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) ? ???
# 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 (len? ?? ????? ????? index? ??)
    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_skips? ??? batch? ?????.
    assert num_skips <= 2 * skip_window # num_skips == 2*skip_window ?? ?? context window? context? ?? pair? ????.
    # ?, ? ?? ??? ? ?.

    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".
    # ??? ?? push(append) & pop ? ? ? ??.

    # 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)

    # // ? ??? ?? ??? ??? ??? ???
    # skip-gram? ?? ????? ??? ???? ??? ???? ????.
    # skip-gram model? ???? ??, words? (target, context) ??? ??? ??? ??.
    # ?? ??? ? ??? batch_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:
                # context window?? context? ???? ??? ???? ?????.
                # ?, skip_window*2 == num_skips ? ??, ??? ?? context? ? ????? ??? ? ??? ??. ??? ???? ? ?.
                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? ??? ?????? ?? ?? ?? ??? ??:
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] ? numpy array? ??? ??? valid_size ?? ????.
# ?, ???? 0~99 ??? ? ? ???? 16?? ?? ?? valid_examples ?.
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? GPU implementation? ??? ??? ??? CPU? ???.
    # default? GPU?? ????? 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))
        # ?? embedding matrix?? train_inputs (mini-batch; indices) ? ???? ??? ???? ??
        embed = tf.nn.embedding_lookup(embeddings, train_inputs)

        # Construct the variables for the NCE loss
        # NCE loss ? logistic regression model ? ???? ????.
        # ?, logistic regression ? ??, vocabulary? ? ???? ?? weight? bias? ???.
        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.
    # minibatch (valid_embeddings) ? all embeddings ??? cosine similarity? ????.
    # ? ??? ??? ????? ? valid_example ??? ?? ??? ??? ?? ???? ???? ???? (? ?? ??? ???? ??).
    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()
        # feed_dict? ???? placeholder? ???? ???? ????.
        _, 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:
    # ?? ??? ??? ???, scikit-learn ? matplotlib ? ?????? ??????.
    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.")

  

时间: 2024-10-29 19:06:21

tensorflow实现Word2vec的相关文章

文本分布式表示(二):用tensorflow和word2vec训练词向量

博客园的markdown用起来太心塞了,现在重新用其他编辑器把这篇博客整理了一下. 目前用word2vec算法训练词向量的工具主要有两种:gensim 和 tensorflow.gensim中已经封装好了word2vec这个包,用起来很方便,只要把文本处理成规范的输入格式,寥寥几行代码就能训练词向量.这样比较适合在做项目时提高效率,但是对理解算法的原理帮助不大.相比之下,用tensorflow来训练word2vec比较麻烦,生成batch.定义神经网络的各种参数,都要自己做,但是对于理解算法原理

Tensorflow 的Word2vec demo解析

简单demo的代码路径在tensorflow\tensorflow\g3doc\tutorials\word2vec\word2vec_basic.py Sikp gram方式的model思路 http://tensorflow.org/tutorials/word2vec/index.md 另外可以参考cs224d课程的课件. ? ? 窗口设置为左右1个词 对应skip gram模型 就是一个单词预测其周围单词(cbow模型是 输入一系列context词,预测一个中心词) ? ? Quick

第十四章——循环神经网络(Recurrent Neural Networks)(第二部分)

本章共两部分,这是第二部分: 第十四章--循环神经网络(Recurrent Neural Networks)(第一部分) 第十四章--循环神经网络(Recurrent Neural Networks)(第二部分) 14.4 深度RNN 堆叠多层cell是很常见的,如图14-12所示,这就是一个深度RNN. 图14-12 深度RNN(左),随时间展开(右) 在TensorFlow中实现深度RNN,需要创建多个cell并将它们堆叠到一个MultiRNNCell中.下面的代码创建了三个完全相同的cel

唐宇迪-人工智能学习路线(下篇)

唐宇迪:51CTO微职位讲师,计算机博士,人工智能专家 内容梗概:介绍人工智能学习步骤和知识框架.人工智能的学习可以拆分为7步,此为4-7步:下篇. 第4步:深度学习 1)深度学习概述 终于说到深度学习了,都需要学什么呢?深度学习可以说是当下最好用的算法了,各个领域都能吃得开.其实最核心的还是在计算机视觉和自然语言处理中,因为神经网络算法更适用于图像和文本数据. 主要需要掌握的就是算法和框架了,算法就是CNN,RNN这些经典网络模型,框架就是实战的工具了例如tenorflow,Pytorch等,

Python Tensorflow下的Word2Vec代码解释

前言: 作为一个深度学习的重度狂热者,在学习了各项理论后一直想通过项目练手来学习深度学习的框架以及结构用在实战中的知识.心愿是好的,但机会却不好找.最近刚好有个项目,借此机会练手的过程中,我发现其实各大机器学习以及tensorflow框架群里的同学们也有类似的问题.于是希望借项目之手分享一点本人运行过程中的理解以及经验,希望在有益大家工作的基础上抛砖引玉,得到行业内各位专业人士的批评指点,多谢大家支持! 第一章博客我将会分为两个部分,这一部分将讲述Word2Vec在tensorflow中官方提供

tensorflow学习记录-word2vec

好吧,我承认,我又开始搞word2vec了,好多次了. 上一篇是: http://www.cnblogs.com/charlesblc/p/7615825.html 先在阡陌机器上面训练了: [email protected] tensorflow-word2vec]$ pwd/home/work/data/code/dl/tensorflow/tensorflow-word2vec 然后把生成的vec拷贝到自己机器,准备画图.可是发现的air没有安装tensorflow.那么就再安装一下吧.

利用Tensorflow进行自然语言处理(NLP)系列之一Word2Vec

写在前面的话(可略过): 一直想写下.整理下利用Tensorflow或Keras工具进行自然语言处理(NLP)方面的文章,对比和纠结了一段时间,发现博众家之长不如静下心来一步一个脚印地去看一本书来得更实在,虽然慢但是心里相对踏实些.近期刚把Thushan Ganegedara写的<Natural Language Processing with TensorFlow>(2018年5月第一次出版),目前没看到中文版.讲真,看原版书确实很耗费精力,但原版书的好处是可以原汁原味地探索.写博文的过程中

Tensorflow word2vec+manage experiments

Lecture note 5: word2vec + manage experiments Word2vec Most of you are probably already familiar with word embedding and understand the importance of a model like word2vec. For those who aren't, Stanford CS 224N's lecture on word vectors is a great r

Word2vec多线程(tensorflow)

workers = [] for _ in xrange(opts.concurrent_steps): t = threading.Thread(target=self._train_thread_body) t.start() workers.append(t) ? ? ? ? Word2vec.py使用了多线程 一般认为python多线程其实是单线程 由于python的设计 GPL 内存不是现成安全的 但是这里由于内部是调用c++代码 所以还是能起到多线程作用 ? ? 而 Word2vec