rnn_multi_gpu.py

#!/usr/bin/env python
from six.moves import xrange  # pylint: disable=redefined-builtin
import re
import math
import time
import numpy as np
from datetime import datetime

import reader
import tensorflow as tf
from tensorflow.python.ops import rnn

FLAGS = tf.app.flags.FLAGS

tf.app.flags.DEFINE_integer('batch_size', 64, """Batch size.""")
tf.app.flags.DEFINE_integer('num_batches', 100, """Number of batches to run.""")
tf.app.flags.DEFINE_integer('num_layers', 1, """Number of batches to run.""")
tf.app.flags.DEFINE_integer('max_len', 100, """Number of batches to run.""")
tf.app.flags.DEFINE_integer('hidden_size', 128, """Number of batches to run.""")
tf.app.flags.DEFINE_integer('emb_size', 64, """Number of batches to run.""")
tf.app.flags.DEFINE_boolean('log_device_placement', False,
                            """Whether to log device placement.""")
tf.app.flags.DEFINE_integer('num_gpus', 4, """How many GPUs to use.""")

VOCAB_SIZE = 30000
NUM_CLASS = 2

NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN = 50000
NUM_EPOCHS_PER_DECAY = 50
INITIAL_LEARNING_RATE = 0.1
LEARNING_RATE_DECAY_FACTOR = 0.1
TOWER_NAME = 'tower'

train_dataset = reader.create_datasets("imdb.pkl", VOCAB_SIZE)


def get_incoming_shape(incoming):
    """ Returns the incoming data shape """
    if isinstance(incoming, tf.Tensor):
        return incoming.get_shape().as_list()
    elif type(incoming) in [np.array, list, tuple]:
        return np.shape(incoming)
    else:
        raise Exception("Invalid incoming layer.")


# Note input * W is done in LSTMCell, 
# which is different from PaddlePaddle
def single_lstm(name,
                incoming,
                n_units,
                use_peepholes=True,
                return_seq=False,
                return_state=False):
    with tf.name_scope(name) as scope:
        cell = tf.nn.rnn_cell.LSTMCell(n_units, use_peepholes=use_peepholes)
        output, _cell_state = rnn.rnn(cell, incoming, dtype=tf.float32)
        out = output if return_seq else output[-1]
        return (out, _cell_state) if return_state else out


def lstm(name,
         incoming,
         n_units,
         use_peepholes=True,
         return_seq=False,
         return_state=False,
         num_layers=1):
    with tf.name_scope(name) as scope:
        lstm_cell = tf.nn.rnn_cell.LSTMCell(
            n_units, use_peepholes=use_peepholes)
        cell = tf.nn.rnn_cell.MultiRNNCell([lstm_cell] * num_layers)
        initial_state = cell.zero_state(FLAGS.batch_size, dtype=tf.float32)
        if not isinstance(incoming, list):
            # if the input is embeding, the Tensor shape : [None, time_step, emb_size]
            incoming = [
                tf.squeeze(input_, [1])
                for input_ in tf.split(1, FLAGS.max_len, incoming)
            ]
        outputs, state = tf.nn.rnn(cell,
                                   incoming,
                                   initial_state=initial_state,
                                   dtype=tf.float32)
        out = outputs if return_seq else outputs[-1]
        return (out, _cell_state) if return_state else out


def embedding(name, incoming, vocab_size, emb_size):
    with tf.name_scope(name) as scope:
        #with tf.device("/cpu:0"):
        embedding = tf.get_variable(
            name + '_emb', [vocab_size, emb_size], dtype=tf.float32)
        out = tf.nn.embedding_lookup(embedding, incoming)
        return out


def fc(name, inpOp, nIn, nOut, act=True):
    with tf.name_scope(name) as scope:
        kernel = tf.get_variable(
            name + '_w', [nIn, nOut],
            initializer=tf.truncated_normal_initializer(
                stddev=0.01, dtype=tf.float32),
            dtype=tf.float32)

        biases = tf.get_variable(
            name + '_b', [nOut],
            initializer=tf.constant_initializer(
                value=0.0, dtype=tf.float32),
            dtype=tf.float32,
            trainable=True)

