def train():
with tf.Graph().as_default():
# get global step
global_step = tf.train.get_or_create_global_step()
# get data through cpu
with tf.device('/cpu:0'):
images, labels = cifar10.distorted_inputs()
# get loss and logit
# logits = cifar10.inference(images=images, r=low_ranks)
logits = cifar10.inference(images=images, r=low_ranks)
loss = cifar10.loss(logits=logits, labels=labels)
# set train_op
train_op = cifar10.train(loss, global_step)
for v in tf.trainable_variables():
print(v)
nonzero = tf.count_nonzero(tf.get_collection('sparse_components')[-1])
# define a LoggerHook to log something
# clean_list = tf.get_collection('sparse_components')
# clean_list = clean_s(clean_list)
# clean_op = [c.op for c in clean_list]
class _LoggerHook(tf.train.SessionRunHook):
"""
log session and runtime info
"""
def begin(self):
self._step = -1
self._start_time = time.time()
def before_run(self, run_context):
self._step += 1
return tf.train.SessionRunArgs(loss)
def after_run(self, run_context, run_values):
if self._step % FLAGS.log_frequency == 0:
current_time = time.time()
duration = current_time - self._start_time
self._start_time = current_time
loss_value = run_values.results
example_per_sec = FLAGS.log_frequency * FLAGS.batch_size / duration
sec_per_batch = float(duration / FLAGS.log_frequency)
format_str = (
'%s: step %d, loss = %.6f (%.1f examples/sec;'
'%.3f sec/batch)')
print(format_str % (datetime.now(), self._step, loss_value,
example_per_sec, sec_per_batch))
with tf.train.MonitoredTrainingSession(
checkpoint_dir=FLAGS.train_dir,
hooks=[
tf.train.StopAtStepHook(last_step=FLAGS.max_steps),
tf.train.NanTensorHook(loss),
_LoggerHook()
],
config=tf.ConfigProto(log_device_placement=FLAGS.
log_device_placement)) as mon_sess:
while not mon_sess.should_stop():
mon_sess.run(train_op)
def main(_):
with tf.Graph().as_default():
global_step = tf.contrib.framework.get_or_create_global_step()
images, labels = cifar10.distorted_inputs()
logits = cifar10.inference(images)
loss = cifar10.loss(logits, labels)
train_op = cifar10.train(loss, global_step)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(coord=coord)
summary = tf.summary.merge_all()
writer = tf.summary.FileWriter(FLAGS.log_dir, sess.graph)
print('Training started')
for i in range(FLAGS.max_step):
start_time = time.time()
_, loss_value = sess.run([train_op, loss])
duration = time.time() - start_time
if i % 100 == 0:
print('Step %6d: loss = %.2f (%.3f sec)' %
(i, loss_value, duration))
summary_str = sess.run(summary)
writer.add_summary(summary_str, i)
writer.flush()
coord.request_stop()
coord.join(threads)
def evaluate():
"""Eval CIFAR-10 for a number of steps."""
with tf.Graph().as_default() as g:
# Get images and labels for CIFAR-10.
eval_data = FLAGS.eval_data == 'test'
images, labels = cifar10.distorted_inputs(eval_data=eval_data)
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(images)
# Calculate predictions.
top_k_op = tf.nn.in_top_k(logits, labels, 1)
# Normalize predictions with softmax
preds = tf.nn.softmax(logits, name='predictions')
# Restore the moving average version of the learned variables for eval.
variable_averages = tf.train.ExponentialMovingAverage(
cifar10.MOVING_AVERAGE_DECAY)
variables_to_restore = variable_averages.variables_to_restore()
saver = tf.train.Saver(variables_to_restore)
# Build the summary operation based on the TF collection of Summaries.
summary_op = tf.merge_all_summaries()
summary_writer = tf.train.SummaryWriter(FLAGS.eval_dir, g)
while True:
eval_once(saver, summary_writer, top_k_op, summary_op, preds, labels)
if FLAGS.run_once:
break
time.sleep(FLAGS.eval_interval_secs)
def train():
"""Train CIFAR-10 for a number of steps."""
with tf.Graph().as_default():
global_step = tf.Variable(0, trainable=False)
# Get images and labels for CIFAR-10.
images, labels = cifar10.distorted_inputs()
testImg, testlabels = cifar10.inputs(eval_data=True)
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(images)
test_pre = cifar10.inference(testImg,test=True)
# Calculate loss.
loss = cifar10.loss(logits, labels)
# Build a Graph that trains the model with one batch of examples and
# updates the model parameters.
train_op = cifar10.train(loss, global_step)
# Create a saver.
saver = tf.train.Saver(tf.all_variables())
# Build the summary operation based on the TF collection of Summaries.
summary_op = tf.merge_all_summaries()
# Build an initialization operation to run below.
init = tf.initialize_all_variables()
# Start running operations on the Graph.
sess = tf.Session(config=tf.ConfigProto(
log_device_placement=FLAGS.log_device_placement))
sess.run(init)
# Start the queue runners.
tf.train.start_queue_runners(sess=sess)
summary_writer = tf.train.SummaryWriter(FLAGS.train_dir, sess.graph)
for step in xrange(FLAGS.max_steps):
start_time = time.time()
_, loss_value = sess.run([train_op, loss])
duration = time.time() - start_time
if step % 10 == 0:
print ('loss '+str(loss_value))
if step % 100 == 0:
summary_str = sess.run(summary_op)
summary_writer.add_summary(summary_str, step)
# Save the model checkpoint periodically.
if step % 10 == 0 or (step + 1) == FLAGS.max_steps:
checkpoint_path = os.path.join(FLAGS.train_dir, 'model.ckpt')
saver.save(sess, checkpoint_path, global_step=step)
#eval
if step%10==0:
cifar10.accuracy(test_pre,testlabels)
def evaluate():
"""Eval CIFAR-10 for a number of steps."""
with tf.Graph().as_default() as g:
# Get images and labels for CIFAR-10.
eval_data = FLAGS.eval_data == 'test'
images, labels = cifar10.distorted_inputs(eval_data=eval_data)
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(images)
# Calculate predictions.
top_k_op = tf.nn.in_top_k(logits, labels, 1)
# Normalize predictions with softmax
preds = tf.nn.softmax(logits, name='predictions')
# Restore the moving average version of the learned variables for eval.
variable_averages = tf.train.ExponentialMovingAverage(
cifar10.MOVING_AVERAGE_DECAY)
variables_to_restore = variable_averages.variables_to_restore()
saver = tf.train.Saver(variables_to_restore)
# Build the summary operation based on the TF collection of Summaries.
summary_op = tf.merge_all_summaries()
summary_writer = tf.train.SummaryWriter(FLAGS.eval_dir, g)
while True:
eval_once(saver, summary_writer, top_k_op, summary_op, preds,
labels)
if FLAGS.run_once:
break
time.sleep(FLAGS.eval_interval_secs)
def tower_loss(scope):
"""Calculate the total loss on a single tower running the CIFAR model.
Args:
scope: unique prefix string identifying the CIFAR tower, e.g. 'tower_0'
Returns:
Tensor of shape [] containing the total loss for a batch of data
"""
# Get images and labels for CIFAR-10.
images, labels = cifar10.distorted_inputs()
# Build inference Graph.
logits = cifar10.inference(images)
# Build the portion of the Graph calculating the losses. Note that we will
# assemble the total_loss using a custom function below.
_ = cifar10.loss(logits, labels)
# Assemble all of the losses for the current tower only.
losses = tf.get_collection('losses', scope)
# Calculate the total loss for the current tower.
total_loss = tf.add_n(losses, name='total_loss')
# Attach a scalar summary to all individual losses and the total loss; do the
# same for the averaged version of the losses.
for l in losses + [total_loss]:
# Remove 'tower_[0-9]/' from the name in case this is a multi-GPU training
# session. This helps the clarity of presentation on tensorboard.
loss_name = re.sub('%s_[0-9]*/' % cifar10.TOWER_NAME, '', l.op.name)
tf.summary.scalar(loss_name, l)
return total_loss
def train():
"""Train CIFAR-10 for a number of steps."""
with tf.Graph().as_default():
global_step = tf.contrib.framework.get_or_create_global_step()
# Get images and labels for CIFAR-10.
images, labels = cifar10.distorted_inputs()
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(images, labels)
# Calculate loss.
loss = cifar10.loss(logits, labels)
# Calculate accuracy.
accuracy = cifar10.accuracy(logits, labels)
# Build a Graph that trains the model with one batch of examples and
# updates the model parameters.
train_op = cifar10.train(loss, global_step)
class _LoggerHook(tf.train.SessionRunHook):
"""Logs loss and runtime."""
def begin(self):
self._step = -1
self._start_time = time.time()
def before_run(self, run_context):
self._step += 1
return tf.train.SessionRunArgs(
accuracy) # Asks for loss value.
def after_run(self, run_context, run_values):
if self._step % FLAGS.log_frequency == 0:
current_time = time.time()
duration = current_time - self._start_time
self._start_time = current_time
loss_value = run_values.results
examples_per_sec = FLAGS.log_frequency * FLAGS.batch_size / duration
sec_per_batch = float(duration / FLAGS.log_frequency)
format_str = (
'%s: step %d, accuracy = %.4f, (%.1f examples/sec; %.3f '
'sec/batch)')
print(format_str % (datetime.now(), self._step, loss_value,
examples_per_sec, sec_per_batch))
with tf.train.MonitoredTrainingSession(
checkpoint_dir=FLAGS.train_dir,
hooks=[
tf.train.StopAtStepHook(last_step=FLAGS.max_steps),
tf.train.NanTensorHook(loss),
_LoggerHook()
],
config=tf.ConfigProto(log_device_placement=FLAGS.
log_device_placement)) as mon_sess:
while not mon_sess.should_stop():
mon_sess.run(train_op)
def evaluate():
"""Eval CIFAR-10 for a number of steps."""
with tf.Graph().as_default() as g:
with tf.device('/cpu:0'):
images, labels = cifar10.distorted_inputs()
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(images)
# Calculate predictions.
top_k_op = tf.nn.in_top_k(logits, labels, 1)
# Restore the moving average version of the learned variables for eval.
variable_averages = tf.train.ExponentialMovingAverage(
cifar10.MOVING_AVERAGE_DECAY)
variables_to_restore = variable_averages.variables_to_restore()
saver = tf.train.Saver(variables_to_restore)
# Build the summary operation based on the TF collection of Summaries.
summary_op = tf.summary.merge_all()
summary_writer = tf.summary.FileWriter(FLAGS.train_dir, g)
eval_once(saver, summary_writer, top_k_op, summary_op)
def tower_loss(scope):
"""Calculate the total loss on a single tower running the CIFAR model.
Args:
scope: unique prefix string identifying the CIFAR tower, e.g. 'tower_0'
Returns:
Tensor of shape [] containing the total loss for a batch of data
"""
# Get images and labels for CIFAR-10.
images, labels = cifar10.distorted_inputs(FLAGS.contrastcase)
# Build inference Graph.
logits = cifar10.inference1_2(images)
# Build the portion of the Graph calculating the losses. Note that we will
# assemble the total_loss using a custom function below.
_ = cifar10.loss(logits, labels)
# Assemble all of the losses for the current tower only.
losses = tf.get_collection('losses', scope)
# Calculate the total loss for the current tower.
total_loss = tf.add_n(losses, name='total_loss')
# Attach a scalar summary to all individual losses and the total loss; do the
# same for the averaged version of the losses.
for l in losses + [total_loss]:
# Remove 'tower_[0-9]/' from the name in case this is a multi-GPU training
# session. This helps the clarity of presentation on tensorboard.
loss_name = re.sub('%s_[0-9]*/' % cifar10.TOWER_NAME, '', l.op.name)
tf.summary.scalar(loss_name, l)
return total_loss
def train():
with tf.Graph().as_default():
global_step = tf.train.get_or_create_global_step()
images, labels = cifar10.distorted_inputs()
logits = cifar10.inference(images, train = True)
loss = cifar10.loss(logits, labels)
accuracy = cifar10.accuracy(logits, labels)
train_op = cifar10.train(loss, global_step)
class _LoggerHook(tf.train.SessionRunHook):
def begin(self):
self._step = -1
def before_run(self, run_context):
self._step += 1
return tf.train.SessionRunArgs([loss, accuracy])
def after_run(self, run_context, run_values):
if self._step % 10 == 0:
loss_value, acc_value = run_values.results
format_str = ('step %d, loss = %.2f, accuracy = %.2f ')
print (format_str %(self._step, loss_value, acc_value))
with tf.train.MonitoredTrainingSession(
checkpoint_dir=train_dir,
hooks=[tf.train.StopAtStepHook(last_step=max_step),
tf.train.NanTensorHook(loss),
_LoggerHook()],
config=tf.ConfigProto(
log_device_placement=False)) as mon_sess:
while not mon_sess.should_stop():
mon_sess.run(train_op)
def train():
"""Train CIFAR-10 for a number of steps."""
with tf.Graph().as_default():
global_step = tf.Variable(0, trainable=False)
images, labels = cifar10.distorted_inputs()
logits = cifar10.inference(images)
loss = cifar10.loss(logits, labels)
# loss = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits, labels))
# train_op = tf.train.GradientDescentOptimizer(1e-2).minimize(loss)
train_op = cifar10.train(loss, global_step)
top_k_op = tf.nn.in_top_k(logits, labels, 1)
saver = tf.train.Saver(tf.all_variables())
init = tf.initialize_all_variables()
sess = tf.Session(config=tf.ConfigProto(
log_device_placement=FLAGS.log_device_placement))
sess.run(init)
tf.train.start_queue_runners(sess=sess)
true_count = 0
for step in xrange(FLAGS.max_steps):
start_time = time.time()
_, loss_value, precisions = sess.run([train_op, loss, top_k_op])
true_count += np.sum(precisions)
if step % 10 == 0:
checkpoint_path = os.path.join(FLAGS.train_dir, 'model.ckpt')
saver.save(sess, checkpoint_path, global_step=step)
duration = time.time() - start_time
print(' step %d, loss = %.3f, acc = %.3f, dur = %.2f' %
(step, loss_value, true_count/(FLAGS.batch_size*10), duration))
true_count = 0
def tower_loss(scope):
"""Calculate the total loss on a single tower running the CIFAR model.
Args:
scope: unique prefix string identifying the CIFAR tower, e.g. 'tower_0'
Returns:
Tensor of shape [] containing the total loss for a batch of data
"""
# Get images and labels for CIFAR-10.
images, labels = cifar10.distorted_inputs()
# Build inference Graph.
logits = cifar10.inference(images)
# Build the portion of the Graph calculating the losses. Note that we will
# assemble the total_loss using a custom function below.
_ = cifar10.loss(logits, labels)
# Assemble all of the losses for the current tower only.
losses = tf.get_collection('losses', scope)
# Calculate the total loss for the current tower.
total_loss = tf.add_n(losses, name='total_loss')
# Compute the moving average of all individual losses and the total loss.
loss_averages = tf.train.ExponentialMovingAverage(0.9, name='avg')
loss_averages_op = loss_averages.apply(losses + [total_loss])
# Attach a scalar summmary to all individual losses and the total loss; do the
# same for the averaged version of the losses.
for l in losses + [total_loss]:
# Remove 'tower_[0-9]/' from the name in case this is a multi-GPU training
# session. This helps the clarity of presentation on tensorboard.
loss_name = re.sub('%s_[0-9]*/' % cifar10.TOWER_NAME, '', l.op.name)
# Name each loss as '(raw)' and name the moving average version of the loss
# as the original loss name.
tf.scalar_summary(loss_name +' (raw)', l)
tf.scalar_summary(loss_name, loss_averages.average(l))
with tf.control_dependencies([loss_averages_op]):
total_loss = tf.identity(total_loss)
return total_loss
def tower_loss(scope):
images, labels = cifar10.distorted_inputs()
logits = cifar10.inference(images)
_ = cifar10.loss(logits, labels)
losses = tf.get_collection('losses', scope)
total_loss = tf.add_n(losses, name = 'total_loss')
return total_loss
def train():
with tf.Graph().as_default():
global_step = tf.contrib.framework.get_or_create_global_step()
# Get images and labels for training data.
# Force input pipeline to CPU:0 to avoid operations sometimes ending up on
# GPU and resulting in a slow down.
with tf.device('/cpu:0'):
images, labels = cifar10.distorted_inputs(
custom_data_dir="./train_data")
# Build a Graph that computes the logits predictions from the
# inference model.
# ccen: there're 128 classes instead of 10
logits = cifar10.inference(images, 128)
# Calculate loss.
loss = cifar10.loss(logits, labels)
# Build a Graph that trains the model with one batch of examples and
# updates the model parameters.
train_op = cifar10.train(loss, global_step)
class _LoggerHook(tf.train.SessionRunHook):
"""Logs loss and runtime."""
def begin(self):
self._step = -1
self._start_time = time.time()
def before_run(self, run_context):
self._step += 1
return tf.train.SessionRunArgs(loss) # Asks for loss value.
def after_run(self, run_context, run_values):
if self._step % FLAGS.log_frequency == 0:
current_time = time.time()
duration = current_time - self._start_time
self._start_time = current_time
loss_value = run_values.results
examples_per_sec = FLAGS.log_frequency * FLAGS.batch_size / duration
sec_per_batch = float(duration / FLAGS.log_frequency)
format_str = (
'%s: step %d, loss = %.2f (%.1f examples/sec; %.3f '
'sec/batch)')
print(format_str % (datetime.now(), self._step, loss_value,
examples_per_sec, sec_per_batch))
with tf.train.MonitoredTrainingSession(
checkpoint_dir=FLAGS.train_dir,
hooks=[
tf.train.StopAtStepHook(last_step=FLAGS.max_steps),
tf.train.NanTensorHook(loss),
_LoggerHook()
],
config=tf.ConfigProto(log_device_placement=FLAGS.
log_device_placement)) as mon_sess:
while not mon_sess.should_stop():
mon_sess.run(train_op)
def train():
"""Train CIFAR-10 for a number of steps."""
with tf.Graph().as_default():
global_step = tf.contrib.framework.get_or_create_global_step()
# Get images and labels for CIFAR-10.
# Force input pipeline to CPU:0 to avoid operations sometimes ending up on
# GPU and resulting in a slow down.
with tf.device('/cpu:0'):
images, labels = cifar10.distorted_inputs()
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(images)
# Calculate loss.
loss = cifar10.loss(logits, labels)
# Build a Graph that trains the model with one batch of examples and
# updates the model parameters.
train_op = cifar10.train(loss, global_step)
saver = tf.train.Saver(tf.global_variables())
# Build an initialization operation to run below.
init = tf.global_variables_initializer()
# Start running operations on the Graph. allow_soft_placement must be set to
# True to build towers on GPU, as some of the ops do not have GPU
# implementations.
sess = tf.Session(config=tf.ConfigProto(
allow_soft_placement=True,
log_device_placement=FLAGS.log_device_placement))
sess.run(init)
# Start the queue runners.
tf.train.start_queue_runners(sess=sess)
for step in xrange(FLAGS.max_steps):
start_time = time.time()
_, loss_value = sess.run([train_op, loss])
duration = time.time() - start_time
assert not np.isnan(loss_value), 'Model diverged with loss = NaN'
if step % 10 == 0:
num_examples_per_step = FLAGS.batch_size
examples_per_sec = num_examples_per_step / duration
sec_per_batch = duration
format_str = ('%s: step %d, loss = %.2f (%.1f examples/sec; %.3f '
'sec/batch)')
print (format_str % (datetime.now(), step, loss_value,
examples_per_sec, sec_per_batch))
# Save the model checkpoint periodically.
if step % 1000 == 0 or (step + 1) == FLAGS.max_steps:
checkpoint_path = os.path.join(FLAGS.train_dir, 'model.ckpt')
saver.save(sess, checkpoint_path, global_step=step)
evaluate()
def tower_loss(scope):
"""Calculate the total loss on a single tower running the CIFAR model.
Args:
scope: unique prefix string identifying the CIFAR tower, e.g. 'tower_0'
Returns:
Tensor of shape [] containing the total loss for a batch of data
"""
# Get images and labels for CIFAR-10.
images, labels = cifar10.distorted_inputs()
# Build inference Graph.
logits = cifar10.inference(images)
# Build the portion of the Graph calculating the losses. Note that we will
# assemble the total_loss using a custom function below.
_ = cifar10.loss(logits, labels)
# Assemble all of the losses for the current tower only.
losses = tf.get_collection('losses', scope)
# Calculate the total loss for the current tower.
total_loss = tf.add_n(losses, name='total_loss')
# Compute the moving average of all individual losses and the total loss.
# loss_averages = tf.train.ExponentialMovingAverage(0.9, name='avg')
# loss_averages_op = loss_averages.apply(losses + [total_loss])
# Attach a scalar summary to all individual losses and the total loss; do the
# same for the averaged version of the losses.
# for l in losses + [total_loss]:
# Remove 'tower_[0-9]/' from the name in case this is a multi-GPU training
# session. This helps the clarity of presentation on tensorboard.
# loss_name = re.sub('%s_[0-9]*/' % cifar10.TOWER_NAME, '', l.op.name)
# Name each loss as '(raw)' and name the moving average version of the loss
# as the original loss name.
# tf.scalar_summary(loss_name +' (raw)', l)
# tf.scalar_summary(loss_name, loss_averages.average(l))
# with tf.control_dependencies([loss_averages_op]):
# total_loss = tf.identity(total_loss)
return total_loss
def train():
with tf.Graph().as_default():
images, labels = cifar10.distorted_inputs()
logits = cifar10.inference(images)
loss = cifar10.loss(logits, labels)
global_step = tf.Variable(0, trainable=False)
train_op = cifar10.train(loss, global_step=global_step)
summary_op = tf.merge_all_summaries()
init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)
tf.train.start_queue_runners(sess=sess)
summary_writer = tf.train.SummaryWriter(FLAGS.train_dir, sess.graph)
trainable_var = tf.trainable_variables()
print([v.name for v in trainable_var])
var = trainable_var[0]
res = sess.run(var)
print(type(res))
def train():
"""Train CIFAR-10 for a number of steps."""
with tf.Graph().as_default():
# with tf.variable_scope("cifar10", reuse=tf.AUTO_REUSE) as scope:
global_step = tf.train.get_or_create_global_step()
# Get images and labels for CIFAR-10.
train_flag = tf.placeholder(tf.bool, shape = ())
trX, trY = cifar10.distorted_inputs()
teX, teY = cifar10.inputs(eval_data = True)
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(trX)
# Calculate accuracy
tr_acc = cifar10.accuracy(logits, trY)[1]
print(tr_acc, "tr_acc\n")
# tr_acc_sum = tf.summary.scalar('train/accuracy', tr_acc)
# Calculate loss.
loss = cifar10.loss(logits, trY)
# Build a Graph that trains the model with one batch of examples and
# updates the model parameters.
train_op = cifar10.train(loss, global_step)
tf.get_variable_scope().reuse_variables()
eval_logits = cifar10.inference(teX)
te_acc = cifar10.accuracy(eval_logits, teY)[1]
print(te_acc, "te_acc\n")
# te_acc_sum = tf.summary.scalar('test/accuracy', te_acc)
accuracy = tf.cond(train_flag, lambda: tr_acc, lambda: te_acc)
tf.summary.scalar("accuracy", accuracy)
merged = tf.summary.merge_all()
train_writer = tf.summary.FileWriter('tmp/cifar10/train')
test_writer = tf.summary.FileWriter('tmp/cifar10/test')
print("Training Starts")
# Configs
config = tf.ConfigProto(log_device_placement=FLAGS.log_device_placement)
config.gpu_options.allow_growth=True
mon_sess = tf.train.MonitoredTrainingSession(
hooks=[tf.train.StopAtStepHook(last_step=FLAGS.max_steps),
tf.train.NanTensorHook(loss)],config=config)
step = -1
while not mon_sess.should_stop():
step += 1
_,loss_value = mon_sess.run([train_op,loss])
if step % FLAGS.log_frequency == 0:
tr_acc,summary = mon_sess.run([accuracy,merged], feed_dict = {train_flag : True})
train_writer.add_summary(summary, step)
te_acc, summary = mon_sess.run([accuracy, merged], feed_dict = {train_flag : False})
test_writer.add_summary(summary, step)
format_str = ('%s: step %d, loss = %.2f, test accuracy = %.2f, train accuracy = %.2f')
print (format_str % (datetime.now(), step, loss_value, te_acc, tr_acc))
def train():
"""Train CIFAR-10 for a number of steps."""
g1 = tf.Graph()
with g1.as_default():
global_step = tf.contrib.framework.get_or_create_global_step()
# Get images and labels for CIFAR-10.
images, labels = cifar10.distorted_inputs()
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(images)
# Calculate loss.
loss = cifar10.loss(logits, labels)
grads = cifar10.train_part1(loss, global_step)
only_gradients = [g for g, _ in grads]
only_vars = [v for _, v in grads]
placeholder_gradients = []
#with tf.device("/gpu:0"):
for grad_var in grads:
placeholder_gradients.append(
(tf.placeholder('float',
shape=grad_var[0].get_shape()), grad_var[1]))
feed_dict = {}
for i, grad_var in enumerate(grads):
feed_dict[placeholder_gradients[i][0]] = np.zeros(
placeholder_gradients[i][0].shape)
train_op = cifar10.train_part2(global_step, placeholder_gradients)
sess = tf.Session()
sess.run(tf.global_variables_initializer())
feeds = []
print("Reached here")
for i, grad_var in enumerate(grads):
feeds.append(placeholder_gradients[i][0])
# Partial Run
print("Reached here", len(feeds))
for x in feeds:
print(x, )
h = sess.partial_run_setup([only_gradients, train_op], feeds)
print("Reached here")
for i in xrange(10):
res_grads = sess.partial_run(h,
only_gradients,
feed_dict=feed_dict)
feed_dict = {}
for i, grad_var in enumerate(res_grads):
feed_dict[placeholder_gradients[i][0]] = res_grads[i]
res_train_op = sess.partial_run(h, train_op, feed_dict=feed_dict)
def train():
"""Train CIFAR-10 for a number of steps."""
with tf.Graph().as_default():
global_step = tf.Variable(0, trainable=False)
images, labels = cifar10.distorted_inputs()
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(images)
# Calculate loss.
loss = cifar10.loss(logits, labels)
# Build a Graph that trains the model with one batch of examples and
# updates the model parameters.
train_op = cifar10.train(loss, global_step)
# Create a saver.
saver = tf.train.Saver(tf.all_variables())
# Build the summary operation based on the TF collection of Summaries.
summary_op = tf.merge_all_summaries()
# Build an initialization operation to run below.
init = tf.initialize_all_variables()
# Start running operations on the Graph.
sess = tf.Session(config=tf.ConfigProto(log_device_placement=FLAGS.log_device_placement))
sess.run(init)
# Start the queue runners.
tf.train.start_queue_runners(sess=sess)
summary_writer = tf.train.SummaryWriter(FLAGS.train_dir, sess.graph)
for step in xrange(FLAGS.max_steps):
start_time = time.time()
_, loss_value = sess.run([train_op, loss])
duration = time.time() - start_time
assert not np.isnan(loss_value), "Model diverged with loss = NaN"
if step % 10 == 0:
num_examples_per_step = FLAGS.batch_size
examples_per_sec = num_examples_per_step / duration
sec_per_batch = float(duration)
format_str = "%s: step %d, loss = %.2f (%.1f examples/sec; %.3f " "sec/batch)"
print(format_str % (datetime.now(), step, loss_value, examples_per_sec, sec_per_batch))
if step % 100 == 0:
summary_str = sess.run(summary_op)
summary_writer.add_summary(summary_str, step)
# Save the model checkpoint periodically.
if step % 1000 == 0 or (step + 1) == FLAGS.max_steps:
checkpoint_path = os.path.join(FLAGS.train_dir, "model.ckpt")
saver.save(sess, checkpoint_path, global_step=step)
def train():
"""Train CIFAR-10 for a number of steps."""
with tf.Graph().as_default():
global_step = tf.contrib.framework.get_or_create_global_step()
# Get images and labels for CIFAR-10.
images, labels = cifar10.distorted_inputs()
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(images)
# Calculate loss.
loss = cifar10.loss(logits, labels)
# Build a Graph that trains the model with one batch of examples and
# updates the model parameters.
train_op = cifar10.train(loss, global_step)
class _LoggerHook(tf.train.SessionRunHook):
"""Logs loss and runtime."""
def begin(self):
self._step = -1
def before_run(self, run_context):
self._step += 1
self._start_time = time.time()
return tf.train.SessionRunArgs(loss) # Asks for loss value.
def after_run(self, run_context, run_values):
duration = time.time() - self._start_time
loss_value = run_values.results
if self._step % 10 == 0:
num_examples_per_step = FLAGS.batch_size
examples_per_sec = num_examples_per_step / duration
sec_per_batch = float(duration)
format_str = (
'%s: step %d, loss = %.2f (%.1f examples/sec; %.3f '
'sec/batch)')
print(format_str % (datetime.now(), self._step, loss_value,
examples_per_sec, sec_per_batch))
global step_no
step_no = self._step
with tf.train.MonitoredTrainingSession(
checkpoint_dir=FLAGS.train_dir,
hooks=[
tf.train.StopAtStepHook(last_step=FLAGS.max_steps),
tf.train.NanTensorHook(loss),
_LoggerHook()
],
config=tf.ConfigProto(
log_device_placement=FLAGS.log_device_placement,
inter_op_parallelism_threads=4,
intra_op_parallelism_threads=0)) as mon_sess:
while not mon_sess.should_stop():
mon_sess.run(train_op)
"""run_options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE)
def train():
""" 多步训练CIFAR-10数据集 """
with tf.Graph().as_default():
global_step = tf.contrib.framework.get_or_create_global_step()
# 指定 CPU:0 进行数据的变形处理
with tf.device('/cpu:0'):
images, labels = cifar10.distorted_inputs()
# inference 前向预测，定义在cifar10.py文件
logits = cifar10.inference(images)
# loss.
loss = cifar10.loss(logits, labels)
# train
train_op = cifar10.train(loss, global_step)
class _LoggerHook(tf.train.SessionRunHook):
"""记录 loss 和 runtime."""
def __init__(self):
self._start_time = time.time()
self._step = -1
def begin(self):
pass
def before_run(self, run_context):
self._step += 1
return tf.train.SessionRunArgs(
loss) # 将loss 操作加到Session.run()调用
def after_run(self, run_context, run_values):
if self._step % FLAGS.log_frequency == 0:
current_time = time.time()
duration = current_time - self._start_time
self._start_time = current_time
loss_value = run_values.results
examples_per_sec = FLAGS.log_frequency * FLAGS.batch_size / duration
sec_per_batch = float(duration / FLAGS.log_frequency)
format_str = '%s: step %d, loss = %.2f (%.1f examples/sec; %.3f ' 'sec/batch)'
print(format_str % (datetime.now(), self._step, loss_value,
examples_per_sec, sec_per_batch))
with tf.train.MonitoredTrainingSession(
checkpoint_dir=FLAGS.train_dir,
hooks=[
tf.train.StopAtStepHook(
last_step=FLAGS.max_steps), # 在运行指定步长后停止
tf.train.NanTensorHook(loss), # 监视loss，如果loss==NaN,停止训练
_LoggerHook()
],
# log_device_placement是否打印设备分配日志
config=tf.ConfigProto(log_device_placement=FLAGS.
log_device_placement)) as mon_sess:
while not mon_sess.should_stop():
mon_sess.run(train_op)
def tower_loss(scope):
"""Calculate the total loss on a single tower running the CIFAR model.
