欢迎您访问程序员文章站本站旨在为大家提供分享程序员计算机编程知识!
您现在的位置是: 首页

使用Tensorflow构建一个性能超差的word2vec

程序员文章站 2022-06-03 21:13:40
...

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?