tensorflow实现Word2vec

# coding: utf-8
'''
Note: Step 3 is missing. That's why I left it.
'''

from __future__ import absolute_import
from __future__ import print_function

import collections
import math
import numpy as np
import os
import random
from six.moves import urllib
from six.moves import xrange  # pylint: disable=redefined-builtin
import tensorflow as tf
import zipfile

# Step 1: Download the data.
# Downloading data. If the file already exists, check that it was received correctly (the file size is the same).
# Return filename after download.
print("Step 1: Download the data.")
url = 'http://mattmahoney.net/dc/'

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, 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 ' + filename + '. Can you get to it with a browser?')
    return filename

filename = maybe_download('text8.zip', 31344016)

# Read the data into a string.
# file (zipfile) 을 읽어옴
# text8.zip contains only one file. Looking at the code, it seems to be words separated by ''.
def read_data(filename):
    f = zipfile.ZipFile(filename)
    for name in f.namelist():
        return f.read(name).split()
    f.close()

words = read_data(filename)
print('Data size', len(words))
print('Sample words: ', words[:10])

# Step 2: Build the dictionary and replace rare words with UNK token.
print("\nStep 2: Build the dictionary and replace rare words with UNK token.")
vocabulary_size = 50000

def build_dataset(words):
#vocabulary_size is the number of frequent words to use.
#all words that do not fit within the top 50000 (vocabulary_size) are treated as UNK.
#Param words: literally a list of words
#Return data: indices of words including UNK. That is, words index list.
#Return count: collections.Counter which counts the frequency of occurrence of each word
#:return dictionary: {"word": "index"}
#:return reverse_dictionary: {"index": "word"}. e.g.) {0: 'UNK', 1: 'the', ...}
    count = [['UNK', -1]]
    count.extend(collections.Counter(words).most_common(vocabulary_size - 1))
    dictionary = dict()
    for word, _ in count:
        dictionary[word] = len(dictionary) # insert index to dictionary (len이 계속 증가하므로 결과적으로 index의 효과)
    data = list()
    unk_count = 0
    for word in words:
       if word in dictionary:
          index = dictionary[word]
       else:
          index = 0  # dictionary['UNK']
    unk_count = unk_count + 1
    data.append(index]
    count[0][1] = unk_count
    reverse_dictionary = dict(zip(dictionary.values(), dictionary.keys()))
    return data, count, dictionary, reverse_dictionary

data, count, dictionary, reverse_dictionary = build_dataset(words)
del words  # Hint to reduce memory.
print('Most common words (+UNK)', count[:5])
print('Sample data: ', data[:10])
print('Sample count: ', count[:10])
print('Sample dict: ', dictionary.items()[:10])
print('Sample reverse dict: ', reverse_dictionary.items()[:10])

data_index = 0

# Step 4: Function to generate a training batch for the skip-gram model.
print("\nStep 4: Function to generate a training batch for the skip-gram model.")
def generate_batch(batch_size, num_skips, skip_window):

#Function to generate minibatch.
#Data_index is declared as global, which acts as static here. That is, the value of data_index is retained even if this function is continually recalled.
#Param batch_size: batch_size.
#Param num_skips: how many (target, context) pairs to generate in the context window.
#Param skip_window: context window size. The skip-gram model predicts the surrounding words from the target word, and skip_window defines the range of the surrounding words.
#Return batch: mini-batch of data.
#Return labels: labels of mini-batch. 2d array of [batch_size] [1].

    global data_index
    assert batch_size % num_skips == 0  # num_skips의 배수로 batch가 생성되므로.
    assert num_skips <= 2 * skip_window # num_skips == 2*skip_window 이면 모든 context window의 context에 대해 pair가 생성된다.
    # That is, it should not be larger than that.
    batch = np.ndarray(shape=(batch_size), dtype=np.int32)
    labels = np.ndarray(shape=(batch_size, 1), dtype=np.int32)
    span = 2 * skip_window + 1 # [ skip_window target skip_window ]
    buffer = collections.deque(maxlen=span)
    # Deques are a generalization of stacks and queues.
    # The name is pronounced "deck" and is short for "double-ended queue".
    # You can both push (append) & pop both sides.
    # buffer = data[data_index:data_index+span] with circling
    for _ in range(span):
        buffer.append(data[data_index])
        data_index = (data_index + 1) % len(data)

# operator to discard the remainder or decimal point
#Skip-gram is a model for predicting surrounding context words from target words.
#Before we learn skip-gram model, we need to convert words to (target, context) type.
#The code below will do the job with batch_size size.