Args:
scope: unique prefix string identifying the CIFAR tower, e.g. 'tower_0'
Returns:
Tensor of shape [] containing the total loss for a batch of data
"""
# Get images and labels for CIFAR-10.
images, labels = cifar10.distorted_inputs()
# Build inference Graph.
if FLAGS.model == 'small_cnn':
logits = cifar10.inference_small(images)
elif FLAGS.model == 'resnet_18':
logits = cifar10.inference_resnet(images, is_train=True)
elif FLAGS.model == 'lcnn_resnet_18':
logits = cifar10.inference_lcnn_resnet(images, is_train=True)
elif FLAGS.model == 'resnet_lcnn_hybrid':
logits = cifar10.inference_resnet_lcnn_hybrid(images, is_train=True)
else:
raise Exception('Unknown model type: {}'.format(FLAGS.model))
# Build the portion of the Graph calculating the losses. Note that we will
# assemble the total_loss using a custom function below.
_ = cifar10.loss(logits, labels)
# Assemble all of the losses for the current tower only.
losses = tf.get_collection('losses', scope)
# Calculate the total loss for the current tower.
total_loss = tf.add_n(losses, name='total_loss')
l1_loss = tf.add_n([0.] + [l for l in losses if l.op.name.find('l1_loss') != -1],
name='l1_sum_loss')
l2_loss = tf.add_n([0.] + [l for l in losses if l.op.name.find('weight_loss') != -1],
name='l2_sum_loss')
all_losses = losses + [total_loss, l1_loss, l2_loss]
# Compute the moving average of all individual losses and the total loss.
loss_averages = tf.train.ExponentialMovingAverage(0.9, name='avg')
loss_averages_op = loss_averages.apply(all_losses)
# Attach a scalar summary to all individual losses and the total loss; do the
# same for the averaged version of the losses.
for l in all_losses:
# Remove 'tower_[0-9]/' from the name in case this is a multi-GPU training
# session. This helps the clarity of presentation on tensorboard.
loss_name = re.sub('%s_[0-9]*/' % cifar10.TOWER_NAME, '', l.op.name)
# Name each loss as '(raw)' and name the moving average version of the loss
# as the original loss name.
tf.scalar_summary(loss_name +' (raw)', l)
tf.scalar_summary(loss_name, loss_averages.average(l))
update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
updates = tf.group(*update_ops) # For correct batch norm execution
with tf.control_dependencies([updates, loss_averages_op]):
total_loss = tf.identity(total_loss)
return total_loss
def build_model_func():
images, labels = cifar10.distorted_inputs()
logits = cifar10.inference(images)
loss = cifar10.loss(logits, labels)
return loss
def train():
"""Train CIFAR-10 for a number of steps."""
with tf.Graph().as_default():
global_step = tf.train.get_or_create_global_step()
# Get images and labels for CIFAR-10.
# Force input pipeline to CPU:0 to avoid operations sometimes ending up on
# GPU and resulting in a slow down.
with tf.device('/cpu:0'):
images, labels = cifar10.distorted_inputs()
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(images)
# Calculate loss.
loss = cifar10.loss(logits, labels)
# Build a Graph that trains the model with one batch of examples and
# updates the model parameters.
train_op = cifar10.train(loss, global_step)
class _LoggerHook(tf.train.SessionRunHook):
"""Logs loss and runtime."""
def begin(self):
self._step = -1
self._start_time = time.time()
def before_run(self, run_context):
self._step += 1
return tf.train.SessionRunArgs(loss) # Asks for loss value.
def after_run(self, run_context, run_values):
if self._step % FLAGS.log_frequency == 0:
current_time = time.time()
duration = current_time - self._start_time
self._start_time = current_time
loss_value = run_values.results
examples_per_sec = FLAGS.log_frequency * FLAGS.batch_size / duration
sec_per_batch = float(duration / FLAGS.log_frequency)
format_str = ('%s: step %d, loss = %.2f (%.1f examples/sec; %.3f '
'sec/batch)')
print (format_str % (datetime.now(), self._step, loss_value,
examples_per_sec, sec_per_batch))
with tf.train.MonitoredTrainingSession(
checkpoint_dir=FLAGS.train_dir,
hooks=[tf.train.StopAtStepHook(last_step=FLAGS.max_steps),
tf.train.NanTensorHook(loss),
_LoggerHook()],
config=tf.ConfigProto(
log_device_placement=FLAGS.log_device_placement)) as mon_sess:
while not mon_sess.should_stop():
mon_sess.run(train_op)
def tower_loss(scope):
"""Calculate the total loss on a single tower running the CIFAR model.
Args:
scope: unique prefix string identifying the CIFAR tower, e.g. 'tower_0'
Returns:
Tensor of shape [] containing the total loss for a batch of data
"""
# Build a Graph that computes the logits predictions from the
# inference model.
if tfFLAGS.network == 1:
images, labels = cifar10.distorted_inputs()
logits, fc1_w, fc1_b, fc2_w, fc2_b = MyModel.inference(images)
else:
images, labels = cifar10.distorted_inputs()
logits, fc1_w, fc1_b, fc2_w, fc2_b = MyModel2.inference(images)
# Calculate loss.
loss = cifar10.loss(logits, labels)
# L2 regularization for the fully connected parameters.
regularizers = (tf.nn.l2_loss(fc1_w) + tf.nn.l2_loss(fc1_b) + tf.nn.l2_loss(fc2_w) + tf.nn.l2_loss(fc2_b))
# Add the regularization term to the loss.
loss += 5e-4 * regularizers
# Build the portion of the Graph calculating the losses. Note that we will
# assemble the total_loss using a custom function below.
_ = cifar10.loss(logits, labels)
# Assemble all of the losses for the current tower only.
losses = tf.get_collection('losses', scope)
# Calculate the total loss for the current tower.
total_loss = tf.add_n(losses, name='total_loss')
# Attach a scalar summary to all individual losses and the total loss; do the
# same for the averaged version of the losses.
for l in losses + [total_loss]:
# Remove 'tower_[0-9]/' from the name in case this is a multi-GPU training
# session. This helps the clarity of presentation on tensorboard.
loss_name = re.sub('%s_[0-9]*/' % tfFLAGS.TOWER_NAME, '', l.op.name)
tf.summary.scalar(loss_name, l)
return total_loss
def train():
"""Train CIFAR-10 for a number of steps."""
with tf.Graph().as_default():
global_step = tf.contrib.framework.get_or_create_global_step()
# Get images and labels for CIFAR-10.
images, labels = cifar10.distorted_inputs()
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(images)
# Calculate loss.
loss = cifar10.loss(logits, labels)
# Build a Graph that trains the model with one batch of examples and
# updates the model parameters.
train_op = cifar10.train(loss, global_step)
class _LoggerHook(tf.train.SessionRunHook):
"""Logs loss and runtime."""
def begin(self):
self._step = -1
def before_run(self, run_context):
self._step += 1
self._start_time = time.time()
return tf.train.SessionRunArgs(loss) # Asks for loss value.
def after_run(self, run_context, run_values):
duration = time.time() - self._start_time
loss_value = run_values.results
if self._step % 10 == 0:
num_examples_per_step = FLAGS.batch_size
examples_per_sec = num_examples_per_step / duration
sec_per_batch = float(duration)
format_str = ('%s: step %d, loss = %.2f (%.1f examples/sec; %.3f '
'sec/batch)')
print (format_str % (datetime.now(), self._step, loss_value,
examples_per_sec, sec_per_batch))
global step_no
step_no = self._step
with tf.train.MonitoredTrainingSession(
checkpoint_dir=FLAGS.train_dir,
hooks=[tf.train.StopAtStepHook(last_step=FLAGS.max_steps),
tf.train.NanTensorHook(loss),
_LoggerHook()],
config=tf.ConfigProto(
log_device_placement=FLAGS.log_device_placement, inter_op_parallelism_threads=4,intra_op_parallelism_threads=0)) as mon_sess:
while not mon_sess.should_stop():
mon_sess.run(train_op)
"""run_options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE)
def train():
with g.as_default():
global_step = tf.train.get_or_create_global_step()
with tf.device('/cpu:0'):
images, labels = cifar10.distorted_inputs()
logits = cifar10.inference3model(images)
stats_graph(g)
loss = cifar10.loss(logits, labels)
train_op = cifar10.train(loss, global_step)
class _LoggerHook(tf.train.SessionRunHook):
def begin(self):
self._step = -1
self._start_time = time.time()
def before_run(self, run_context):
self._step += 1
# 在这里返回你想在运行过程中产看的信息，以list的形式传递,如:[loss, accuracy]
return tf.train.SessionRunArgs(loss)
def after_run(self, run_context, run_values):
if self._step % FLAGS.log_frequency == 0:
current_time = time.time()
duration = current_time - self._start_time
self._start_time = current_time
# results返回的是上面before_run()的返回结果，上面是loss所以返回loss
# 如若上面返回的是个list,则返回的也是个list
loss_value = run_values.results
examples_per_sec = FLAGS.log_frequency * FLAGS.batch_size / duration
sec_per_batch = float(duration / FLAGS.log_frequency)
print(
'%s: step %d, loss = %.2f (%.1f examples/sec; %.3f sec/batch)'
% (datetime.now(), self._step, loss_value,
examples_per_sec, sec_per_batch))
with tf.train.MonitoredTrainingSession(
checkpoint_dir=FLAGS.train_dir,
hooks=[
tf.train.StopAtStepHook(last_step=FLAGS.max_steps),
tf.train.NanTensorHook(loss),
_LoggerHook()
],
save_checkpoint_secs=60,
config=tf.ConfigProto(log_device_placement=FLAGS.
log_device_placement)) as mon_sess:
while not mon_sess.should_stop():
mon_sess.run(train_op)
def train():
"""Train CIFAR-10 for a number of steps."""
with tf.Graph().as_default():
global_step = tf.contrib.framework.get_or_create_global_step()
#为CIFAR-10获取（图像，label）样本对
images, labels = cifar10.distorted_inputs()
#创建图Graph用于从推断模型计算logits predictions
logits = cifar10.inference(images)
#计算损失
loss = cifar10.loss(logits, labels)
#创建图：用一个batch的样本来训练模型
#更新模型参数
train_op = cifar10.train(loss, global_step)
class _LoggerHook(tf.train.SessionRunHook):
"""Logs loss and runtime"""
def begin(self):
self._step = -1
self._start_time = time.time()
def before_run(self, run_context):
self._step += 1
return tf.train.SessionRunArgs(loss) #请求loss值
def after_run(self, run_context, run_values):
if self._step % FLAGS.log_frequency == 0:
current_time = time.time()
duration = current_time - _start_time
self._start_time = current_time
loss_value = run_values.results
examples_per_sec = FLAGS.batch_size * log_frequency / duration
sec_per_batch = float(duration / log_frequency)
format_str = (
'%s: step %d, loss = %.2f(%.1f examples/sec, %.3f sec/batch)'
)
print(format_str % (datetime.now(), self._step, loss_value,
examples_per_sec, sec_per_batch))
with tf.train.MonitoredTrainingSession(
checkpoint_dir=FLAGS.train_dir,
hooks=[
tf.train.StopAtStepHook(last_step=FLAGS.max_steps),
tf.train.NanTensorHook(loss),
_LoggerHook()
],
config=tf.ConfigProto(
log_device_placement=FLAGS.log_device_placement)) as mon_sess:
while not mon_sess.should_stop():
mon_sess.run(train_op)
def train():
""" CIFAR10训练函数 """
with tf.Graph().as_default():
global_step = tf.contrib.framework.get_or_create_global_step()
# 从数据中读取图像和标签.
images, labels = cifar10.distorted_inputs()
# 建立一个图用来计算模型的预测结果
logits = cifar10.inference(images)
# 计算误差
loss = cifar10.loss(logits, labels)
# 构建图用一个batch的训练数据来训练模型并更新参数
train_op = cifar10.train(loss, global_step)
class _LoggerHook(tf.train.SessionRunHook):
""" 记录误差和运行时间 """
def begin(self):
self._step = -1
self._start_time = time.time()
def before_run(self, run_context):
self._step += 1
return tf.train.SessionRunArgs(loss) # 计算误差值
def after_run(self, run_context, run_values):
if self._step % FLAGS.log_frequency == 0:
current_time = time.time()
duration = current_time - self._start_time
self._start_time = current_time
loss_value = run_values.results
examples_per_sec = FLAGS.log_frequency * FLAGS.batch_size / duration
sec_per_batch = float(duration / FLAGS.log_frequency)
format_str = (
'%s: step %d, loss = %.5f (%.1f examples/sec; %.3f sec/batch)'
)
print(format_str % (datetime.now(), self._step, loss_value,
examples_per_sec, sec_per_batch))
with tf.train.MonitoredTrainingSession(
checkpoint_dir=FLAGS.train_dir,
hooks=[
tf.train.StopAtStepHook(last_step=FLAGS.max_steps),
tf.train.NanTensorHook(loss),
_LoggerHook()
],
config=tf.ConfigProto(log_device_placement=FLAGS.
log_device_placement)) as mon_sess:
while not mon_sess.should_stop():
mon_sess.run(train_op)
def train():
"""
Train CIFAR-10 for a number of steps
:return:
"""
with tf.Graph().as_default():
global_step = tf.train.get_or_create_global_step()
# 获取CIFAR-10的images和labels
# 让CPU专注流输入，避免GPU处理完，导致停顿
with tf.device('/cpu:0'):
images, labels = cifar10.distorted_inputs()
# 预测结果 以及 loss
logits = cifar10.inference(images)
loss = cifar10.loss(logits, labels)
train_op = cifar10.train(loss, global_step)
class _LoggerHook(tf.train.SessionRunHook):
""" log loss and runtime. """
def begin(self):
self._step = -1
self._start_time = time.time()
def before_run(self, run_context):
self._step += 1
return tf.train.SessionRunArgs(loss)
def after_run(self, run_context, run_values):
if self._step % FLAGS.log_frequency == 0:
current_time = time.time()
duration = current_time - self._start_time
self._start_time = current_time
loss_value = run_values.results
examples_per_sec = FLAGS.log_frequency * FLAGS.batch_size / duration
sec_per_batch = float(duration / FLAGS.log_frequency)
format_str = ('%s: step %d, loss = %.2f (%.1f examples/sec; %.3f '
'sec/batch)')
print(format_str % (datetime.now(), self._step, loss_value,
examples_per_sec, sec_per_batch))
with tf.train.MonitoredTrainingSession(
checkpoint_dir=FLAGS.train_dir,
hooks=[tf.train.StopAtStepHook(last_step=FLAGS.max_steps),
tf.train.NanTensorHook(loss),
_LoggerHook()],
config=tf.ConfigProto(log_device_placement=FLAGS.log_device_placement)) as mon_sess:
while not mon_sess.should_stop():
mon_sess.run(train_op)
def train():
"""Train CIFAR-10 for a number of steps."""
with tf.Graph().as_default():
global_step = tf.train.get_or_create_global_step()
# Get images and labels for CIFAR-10.
# Force input pipeline to CPU:0 to avoid operations sometimes ending up on
# GPU and resulting in a slow down.
with tf.device('/cpu:0'):
images, labels = cifar10.distorted_inputs()
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(images)
# Calculate loss.
loss = cifar10.loss(logits, labels)
# Build a Graph that trains the model with one batch of examples and
# updates the model parameters.
train_op = cifar10.train(loss, global_step)
class _LoggerHook(tf.train.SessionRunHook):
"""Logs loss and runtime."""
def begin(self):
self._step = -1
self._start_time = time.time()
def before_run(self, run_context):
self._step += 1
return tf.train.SessionRunArgs(loss) # Asks for loss value.
def after_run(self, run_context, run_values):
if self._step % FLAGS.log_frequency == 0:
current_time = time.time()
self._start_time = current_time
for step in range(0, FLAGS.max_steps + 1, FLAGS.log_frequency):
print(str(step))
with tf.train.MonitoredTrainingSession(
checkpoint_dir=FLAGS.train_dir,
hooks=[
tf.train.StopAtStepHook(last_step=step),
tf.train.NanTensorHook(loss),
_LoggerHook()
],
config=tf.ConfigProto(log_device_placement=FLAGS.
log_device_placement)) as mon_sess:
while not mon_sess.should_stop():
mon_sess.run(train_op)
# evaluate test data
cifar10_eval.evaluate()
# evaluate train data
evaluate()
def train():
"""Train CIFAR-10 for a number of steps."""
with tf.Graph().as_default():
global_step = tf.contrib.framework.get_or_create_global_step()
# Get images and labels for CIFAR-10.
images, labels = cifar10.distorted_inputs()
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(images)
# Calculate loss.
loss = cifar10.loss(logits, labels)
# Build a Graph that trains the model with one batch of examples and
# updates the model parameters.
train_op = cifar10.train(loss, global_step)
class _LoggerHook(tf.train.SessionRunHook):
"""Logs loss and runtime."""
def begin(self):
self._step = -1
def before_run(self, run_context):
self._step += 1
self._start_time = time.time()
return tf.train.SessionRunArgs([loss, logits, labels]) # Asks for loss value.
def after_run(self, run_context, run_values):
duration = time.time() - self._start_time
loss_value, predictions, targets = run_values.results
if self._step % FLAGS.print_eval_interval == 0:
num_examples_per_step = FLAGS.batch_size
examples_per_sec = num_examples_per_step / duration
sec_per_batch = float(duration)
accuracy = 100.0 * np.mean(np.argmax(predictions, 1) == targets)
format_str = ('%s: step %d, loss = %.2f, train acc = %.1f%%, '
'(%.1f examples/sec; %.3f sec/batch)')
print (format_str % (datetime.now(), self._step, loss_value,
accuracy, examples_per_sec, sec_per_batch))
with tf.train.MonitoredTrainingSession(
checkpoint_dir=FLAGS.train_dir,
hooks=[tf.train.StopAtStepHook(last_step=FLAGS.max_steps),
tf.train.NanTensorHook(loss),
_LoggerHook()],
save_checkpoint_secs=FLAGS.save_checkpoint_secs,
config=tf.ConfigProto(
log_device_placement=FLAGS.log_device_placement)) as mon_sess:
while not mon_sess.should_stop():
mon_sess.run(train_op)
def train():
global updated_batch_size_num
global passed_info
global shall_update
ps_hosts = FLAGS.ps_hosts.split(',')
worker_hosts = FLAGS.worker_hosts.split(',')
print ('PS hosts are: %s' % ps_hosts)
print ('Worker hosts are: %s' % worker_hosts)
issync = FLAGS.sync
#server = tf.train.Server({'ps': ps_hosts, 'worker': worker_hosts},
# job_name = FLAGS.job_name,
# task_index=FLAGS.task_id)
cluster = tf.train.ClusterSpec({'ps': ps_hosts, 'worker': worker_hosts})
server = tf.train.Server(cluster, job_name=FLAGS.job_name, task_index=FLAGS.task_index)
# batchSizeManager = BatchSizeManager(FLAGS.batch_size, len(worker_hosts))
if FLAGS.job_name == 'ps':
# rpcServer = batchSizeManager.create_rpc_server(ps_hosts[0].split(':')[0])
# rpcServer.serve()
server.join()
elif FLAGS.job_name == "worker":
time.sleep(10)
# rpcClient = batchSizeManager.create_rpc_client(ps_hosts[0].split(':')[0])
is_chief = (FLAGS.task_index == 0)
if is_chief:
if tf.gfile.Exists(FLAGS.train_dir):
tf.gfile.DeleteRecursively(FLAGS.train_dir)
tf.gfile.MakeDirs(FLAGS.train_dir)
# modified by faye
# device_setter = tf.train.replica_device_setter(ps_tasks=1)
# with tf.device('/job:worker/task:%d' % FLAGS.task_index):
# with tf.device(device_setter):
# global_step = tf.Variable(0, trainable=False)
with tf.device(tf.train.replica_device_setter(
worker_device="/job:worker/task:%d" % FLAGS.task_index,
cluster=cluster
)):
global_step = tf.get_variable(
'global_step', [],
initializer=tf.constant_initializer(0), trainable=False)
decay_steps = 50000*350.0/FLAGS.batch_size
batch_size = tf.placeholder(dtype=tf.int32, shape=(), name='batch_size')
images, labels = cifar10.distorted_inputs(batch_size)
print('zx0')
print(images.get_shape().as_list())
def train():
with tf.Graph().as_default(), tf.device('/cpu:0'):
global_step = tf.get_variable('global_step', [],
initializer=tf.constant_initializer(0),
trainable=False)
num_batches_per_epoch = cifar10.NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN / batch_size
decay_steps = int(num_batches_per_epoch * cifar10.NUM_EPOCHS_PER_DECAY)
lr = tf.train.exponential_decay(cifar10.INITIAL_LEARNING_RATE,
global_step,
decay_steps,
cifar10.LEARNING_RATE_DECAY_FACTOR,
staircase=True)
opt = tf.train.GradientDescentOptimizer(lr)
images, labels = cifar10.distorted_inputs()
# 使用预加载的队列,使用这个会速度可以提高五倍，亲测
batch_queue = tf.contrib.slim.prefetch_queue.prefetch_queue(
[images, labels], capacity=2 * FLAGS.num_gpus)
tower_grads = []
for i in range(num_gpus):
with tf.device('/gpu:%d' % i):
with tf.name_scope('%s_%d' % (cifar10.TOWER_NAME, i)) as scope:
loss = tower_loss(scope, images, labels)
tf.get_variable_scope().reuse_variables()
grads = opt.compute_gradients(loss)
tower_grads.append(grads)
grads = average_gradients(tower_grads)
apply_gradient_op = opt.apply_gradients(grads, global_step=global_step)
saver = tf.train.Saver(tf.all_variables())
init = tf.global_variables_initializer()
sess = tf.Session(config=tf.ConfigProto(allow_soft_placement=True))
sess.run(init)
tf.train.start_queue_runners(sess=sess)
for step in range(max_steps):
start_time = time.time()
_, loss_value = sess.run([apply_gradient_op, loss])
duration = time.time() - start_time
if step % 10 == 0:
num_examples_per_step = batch_size * num_gpus
examples_per_sec = num_examples_per_step / duration
sec_per_batch = duration / num_gpus
format_str = ('step %d, loss= %.2f (%.1f example/sec; %.3f'
'sec/batch)')
print(format_str %
(step, loss_value, examples_per_sec, sec_per_batch))
if step % 1000 == 0 or (step + 1) == max_steps:
saver.save(sess, '/', global_step=step)
def train():
with tf.Graph().as_default():
images, labels = cifar10.distorted_inputs()
sess = tf.Session()
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(coord=coord, sess=sess)
for i in range(10):
im, label = sess.run([images, labels])
print(im.shape)
coord.request_stop()
coord.join(threads)
def train():
# todo:这句话啥意识
with tf.Graph().as_default():
# todo:啥意思
global_step = tf.train.get_or_create_global_step()
with tf.device('/cpu:0'):
images, labels = cifar10.distorted_inputs()
# 推测模型
logits = cifar10.inference(images)
loss = cifar10.loss(logits, labels)
train_op = cifar10.train(loss, global_step)
class _LoggerHook(tf.train.SessionRunHook):
"""
logs loss and runtime
"""
def begin(self):
self._step = -1
self._start_time = time.time()
def before_run(self, run_context):
self._step += 1
return tf.train.SessionRunArgs(loss)
def after_run(
self,
run_context, # pylint: disable=unused-argument
run_values):
if self._step % FLAGS.log_frequency == 0:
current_time = time.time()
duration = current_time - self._start_time
self._start_time = current_time
loss_value = run_values.results
example_per_sec = FLAGS.log_frequency * FLAGS.batch_size / duration
sec_per_batch = float(duration / FLAGS.log_frequency)
format_str = (
'%s: step %d,loss=%.2f (%.1f example/sec);%.3f sec/batch'
)
print(format_str)
with tf.train.MonitoredTrainingSession(
checkpoint_dir=FLAGS.train_dir,
hooks=[
tf.train.StopAtStepHook(last_step=FLAGS.max_steps),
tf.train.NanTensorHook(loss),
_LoggerHook()
],
config=tf.ConfigProto(log_device_placement=FLAGS.
log_device_placement)) as mon_sess:
while not mon_sess.should_stop():
mon_sess.run(train_op)
def run_training():
with tf.Graph().as_default(), tf.device('/gpu:0'):
global_step = tf.Variable(0, trainable=False)
with tf.device('/cpu:0'):
images, labels = cifar10.distorted_inputs()
images_placeholder, labels_placeholder = placeholder_inputs(
FLAGS.batch_size, cifar10.IMAGE_SIZE)
logits = cifar10.inference(images_placeholder)
losses_dict = cifar10.loss(logits, labels_placeholder)
moving_averages_op = cifar10.add_summaries_and_moving_avgs(
losses_dict, global_step)
lbfgs_optimizer = customized_optimizer.CustomizedOptimizerInterface(
global_step=global_step,
loss_dict=losses_dict,
data_fetches=[images, labels],
data_placeholders=(images_placeholder, labels_placeholder),
maxiter=FLAGS.max_steps)
saver = tf.train.Saver(tf.global_variables(), max_to_keep=25)
summary_op = tf.summary.merge_all()
init = tf.global_variables_initializer()
with tf.Session(config=tf.ConfigProto(
allow_soft_placement=True,
log_device_placement=FLAGS.log_device_placement,
gpu_options=tf.GPUOptions(
per_process_gpu_memory_fraction=0.4))) as sess:
sess.run(init)
coordinator = tf.train.Coordinator()
try:
threads = tf.train.start_queue_runners(sess=sess,
coord=coordinator)
lbfgs_optimizer.minimize(session=sess,
moving_averages_op=moving_averages_op,
summary_op=summary_op,
saver=saver,
step_callback=step_callback)
except Exception as e:
coordinator.request_stop(e)
coordinator.request_stop()
coordinator.join(threads, stop_grace_period_secs=10)
def train():
"""Train CIFAR-10 for a number of steps."""
sess = tf.InteractiveSession()
#sess = tf_debug.LocalCLIDebugWrapperSession(sess)
# Get images and labels for CIFAR-10.
with tf.device('/cpu:0'):
images, labels = cifar10.distorted_inputs()
# Define all the fixed point variables we will be using later
cifar10.initialize_fix_point_variables()
# Build a Graph that computes the logits predictions from the inference model
logits = cifar10.inference(images)
# Calculate loss.
loss = cifar10.loss(logits, labels)
# Build a Graph that trains the model with one batch of examples and
# updates the model parameters.
train_op = cifar10.train(loss, 0.05)
# Update fixed point conversion parameters when needed
update_fix_pt_ops = cifar10.update_fix_point_accuracy()
# Merge all the summaries and write them out to
# FLAGS.log_dir
merged_summary = tf.summary.merge_all()
train_writer = tf.summary.FileWriter(FLAGS.log_dir + '/train', sess.graph)
test_writer = tf.summary.FileWriter(FLAGS.log_dir + '/test')
# init all variables
tf.global_variables_initializer().run()
# create a saver for checkpoints
saver = tf.train.Saver(tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES))
# needed on interactive session so it doesn't hang
tf.train.start_queue_runners()
for i in range(FLAGS.max_steps):
summary, _ = sess.run([merged_summary, train_op])
train_writer.add_summary(summary, i) # summary
if (i % 10 == 0):
saver.save(sess, FLAGS.log_dir + '/checkpoint', global_step=i)
if (i % 5 == 0):
sess.run([update_fix_pt_ops])
print('Step: %s, Loss: %s' % (i, loss.eval()))
train_writer.close()
def train():
with tf.Graph().as_default():
global_step = tf.Variable(0, trainable=False)
images, labels = cifar10.distorted_inputs()
logits = cifar10.inference(images)
loss = cifar10.loss(logits, labels)
train_op = cifar10.train(loss, global_step)
saver = tf.train.Saver(tf.all_variables())
summary_op = tf.merge_all_summaries()
init = tf.initialize_all_variables()
sess = tf.Session(config=tf.ConfigProto(log_device_placement=FLAGS.log_device_placement))
sess.run(init)
tf.train.start_queue_runners(sess=sess)
summary_writer = tf.train.SummaryWriter(FLAGS.train_dir, graph_def=sess.graph_def)
for step in xrange(FLAGS.max_steps):
start_time = time.time()
_, loss_value = sess.run([train_op, loss])
duration = time.time() - start_time
assert not np.isnan(loss_value), "Model diverged with loss = NaN"
if step % 10 == 0:
num_examples_per_step = FLAGS.batch_size
examples_per_sec = num_examples_per_step / duration
sec_per_batch = float(duration)
format_str = "%s: step %d, loss = %.2f (%.1f examples/sec; %.3f sec/batch)"
print(format_str % (datetime.now(), step, loss_value, examples_per_sec, sec_per_batch))
if step % 100 == 0:
summary_str = sess.run(summary_op)
summary_writer.add_summary(summary_str, step)
if step % 1000 == 0 or (step + 1) == FLAGS.max_steps:
checkpoint_path = os.path.join(FLAGS.train_dir, "model.ckpt")
saver.save(sess, checkpoint_path, global_step=step)
def train_vgg16():
with tf.Graph().as_default():
image_size = 224 # 输入图像尺寸
# 生成随机数测试是否能跑通
#images = tf.Variable(tf.random_normal([batch_size, image_size, image_size, 3], dtype=tf.float32, stddev=1e-1))
with tf.device('/cpu:0'):
images, labels = cifar10.distorted_inputs()
keep_prob = tf.placeholder(tf.float32)
prediction,softmax,fc8,p = inference_op(images,keep_prob)
init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)
time_tensorflow_run(sess, prediction,{keep_prob:1.0}, "Forward")
# 用以模拟训练的过程
objective = tf.nn.l2_loss(fc8) # 给一个loss
grad = tf.gradients(objective, p) # 相对于loss的 所有模型参数的梯度
time_tensorflow_run(sess, grad, {keep_prob:0.5},"Forward-backward")
def train():
"""Train CIFAR-10 for a number of steps."""
with tf.Graph().as_default():
global_step = tf.Variable(0, trainable=False)
# Get images and labels for CIFAR-10.
images, labels = cifar10.distorted_inputs()
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(images)
# Calculate loss.
loss = cifar10.loss(logits, labels)
# Build a Graph that trains the model with one batch of examples and
# updates the model parameters.
train_op = cifar10.train(loss, global_step)
# Create a saver.
saver = tf.train.Saver(tf.all_variables())
# Build the summary operation based on the TF collection of Summaries.
summary_op = tf.merge_all_summaries()
# # Visualize conv1 features
# with tf.variable_scope('conv1') as scope_conv:
# #tf.get_variable_scope().reuse_variables()
# scope_conv.reuse_variables()
# weights = tf.get_variable('weights')
# grid_x = grid_y = 8 # to get a square grid for 64 conv1 features
# grid = put_kernels_on_grid (weights, (grid_y, grid_x))
# tf.image_summary('conv1/features', grid, max_images=1)
# Build an initialization operation to run below.
init = tf.initialize_all_variables()
# Start running operations on the Graph.
sess = tf.Session(config=tf.ConfigProto(
log_device_placement=FLAGS.log_device_placement))
sess.run(init)
# Start the queue runners.
tf.train.start_queue_runners(sess=sess)
summary_writer = tf.train.SummaryWriter(FLAGS.train_dir,
graph_def=sess.graph_def)
for step in xrange(FLAGS.max_steps):
start_time = time.time()
_, loss_value = sess.run([train_op, loss])
duration = time.time() - start_time
assert not np.isnan(loss_value), 'Model diverged with loss = NaN'
if step % 10 == 0:
num_examples_per_step = FLAGS.batch_size
examples_per_sec = num_examples_per_step / float(duration)
sec_per_batch = float(duration)
format_str = ('%s: step %d, loss = %.2f (%.1f examples/sec; %.3f '
'sec/batch)')
print (format_str % (datetime.now(), step, loss_value,
examples_per_sec, sec_per_batch))
if step % 100 == 0:
summary_str = sess.run(summary_op)
summary_writer.add_summary(summary_str, step)
# Save the model checkpoint periodically.
if step % 1000 == 0 or (step + 1) == FLAGS.max_steps:
checkpoint_path = os.path.join(FLAGS.train_dir, 'model.ckpt')
saver.save(sess, checkpoint_path, global_step=step)
def train(lambs):
"""Train CIFAR-10 for a number of steps."""
with tf.Graph().as_default():
global_step = tf.Variable(0, trainable=False)
# Get images and labels for CIFAR-10.
images, labels = cifar10.distorted_inputs()
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(images)
# Calculate loss.
lambs = tf.constant(lambs, dtype=tf.float32)
loss = infor.loss(logits, labels, lambs)
# Build a Graph that trains the model with one batch of examples and
# updates the model parameters.
train_op, lr_op = cifar10.train(loss, global_step)
# Create a saver.
saver = tf.train.Saver(tf.all_variables())
sess = tf.Session(config=tf.ConfigProto(
log_device_placement=FLAGS.log_device_placement))
# Build the summary operation based on the TF collection of Summaries.
summary_op = tf.merge_all_summaries()
ckpt = tf.train.get_checkpoint_state(FLAGS.checkpoint_dir)
if ckpt and ckpt.model_checkpoint_path:
# Restores from checkpoint
saver.restore(sess, ckpt.model_checkpoint_path)
# Assuming model_checkpoint_path looks something like:
# /my-favorite-path/cifar10_train/model.ckpt-0,
# extract global_step from it.
previous_step = ckpt.model_checkpoint_path.split('/')[-1].split('-')[-1]
print ('Start training from previous')
else:
print('Strart training from step 0')
previous_step = 0
init = tf.initialize_all_variables()
sess.run(init)
# Start the queue runners.
tf.train.start_queue_runners(sess=sess)
summary_writer = tf.train.SummaryWriter(FLAGS.train_dir,
graph_def=sess.graph_def)
step = previous_step
while step <= FLAGS.max_steps:
start_time = time.time()
_, loss_value, lr_value = sess.run([train_op, loss, lr_op])
duration = time.time() - start_time
assert not np.isnan(loss_value), 'Model diverged with loss = NaN'
if step % 10 == 0:
num_examples_per_step = FLAGS.batch_size
examples_per_sec = num_examples_per_step / duration
sec_per_batch = float(duration)
format_str = ('%s: step %d, loss = %.2f, lr_value = %.4f (%.1f examples/sec; %.3f '
'sec/batch)')
print(format_str % (datetime.now(), step, loss_value, lr_value,
examples_per_sec, sec_per_batch))
if step % 100 == 0:
summary_str = sess.run(summary_op)
summary_writer.add_summary(summary_str, step)
# Save the model checkpoint periodically.
if step % 1000 == 0 or (step + 1) == FLAGS.max_steps:
checkpoint_path = os.path.join(FLAGS.train_dir, 'model.ckpt')
saver.save(sess, checkpoint_path, global_step=step)
step += 1
def train():
"""Train a model for a number of steps."""
with tf.Graph().as_default():
global_step = tf.Variable(0, trainable=False)
# Get images and labels for a segmentation model.
images, labels, ground_truth = cifar10.distorted_inputs()
tf.histogram_summary('label_hist/with_ignore', labels)
tf.histogram_summary('label_hist/ground_truth', ground_truth)
# Build a Graph that computes the logits predictions from the
# inference model.
print("before inference")
print(images.get_shape())
logits, nr_params = cifar10.inference(images)
print("nr_params: "+str(nr_params) )
print("after inference")
# Calculate loss.
loss = cifar10.loss(logits, labels)
accuracy, precision, cat_accs = cifar10.accuracy(logits, ground_truth)
# Build a Graph that trains the model with one batch of examples and
# updates the model parameters.
train_op = cifar10.train(loss, global_step)
# Create a saver.
saver = tf.train.Saver(tf.all_variables())
# tf.image_summary('images2', images)
print (logits)
# tf.image_summary('predictions', logits)
# Build the summary operation based on the TF collection of Summaries.
summary_op = tf.merge_all_summaries()
# Build an initialization operation to run below.
init = tf.initialize_all_variables()
# Start running operations on the Graph.
sess = tf.Session(config=tf.ConfigProto(
log_device_placement=FLAGS.log_device_placement))
ckpt = tf.train.get_checkpoint_state(FLAGS.checkpoint_dir)
if ckpt and ckpt.model_checkpoint_path:
# Restores from checkpoint
saver.restore(sess, ckpt.model_checkpoint_path)
# Assuming model_checkpoint_path looks something like:
# /my-favorite-path/cifar10_train/model.ckpt-0,
# extract global_step from it.
global_step = int(ckpt.model_checkpoint_path.split('/')[-1].split('-')[-1])
else:
print('No checkpoint file found')
print('Initializing new model')
sess.run(init)
global_step = 0
# Start the queue runners.
tf.train.start_queue_runners(sess=sess)
summary_writer = tf.train.SummaryWriter(FLAGS.train_dir, sess.graph)
for step in xrange(global_step, FLAGS.max_steps):
start_time = time.time()
_, loss_value, accuracy_value, precision_value, cat_accs_val = sess.run([train_op,
loss,
accuracy,
precision,
cat_accs])
duration = time.time() - start_time
print (precision_value)
print (cat_accs_val)
assert not np.isnan(loss_value), 'Model diverged with loss = NaN'
#precision_value = [0 if np.isnan(p) else p for p in precision_value]
#print (precision_value)
if step % 10 == 0:
num_examples_per_step = FLAGS.batch_size
examples_per_sec = num_examples_per_step / duration
sec_per_batch = float(duration)
format_str = ('%s: step %d, loss = %.2f (%.1f examples/sec; %.3f '
'sec/batch)\n Accuracy = %.4f, mean average precision = %.4f')
print (format_str % (datetime.now(), step, loss_value,
examples_per_sec, sec_per_batch,
accuracy_value, np.mean(precision_value)))
if step % 100 == 0:
summary_str = sess.run(summary_op)
summary_writer.add_summary(summary_str, step)
summary = tf.Summary()
summary.value.add(tag='Accuracy (raw)', simple_value=float(accuracy_value))
for i,s in enumerate(CLASSES):
summary.value.add(tag="precision/"+s+" (raw)",simple_value=float(precision_value[i]))
summary.value.add(tag="accs/"+s+" (raw)",simple_value=float(cat_accs_val[i]))
# summary.value.add(tag='Human precision (raw)', simple_value=float(precision_value))
summary_writer.add_summary(summary, step)
print("hundred steps")
# Save the model checkpoint periodically.
if step % 1000 == 0 or (step + 1) == FLAGS.max_steps:
print("thousand steps")
checkpoint_path = os.path.join(FLAGS.train_dir, 'model.ckpt')
saver.save(sess, checkpoint_path, global_step=step)
def main(_):
class _LoggerHook(tf.train.SessionRunHook):
"""Logs loss and runtime."""
def begin(self):
self._step = -1
def before_run(self, run_context):
self._step += 1
self._start_time = time.time()
return tf.train.SessionRunArgs(loss) # Asks for loss value.
def after_run(self, run_context, run_values):
duration = time.time() - self._start_time
loss_value = run_values.results
if self._step % 10 == 0:
num_examples_per_step = FLAGS.batch_size
examples_per_sec = num_examples_per_step / duration
sec_per_batch = float(duration)
format_str = ('%s: step %d, loss = %.2f (%.1f examples/sec; %.3f '
'sec/batch)')
print (format_str % (datetime.now(), self._step, loss_value,
examples_per_sec, sec_per_batch))
ps_hosts = FLAGS.ps_hosts.split(",")
worker_hosts = FLAGS.worker_hosts.split(",")
# Create a cluster from the parameter server and worker hosts.
cluster = tf.train.ClusterSpec({"ps": ps_hosts, "worker": worker_hosts})
# Create and start a server for the local task.
server = tf.train.Server(cluster,
job_name=FLAGS.job_name,
task_index=FLAGS.task_index)
if FLAGS.job_name == "ps":
server.join()
elif FLAGS.job_name == "worker":
# Assigns ops to the local worker by default.
with tf.device(tf.train.replica_device_setter(
worker_device="/job:worker/task:%d" % FLAGS.task_index,
cluster=cluster)):
global_step = tf.contrib.framework.get_or_create_global_step()
# Get images and labels for CIFAR-10.
images, labels = cifar10.distorted_inputs()
# Build inference Graph.
logits = cifar10.inference(images)
# Build the portion of the Graph calculating the losses. Note that we will
# assemble the total_loss using a custom function below.
loss = cifar10.loss(logits, labels)
# Build a Graph that trains the model with one batch of examples and
# updates the model parameters.
train_op = cifar10.train(loss,global_step)
# The StopAtStepHook handles stopping after running given steps.
hooks=[tf.train.StopAtStepHook(last_step=FLAGS.max_steps), _LoggerHook()]
# The MonitoredTrainingSession takes care of session initialization,
# restoring from a checkpoint, saving to a checkpoint, and closing when done
# or an error occurs.
with tf.train.MonitoredTrainingSession(master=server.target,
is_chief=(FLAGS.task_index == 0),
checkpoint_dir=FLAGS.train_dir,
save_checkpoint_secs=60,
hooks=hooks) as mon_sess:
while not mon_sess.should_stop():
# Run a training step asynchronously.
# See `tf.train.SyncReplicasOptimizer` for additional details on how to
# perform *synchronous* training.
# mon_sess.run handles AbortedError in case of preempted PS.
mon_sess.run(train_op)
def train():
"""Train CIFAR-10 for a number of steps."""
with tf.Graph().as_default():
global_step = tf.contrib.framework.get_or_create_global_step()
# Get images and labels for CIFAR-10.
images, labels = cifar10.distorted_inputs()
# Build a Graph that computes the logits predictions from the
# inference model.
if tfFLAGS.network == 1:
images, labels = cifar10.distorted_inputs()
logits, fc1_w, fc1_b, fc2_w, fc2_b = MyModel.inference(images)
else:
images, labels = cifar10.distorted_inputs()
logits, fc1_w, fc1_b, fc2_w, fc2_b = MyModel2.inference(images)
# Calculate loss.
loss = cifar10.loss(logits, labels)
# L2 regularization for the fully connected parameters.
regularizers = (tf.nn.l2_loss(fc1_w) + tf.nn.l2_loss(fc1_b) + tf.nn.l2_loss(fc2_w) + tf.nn.l2_loss(fc2_b))
# Add the regularization term to the loss.
loss += 5e-4 * regularizers
# Build a Graph that trains the model with one batch of examples and
# updates the model parameters.
train_op = cifar10.train(loss, global_step)
class _LoggerHook(tf.train.SessionRunHook):
"""Logs loss and runtime."""
def begin(self):
self._step = -1
self._start_time = time.time()
def before_run(self, run_context):
self._step += 1
return tf.train.SessionRunArgs(loss) # Asks for loss value.
def after_run(self, run_context, run_values):
if self._step % tfFLAGS.log_frequency == 0:
current_time = time.time()
duration = current_time - self._start_time
self._start_time = current_time
loss_value = run_values.results
examples_per_sec = tfFLAGS.log_frequency * tfFLAGS.batch_size / duration
sec_per_batch = float(duration / tfFLAGS.log_frequency)
format_str = ('%s: step %d, loss = %.2f (%.1f examples/sec; %.3f '
'sec/batch)')
print_(format_str % (datetime.now(), self._step, loss_value, examples_per_sec, sec_per_batch))
texts = ['conv1:', 'conv1Biases:', 'conv2:', 'conv2Biases:', 'local3:', 'local3Biases:', 'local4:', 'local4Biases:', 'softmax:', 'softmaxBiases:']
total_parameters = 0; count = 0
for variable in tf.trainable_variables():
variable_parametes = 1
for dim in variable.get_shape():
variable_parametes *= dim.value
print('Number of hidden parameters of ' + texts[count], variable_parametes)
total_parameters += variable_parametes
count += 1
print('Total Number of hidden parameters:', total_parameters)
with tf.train.MonitoredTrainingSession(checkpoint_dir=tfFLAGS.train_dir,
hooks=[tf.train.StopAtStepHook(last_step=tfFLAGS.max_steps), tf.train.NanTensorHook(loss),_LoggerHook()],
config=tf.ConfigProto( device_count = {'GPU': 0}, log_device_placement=tfFLAGS.log_device_placement)) as mon_sess:
while not mon_sess.should_stop():
mon_sess.run(train_op)
def main_fun(argv, ctx):
import tensorflow as tf
import cifar10
sys.argv = argv
FLAGS = tf.app.flags.FLAGS
tf.app.flags.DEFINE_string('train_dir', '/tmp/cifar10_train',
"""Directory where to write event logs """
"""and checkpoint.""")
tf.app.flags.DEFINE_integer('max_steps', 1000000,
"""Number of batches to run.""")
tf.app.flags.DEFINE_boolean('log_device_placement', False,
"""Whether to log device placement.""")
tf.app.flags.DEFINE_boolean('rdma', False, """Whether to use rdma.""")
