使用Tensorflow构建一个性能超差的word2vec

使用Tensorflow构建一个性能超差的word2vec

https://www.jianshu.com/p/25287d969a30
首先是对word2vec的原理（只针对：https://github.com/tensorflow/tensorflow/blob/r1.2/tensorflow/examples/tutorials/word2vec/word2vec_basic.py的实现）的一个直觉层面的理解：
word2vec第一步随机生成每个单词的embedding，存到一个变量池里。如果你的词汇表长度是vocabulary_size（也就是你一共就研究这么几个单词，别的单词都归到unknown）， 而对每个单词，你想用embedding_size维的向量来表示的话，那这个变量池自然是一个有vocabulary_size行，embedding_size列的矩阵啦。
然后就是给每个单词打标签，如果是Skip-gram Model的话，单词w的标签，就是出现在它前面的（紧挨着）单词（的词汇表id），或者是出现在他后面的单词。对，一个单词会有两个标签（参数不一样的话，label还会更多）。也就是一张图片既是猫，又是狗。似乎有点矛盾，但计算机会用数学的方法消化掉这种矛盾。
现在输入，输出，就明了了，输入是一个单词对应的embedding变量池中相应的一行，输出是它对应的label（label不止一个的话，就多次输入输出）。
如果你的embedding_size是128的话，那你的模型按理说应该是128进1出的一个模型。
但是，跟识别数字一样，对于输出，你不能用0表示0，用1表示1，
你得用【1，0,0，...】这个onehot的向量表示0,1的话类似
所以，你的输出不是1维的，而是你要分几类就是几维的。数字识别是分10类，所以y是10维的，word2vec要分的类和你的词汇表数量一样多，所以y是vocabulary_size维（以5000为例）的（onehot形式）。
恩，你就是要训练一个128进5000出的分类器。在不断训练的过程中，通过梯度下降，不仅更新分类器的参数，还要更新embedding变量池中的每一个数。训练结束后，我们反而并不太关心这个分类器，而是更希望得到性能更好的embedding变量池。
所以代码的话：

'''
Created on 2017-6-19

@author: Administrator
'''
import collections
import math
import os

import numpy as np

import tensorflow as tf
import random

path="E:\\NLPdata\\"

filename = "text8"

with open(path+filename) as f:
    data=tf.compat.as_str(f.read()).split()

words=data
print("data size:"+str(len(words)))

vocabulary_size=5000

count=[['UNK',-1]]

count.extend(collections.Counter(words).most_common(vocabulary_size - 1))
print(count[:5])

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
        unk_count+=1
    data.append(index)
count[0][1]=unk_count

reverse_dictionary = dict(zip(dictionary.values(), dictionary.keys()))
print('Most common words (+UNK)', count[:5])
print('Sample data', data[:10], [reverse_dictionary[i] for i in data[:10]])

data_index = 0

def one_hot(index,ds):
    oh=np.zeros(ds)
    oh[index]=1
    return oh

def generate_batch(batch_size, num_skips, skip_window):
    global data_index
    assert batch_size % num_skips == 0
    assert num_skips <= 2 * skip_window
    batch = np.ndarray(shape=(batch_size), dtype=np.int32)
    labels = np.ndarray(shape=(batch_size, vocabulary_size), dtype=np.int32)
    span = 2 * skip_window + 1 # [ skip_window target skip_window ]
    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] = one_hot(buffer[target],vocabulary_size)
        buffer.append(data[data_index])
        data_index = (data_index + 1) % len(data)
    return batch, labels

print('data:', [reverse_dictionary[di] for di in data[:8]])

for num_skips, skip_window in [(2, 1), (4, 2)]:
    data_index = 0
    batch, labels = generate_batch(batch_size=8, num_skips=num_skips, skip_window=skip_window)
    print('\nwith num_skips = %d and skip_window = %d:' % (num_skips, skip_window))
    print('    batch:', [reverse_dictionary[bi] for bi in batch])
    print('    labels:',labels)



# batch_size = 128
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(range(valid_window), valid_size))

# num_sampled = 64 # Number of negative examples to sample.
 
graph = tf.Graph()

with graph.as_default(), tf.device('/cpu:0'):
    train_dataset = tf.placeholder(tf.int32, shape=[batch_size])
    train_labels = tf.placeholder(tf.float32, shape=[batch_size, vocabulary_size])
    valid_dataset = tf.constant(valid_examples, dtype=tf.int32)
    embeddings = tf.Variable(tf.random_uniform([vocabulary_size, embedding_size], -1.0, 1.0))
    #softmax_weights = tf.Variable(tf.truncated_normal([vocabulary_size, embedding_size], stddev=1.0 / math.sqrt(embedding_size)))
    #softmax_biases = tf.Variable(tf.zeros([vocabulary_size]))
    W = tf.Variable(tf.zeros([embedding_size, vocabulary_size]))
    b = tf.Variable(tf.zeros([vocabulary_size]))
    embed = tf.nn.embedding_lookup(embeddings, train_dataset)
    y=tf.nn.softmax(tf.matmul(embed,W)+b)
    loss = tf.reduce_mean(-tf.reduce_sum(train_labels * tf.log(y), reduction_indices=[1]))
    optimizer = tf.train.AdagradOptimizer(1.0).minimize(loss)
    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, tf.transpose(normalized_embeddings))
 
num_steps = 100001
 
with tf.Session(graph=graph) as session:
    tf.global_variables_initializer().run()
    print('Initialized')
    average_loss = 0
    for step in range(num_steps):
        batch_data, batch_labels = generate_batch(
                                                  batch_size, num_skips, skip_window)
        feed_dict = {train_dataset : batch_data, train_labels : batch_labels}
        _, l = session.run([optimizer, loss], feed_dict=feed_dict)
        average_loss += l
        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 %d: %f' % (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 range(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 = 'Nearest to %s:' % valid_word
                for k in range(top_k):
                    close_word = reverse_dictionary[nearest[k]]
                    log = '%s %s,' % (log, close_word)
                print(log)
    final_embeddings = normalized_embeddings.eval()

其实就是基于原来的代码改的，主要改动包括，
把词汇表降到5000，降低y的维度
label直接变成one-hot
把训练模型直接改成softmax分类器，更容易看懂，
loss函数就是之前的交叉熵，是不是很亲切，

这段代码训练慢，效果一般，但还是有一定效果。不过它说明了word2vec就是个分类器的本质，对理解原理有帮助。

事实上，这里实现的就是官方文档里，说的那个因为效果差被对比的方法（下图）。哈哈，就是这样。不过我们实现了这个的基础上，理解了原理，再去亲手实现NCE（而不是直接调函数），岂不是对NCE掌握的更好了？