        net = tf.nn.relu_layer(inpOp, kernel, biases, name=name) if act else \
                  tf.matmul(inpOp, kernel) + biases

        return net


def inference(seq):
    net = embedding('emb', seq, VOCAB_SIZE, FLAGS.emb_size)
    print "emb:", get_incoming_shape(net)
    net = lstm('lstm', net, FLAGS.hidden_size, num_layers=FLAGS.num_layers)
    print "lstm:", get_incoming_shape(net)
    net = fc('fc1', net, FLAGS.hidden_size, 2)
    return net


def loss(logits, labels):
    # one label index for one sample
    #labels = tf.cast(labels, tf.int64)
    # cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(
    #                logits, labels, name='cross_entropy_per_example')
    labels = tf.cast(labels, tf.float32)
    cross_entropy = tf.nn.softmax_cross_entropy_with_logits(
        logits, labels, name='cross_entropy_per_example')
    cross_entropy_mean = tf.reduce_mean(cross_entropy, name='cross_entropy')
    tf.add_to_collection('losses', cross_entropy_mean)
    return tf.add_n(tf.get_collection('losses'), name='total_loss')


def tower_loss(scope):
    """Calculate the total loss on a single tower running the model.
    Args:
        scope: unique prefix string identifying the tower, e.g. 'tower_0'
    Returns:
        Tensor of shape [] containing the total loss for a batch of data
    """
    data, label = train_dataset.next_batch(FLAGS.batch_size)

    # Build a Graph that computes the logits predictions from the
    # inference model.
    last_layer = inference(data)

    # Build the portion of the Graph calculating the losses. Note that we will
    # assemble the total_loss using a custom function below.
    #_ = loss(last_layer, label)
    _ = loss(last_layer, label)

    # 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]*/' % 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 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(0, 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 time_tensorflow_run(session, target):
    num_steps_burn_in = 80
    total_duration = 0.0
    total_duration_squared = 0.0
    for i in xrange(FLAGS.num_batches + num_steps_burn_in):
        start_time = time.time()
        _ = session.run(target, feed_dict={x_input: data, y_input: label})
        _, loss_value = session.run(target)
        duration = time.time() - start_time
        if i > num_steps_burn_in:
            if not i % 10:
                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
                sec_per_batch = duration

                format_str = (
                    '%s: step %d, loss= %.2f (%.1f examples/sec; %.3f '
                    'sec/batch batch_size= %d)')
                print(format_str %
                      (datetime.now(), i - num_steps_burn_in, loss_value,
                       duration, sec_per_batch, num_examples_per_step))

            total_duration += duration
            total_duration_squared += duration * duration

    mn = total_duration / FLAGS.num_batches
    vr = total_duration_squared / FLAGS.num_batches - mn * mn
    sd = math.sqrt(vr)
    print('%s: FwdBwd across %d steps, %.3f +/- %.3f sec / batch' %
          (datetime.now(), FLAGS.num_batches, mn, sd))


def run_benchmark():
    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 = (NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN /
                                 FLAGS.batch_size)
        decay_steps = int(num_batches_per_epoch * NUM_EPOCHS_PER_DECAY)

        # Create an optimizer that performs gradient descent.
        opt = tf.train.AdamOptimizer(0.001)

        #train_dataset = reader.create_datasets("imdb.pkl", VOCAB_SIZE)

        # Calculate the gradients for each model tower.
        tower_grads = []
        for i in xrange(FLAGS.num_gpus):
            with tf.device('/gpu:%d' % i):
                with tf.name_scope('%s_%d' % (TOWER_NAME, i)) as scope:
                    # Calculate the loss for one tower of the model. This function
                    # constructs the entire model but shares the variables across
                    # all towers.
                    loss = tower_loss(scope)

                    # 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 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)

        # Apply the gradients to adjust the shared variables.
        apply_gradient_op = opt.apply_gradients(grads, global_step=global_step)

        # Group all updates to into a single train op.
        train_op = tf.group(apply_gradient_op)

        # Build an initialization operation.
        init = tf.initialize_all_variables()

        # 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)
        time_tensorflow_run(sess, [train_op, loss])


def main(_):
    run_benchmark()


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