tensorflow_word2vec.py

# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
# Modifications copyright (C) 2017 Uber Technologies, Inc.
#
# 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.
# ==============================================================================
"""Basic word2vec example."""

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

import collections
import math
import os
import random
import zipfile

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

# Horovod: initialize Horovod.
hvd.init()


# Step 1: Download the data.
url = 'http://mattmahoney.net/dc/text8.zip'


def maybe_download(filename, expected_bytes):
    """Download a file if not present, and make sure it's the right size."""
    if not os.path.exists(filename):
        filename, _ = urllib.request.urlretrieve(url, filename)
    statinfo = os.stat(filename)
    if statinfo.st_size == expected_bytes:
        print('Found and verified', filename)
    else:
        print(statinfo.st_size)
        raise Exception(
            'Failed to verify ' + url + '. Can you get to it with a browser?')
    return filename

filename = maybe_download('text8-%d.zip' % hvd.rank(), 31344016)


# Read the data into a list of strings.
def read_data(filename):
    """Extract the first file enclosed in a zip file as a list of words."""
    with zipfile.ZipFile(filename) as f:
        data = tf.compat.as_str(f.read(f.namelist()[0])).split()
    return data

vocabulary = read_data(filename)
print('Data size', len(vocabulary))

# Step 2: Build the dictionary and replace rare words with UNK token.
vocabulary_size = 50000


def build_dataset(words, n_words):
    """Process raw inputs into a dataset."""
    count = [['UNK', -1]]
    count.extend(collections.Counter(words).most_common(n_words - 1))
    dictionary = dict()
    for word, _ in count:
        dictionary[word] = len(dictionary)
    data = list()
    unk_count = 0
    for word in words:
        if word in dictionary:
            index = dictionary[word]
        else:
            index = 0  # dictionary['UNK']
            unk_count += 1
        data.append(index)
    count[0][1] = unk_count
    reversed_dictionary = dict(zip(dictionary.values(), dictionary.keys()))
    return data, count, dictionary, reversed_dictionary

data, count, dictionary, reverse_dictionary = build_dataset(vocabulary,
                                                            vocabulary_size)
del vocabulary  # Hint to reduce memory.
print('Most common words (+UNK)', count[:5])
print('Sample data', data[:10], [reverse_dictionary[i] for i in data[:10]])


# Step 3: Function to generate a training batch for the skip-gram model.
def generate_batch(batch_size, num_skips, skip_window):
    assert num_skips <= 2 * skip_window
    # Adjust batch_size to match num_skips
    batch_size = batch_size // num_skips * num_skips
    span = 2 * skip_window + 1  # [ skip_window target skip_window ]
    # Backtrack a little bit to avoid skipping words in the end of a batch
    data_index = random.randint(0, len(data) - span - 1)
    batch = np.ndarray(shape=(batch_size), dtype=np.int32)
    labels = np.ndarray(shape=(batch_size, 1), dtype=np.int32)
    buffer = collections.deque(maxlen=span)
    for _ in range(span):
        buffer.append(data[data_index])
        data_index = (data_index + 1) % len(data)
    for i in range(batch_size // num_skips):
        target = skip_window  # target label at the center of the buffer
        targets_to_avoid = [skip_window]
        for j in range(num_skips):
            while target in targets_to_avoid:
                target = random.randint(0, span - 1)
            targets_to_avoid.append(target)
            batch[i * num_skips + j] = buffer[skip_window]
            labels[i * num_skips + j, 0] = buffer[target]
        buffer.append(data[data_index])
        data_index = (data_index + 1) % len(data)
    return batch, labels

batch, labels = generate_batch(batch_size=8, num_skips=2, skip_window=1)
for i in range(8):
    print(batch[i], reverse_dictionary[batch[i]],
          '->', labels[i, 0], reverse_dictionary[labels[i, 0]])

# Step 4: Build and train a skip-gram model.

max_batch_size = 128
embedding_size = 128  # Dimension of the embedding vector.
skip_window = 1       # How many words to consider left and right.
num_skips = 2         # How many times to reuse an input to generate a label.

# We pick a random validation set to sample nearest neighbors. Here we limit the
# validation samples to the words that have a low numeric ID, which by
# construction are also the most frequent.
valid_size = 16     # Random set of words to evaluate similarity on.
valid_window = 100  # Only pick dev samples in the head of the distribution.
valid_examples = np.random.choice(valid_window, valid_size, replace=False)
num_sampled = 64    # Number of negative examples to sample.

graph = tf.Graph()

with graph.as_default():

    # Input data.
    train_inputs = tf.placeholder(tf.int32, shape=[None])
    train_labels = tf.placeholder(tf.int32, shape=[None, 1])
    valid_dataset = tf.constant(valid_examples, dtype=tf.int32)

    # Look up embeddings for inputs.
    embeddings = tf.Variable(
        tf.random_uniform([vocabulary_size, embedding_size], -1.0, 1.0))
    embed = tf.nn.embedding_lookup(embeddings, train_inputs)

    # Construct the variables for the NCE loss
    nce_weights = tf.Variable(
        tf.truncated_normal([vocabulary_size, embedding_size],
                            stddev=1.0 / math.sqrt(embedding_size)))
    nce_biases = tf.Variable(tf.zeros([vocabulary_size]))

    # Compute the average NCE loss for the batch.
    # tf.nce_loss automatically draws a new sample of the negative labels each
    # time we evaluate the loss.
    loss = tf.reduce_mean(
        tf.nn.nce_loss(weights=nce_weights,
                       biases=nce_biases,
                       labels=train_labels,
                       inputs=embed,
                       num_sampled=num_sampled,
                       num_classes=vocabulary_size))

    # Horovod: adjust learning rate based on number of GPUs.
    optimizer = tf.train.GradientDescentOptimizer(1.0 * hvd.size())

    # Horovod: add Horovod Distributed Optimizer.
    optimizer = hvd.DistributedOptimizer(optimizer)

    train_op = optimizer.minimize(loss)

    # Compute the cosine similarity between minibatch examples and all embeddings.
    norm = tf.sqrt(tf.reduce_sum(tf.square(embeddings), 1, keep_dims=True))
    normalized_embeddings = embeddings / norm
    valid_embeddings = tf.nn.embedding_lookup(
        normalized_embeddings, valid_dataset)
    similarity = tf.matmul(
        valid_embeddings, normalized_embeddings, transpose_b=True)

    # Add variable initializer.
    init = tf.global_variables_initializer()

    # Horovod: broadcast initial variable states from rank 0 to all other processes.
    # This is necessary to ensure consistent initialization of all workers when
    # training is started with random weights or restored from a checkpoint.
    bcast = hvd.broadcast_global_variables(0)

# Step 5: Begin training.

# Horovod: adjust number of steps based on number of GPUs.
num_steps = 100000 // hvd.size() + 1

# Horovod: pin GPU to be used to process local rank (one GPU per process)
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
config.gpu_options.visible_device_list = str(hvd.local_rank())

with tf.Session(graph=graph, config=config) as session:
    # We must initialize all variables before we use them.
    init.run()
    bcast.run()
    print('Initialized')

    average_loss = 0
    for step in xrange(num_steps):
        # simulate various sentence length by randomization
        batch_size = random.randint(max_batch_size // 2, max_batch_size)
        batch_inputs, batch_labels = generate_batch(
            batch_size, num_skips, skip_window)
        feed_dict = {train_inputs: batch_inputs, train_labels: batch_labels}

        # We perform one update step by evaluating the optimizer op (including it
        # in the list of returned values for session.run()
        _, loss_val = session.run([train_op, loss], feed_dict=feed_dict)
        average_loss += loss_val

        if step % 2000 == 0:
            if step > 0:
                average_loss /= 2000
            # The average loss is an estimate of the loss over the last 2000 batches.
            print('Average loss at step ', step, ': ', average_loss)
            average_loss = 0
    final_embeddings = normalized_embeddings.eval()

    # Evaluate similarity in the end on worker 0.
    if hvd.rank() == 0:
        sim = similarity.eval()
        for i in xrange(valid_size):
            valid_word = reverse_dictionary[valid_examples[i]]
            top_k = 8  # number of nearest neighbors
            nearest = (-sim[i, :]).argsort()[1:top_k + 1]
            log_str = 'Nearest to %s:' % valid_word
            for k in xrange(top_k):
                close_word = reverse_dictionary[nearest[k]]
                log_str = '%s %s,' % (log_str, close_word)
            print(log_str)