# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================

"""A binary to train CIFAR-10 using multiple GPUs with synchronous updates.
Accuracy:
cifar10_multi_gpu_train.py achieves ~86% accuracy after 100K steps (256
epochs of data) as judged by cifar10_eval.py.
Speed: With batch_size 128.
System        | Step Time (sec/batch)  |     Accuracy
--------------------------------------------------------------------
1 Tesla K20m  | 0.35-0.60              | ~86% at 60K steps  (5 hours)
1 Tesla K40m  | 0.25-0.35              | ~86% at 100K steps (4 hours)
2 Tesla K20m  | 0.13-0.20              | ~84% at 30K steps  (2.5 hours)
3 Tesla K20m  | 0.13-0.18              | ~84% at 30K steps
4 Tesla K20m  | ~0.10                  | ~84% at 30K steps
Usage:
Please see the tutorial and website for how to download the CIFAR-10
data set, compile the program and train the model.
http://tensorflow.org/tutorials/deep_cnn/
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

from datetime import datetime
import os.path
import re
import time

import numpy as np
from six.moves import xrange  # pylint: disable=redefined-builtin
import tensorflow as tf
import cifar10

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_integer('num_gpus', 1,
                            """How many GPUs to use.""")
tf.app.flags.DEFINE_boolean('log_device_placement', False,
                            """Whether to log device placement.""")


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 average_gradients(tower_grads):
  """Calculate the average gradient for each shared variable across all towers.
  Note that this function provides a synchronization point across all towers.
  Args:
    tower_grads: List of lists of (gradient, variable) tuples. The outer list
      is over individual gradients. The inner list is over the gradient
      calculation for each tower.
  Returns:
     List of pairs of (gradient, variable) where the gradient has been averaged
     across all towers.
  """
  average_grads = []
  for grad_and_vars in zip(*tower_grads):
    # Note that each grad_and_vars looks like the following:
    #   ((grad0_gpu0, var0_gpu0), ... , (grad0_gpuN, var0_gpuN))
    grads = []
    for g, _ in grad_and_vars:
      # Add 0 dimension to the gradients to represent the tower.
      expanded_g = tf.expand_dims(g, 0)

      # Append on a 'tower' dimension which we will average over below.
      grads.append(expanded_g)

    # Average over the 'tower' dimension.
    grad = tf.concat(axis=0, values=grads)
    grad = tf.reduce_mean(grad, 0)

    # Keep in mind that the Variables are redundant because they are shared
    # across towers. So .. we will just return the first tower's pointer to
    # the Variable.
    v = grad_and_vars[0][1]
    grad_and_var = (grad, v)
    average_grads.append(grad_and_var)
  return average_grads


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 main(argv=None):  # pylint: disable=unused-argument
  cifar10.maybe_download_and_extract()
  if tf.gfile.Exists(FLAGS.train_dir):
    tf.gfile.DeleteRecursively(FLAGS.train_dir)
  tf.gfile.MakeDirs(FLAGS.train_dir)
  train()


if __name__ == '__main__':
  tf.app.run()

# Copyright 2016 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Distributed MNIST training and validation, with model replicas.
A simple softmax model with one hidden layer is defined. The parameters
(weights and biases) are located on one parameter server (ps), while the ops
are executed on two worker nodes by default. The TF sessions also run on the
worker node.
Multiple invocations of this script can be done in parallel, with different
values for --task_index. There should be exactly one invocation with
--task_index, which will create a master session that carries out variable
initialization. The other, non-master, sessions will wait for the master
session to finish the initialization before proceeding to the training stage.
The coordination between the multiple worker invocations occurs due to
the definition of the parameters on the same ps devices. The parameter updates
from one worker is visible to all other workers. As such, the workers can
perform forward computation and gradient calculation in parallel, which
should lead to increased training speed for the simple model.
"""

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import math
import sys
import tempfile
import time

