分为4个步骤:
1)下载数据文件。即语料库。
2)处理语料库:把语料库转化为数据字典。选取语料库中频率最高的前49999个word,建立索引1:49999,其它word标为UNK(unknown),索引值为0。
3)为skip-gram model 准备数据。即一个input(target)需要有2*skip_window个output(上下文词汇)
4)建立skip-gram model graph
5)training skip-gram model
6)可视化
tf code 如下:
import tensorflow as tf
# encoding=utf8
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
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
# 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)
# 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)
# Data size
print('Data size', len(words))
Found and verified text8.zip
Data size 17005207
# 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]]
# extend追加一个列表
# Counter用来统计每个词出现的次数
# most_common返回一个TopN列表,只留50000个单词包括UNK
# c = Counter('abracadabra')
# c.most_common()
# [('a', 5), ('r', 2), ('b', 2), ('c', 1), ('d', 1)]
# c.most_common(3)
# [('a', 5), ('r', 2), ('b', 2)]
# 前50000个出现次数最多的词
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:把字典的键和值对换位置
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.
print('Most common words (+UNK)', count[:5])
print('Sample data', data[:10], [reverse_dictionary[i] for i in data[:10]])
Most common words (+UNK) [['UNK', 418391], ('the', 1061396), ('of', 593677), ('and', 416629), ('one', 411764)]
Sample data [5236, 3083, 12, 6, 195, 2, 3136, 46, 59, 156] ['anarchism', 'originated', 'as', 'a', 'term', 'of', 'abuse', 'first', 'used', 'against']
# Step 3: Function to generate a training batch for the skip-gram model.
data_index = 0
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 ]
#deque:双向队列
buffer = collections.deque(maxlen=span)
# [ skip_window target skip_window ]
# [ skip_window target skip_window ]
# [ skip_window target skip_window ]
# [0 1 2 3 4 5 6 7 8 9 ...]
# t i
# 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)#这里是从0,1,2中随机取一个,它包含两端的值,注意和np.random.randint的区别
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)
for i in range(8):
print(batch[i], reverse_dictionary[batch[i]],
'->', labels[i, 0], reverse_dictionary[labels[i, 0]])
3083 originated -> 5236 anarchism
3083 originated -> 12 as
12 as -> 6 a
12 as -> 3083 originated
6 a -> 12 as
6 a -> 195 term
195 term -> 6 a
195 term -> 2 of
# Step 4: Build and train a skip-gram model.
batch_size = 128
# 词向量维度
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。
'''
注:num_skips = skip_window*2。这是必然的关系。如对应一段文本 a b c d e
如果skip_window = 1,num_skips = 2 对于c生成如下的上下文字符串
c -> b
c -> d
如果skip_window = 2,num_skips = 4 对于c生成如下的上下文字符串
c -> b
c -> d
c -> a
c -> e
注意这些字符串是没有先后顺序的。所以从这点而言,该模型已经遗漏了相对位置信息,有改进的空间。
'''
# 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)
# Ops and variables pinned to the CPU because of missing GPU implementation
# with tf.device('/cpu:0'):
# 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(余弦相似度) 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 = tf.nn.embedding_lookup(
normalized_embeddings, valid_dataset)
# valid_size == 16
# [16,1] * [1*50000] = [16,50000]
similarity = tf.matmul(
valid_embeddings, normalized_embeddings, transpose_b=True)
