本文为原创文章,未经本人允许,禁止转载。转载请注明出处。
1.Word2Vec
Word2Vec相关知识请见:【深度学习基础】第四十五课:自然语言处理与词嵌入。
2.代码实现
👉载入包:
1
2
3
4
5
6
7
8
9
10
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
👉下载数据集:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
# Step 1: Download the data.
url = 'http://mattmahoney.net/dc/'
# 下载数据集
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)
# 获取文件相关属性
statinfo = os.stat(filename)
# 比对文件的大小是否正确
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)
os.path.exists()
用于判断文件是否存在,如果存在则返回true;如果不存在,则返回false。
urllib.request.urlretrieve(url,filename)
用于将url(可以是本地路径也可以是网络链接)表示的对象复制到filename(保存到本地的路径)。
os.stat(path)
用于在给定的路径上执行一个系统stat的调用。返回值:
st_mode
:inode保护模式。st_ino
:inode节点号。st_dev
:inode驻留的设备。st_nlink
:inode的链接数。st_uid
:所有者的用户ID。st_gid
:所有者的组ID。st_size
:普通文件以字节为单位的大小;包含等待某些特殊文件的数据。st_atime
:上次访问的时间。st_mtime
:最后一次修改的时间。st_ctime
:由操作系统报告的“ctime”。在某些系统上(如Unix)是最新的元数据更改的时间,在其它系统上(如Windows)是创建时间(详细信息参见平台的文档)。
👉读取数据:
1
2
3
4
5
6
7
8
9
# 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
# 单词表
words = read_data(filename)
zipfile.ZipFile(file,mode)
:如果mode=’r’,则为读取压缩文件file中的内容;如果mode=’w’,则为向压缩文件file中写入内容。
ZipFile.namelist()
:获取压缩文件内所有文件的名称列表。
tf.compat.as_str
:将目标转化为字符串格式。
👉创建一个单词表(共50000个最常见的单词,包含UNK
):
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
# Step 2: Build the dictionary and replace rare words with UNK token.
# 只留50000个单词,其他的词都归为UNK
vocabulary_size = 50000
def build_dataset(words, vocabulary_size):
count = [['UNK', -1]]
count.extend(collections.Counter(words).most_common(vocabulary_size - 1))
# 生成 dictionary,词对应编号, word:id(0-49999)
# 词频越高编号越小
dictionary = dict()
for word, _ in count:
dictionary[word] = len(dictionary)
# data把数据集的词都编号
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)
# 记录UNK词的数量
count[0][1] = unk_count
# 编号对应词的字典
reverse_dictionary = dict(zip(dictionary.values(), dictionary.keys()))
return data, count, dictionary, reverse_dictionary
# data 数据集,编号形式
# count 前50000个出现次数最多的词
# dictionary 词对应编号
# reverse_dictionary 编号对应词
data, count, dictionary, reverse_dictionary = build_dataset(words, vocabulary_size)
del words # Hint to reduce memory.
👉产生batch:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
data_index = 0
# Step 3: Function to generate a training batch for the skip-gram model.
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, 1), dtype=np.int32)
span = 2 * skip_window + 1 # [ skip_window target skip_window ]
buffer = collections.deque(maxlen=span)
# 循环3次
for _ in range(span):
buffer.append(data[data_index])
data_index = (data_index + 1) % len(data)
# 获取batch和labels
for i in range(batch_size // num_skips):
target = skip_window # target label at the center of the buffer
targets_to_avoid = [skip_window]
# 循环2次,一个目标单词对应两个上下文单词
for j in range(num_skips):
while target in targets_to_avoid:
# 可能先拿到前面的单词也可能先拿到后面的单词
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)
# Backtrack a little bit to avoid skipping words in the end of a batch
# 回溯3个词。因为执行完一个batch的操作之后,data_index会往右多偏移span个位置
data_index = (data_index + len(data) - span) % len(data)
return batch, labels
# 打印sample data
batch, labels = generate_batch(batch_size=8, num_skips=2, skip_window=1)
举个例子解释一下,数据集中前6个单词在单词表中的索引见下:
如果以第1个单词(3081)为中心,则其上下文为第0个单词(5234)和第2个单词(12);如果以第2个单词(12)为中心,则其上下文为第1个单词(3081)和第3个单词(6);剩余以此类推,则此时generate_batch
函数返回的batch为:
1
[3081 3081 12 12 6 6 195 195]
返回的labels为:
1
2
3
4
5
6
7
8
9
10
[
[5234]
[12]
[3081]
[6]
[12]
[195]
[6]
[2]
]
对应关系为:
1
2
3
4
5
6
7
8
3081 originated -> 5234 anarchism
3081 originated -> 12 as
12 as -> 3081 originated
12 as -> 6 a
6 a -> 12 as
6 a -> 195 term
195 term -> 6 a
195 term -> 2 of
collections.deque
用于产生一个双向队列,可以从两端append、extend或pop:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
import collections
d = collections.deque([])
d.append('a') # 在最右边添加一个元素,此时 d=deque('a')
d.appendleft('b') # 在最左边添加一个元素,此时 d=deque(['b', 'a'])
d.extend(['c','d']) # 在最右边添加所有元素,此时 d=deque(['b', 'a', 'c', 'd'])
d.extendleft(['e','f']) # 在最左边添加所有元素,此时 d=deque(['f', 'e', 'b', 'a', 'c', 'd'])
d.pop() # 将最右边的元素取出,返回 'd',此时 d=deque(['f', 'e', 'b', 'a', 'c'])
d.popleft() # 将最左边的元素取出,返回 'f',此时 d=deque(['e', 'b', 'a', 'c'])
d.rotate(-2) # 向左旋转两个位置(正数则向右旋转),此时 d=deque(['a', 'c', 'e', 'b'])
d.count('a') # 队列中'a'的个数,返回 1
d.remove('c') # 从队列中将'c'删除,此时 d=deque(['a', 'e', 'b'])
d.reverse() # 将队列倒序,此时 d=deque(['b', 'e', 'a'])
f=d.copy()
print(f)#deque(['b', 'e', 'a'])
f.clear()
print(f)#deque([])
#可以指定队列的长度,如果添加的元素超过指定长度,则原元素会被挤出。
e=collections.deque(maxlen=5)
e.extend([1,2,3,4,5])
e.append("a")
print(e)
#deque([2, 3, 4, 5, 'a'], maxlen=5)
e.appendleft("b")
print(e)
#deque(['b', 2, 3, 4, 5], maxlen=5)
e.extendleft(["c","d"])
print(e)
#deque(['d', 'c', 'b', 2, 3], maxlen=5)
random.randint(a,b)