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data

flags = tf.app.flags
flags.DEFINE_string("data_dir", "/tmp/mnist-data",
                    "Directory for storing mnist data")
flags.DEFINE_boolean("download_only", False,
                     "Only perform downloading of data; Do not proceed to "
                     "session preparation, model definition or training")
flags.DEFINE_integer("task_index", None,
                     "Worker task index, should be >= 0. task_index=0 is "
                     "the master worker task the performs the variable "
                     "initialization ")
flags.DEFINE_integer("num_gpus", 1, "Total number of gpus for each machine."
                     "If you don't use GPU, please set it to '0'")
flags.DEFINE_integer("replicas_to_aggregate", None,
                     "Number of replicas to aggregate before parameter update "
                     "is applied (For sync_replicas mode only; default: "
                     "num_workers)")
flags.DEFINE_integer("hidden_units", 100,
                     "Number of units in the hidden layer of the NN")
flags.DEFINE_integer("train_steps", 200,
                     "Number of (global) training steps to perform")
flags.DEFINE_integer("batch_size", 100, "Training batch size")
flags.DEFINE_float("learning_rate", 0.01, "Learning rate")
flags.DEFINE_boolean(
    "sync_replicas", False,
    "Use the sync_replicas (synchronized replicas) mode, "
    "wherein the parameter updates from workers are aggregated "
    "before applied to avoid stale gradients")
flags.DEFINE_boolean(
    "existing_servers", False, "Whether servers already exists. If True, "
    "will use the worker hosts via their GRPC URLs (one client process "
    "per worker host). Otherwise, will create an in-process TensorFlow "
    "server.")
flags.DEFINE_string("ps_hosts", "localhost:2222",
                    "Comma-separated list of hostname:port pairs")
flags.DEFINE_string("worker_hosts", "localhost:2223,localhost:2224",
                    "Comma-separated list of hostname:port pairs")
flags.DEFINE_string("job_name", None, "job name: worker or ps")

FLAGS = flags.FLAGS

IMAGE_PIXELS = 28


def main(unused_argv):
  mnist = input_data.read_data_sets(FLAGS.data_dir, one_hot=True)
  if FLAGS.download_only:
    sys.exit(0)

  if FLAGS.job_name is None or FLAGS.job_name == "":
    raise ValueError("Must specify an explicit `job_name`")
  if FLAGS.task_index is None or FLAGS.task_index == "":
    raise ValueError("Must specify an explicit `task_index`")

  print("job name = %s" % FLAGS.job_name)
  print("task index = %d" % FLAGS.task_index)

  #Construct the cluster and start the server
  ps_spec = FLAGS.ps_hosts.split(",")
  worker_spec = FLAGS.worker_hosts.split(",")

  # Get the number of workers.
  num_workers = len(worker_spec)

  cluster = tf.train.ClusterSpec({"ps": ps_spec, "worker": worker_spec})

  if not FLAGS.existing_servers:
    # Not using existing servers. Create an in-process server.
    server = tf.train.Server(
        cluster, job_name=FLAGS.job_name, task_index=FLAGS.task_index)
    if FLAGS.job_name == "ps":
      server.join()

  is_chief = (FLAGS.task_index == 0)
  if FLAGS.num_gpus > 0:
    # Avoid gpu allocation conflict: now allocate task_num -> #gpu
    # for each worker in the corresponding machine
    gpu = (FLAGS.task_index % FLAGS.num_gpus)
    worker_device = "/job:worker/task:%d/gpu:%d" % (FLAGS.task_index, gpu)
  elif FLAGS.num_gpus == 0:
    # Just allocate the CPU to worker server
    cpu = 0
    worker_device = "/job:worker/task:%d/cpu:%d" % (FLAGS.task_index, cpu)
  # The device setter will automatically place Variables ops on separate
  # parameter servers (ps). The non-Variable ops will be placed on the workers.
  # The ps use CPU and workers use corresponding GPU
  with tf.device(
      tf.train.replica_device_setter(
          worker_device=worker_device,
          ps_device="/job:ps/cpu:0",
          cluster=cluster)):
    global_step = tf.Variable(0, name="global_step", trainable=False)

    # Variables of the hidden layer
    hid_w = tf.Variable(
        tf.truncated_normal(
            [IMAGE_PIXELS * IMAGE_PIXELS, FLAGS.hidden_units],
            stddev=1.0 / IMAGE_PIXELS),
        name="hid_w")
    hid_b = tf.Variable(tf.zeros([FLAGS.hidden_units]), name="hid_b")

    # Variables of the softmax layer
    sm_w = tf.Variable(
        tf.truncated_normal(
            [FLAGS.hidden_units, 10],
            stddev=1.0 / math.sqrt(FLAGS.hidden_units)),
        name="sm_w")
    sm_b = tf.Variable(tf.zeros([10]), name="sm_b")

    # Ops: located on the worker specified with FLAGS.task_index
    x = tf.placeholder(tf.float32, [None, IMAGE_PIXELS * IMAGE_PIXELS])
    y_ = tf.placeholder(tf.float32, [None, 10])

    hid_lin = tf.nn.xw_plus_b(x, hid_w, hid_b)
    hid = tf.nn.relu(hid_lin)

    y = tf.nn.softmax(tf.nn.xw_plus_b(hid, sm_w, sm_b))
    cross_entropy = -tf.reduce_sum(y_ * tf.log(tf.clip_by_value(y, 1e-10, 1.0)))

    opt = tf.train.AdamOptimizer(FLAGS.learning_rate)

    if FLAGS.sync_replicas:
      if FLAGS.replicas_to_aggregate is None:
        replicas_to_aggregate = num_workers
      else:
        replicas_to_aggregate = FLAGS.replicas_to_aggregate

      opt = tf.train.SyncReplicasOptimizer(
          opt,
          replicas_to_aggregate=replicas_to_aggregate,
          total_num_replicas=num_workers,
          name="mnist_sync_replicas")

    train_step = opt.minimize(cross_entropy, global_step=global_step)

    if FLAGS.sync_replicas:
      local_init_op = opt.local_step_init_op
      if is_chief:
        local_init_op = opt.chief_init_op

      ready_for_local_init_op = opt.ready_for_local_init_op

      # Initial token and chief queue runners required by the sync_replicas mode
      chief_queue_runner = opt.get_chief_queue_runner()
      sync_init_op = opt.get_init_tokens_op()

    init_op = tf.global_variables_initializer()
    train_dir = tempfile.mkdtemp()

    if FLAGS.sync_replicas:
      sv = tf.train.Supervisor(
          is_chief=is_chief,
          logdir=train_dir,
          init_op=init_op,
          local_init_op=local_init_op,
          ready_for_local_init_op=ready_for_local_init_op,
          recovery_wait_secs=1,
          global_step=global_step)
    else:
      sv = tf.train.Supervisor(
          is_chief=is_chief,
          logdir=train_dir,
          init_op=init_op,
          recovery_wait_secs=1,
          global_step=global_step)

    sess_config = tf.ConfigProto(
        allow_soft_placement=True,
        log_device_placement=False,
        device_filters=["/job:ps",
                        "/job:worker/task:%d" % FLAGS.task_index])

    # The chief worker (task_index==0) session will prepare the session,
    # while the remaining workers will wait for the preparation to complete.
    if is_chief:
      print("Worker %d: Initializing session..." % FLAGS.task_index)
    else:
      print("Worker %d: Waiting for session to be initialized..." %
            FLAGS.task_index)

    if FLAGS.existing_servers:
      server_grpc_url = "grpc://" + worker_spec[FLAGS.task_index]
      print("Using existing server at: %s" % server_grpc_url)

      sess = sv.prepare_or_wait_for_session(server_grpc_url, config=sess_config)
    else:
      sess = sv.prepare_or_wait_for_session(server.target, config=sess_config)

    print("Worker %d: Session initialization complete." % FLAGS.task_index)

    if FLAGS.sync_replicas and is_chief:
      # Chief worker will start the chief queue runner and call the init op.
      sess.run(sync_init_op)
      sv.start_queue_runners(sess, [chief_queue_runner])

    # Perform training
    time_begin = time.time()
    print("Training begins @ %f" % time_begin)

    local_step = 0
    while True:
      # Training feed
      batch_xs, batch_ys = mnist.train.next_batch(FLAGS.batch_size)
      train_feed = {x: batch_xs, y_: batch_ys}

      _, step = sess.run([train_step, global_step], feed_dict=train_feed)
      local_step += 1

      now = time.time()
      print("%f: Worker %d: training step %d done (global step: %d)" %
            (now, FLAGS.task_index, local_step, step))

      if step >= FLAGS.train_steps:
        break

    time_end = time.time()
    print("Training ends @ %f" % time_end)
    training_time = time_end - time_begin
    print("Training elapsed time: %f s" % training_time)

    # Validation feed
    val_feed = {x: mnist.validation.images, y_: mnist.validation.labels}
    val_xent = sess.run(cross_entropy, feed_dict=val_feed)
    print("After %d training step(s), validation cross entropy = %g" %
          (FLAGS.train_steps, val_xent))


if __name__ == "__main__":
  tf.app.run()