    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:
                # Extracting the context from the context window is done randomly.
                # However, if skip_window * 2 == num_skips, all contexts are taken out, so random is not meaningful. The order is random only.
                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

# To see how the batch is configured, output it once:
print("Generating batch ... ")
batch, labels = generate_batch(batch_size=8, num_skips=2, skip_window=1)
print("Sample batches: ", batch[:10])
print("Sample labels: ", labels[:10])
for i in range(8):
    print(batch[i], '->', labels[i, 0])
    print(reverse_dictionary[batch[i]], '->', reverse_dictionary[labels[i, 0]])

# Step 5: Build and train a skip-gram model.
print("\nStep 5: Build and train a skip-gram model.")
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.array(random.sample(np.arange(valid_window), valid_size))
# [0 ~ valid_window] And then sampled by valid_size.
# In other words, valid_examples is a random selection of 16 from 0 to 99.
num_sampled = 64    # Number of negative examples to sample.

print("valid_examples: ", valid_examples)

graph = tf.Graph()

with graph.as_default():

    # Input data.
    train_inputs = tf.placeholder(tf.int32, shape=[batch_size])
    train_labels = tf.placeholder(tf.int32, shape=[batch_size, 1])
    valid_dataset = tf.constant(valid_examples, dtype=tf.int32)

    # Ops and variables pinned to the CPU because of missing GPU implementation
    # Embedding_lookup This GPU implementation is not implemented, so it must be a CPU.
    # Since default is GPU, it explicitly specifies CPU.
    with tf.device('/cpu:0'):
        # Look up embeddings for inputs.
        # embedding matrix (vectors)
        embeddings = tf.Variable(tf.random_uniform([vocabulary_size, embedding_size], -1.0, 1.0))
        # Only the embedding vectors pointed to by train_inputs (mini-batch; indices) in the entire embedding matrix are extracted
        embed = tf.nn.embedding_lookup(embeddings, train_inputs)

        # Construct the variables for the NCE loss
        # NCE loss is defined using a logistic regression model.
        # That is, for logistic regression, weight and bias are needed for each word of vocabulary.
        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(nce_weights, nce_biases, embed, train_labels,
                       num_sampled, vocabulary_size))

    # Construct the SGD optimizer using a learning rate of 1.0.
    optimizer = tf.train.GradientDescentOptimizer(1.0).minimize(loss)

    # Compute the cosine similarity between minibatch examples and all embeddings.
    # Calculate the cosine similarity between minibatch (valid_embeddings) and all embeddings.
    # This process is to show which words are closest to each valid_example as the learning progresses (ie to show the learning process).
    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)

# Step 6: Begin training
print("\nStep 6: Begin training")
num_steps = 100001

with tf.Session(graph=graph) as session:
    # We must initialize all variables before we use them.
    tf.initialize_all_variables().run()
    print("Initialized")

    average_loss = 0
    for step in xrange(num_steps):
        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()
        # Use feed_dict to put data into the placeholder and learn it.
        _, loss_val = session.run([optimizer, loss], feed_dict=feed_dict)
        average_loss += loss_val

        if step % 2000 == 0:
            if step > 0:
                average_loss = 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

        # note that this is expensive (~20% slowdown if computed every 500 steps)
        if step % 10000 == 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)
    final_embeddings = normalized_embeddings.eval()

# Step 7: Visualize the embeddings.
print("\nStep 7: Visualize the embeddings.")
def plot_with_labels(low_dim_embs, labels, filename='tsne.png'):
    assert low_dim_embs.shape[0] >= len(labels), "More labels than embeddings"
    plt.figure(figsize=(18, 18))  #in inches
    for i, label in enumerate(labels):
        x, y = low_dim_embs[i,:]
        plt.scatter(x, y)
        plt.annotate(label,
                     xy=(x, y),
                     xytext=(5, 2),
                     textcoords='offset points',
                     ha='right',
                     va='bottom')

    plt.savefig(filename)

try:
    # If you get an error here, update scikit-learn and matplotlib to the latest version.
    from sklearn.manifold import TSNE
    import matplotlib.pyplot as plt

    tsne = TSNE(perplexity=30, n_components=2, init='pca', n_iter=5000)
    plot_only = 500

    low_dim_embs = tsne.fit_transform(final_embeddings[:plot_only,:])
    labels = [reverse_dictionary[i] for i in xrange(plot_only)]
    plot_with_labels(low_dim_embs, labels)

except ImportError:
    print("Please install sklearn and matplotlib to visualize embeddings.")