# cifar10.maybe_download_and_extract()
if tf.gfile.Exists(FLAGS.train_dir):
tf.gfile.DeleteRecursively(FLAGS.train_dir)
tf.gfile.MakeDirs(FLAGS.train_dir)
cluster_spec, server = TFNode.start_cluster_server(ctx, 1, FLAGS.rdma)
# Train CIFAR-10 for a number of steps.
with tf.Graph().as_default():
global_step = tf.contrib.framework.get_or_create_global_step()
# Get images and labels for CIFAR-10.
images, labels = cifar10.distorted_inputs()
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(images)
# Calculate loss.
loss = cifar10.loss(logits, labels)
# Build a Graph that trains the model with one batch of examples and
# updates the model parameters.
train_op = cifar10.train(loss, global_step)
class _LoggerHook(tf.train.SessionRunHook):
"""Logs loss and runtime."""
def begin(self):
self._step = -1
def before_run(self, run_context):
self._step += 1
self._start_time = time.time()
return tf.train.SessionRunArgs(loss) # Asks for loss value.
def after_run(self, run_context, run_values):
duration = time.time() - self._start_time
loss_value = run_values.results
if self._step % 10 == 0:
num_examples_per_step = FLAGS.batch_size
examples_per_sec = num_examples_per_step / duration
sec_per_batch = float(duration)
format_str = ('%s: step %d, loss = %.2f (%.1f examples/sec; %.3f '
'sec/batch)')
print (format_str % (datetime.now(), self._step, loss_value,
examples_per_sec, sec_per_batch))
with tf.train.MonitoredTrainingSession(
checkpoint_dir=FLAGS.train_dir,
hooks=[tf.train.StopAtStepHook(last_step=FLAGS.max_steps),
tf.train.NanTensorHook(loss),
_LoggerHook()],
config=tf.ConfigProto(
log_device_placement=FLAGS.log_device_placement)) as mon_sess:
while not mon_sess.should_stop():
mon_sess.run(train_op)
def train():
"""Train CIFAR-10 for a number of steps."""
with tf.Graph().as_default(), tf.device('/cpu:0'):
# Create a variable to count the number of train() calls. This equals the
# number of batches processed * FLAGS.num_gpus.
global_step = tf.get_variable(
'global_step', [],
initializer=tf.constant_initializer(0), trainable=False)
# Calculate the learning rate schedule.
num_batches_per_epoch = (cifar10.NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN /
FLAGS.batch_size)
decay_steps = int(num_batches_per_epoch * cifar10.NUM_EPOCHS_PER_DECAY)
# Decay the learning rate exponentially based on the number of steps.
lr = tf.train.exponential_decay(cifar10.INITIAL_LEARNING_RATE,
global_step,
decay_steps,
cifar10.LEARNING_RATE_DECAY_FACTOR,
staircase=True)
# Create an optimizer that performs gradient descent.
opt = tf.train.GradientDescentOptimizer(lr)
# Get images and labels for CIFAR-10.
images, labels = cifar10.distorted_inputs()
batch_queue = tf.contrib.slim.prefetch_queue.prefetch_queue(
[images, labels], capacity=2 * FLAGS.num_gpus)
# Calculate the gradients for each model tower.
tower_grads = []
with tf.variable_scope(tf.get_variable_scope()):
for i in xrange(FLAGS.num_gpus):
with tf.device('/gpu:%d' % i):
with tf.name_scope('%s_%d' % (cifar10.TOWER_NAME, i)) as scope:
# Dequeues one batch for the GPU
image_batch, label_batch = batch_queue.dequeue()
# Calculate the loss for one tower of the CIFAR model. This function
# constructs the entire CIFAR model but shares the variables across
# all towers.
loss = tower_loss(scope, image_batch, label_batch)
# Reuse variables for the next tower.
tf.get_variable_scope().reuse_variables()
# Retain the summaries from the final tower.
summaries = tf.get_collection(tf.GraphKeys.SUMMARIES, scope)
# Calculate the gradients for the batch of data on this CIFAR tower.
grads = opt.compute_gradients(loss)
# Keep track of the gradients across all towers.
tower_grads.append(grads)
# We must calculate the mean of each gradient. Note that this is the
# synchronization point across all towers.
grads = average_gradients(tower_grads)
# Add a summary to track the learning rate.
summaries.append(tf.summary.scalar('learning_rate', lr))
# Add histograms for gradients.
for grad, var in grads:
if grad is not None:
summaries.append(tf.summary.histogram(var.op.name + '/gradients', grad))
# Apply the gradients to adjust the shared variables.
apply_gradient_op = opt.apply_gradients(grads, global_step=global_step)
# Add histograms for trainable variables.
for var in tf.trainable_variables():
summaries.append(tf.summary.histogram(var.op.name, var))
# Track the moving averages of all trainable variables.
variable_averages = tf.train.ExponentialMovingAverage(
cifar10.MOVING_AVERAGE_DECAY, global_step)
variables_averages_op = variable_averages.apply(tf.trainable_variables())
# Group all updates to into a single train op.
train_op = tf.group(apply_gradient_op, variables_averages_op)
# Create a saver.
saver = tf.train.Saver(tf.global_variables())
# Build the summary operation from the last tower summaries.
summary_op = tf.summary.merge(summaries)
# Build an initialization operation to run below.
init = tf.global_variables_initializer()
# Start running operations on the Graph. allow_soft_placement must be set to
# True to build towers on GPU, as some of the ops do not have GPU
# implementations.
sess = tf.Session(config=tf.ConfigProto(
allow_soft_placement=True,
log_device_placement=FLAGS.log_device_placement))
sess.run(init)
# Start the queue runners.
tf.train.start_queue_runners(sess=sess)
summary_writer = tf.summary.FileWriter(FLAGS.train_dir, sess.graph)
for step in xrange(FLAGS.max_steps):
start_time = time.time()
_, loss_value = sess.run([train_op, loss])
duration = time.time() - start_time
assert not np.isnan(loss_value), 'Model diverged with loss = NaN'
if step % 10 == 0:
num_examples_per_step = FLAGS.batch_size * FLAGS.num_gpus
examples_per_sec = num_examples_per_step / duration
sec_per_batch = duration / FLAGS.num_gpus
format_str = ('%s: step %d, loss = %.2f (%.1f examples/sec; %.3f '
'sec/batch)')
print (format_str % (datetime.now(), step, loss_value,
examples_per_sec, sec_per_batch))
if step % 100 == 0:
summary_str = sess.run(summary_op)
summary_writer.add_summary(summary_str, step)
# Save the model checkpoint periodically.
if step % 1000 == 0 or (step + 1) == FLAGS.max_steps:
checkpoint_path = os.path.join(FLAGS.train_dir, 'model.ckpt')
saver.save(sess, checkpoint_path, global_step=step)
def train():
# # debug
# true_classes = np.ndarray(shape=(FLAGS.batch_size, 1), dtype=int)
# true_classes.fill(2)
# # Create a pair of constant ops, add the numpy
# # array matrices.
# true_classes_tf_matrix = tf.constant(true_classes, dtype=tf.int64)
# # playing with introducing the sampler
# classes_sampler = tf.nn.learned_unigram_candidate_sampler(
# true_classes_tf_matrix,
# 1, # true_classes
# 5, # num_sampled
# False, # unique
# 10, # range_max
# seed=None,
# name="my_classes_sampler")
# # print(classes_sampler)
# # print("debug")
# # print(classes_sampler.set_sampler)
# # exit()
"""Train CIFAR-10 for a number of steps."""
with tf.Graph().as_default():
global_step = tf.Variable(0, trainable=False)
# Get images and labels for CIFAR-10.
images, labels = cifar10.distorted_inputs()
# print("images")
# print(images)
# images = tf.Print(images, [images])
# print()
# print(images[1])
print("------------------- train calling interference ---------------------")
print(cifar10.__file__)
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(images)
# Calculate loss.
loss = cifar10.loss(logits, labels)
# Build a Graph that trains the model with one batch of examples and
# updates the model parameters.
train_op = cifar10.train(loss, global_step)
# Create a saver.
saver = tf.train.Saver(tf.all_variables())
# Build the summary operation based on the TF collection of Summaries.
summary_op = tf.merge_all_summaries()
# Build an initialization operation to run below.
init = tf.initialize_all_variables()
# Start running operations on the Graph.
sess = tf.Session(config=tf.ConfigProto(
log_device_placement=FLAGS.log_device_placement))
sess.run(init)
# Start the queue runners.
tf.train.start_queue_runners(sess=sess)
summary_writer = tf.train.SummaryWriter(FLAGS.train_dir, sess.graph)
for step in xrange(FLAGS.max_steps):
# manually load the contents of images and labels
# before calling this sess.run()
# 1. have Cifar10 dataset in memory
# 2. create a mini-batch
# 3. set the placeholders/vars to the the mini-batch data
# 4. run one forward-backward step
# print("training step: " + str(step))
start_time = time.time()
_, loss_value = sess.run([train_op, loss])
duration = time.time() - start_time
assert not np.isnan(loss_value), 'Model diverged with loss = NaN'
if step % 10 == 0:
num_examples_per_step = FLAGS.batch_size
examples_per_sec = num_examples_per_step / duration
sec_per_batch = float(duration)
format_str = ('%s: step %d, loss = %.2f (%.1f examples/sec; %.3f '
'sec/batch)')
print (format_str % (datetime.now(), step, loss_value,
examples_per_sec, sec_per_batch))
# debug, temp change, go back to the one below
summary_str = sess.run(summary_op)
# print("summary: " + summary_str)
summary_writer.add_summary(summary_str, step)
summary_writer.flush()
# if step % 100 == 0:
# summary_str = sess.run(summary_op)
# # print("summary: " + summary_str)
# summary_writer.add_summary(summary_str, step)
# summary_writer.flush()
# Save the model checkpoint periodically.
if step % 1000 == 0 or (step + 1) == FLAGS.max_steps:
checkpoint_path = os.path.join(FLAGS.train_dir, 'model.ckpt')
saver.save(sess, checkpoint_path, global_step=step)