def main(_):
'''Building the graph, opening of a session and starting the training od the neural network.'''
num_batches=int(FLAGS.num_samples/FLAGS.batch_size)
with tf.Graph().as_default():
train_data, train_data_infer=_get_training_data(FLAGS)
test_data=_get_test_data(FLAGS)
iter_train = train_data.make_initializable_iterator()
iter_train_infer=train_data_infer.make_initializable_iterator()
iter_test=test_data.make_initializable_iterator()
x_train= iter_train.get_next()
x_train_infer=iter_train_infer.get_next()
x_test=iter_test.get_next()
model=TrainModel(FLAGS, 'training')
train_op, train_loss_op=model.train(x_train)
prediction, labels, test_loss_op, mae_ops=model._validation_loss(x_train_infer, x_test)
saver=tf.train.Saver()
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
train_loss=0
test_loss=[]
mae=[]
for epoch in range(FLAGS.num_epoch):
sess.run(iter_train.initializer)
sess.run(iter_train_infer.initializer)
sess.run(iter_test.initializer)
for batch_nr in range(num_batches):
_, loss_=sess.run((train_op, train_loss_op))
train_loss+=loss_
for i in range(FLAGS.num_samples):
pred, labels_, loss_, mae_=sess.run((prediction, labels, test_loss_op,mae_ops))
test_loss.append(loss_)
mae.append(mae_)
print('epoch_nr: %i, train_loss: %.3f, test_loss: %.3f, mean_abs_error: %.3f'
%(epoch,(train_loss/num_batches),np.mean(test_loss), np.mean(mae)))
if np.mean(mae)<0.9:
saver.save(sess, FLAGS.checkpoints_path)
train_loss=0
test_loss=[]
mae=[]
def main(_):
'''Building the graph, opening of a session and starting the training od the neural network.'''
num_batches = int(FLAGS.num_samples / FLAGS.batch_size)
train_loss_summary = []
test_loss_summary = []
with tf.Graph().as_default():
train_data, train_data_infer = _get_training_data(FLAGS)
test_data = _get_test_data(FLAGS)
iter_train = train_data.make_initializable_iterator()
iter_train_infer = train_data_infer.make_initializable_iterator()
iter_test = test_data.make_initializable_iterator()
x_train = iter_train.get_next()
x_train_infer = iter_train_infer.get_next()
x_test = iter_test.get_next()
model = DAE(FLAGS)
train_op, train_loss_op = model._optimizer(x_train)
pred_op, test_loss_op = model._validation_loss(x_train_infer, x_test)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
train_loss = 0
test_loss = 0
for epoch in range(FLAGS.num_epoch):
sess.run(iter_train.initializer)
for batch_nr in range(num_batches):
_, loss_ = sess.run((train_op, train_loss_op))
train_loss += loss_
sess.run(iter_train_infer.initializer)
sess.run(iter_test.initializer)
for i in range(FLAGS.num_samples):
pred, loss_ = sess.run((pred_op, test_loss_op))
test_loss += loss_
print('epoch_nr: %i, train_loss: %.3f, test_loss: %.3f' %
(epoch, (train_loss / num_batches),
(test_loss / FLAGS.num_samples)))
train_loss_summary.append(train_loss / num_batches)
test_loss_summary.append(test_loss / FLAGS.num_samples)
train_loss = 0
test_loss = 0
def main(_):
'''Building the graph, opening of a session and starting the training od the neural network.'''
num_batches = int(FLAGS.num_samples / FLAGS.batch_size)
with tf.Graph().as_default():
train_data, train_data_infer = _get_training_data(FLAGS)
test_data = _get_test_data(FLAGS)
iter_train = train_data.make_initializable_iterator()
iter_train_infer = train_data_infer.make_initializable_iterator()
iter_test = test_data.make_initializable_iterator()
x_train = iter_train.get_next()
x_train_infer = iter_train_infer.get_next()
x_test = iter_test.get_next()
model = RBM(FLAGS)
update_op, accuracy = model.optimize(x_train)
v_infer = model.inference(x_train_infer)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
for epoch in range(FLAGS.num_epoch):
acc_train = 0
acc_infer = 0
sess.run(iter_train.initializer)
for batch_nr in range(num_batches):
_, acc = sess.run((update_op, accuracy))
acc_train += acc
if batch_nr > 0 and batch_nr % FLAGS.eval_after == 0:
sess.run(iter_train_infer.initializer)
sess.run(iter_test.initializer)
num_valid_batches = 0
for i in range(FLAGS.num_samples):
v_target = sess.run(x_test)[0]
if len(v_target[v_target >= 0]) > 0:
v_ = sess.run(v_infer)[0]
acc = 1.0 - np.mean(
np.abs(v_[v_target >= 0] -
v_target[v_target >= 0]))
acc_infer += acc
num_valid_batches += 1
print(
'epoch_nr: %i, batch: %i/%i, acc_train: %.3f, acc_test: %.3f'
% (epoch, batch_nr, num_batches,
(acc_train / FLAGS.eval_after),
(acc_infer / num_valid_batches)))
acc_train = 0
acc_infer = 0
'num_v', 3952,
'Number of visible neurons (Number of movies the users rated.)')
tf.app.flags.DEFINE_integer('num_h', 128, 'Number of hidden neurons.)')
tf.app.flags.DEFINE_integer(
'num_samples', 5953,
'Number of training samples (Number of users, who gave a rating).')
FLAGS = tf.app.flags.FLAGS
num_batches = int(FLAGS.num_samples / FLAGS.batch_size)
with tf.Graph().as_default():
train_loss_summary = []
test_loss_summary = []
train_data, train_data_infer = _get_training_data(FLAGS)
test_data = _get_test_data(FLAGS)
iter_train = train_data.make_initializable_iterator()
iter_train_infer = train_data_infer.make_initializable_iterator()
iter_test = test_data.make_initializable_iterator()
x_train = iter_train.get_next()
x_train_infer = iter_train_infer.get_next()
x_test = iter_test.get_next()
model = DAE(FLAGS)
train_op, train_loss_op = model._optimizer(x_train)
pred_op, test_loss_op = model._validation_loss(x_train_infer, x_test)