def analyze():
# Initialize session and trigger debugging mode
session = tf.Session()
if args.debug:
session = tf_debug.LocalCLIDebugWrapperSession(session)
# Load data and define placeholders
print('Loading data from: ' + args.data_dir)
data = np.load(args.data_dir)
placeholders = define_placeholders(args, data.shape)
# Create model and optimizer
model = define_model(args, data.shape, placeholders)
model_name = "%s_%s_%s_%s_%s_tol=%s" % (
args.data_dir[8:-10], args.model_type, str(args.hidden_dim_1),
str(args.hidden_dim_2), str(args.hidden_dim_3), str(args.tol))
model_path = "../models/%s.ckpt" % (model_name)
saver = tf.train.Saver()
with session as sess:
saver.restore(sess, model_path)
start_time = time.ctime(int(time.time()))
print("Analyzing '%s'... \nStart Time: %s" %
(model_name, str(start_time)))
if args.model_type == 'VAE':
analyze_VAE(args, placeholders, data, model, model_name, sess)
score_VAE(args, placeholders, data, model, model_name, sess)
elif args.model_type == 'VGAE':
analyze_VGAE(args, placeholders, data, model, model_name, sess)
def augment():
# Initialize session and trigger debugging mode
session = tf.Session()
if args.debug:
session = tf_debug.LocalCLIDebugWrapperSession(session)
# Load data and define placeholders
print('Loading data from: ' + args.data_dir)
data = np.load(args.data_dir)
placeholders = define_placeholders(args, data.shape)
# Create model and optimizer
model = define_model(args, data.shape, placeholders)
model_name = "%s_%s_%s_%s_%s_tol=%s" % (
args.data_dir[8:-10], args.model_type, str(args.hidden_dim_1),
str(args.hidden_dim_2), str(args.hidden_dim_3), str(args.tol))
model_path = "../models/%s.ckpt" % (model_name)
saver = tf.train.Saver()
with session as sess:
saver.restore(sess, model_path)
start_time = time.ctime(int(time.time()))
print("Augmenting '%s'... \nStart Time: %s" %
(model_name, str(start_time)))
gen_all = []
for i in range(args.num_batches):
randoms = np.random.normal(0.0, 1.0,
(args.batch_size, args.hidden_dim_3))
[gen, gen_preds] = sess.run([model.reconstructions, model.preds],
feed_dict={model.z: randoms})
gen = np.concatenate((gen, gen_preds), axis=1)
gen = gen.reshape(args.batch_size, -1)
gen_all.append(gen)
gen = np.array(gen_all).reshape(args.num_batches, args.batch_size, -1)
gen = gen.reshape(-1, input_dim)
if args.predict_type == 'Classifier':
gen[:, 16110:] = np.clip(tf.round(gen[:, 16110:]).eval(), 0, 1)
augmented_data = np.concatenate((np.transpose(gen), train_data),
axis=1)
print(augmented_data.shape)
print('Saving augmented dataset at ../data/%s_augmented_train.npy' %
(model_name))
np.save('../data/%s_augmented_train.npy' % (model_name),
augmented_data)
def main():
# Initialize session and trigger debugging mode
session = tf.Session()
if args.debug:
session = tf_debug.LocalCLIDebugWrapperSession(session)
# Load data and define placeholders
print('Loading test data from: ' + args.test_data_dir)
data = np.load(args.test_data_dir)
placeholders = define_placeholders(args, data.shape)
# Create model and optimizer
model = define_model(args, data.shape, placeholders)
model_name = "%s_%s_%s_%s_%s" % (
args.data_dir[8:-10], args.model_type, str(
args.hidden_dim_1), str(args.hidden_dim_2), str(args.hidden_dim_3))
model_path = "../models/%s.ckpt" % (model_name)
saver = tf.train.Saver()
with session as sess:
print('Scoring %s' % (model_name))
saver.restore(sess, model_path)
start, accuracy = 0, 0
i = 0
# Get average reconstruction loss on test set
while start + args.batch_size <= data.shape[1]:
batch, labels = get_consecutive_batch(start, args.batch_size, data)
feed_dict = {placeholders['inputs']: batch}
outs = sess.run([model.preds], feed_dict=feed_dict)
if args.model_type == 'Classifier':
preds = tf.nn.sigmoid(outs[0])
correct_pred = tf.equal(tf.round(preds), labels)
print(correct_pred.eval())
accuracy += tf.reduce_mean(tf.cast(correct_pred, tf.float32))
start += args.batch_size
i += 1
else:
accuracy += tf.reduce_mean(tf.square(outs[0] - labels))
start += args.batch_size
i += 1
print("total score:", accuracy.eval() / i)
f = open('./scores/%s.txt' % (model_name), 'w')
f.write("Score on %s: %f" % (args.test_data_dir, accuracy.eval() / i))
f.close()
def main():
# Initialize session and trigger debugging mode
session = tf.Session()
if args.debug:
session = tf_debug.LocalCLIDebugWrapperSession(session)
# Load data and define placeholders
print('Loading data from: ' + args.data_dir)
data = np.load(args.data_dir)
placeholders = define_placeholders(args, data.shape)
# Create model and optimizer
model = define_model(args, data.shape, placeholders)
opt = define_optimizer(args, model, data.shape, placeholders)
model_name = "%s_%s_%s_%s_%s_tol=%s" % (
args.data_dir[8:-10], args.model_type, str(args.hidden_dim_1),
str(args.hidden_dim_2), str(args.hidden_dim_3), str(args.tol))
model_path = "../models/%s.ckpt" % (model_name)
saver = tf.train.Saver()
with session as sess:
sess.run(tf.global_variables_initializer())
if args.restore:
print("Restoring model from: ", model_path)
saver.restore(sess, model_path)
start_time = time.ctime(int(time.time()))
print("Starting to train '%s'... \nStart Time: %s" %
(model_name, str(start_time)))
if args.model_type == 'VAE':
train_VAE(model_path, data, session, saver, placeholders, model,
opt, args)
elif args.model_type == 'VGAE':
train_VGAE(model_path, data, session, saver, placeholders, model,
opt, args)
print("Training Complete. Model name: %s" % (model_name))