StackedAutoEncoder()
""" Tensorflow needs to see the graph before initilize
the variables for the computation """
model.build_graph()
config = tf.ConfigProto(intra_op_parallelism_threads=_PROCESSORS,
inter_op_parallelism_threads=_PROCESSORS)
sess = tf.Session(config=config)
sess.run(tf.global_variables_initializer())
print("Training Stacked Autoencoder...")
for epoch in range(FLAGS.epochs * FLAGS.batch_size):
batch_x = next_batch(FLAGS.batch_size, X_train)
train_dict = {
model.x: batch_x,
model.learning_rate_stacked: FLAGS.learning_rate_stacked,
model.reg_stacked: FLAGS.reg_stacked,
model.noise: FLAGS.noise,
model.fraction: FLAGS.fraction
}
sess.run(model.optimizer, feed_dict=train_dict)
if epoch % 10 == 0:
c = sess.run([model.loss], feed_dict=train_dict)
print("Train Loss: ", c)
""" Computing the encoded version of X_train and X_test """
X_train_dict = {
model.x: X_train,
model.learning_rate_stacked: FLAGS.learning_rate_stacked,
model.reg_stacked: FLAGS.reg_stacked,
model.noise: FLAGS.noise,
model.fraction: FLAGS.fraction
}
X_test_dict = {
model.x: X_test,
model.learning_rate_stacked: FLAGS.learning_rate_stacked,
model.reg_stacked: FLAGS.reg_stacked,
model.noise: FLAGS.noise,
model.fraction: FLAGS.fraction
}
X_train = np.asarray(sess.run([model.x_encoded], feed_dict=X_train_dict))
X_test = np.asarray(sess.run([model.x_encoded], feed_dict=X_test_dict))
X_train = X_train.reshape(-1, 30)
X_test = X_test.reshape(-1, 30)
""" Closing the session to avoid cnflicts with the test """
return X_train, X_test
# Mnist digits dataset
dataset = torchvision.datasets.MNIST(
root='../mnist/',
train=True, # this is training data
transform=torchvision.transforms.ToTensor(
), # Converts a PIL.Image or numpy.ndarray to # torch.FloatTensor of shape (C x H x W) and normalize in the range [0.0, 1.0]
download=DOWNLOAD_DATA, # download it if you don't have it
)
# dataset = MNIST('../data/mnist/', transform=img_transform)
dataloader = DataLoader(dataset,
batch_size=batch_size,
shuffle=True,
num_workers=8)
print("It is DAE")
model = StackedAutoEncoder().cuda()
for epoch in range(num_epochs):
if epoch % 10 == 0:
# Test the quality of our features with a randomly initialzed linear classifier.
classifier = nn.Linear(10, 10).cuda()
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(classifier.parameters(), lr=0.001)
model.train()
total_time = time.time()
correct = 0
for i, data in enumerate(dataloader):
img, target = data
target = Variable(target).cuda()
img = Variable(img).cuda()