# Add variable initializer.
init = tf.global_variables_initializer()
# 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()
Initialized
Average loss at step 0 : 272.700042725
Nearest to than: segal, disgusting, vesicle, malaysians, cipher, plantinga, librarians, beach,
Nearest to there: tuscan, rather, pharsalus, willed, buckingham, handler, salzburg, expanse,
Nearest to no: pipelines, intuitions, burglar, coms, schrader, scalar, chlorine, restrained,
Nearest to a: neko, unmop, drama, madeline, quetzalcoatl, virginian, anschluss, fogo,
Nearest to was: fought, lp, museums, roster, assistant, penitent, hindus, ricardian,
Nearest to and: frail, migratory, rims, jewellery, outsold, rampant, tranquillity, parallax,
Nearest to to: ofcom, fortaleza, mast, libre, forums, bloodshed, vampyre, mummification,
Nearest to their: su, decided, magma, dong, spellings, contrapuntal, ulm, troglodytes,
Nearest to is: expects, dragged, slaughters, taipa, howitzer, conjugal, guessed, paran,
Nearest to i: predisposition, cpr, denard, desktops, crocodile, renewal, regatta, lama,
Nearest to years: amigaos, samplers, toolbars, annexing, joysticks, heraldic, unrecognized, epileptic,
Nearest to these: expressionism, pipes, neighbouring, acheson, superstition, drivetrain, harper, hot,
Nearest to often: status, evangelion, conjoined, declension, scooby, somatic, lpc, despot,
Nearest to however: marquesas, she, ensuring, knuth, footy, ciphertext, loew, workout,
Nearest to new: consultants, mohs, altruism, frasier, hybrid, biking, intrigue, quantity,
Nearest to has: cyanide, amal, frontispiece, stabilization, chad, beet, milestones, orpheus,
Average loss at step 2000 : 113.445801137
Average loss at step 4000 : 52.9759702828
Average loss at step 6000 : 33.313842569
Average loss at step 8000 : 23.3657903601
Average loss at step 10000 : 17.8983022878
Average loss at step 12000 : 14.2972284861
Average loss at step 14000 : 11.6379432209
Average loss at step 16000 : 9.91554711998
Average loss at step 18000 : 8.48712806338
Average loss at step 20000 : 7.9150552063
Nearest to than: nine, or, beach, vesicle, gnat, cipher, estimated, shaded,
Nearest to there: it, also, tuscan, frank, then, salzburg, rather, fathoms,
Nearest to no: moor, nitroglycerin, pipelines, msg, three, schopenhauer, and, phi,
Nearest to a: the, this, circ, agouti, his, rita, and, collectivist,
Nearest to was: is, by, were, as, are, agouti, and, has,
Nearest to and: in, or, of, mya, circ, UNK, agouti, dasyprocta,
Nearest to to: and, nine, for, picks, saints, in, with, traces,
Nearest to their: the, his, agouti, its, levites, a, an, machined,
Nearest to is: was, are, agouti, by, altenberg, has, were, as,
Nearest to i: vaccines, predisposition, gesta, crocodile, four, marx, renewal, advantage,
Nearest to years: gum, year, circ, annexing, dasyprocta, eight, mafia, agouti,
Nearest to these: pipes, content, reaching, suffixes, jude, operatorname, amazons, the,
Nearest to often: status, psi, somatic, agouti, tones, questionable, adelaide, norway,
Nearest to however: she, agouti, seminar, burroughs, and, footy, krebs, coke,
Nearest to new: backslash, quantity, altruism, park, poetic, archie, circ, hybrid,
Nearest to has: is, was, had, agouti, hitchcock, and, flagella, milestones,
……
Average loss at step 82000 : 4.77714890552
Average loss at step 84000 : 4.76479504359
Average loss at step 86000 : 4.76726497281
Average loss at step 88000 : 4.75223901606
Average loss at step 90000 : 4.74017631125
Average loss at step 92000 : 4.65827371514
Average loss at step 94000 : 4.73972713518
Average loss at step 96000 : 4.68014509404
Average loss at step 98000 : 4.58637025118
Average loss at step 100000 : 4.70240500081
Nearest to than: or, much, albury, cartographers, adria, cebus, mla, vesicle,
Nearest to there: they, it, he, albury, she, kapoor, mtsho, now,
Nearest to no: eucalyptus, albury, pipelines, connection, thaler, probably, nine, it,
Nearest to a: the, thaler, circ, ursus, iit, any, agouti, hanlon,
Nearest to was: is, became, had, has, were, thaler, by, being,
Nearest to and: or, but, ursus, cebus, while, dasyprocta, circ, abet,
Nearest to to: ursus, joram, could, coke, would, nine, abet, can,
Nearest to their: its, his, the, her, some, globemaster, agouti, thaler,
Nearest to is: was, has, are, became, agouti, thaler, be, does,
Nearest to i: you, ii, g, lama, thaler, he, gesta, we,
Nearest to years: year, annexing, sept, sunrise, seven, days, peacocks, wct,
Nearest to these: some, many, such, several, all, they, those, jude,
Nearest to often: widely, sometimes, usually, generally, also, commonly, wct, pulau,
Nearest to however: but, when, thaler, kapoor, ursus, although, that, cebus,
Nearest to new: backslash, microcebus, archie, ursus, kifl, circ, callithrix, agouti,
Nearest to has: had, have, is, was, agouti, thaler, circ, microcebus,
# Step 6: 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"
# 设置图片大小
plt.figure(figsize=(15, 15)) # 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:
from sklearn.manifold import TSNE
import matplotlib.pyplot as plt
tsne = TSNE(perplexity=30, n_components=2, init='pca', n_iter=5000, method='exact')# mac:method='exact'
# 画500个点
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, matplotlib, and scipy to visualize embeddings.")