:参数a和参数b必须是整数,该函数返回参数a和参数b之间的任意整数($[a,b]$)。
👉建立skip-gram模型:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
# Step 4: 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.
# 从0-100抽取16个整数,无放回抽样
valid_examples = np.random.choice(valid_window, valid_size, replace=False)
# 负采样样本数
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=[batch_size])
train_labels = tf.placeholder(tf.int32, shape=[batch_size, 1])
valid_dataset = tf.constant(valid_examples, dtype=tf.int32)
# 词向量
# Look up embeddings for inputs.
embeddings = tf.Variable(
tf.random_uniform([vocabulary_size, embedding_size], -1.0, 1.0))
# embedding_lookup(params,ids)其实就是按照ids顺序返回params中的第ids行
# 比如说,ids=[1,7,4],就是返回params中第1,7,4行。返回结果为由params的1,7,4行组成的tensor
# 提取要训练的词
embed = tf.nn.embedding_lookup(embeddings, train_inputs)
# Construct the variables for the noise-contrastive estimation(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))
# Construct the SGD optimizer using a learning rate of 1.0.
optimizer = tf.train.GradientDescentOptimizer(1).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
# 抽取一些常用词来测试余弦相似度
# valid_embeddings维度[16,128]
valid_embeddings = tf.nn.embedding_lookup(
normalized_embeddings, valid_dataset)
# valid_size == 16
# [16,128] * [128,50000] = [16,50000]
# 16个词分别与50000个单词中的每一个计算内积
similarity = tf.matmul(
valid_embeddings, normalized_embeddings, transpose_b=True)
# Add variable initializer.
init = tf.global_variables_initializer()
numpy.random.choice(a, size=None, replace=True, p=None)
用于生成随机样本,参数详解:
a
:一维数组或者一个int型整数。如果a
为数组,则从数组中的元素进行随机采样;如果a
为int型整数,则采样范围为np.arange(a)
。size
:随机采样的样本的数量。replace
:True表示有放回采样;False表示无放回采样。p
:与数组a
对应,为a
中每个元素被采样的概率。
1
2
3
4
5
6
tf.random_uniform(shape,
minval=0,
maxval=None,
dtype=dtypes.float32,
seed=None,
name=None)
用于产生[minval,maxval)范围内服从均匀分布的值。
tf.nn.embedding_lookup
的作用见上述代码注释。
tf.nn.nce_loss
:如果使用softmax函数,则类别数太多,导致计算量太大,所以这里使用NCE loss(原文:Noise-contrastive estimation: A new estimation principle for unnormalized statistical models),将多分类问题转化成二分类。
余弦相似度:
\[similarity = \cos (\theta) = \frac{A \cdot B}{\parallel A \parallel \parallel B \parallel }\]参照内积的计算。
👉训练模型:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
# Step 5: Begin training.
num_steps = 100001
final_embeddings = []
with tf.Session(graph=graph) as session:
# We must initialize all variables before we use them.
init.run()
print("Initialized")
average_loss = 0
for step in xrange(num_steps):
# 获取一个批次的target,以及对应的labels,都是编号形式的
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([optimizer, loss], feed_dict=feed_dict)
average_loss += loss_val
# 计算训练2000次的平均loss
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
# Note that this is expensive (~20% slowdown if computed every 500 steps)
if step % 20000 == 0:
sim = similarity.eval()
# 计算验证集的余弦相似度最高的词
for i in xrange(valid_size):
# 根据id拿到对应单词
valid_word = reverse_dictionary[valid_examples[i]]
top_k = 8 # number of nearest neighbors
# 从大到小排序,排除自己本身,取前top_k个值
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)
# 训练结束得到的词向量
final_embeddings = normalized_embeddings.eval()
xrange
和range
用法完全相同,所不同的是生成的不是一个数组,而是一个生成器。xrange
已在python3中被取消,和range
函数合并为range
。
argsort
将元素从小到大排序并返回其对应的索引:
1
2
3
4
5
6
7
8
import numpy as np
a=np.array([[3,2,5],[6,3,9]])
print(a[0,:]) #array([3, 2, 5])
print(-a[0,:]) #array([-3, -2, -5])
#最小值为-5,对应索引2
#第二小的值为-3,对应索引0
#最大值为-2,对应索引1
print((-a[0,:]).argsort()) #array([2, 0, 1])
👉使用TSNE进行降维可视化: