def evaluate():
"""Eval CIFAR-10 for a number of steps."""
dataset = input_data.read(FLAGS.input_dir)
image_size = dataset.image_size
with tf.Graph().as_default():
# Build a Graph that computes the logits predictions from the
# inference model.
eval_images = tf.placeholder(tf.float32, shape=(2, FLAGS.batch_size, image_size[0], image_size[1], image_size[2]))
labels = tf.placeholder(tf.float32, shape=(FLAGS.batch_size))
images, images_p = tf.split(0, 2, train_images)
with tf.variable_scope('inference') as scope:
logits = cifar10.inference(images)
scope.reuse_variables()
logits2 = cifar10.inference(images_p)
# Calculate predictions.
accuracy = cifar10.accuracy(logits, logits2, labels)
# 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()
graph_def = tf.get_default_graph().as_graph_def()
summary_writer = tf.train.SummaryWriter(FLAGS.eval_dir,
graph_def=graph_def)
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 eval_once(filename):
"""Eval CIFAR-10 for a number of steps."""
with tf.Graph().as_default() as g:
# Get loss.
# Create placeholder.
images = tf.placeholder(tf.float32, shape=(FLAGS.batch_size, IMAGE_SIZE, IMAGE_SIZE, 3))
labels = tf.placeholder(tf.int32, shape = (FLAGS.batch_size,))
domain_labels = tf.placeholder(tf.int32, shape = (FLAGS.batch_size,))
#loss_label, loss_domain = cifar10.inference(images)
logits1, logits2= cifar10.inference(images)
loss_label = cifar10.loss_without_decay(logits1, labels)
loss_domain = cifar10.loss_without_decay(logits2, domain_labels)
# 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)
x_tr, y_tr, d_tr, nl_x_v, nl_y_v, nl_d_v, l_x_te, l_y_te, l_d_te = input.input()
# Calculate losses of training, non-lifelog validation, and lifelog test.
y_tr_loss, d_tr_loss = eval_a_dataset(saver, filename, x_tr, y_tr, d_tr, images, labels, domain_labels, loss_label, loss_domain)
y_nl_loss, d_nl_loss = eval_a_dataset(saver, filename, nl_x_v, nl_y_v, nl_d_v, images, labels, domain_labels, loss_label, loss_domain)
y_l_loss, d_l_loss = eval_a_dataset(saver, filename, l_x_te, l_y_te, l_d_te, images, labels, domain_labels, loss_label, loss_domain)
return y_tr_loss, d_tr_loss, y_nl_loss, d_nl_loss, y_l_loss, d_l_loss
def test_once(filename):
"""Eval CIFAR-10 for a number of steps."""
with tf.Graph().as_default() as g:
# Get loss.
# Create placeholder.
images = tf.placeholder(tf.float32, shape=(FLAGS.batch_size, IMAGE_SIZE, IMAGE_SIZE, 3))
loss_label, loss_domain = cifar10.inference(images)
# 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)
_, _, _, _, _, _, \
x, y, domain= input.input()
# Supplement the number of examples to multiplier of 25.
num_of_examples = np.shape(x)[0]
remainder = FLAGS.batch_size - int(math.ceil(num_of_examples/FLAGS.batch_size))
index = range(num_of_examples) + [0] * remainder
with tf.Session() as sess:
#need to modify for test
saver.restore(sess, filename)
global_step = int(filename.split('-')[1])
# Allocate results in a list.
losses_label = []
losses_domain = []
# Start the queue runners.
step = 0
while step + FLAGS.batch_size <= len(index):
label_loss_value, domain_loss_value = sess.run([loss_label, loss_domain], feed_dict = {images:x[index[step:step+FLAGS.batch_size], :]})
losses_label.append(label_loss_value)
losses_domain.append(domain_loss_value)
step = step + FLAGS.batch_size
# Convert list of lists to numpy array.
losses_label = np.asarray(losses_label)
losses_domain = np.asarray(losses_domain)
losses_label = losses_label.reshape((-1, 21))
losses_domain = losses_domain.reshape((-1, 2))
losses_label = losses_label[:num_of_examples, :]
losses_domain = losses_domain[:num_of_examples, :]
sp.savez('test.npz', losses_label = losses_label, losses_domain = losses_domain, y = y, domain = domain)
return losses_label, losses_domain, y, domain
def tower_loss(scope, images, labels):
# Build inference Graph.
logits = cifar10.inference(images,
None,
None,
None,
None,
None,
None,
None,
train=True)
# 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 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 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, _ = cifar10.inputs(eval_data=eval_data)
logits = cifar10.inference(images)
prediction = tf.nn.softmax(logits)
variable_averages = tf.train.ExponentialMovingAverage(
cifar10.MOVING_AVERAGE_DECAY)
variables_to_restore = variable_averages.variables_to_restore()
saver = tf.train.Saver(variables_to_restore)
with tf.Session() as sess:
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)
else:
print('No checkpoint file found')
return
# Build a Graph that computes the logits predictions from the
# inference model.
result = tf.argmax(prediction, 1)
tf.print(result)
print('DONE')
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 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 tower_loss(scope, images, labels):
"""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'
images: Images. 4D tensor of shape [batch_size, height, width, 3].
labels: Labels. 1D tensor of shape [batch_size].
Returns:
Tensor of shape [] containing the total loss for a batch of data
"""
# 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 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.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)
# 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.eval_dir, g)
while True:
eval_once(saver, summary_writer, top_k_op, summary_op)
if FLAGS.run_once:
break
time.sleep(FLAGS.eval_interval_secs)
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 evaluate(): """ 多次评价CIFAR-10 """ with tf.Graph().as_default() as g: # 从CIFAR-10中读取图像和标签 eval_data = (FLAGS.eval_data == 'test') images, labels = cifar10.input(eval_data = eval_data) # 构建一个图计算推理模型的logits logits = cifar10.inference(images) # 计算准确度 top_k_op = tf.nn.in_top_k(logits, labels, 1) # 读取已经训练好的模型参数 variable_averages = tf.train.ExponentialMovingAverage( cifar10.MOVING_AVERAGE_DECAY) variables_to_restore = variable_averages.variables_to_restore() saver = tf.train.Saver(variables_to_restore) # 用TF collection of Summaries构建summary operation summary_op = tf.summary.merge_all() summary_writer = tf.summary.FileWriter(FLAGS.eval_dir, g) run_time = FLAGS.run_once while True: eval_once(saver, summary_writer, top_k_op, summary_op) if run_time <= 0: break run_time -= 1 time.sleep(FLAGS.eval_interval_secs)
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():
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 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 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.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 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.inputs(eval_data=eval_data)
tf.summary.image('images', images)
# 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.eval_dir, g)
while True:
eval_once(saver, summary_writer, top_k_op, summary_op)
if FLAGS.run_once:
break
time.sleep(FLAGS.eval_interval_secs)
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.inputs(eval_data=eval_data)
f = tf.read_file('/images/input.jpg')
image = tf.image.decode_jpeg(f, channels=3)
image = tf.image.resize_images(image, (24, 24))
tf.summary.image('images', [image])
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference([image])
# 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.eval_dir, g)
eval_once(saver, summary_writer, logits, summary_op)
def train():
"""Train the 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 the input pipeline to CPU:0 to avoid operation sometimes ended 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 the loss.
loss = cifar10.loss(logits, labels)
# Build a Graph that trains the model with one batch of examples and
# updates the model parameter
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 the 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 evaluate():
"""
Evaluate cifar10 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.inputs(eval_data=eval_data)
# build a tf.Graph that computes logits from inference model
logits = cifar10.inference(images)
# calculate prediction
top_k_op = tf.nn.in_top_k(logits, labels, 1)
# restore 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.eval_dir, g)
while True:
eval_once(saver, summary_writer, top_k_op, summary_op)
if FLAGS.run_once:
break
time.sleep(FLAGS.eval_interval_secs)
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 evaluate_image(filename):
with tf.Graph().as_default() as g:
# Number of images to process
FLAGS.batch_size = 1
image = img_read(filename)
logit = cifar10.inference(image)
output = tf.nn.softmax(logit)
top_k_pred = tf.nn.top_k(output, k=1)
variable_averages = tf.train.ExponentialMovingAverage(
cifar10.MOVING_AVERAGE_DECAY)
variables_to_restore = variable_averages.variables_to_restore()
saver = tf.train.Saver(variables_to_restore)
with tf.Session() as sess:
#sess.run(tf.global_variables_initializer())
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)
else:
print("No checkpoint file found")
return
values, indices = sess.run(top_k_pred)
print LABEL_NAMES[indices[0][0] - 1], values[0][0]
def evaluate():
"""Eval CIFAR-10 for a number of steps."""
with tf.Graph().as_default() as g:
# Get images and labels for CIFAR-10.
inputs = cifar10.ram_inputs(unit_variance=True, is_train=False)
images = inputs['images']
labels = inputs['labels']
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(images, 3, use_batchnorm=True,
use_nrelu=False, id_decay=False, add_shortcuts=True, is_train=False)
# 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.merge_all_summaries()
summary_writer = tf.train.SummaryWriter(FLAGS.eval_dir, g)
while True:
eval_once(saver, summary_writer, top_k_op, summary_op, inputs)
if FLAGS.run_once:
break
time.sleep(FLAGS.eval_interval_secs)
def evaluate():
"""Eval CIFAR-10 for a number of steps."""
with tf.Graph().as_default() as g:
# Get images and labels for CIFAR-10.
images, labels = cifar10.inputs(eval_data='test')
# need modify for test
logits, _ = cifar10.inference(images)
# 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()
del variables_to_restore['input_producer/limit_epochs/epochs']
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.test_dir, g)
while True:
eval_once_given_model(saver, summary_writer, logits, labels, summary_op, images)
if FLAGS.run_once:
break
time.sleep(FLAGS.eval_interval_secs)
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.inputs(eval_data=eval_data) # Generate a batch of images and labels
# 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) # k = 1, this outputs a batch_size bool array, an entry out[i] is true if the prediction for the target class is among the top k predictions among all predictions for example i
# 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() # Returns a map of names to Variables to restore.
saver = tf.train.Saver(variables_to_restore) # The Saver class adds ops to save and restore variables to and from checkpoints
# Build the summary operation based on the TF collection of Summaries.
summary_op = tf.summary.merge_all() # Merges all summaries collected in the default graph.
summary_writer = tf.summary.FileWriter(FLAGS.eval_dir, g) # create an event file in a given directory and add summaries and events to it
while True:
eval_once(saver, summary_writer, top_k_op, summary_op)
if FLAGS.run_once:
break
time.sleep(FLAGS.eval_interval_secs) # Suspend execution of the calling thread for the given number of seconds
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 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 evaluate():
"""Eval CIFAR-10 for a number of steps."""
dt = Transformer.Transformer()
dt.checkpoint_version = FLAGS.checkpoint_version
dt.trainable = False
#with dt.graph.as_default() as g:
with tf.Graph().as_default() as g:
if FLAGS.is_compressed:
dt.load_flow('/tmp/cifar10_train_' +
str(dt.checkpoint_version - 1) +
'/pruned_flow_graph.pkl') # prefetch flow graph
dt.flow = dt.cached_flow
else:
dt.flow = NeuralFlow()
with g.device('/cpu:0'):
# Get images and labels for CIFAR-10.
eval_data = FLAGS.eval_data == 'test'
images, labels = cifar10.inputs(eval_data=eval_data)
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(dt,
images,
is_training=False,
is_compressed=FLAGS.is_compressed)
total_params, _ = slim.model_analyzer.analyze_vars(
slim.get_model_variables())
print('total params: %d' % total_params)
# Calculate predictions.
top_k_op = tf.nn.in_top_k(logits, labels, 1)
# Calculate loss.
loss_op = cifar10.loss(dt, logits, labels)
# 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()
variables_to_restore_simple = tf.global_variables()
if FLAGS.is_compressed:
saver = tf.train.Saver(variables_to_restore_simple)
else:
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.eval_dir, g)
while True:
eval_once(dt, saver, summary_writer, top_k_op, loss_op, summary_op)
if FLAGS.run_once:
break
time.sleep(FLAGS.eval_interval_secs)
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.inputs(eval_data=eval_data)
# Build a Graph that computes the logits predictions from the
# # inference model. 推理模型
logits = cifar10.inference(images)
# Calculate predictions.
top_k_op = tf.nn.in_top_k(logits, labels, 1)
###
# tf.nn.in_top_k组要是用于计算预测的结果和实际结果的是否相等，返回一个bool类型的张量，
# tf.nn.in_top_k(prediction, target, K):
# prediction就是表示你预测的结果，大小就是预测样本的数量乘以输出的维度，类型是tf.float32
# 等。target就是实际样本类别的标签，大小就是样本数量的个数。K表示每个样本的预测结果的前K
# 个最大的数里面是否含有target中的值。一般都是取1。
#
# import tensorflow as tf;
# A = [[0.8,0.6,0.3], [0.1,0.6,0.4]]
# B = [1, 1]
# out = tf.nn.in_top_k(A, B, 1)
# with tf.Session() as sess:
# sess.run(tf.initialize_all_variables())
# print sess.run(out)
#
# Running Result:[False, True] ! 注意0起址index
###
# Restore the moving average version of the learned variables for eval.
variable_averages = tf.train.ExponentialMovingAverage(
cifar10.MOVING_AVERAGE_DECAY)
###
# tf.train.ExponentialMovingAverage这个函数用于更新参数，就是采用滑动平均的方法更新参数。
# 这个函数初始化需要提供一个衰减速率（decay），用于控制模型的更新速度。这个函数还会维护一个
# 影子变量（也就是更新参数后的参数值），这个影子变量的初始值就是这个变量的初始值，影子变量
# 值的更新方式如下：
# shadow_variable = decay * shadow_variable + (1-decay) * variable
# shadow_variable是影子变量，variable表示待更新的变量，也就是变量被赋予的值，decay为衰减速率。
# decay一般设为接近于1的数（0.99,0.999）。decay越大模型越稳定，因为decay越大，参数更新的速度
# 就越慢，趋于稳定。
###
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.eval_dir, g)
while True:
eval_once(saver, summary_writer, top_k_op, summary_op)
if FLAGS.run_once:
break
# # 休眠一顿时间后再评估
time.sleep(FLAGS.eval_interval_secs)
def build_model_func():
images, labels = cifar10.distorted_inputs()
logits = cifar10.inference(images)
loss = cifar10.loss(logits, labels)
return loss
def evaluate():
"""Eval CIFAR-10 for a number of steps."""
with tf.Graph().as_default():
# Get images and labels for CIFAR-10.
eval_data = FLAGS.eval_data == 'test'
images, labels = inputs() # cifar10.inputs(eval_data=eval_data)
print "images dimension @ evaluation:", images
print "image label @ evaluation:", labels
# 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)
top_k_predict_op = tf.argmax(logits, 1)
# Restore the moving average version of the learned variables for eval.
variable_averages = tf.train.ExponentialMovingAverage(
cifar10.MOVING_AVERAGE_DECAY)
variables_to_restore = {}
for v in tf.all_variables():
if v in tf.trainable_variables():
restore_name = variable_averages.average_name(v)
else:
restore_name = v.op.name
variables_to_restore[restore_name] = v
saver = tf.train.Saver(variables_to_restore)
# Build the summary operation based on the TF collection of Summaries.
summary_op = tf.merge_all_summaries()
graph_def = tf.get_default_graph().as_graph_def()
summary_writer = tf.train.SummaryWriter(FLAGS.eval_dir,
graph_def=graph_def)
# test_image(saver, logits)
###Test one image
print "#################"
print "##gonna test this:"
print "#################"
###
while True:
eval_once(saver, summary_writer, top_k_op, top_k_predict_op, summary_op, images)
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.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 train():
""" Train the CIFAR10 model for a number of steps"""
global_step = tf.train.get_or_create_global_step()
# Get images and labels for the cifar10
data_dir = os.path.join(FLAGS.data_dir, 'cifar-10-batches-bin')
images_batch, labels_batch = cifar10_input.train_inputs(
data_dir, FLAGS.batch_size)
# Build a Graph that computes the logits predictions from the
# inference model
logits = cifar10.inference(images_batch)
# Compute loss
loss = cifar10.loss(logits, labels_batch)
# 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.checkpoint_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():
# Get images and labels for CIFAR-10.
eval_data = FLAGS.eval_data == 'test'
images, labels = cifar10.inputs(eval_data=eval_data)
with tf.Session() as sess:
# Build a Graph that computes the logits predictions from the
# inference model.
probabilities = tf.nn.softmax(cifar10.inference(images))
# 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)
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)
else:
print('No checkpoint file found')
return
# Start the queue runners.
coord = tf.train.Coordinator()
try:
threads = []
for qr in tf.get_collection(tf.GraphKeys.QUEUE_RUNNERS):
threads.extend(
qr.create_threads(sess,
coord=coord,
daemon=True,
start=True))
num_iter = int(math.ceil(FLAGS.num_examples / FLAGS.batch_size))
submission = []
true_labels = []
step = 0
while step < num_iter and not coord.should_stop():
submission_batch, true_labels_batch = sess.run(
[probabilities, labels])
submission.append(submission_batch)
true_labels.append(true_labels_batch)
step += 1
submission = np.vstack(submission)
true_labels = np.concatenate(true_labels)
except Exception as e: # pylint: disable=broad-except
coord.request_stop(e)
coord.request_stop()
coord.join(threads, stop_grace_period_secs=10)
return submission, true_labels
def run(name):
image_name = 'living.jpg'
image_string = open(image_name, 'rb').read()
image = tf.image.decode_jpeg(image_string, channels=1)
if image.dtype != tf.float32:
image = tf.image.convert_image_dtype(image, dtype=tf.float32)
image = tf.expand_dims(image, 0)
# Save the processed image for Fathom
with tf.Session() as sess:
npimage = image.eval()
np.save('image.npy', npimage.squeeze())
# Run the network with Tensorflow
with tf.Graph().as_default():
image = tf.placeholder("float", [1,45,44,1], name="input")
#with slim.arg_scope(inception.inception_v1_arg_scope()):
logits = cifar10.inference(image)
#probabilities = tf.nn.softmax(logits)
probabilities = logits
#init_fn = slim.assign_from_checkpoint_fn('./model.ckpt-19999', slim.get_model_variables('model'))
variable_averages = tf.train.ExponentialMovingAverage(cifar10.MOVING_AVERAGE_DECAY)
variables_to_restore = variable_averages.variables_to_restore()
saver = tf.train.Saver(variables_to_restore)
with tf.Session() as sess:
ckpt = tf.train.get_checkpoint_state('ecotaxa_train/')
saver.restore(sess, ckpt.model_checkpoint_path)
init_fn = slim.assign_from_checkpoint_fn('./ecotaxa_train/model.ckpt-0', tf.global_variables())
with tf.Session() as sess:
init_fn(sess)
prob = probabilities.eval(feed_dict = {"input:0" : npimage})
# Save the network for Fathom
saver = tf.train.Saver(tf.global_variables())
saver.save(sess, name)
"""
i = 1
for n in tf.get_default_graph().as_graph_def().node:
if i < 100:
print("name:",n.name)
print("op",n.op)
print("input:",n.input)
print(" ")
i = i +1
"""
# Dump output
prob = prob.squeeze()
probidx = np.argsort(-prob)
print('Top-2:')
for i in range(2):
print(probidx[i], prob[probidx[i]])
def evaluate(images, labels=None):
with tf.Graph().as_default():
image_holder = tf.placeholder(tf.uint8, [32, 32, 3])
resized_image = tf.cast(image_holder, tf.float32)
resized_image = tf.image.resize_image_with_crop_or_pad(
image_holder, IMAGE_SIZE, IMAGE_SIZE)
# Subtract off the mean and divide by the variance of the pixels.
float_image = tf.image.per_image_standardization(resized_image)
# batch size = 1
float_image_batch = tf.reshape(float_image,
[1, IMAGE_SIZE, IMAGE_SIZE, 3])
# inference model.
logits = cifar10.inference(float_image_batch, False)
softmax = tf.nn.softmax(logits)
# 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)
with tf.Session() as sess:
ckpt = tf.train.get_checkpoint_state(FLAGS.checkpoint_dir)
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
global_step = ckpt.model_checkpoint_path.split('/')[-1].split(
'-')[-1]
else:
print('No checkpoint file found')
return
total = 0
correct = 0
statics = {}
for cls in CLASSES:
statics[cls] = 0
for i, image in enumerate(images):
total += 1
softmax_out = sess.run(softmax,
feed_dict={image_holder: image})
index = np.argmax(softmax_out)
if labels:
if labels[i] == index:
correct += 1
statics[CLASSES[index]] += 1
for cls in CLASSES:
print('%-12s: %4d/%d' % (cls, statics[cls], total))
if labels:
print('%d/%d, %.1f%%' %
(correct, total, correct * 100 / float(total)))
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 evaluate():
"""Eval CIFAR-10 for a number of steps."""
with tf.Graph().as_default() as g:
eval_data = FLAGS.eval_data == 'test'
images, labels = cifar10.inputs(eval_data=eval_data)
logits = cifar10.inference(images)
top_k_op = tf.nn.in_top_k(logits, labels, 1)
loss = cifar10.loss(logits, labels)
saver = tf.train.Saver(tf.all_variables())
while True:
eval_once(saver, top_k_op, loss)
if FLAGS.run_once:
break
time.sleep(FLAGS.eval_interval_secs)
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 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"
print(eval_data)
images, labels, ground_truth = cifar10.inputs(eval_data=eval_data)
# Build a Graph that computes the logits predictions from the
# inference model.
logits, _ = cifar10.inference(images)
print(logits)
print(logits.get_shape())
print("after inference node creation")
loss = cifar10.loss(logits, labels)
accuracy, precision, accuracies = cifar10.accuracy(logits, ground_truth)
labels = tf.cast(labels, tf.int64)
label_shape = labels.get_shape().as_list()
reshaped_labels = tf.reshape(labels, [label_shape[0] * label_shape[1] * label_shape[2]])
logits_shape = logits.get_shape().as_list()
reshaped_logits = tf.reshape(logits, [logits_shape[0] * logits_shape[1] * logits_shape[2], logits_shape[3]])
# Calculate predictions.
# top_k_op = tf.nn.in_top_k(logits, labels, 1)
# top_k_op = tf.nn.in_top_k(reshaped_logits, reshaped_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.merge_all_summaries()
summary_writer = tf.train.SummaryWriter(FLAGS.eval_dir, g)
while True:
print("evaluate:")
eval_once(saver, summary_writer, summary_op, accuracy, precision, accuracies)
if FLAGS.run_once:
break
time.sleep(FLAGS.eval_interval_secs)
def evaluate():
with tf.Graph().as_default():
eval_data = FLAGS.eval_data == 'test'
images, labels = cifar10.inputs(eval_data=eval_data)
logits = cifar10.inference(images)
top_k_op = tf.nn.in_top_k(logits, labels, 1)
variable_averages = tf.train.ExponentialMovingAverage(cifar10.MOVING_AVERAGE_DECAY)
variables_to_restore = variable_averages.variables_to_restore()
saver = tf.train.Saver(variables_to_restore)
summary_op = tf.merge_all_summaries()
graph_def = tf.get_default_graph().as_graph_def()
summary_writer = tf.train.SummaryWriter(FLAGS.eval_dir, graph_def=graph_def)
while True:
eval_once(saver, summary_writer, top_k_op, summary_op)
if FLAGS.run_once:
break
time.sleep(FLAGS.eval_interval_secs)
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_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 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.Variable(0, trainable=False)
# Get images and labels for CIFAR-10.
# images, labels = cifar10.standard_distorted_inputs()
inputs = cifar10.ram_inputs(unit_variance=True, is_train=True)
images = inputs['images']
labels = inputs['labels']
# Batch generator
batcher = cifar10.Cifar10BatchGenerator(
inputs['data_images'], inputs['data_labels'], True,
FLAGS.max_epochs)
# Build a Graph that computes the logits predictions from the
# inference model.
logits = cifar10.inference(images, 3, use_batchnorm=True,
use_nrelu=False, id_decay=False, add_shortcuts=True, is_train=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.
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.5)
sess = tf.Session(config=tf.ConfigProto(
log_device_placement=FLAGS.log_device_placement,
gpu_options=gpu_options))
sess.run(init)
# Start the queue runners.
tf.train.start_queue_runners(sess=sess)
summary_writer = tf.train.SummaryWriter(FLAGS.train_dir, sess.graph)
step = -1
while not batcher.is_done():
step += 1
batch_im, batch_labs = batcher.next_batch()
feed_dict = {
inputs['images_pl']: batch_im,
inputs['labels_pl']: batch_labs,
}
start_time = time.time()
_, loss_value = sess.run([train_op, loss], feed_dict=feed_dict)
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 % 10 == 0:
summary_str = sess.run(summary_op, feed_dict=feed_dict)
summary_writer.add_summary(summary_str, step)
# Save the model checkpoint periodically.
if step % 10 == 0 or batcher.is_done():
checkpoint_path = os.path.join(FLAGS.train_dir, 'model.ckpt')
saver.save(sess, checkpoint_path, global_step=step)
import cifar10
from PIL import Image
FLAGS = tf.app.flags.FLAGS
FLAGS.batch_size = 1
filename = tf.placeholder(tf.string)
value = tf.read_file(filename)
images = tf.image.decode_png(value, channels = 3)
resized_images = tf.image.resize_images(images, 124, 124)
img = tf.image.per_image_whitening(resized_images)
logits = cifar10.inference(tf.expand_dims(img, 0))
outputs = tf.nn.softmax(logits)
k_pl = tf.placeholder(tf.int32)
top_k_op = tf.nn.top_k(outputs, k = k_pl)
def classify(session, file_name, k = 1):
values, indices = session.run(top_k_op, feed_dict = {
filename: file_name,
k_pl: k
})
return zip(indices.flatten().tolist(), values.flatten().tolist())
def session_with_checkpoint(checkpoint_dir = "./train_old", gpu_enabled = True):
# disable gpu
config = tf.ConfigProto(
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)
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 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)