[关闭]
@Wayne-Z 2017-11-19T10:18:09.000000Z 字数 8505 阅读 3018

word2vec的若干demo在win10平台上的实现

NLP word2dvec

网上找到了很多帖子，但是都千篇一律地相似，但是其中的一个关键点楼主又总觉得没有说清楚，即python3.5.2中，byte和str类型的不同使得直接从github上下载的代码在楼主的机子上并不能跑起来，经过几次迅（蛋）速（疼）地尝试和摸索，还是找到了解决方案的，所以在这里po一下（没错，我就是这么小白）。
好的，还是先从下载语料开始。

平台

语料下载

英文wiki语料链接 https://dumps.wikimedia.org/enwiki/latest/enwiki-latest-pages-articles.xml.bz2
中文wiki语料链接 https://dumps.wikimedia.org/zhwiki/latest/zhwiki-latest-pages-articles.xml.bz2
text8链接 http://mattmahoney.net/dc/text8.zip

代码修改

下载语料的同时，我们进入process_wiki.py，修改这几行代码用与处理wiki，原作是这样的：

  1. __author__ = 'huang'
  3. import os
  4. import logging
  5. import sys
  7. from gensim.corpora import WikiCorpus
  9. if __name__=='__main__':
  11.     program = os.path.basename(sys.argv[0])
  12.     logger = logging.getLogger(program)
  14.     logging.basicConfig(format='%(asctime)s: %(levelname)s: %(message)s')
  15.     logging.root.setLevel(level=logging.INFO)
  17.     if len(sys.argv) < 3:
  18.         print(globals()['__doc__'] %locals())
  19.         sys.exit(1)
  21.     inp, outp = sys.argv[1:3]
  22.     space = ' '
  23.     i = 0
  25.     output = open(outp, 'w')
  26.     wiki = WikiCorpus(inp, lemmatize=False, dictionary={})
  27.     for text in wiki.get_texts():
  28.         output.write（space.join(text) + '\n')
  29.         i = i + 1
  30.         if i % 10000 == 0:
  31.             logger.info('Saved ' + str(i) + ' articles')
  33.     output.close()
  34.     logger.info('Finished ' + str(i) + ' articles')

恩，就删了几行空行，语料下载完以后，放到同一个文件夹里，我们可以先直接尝试在cmd里运行如下代码

  1. python process_wiki.py enwiki-latest-pages-articles.xml.bz2 wiki.en.text

如果没问题，congratulations!但是楼主的平台并不能跑起来，这也是这篇文章存在的原因。问题出在这里

  1. (C:\Anaconda3) E:\NLP\word2vec-for-wiki-master>python process_wiki.py enwiki-latest-pages-articles.xml.bz2 wiki.en.text
  2. Traceback (most recent call last):
  3.   File "process_wiki.py", line 30, in <module>
  4.     output.write(space.join(text).decode() + '\n')
  5. TypeError: sequence item 0: expected str instance, bytes found

度娘了一下，是因为python3里面bytes和str是两种东西，这个在原文里面也说过，但是原文的解决方案，还是让楼主一头雾水（没错，就是这么小白）。根据之前的文章,再查阅了一下文档，原来是因为join()函数返回的类型也是bytes或bytearray，而write()函数只接受str类型。楼主原先尝（zuo）试（si）过如下几种方案，但是都失败了

  1. Traceback (most recent call last):
  2.   File "process_wiki.py", line 30, in <module>
  3.     output.write(bytes.join(space,text).decode() + '\n')
  4. TypeError: descriptor 'join' requires a 'bytes' object but received a 'str'
  6. (C:\Anaconda3) E:\NLP\word2vec-for-wiki-master>python process_wiki.py enwiki-latest-pages-articles.xml.bz2 wiki.en.text
  7. Traceback (most recent call last):
  8.   File "process_wiki.py", line 30, in <module>
  9.     output.write(bytes.join(space.encode(),text).decode() + '\n')
  10. UnicodeEncodeError: 'gbk' codec can't encode character '\u1f00' in position 1714: illegal multibyte sequence
  12. (C:\Anaconda3) E:\NLP\word2vec-for-wiki-master>python process_wiki.py enwiki-latest-pages-articles.xml.bz2 wiki.en.text
  13. Traceback (most recent call last):
  14.   File "process_wiki.py", line 30, in <module>
  15.     output.write(bytes.join(''.encode(),text).decode() + '\n')
  16. UnicodeEncodeError: 'gbk' codec can't encode character '\u1f00' in position 1474: illegal multibyte sequence
  18. (C:\Anaconda3) E:\NLP\word2vec-for-wiki-master>python process_wiki.py enwiki-latest-pages-articles.xml.bz2 wiki.en.text
  19. Traceback (most recent call last):
  20.   File "process_wiki.py", line 30, in <module>
  21.     output.write(bytes.join(b'',text).decode() + '\n')
  22. UnicodeEncodeError: 'gbk' codec can't encode character '\u1f00' in position 1474: illegal multibyte sequence

最终楼主将下面这行代码进行了改写

  1.         space = ' '.encode()

先用下面的代码进行了测试

  1.         data = sapce.join(text)
  2.         print(data)
  3.         #output.write(str(data) + '\n')

在cmd中看到了输出，但是print可以直接输出bytes而wirte只接受str，所以运行代码最终改成了这样

  1.         data = sapce.join(text)
  2.         output.write(str(data) + '\n')

然后结果如下

  1. (C:\Anaconda3) E:\NLP\word2vec-for-wiki-master>python process_wiki.py enwiki-latest-pages-articles.xml.bz2 wiki.en.text
  2. 2016-07-28 10:48:11,057: INFO: Saved 10000 articles
  3. 2016-07-28 10:49:44,660: INFO: Saved 20000 articles
  4. 2016-07-28 10:51:04,023: INFO: Saved 30000 articles
  5. 2016-07-28 10:52:13,199: INFO: Saved 40000 articles
  6. 2016-07-28 10:53:07,548: INFO: Saved 50000 articles
  7. 2016-07-28 10:53:45,695: INFO: Saved 60000 articles
  8. 2016-07-28 10:54:18,993: INFO: Saved 70000 articles
  9. 2016-07-28 10:54:51,188: INFO: Saved 80000 articles
  10. 2016-07-28 10:55:50,520: INFO: Saved 90000 articles
  11. ·
  12. ·
  13. ·
  14. ·
  15. 2016-07-28 15:24:22,182: INFO: Saved 4040000 articles
  16. 2016-07-28 15:25:09,770: INFO: Saved 4050000 articles
  17. 2016-07-28 15:25:46,915: INFO: Saved 4060000 articles
  18. 2016-07-28 15:26:24,892: INFO: Saved 4070000 articles
  19. 2016-07-28 15:27:05,343: INFO: Saved 4080000 articles
  20. 2016-07-28 15:27:48,280: INFO: Saved 4090000 articles
  21. 2016-07-28 15:28:22,146: INFO: finished iterating over Wikipedia corpus of 4099408 documents with 2229304913 positions (total 16753779 articles, 2290359456 positions before pruning articles shorter than 50 words)
  22. 2016-07-28 15:28:22,155: INFO: Finished 4099408 articles

五个小时四十分钟，文章数量已经有409万多了，还真不少···
而训练生成的text文件已经长到无法用记事本打开了,而后输入

  1. python train_word2vec_model.py wiki.en.text wiki.en.text.model wiki.en.text.vector

开始训练英文wiki词向量

  1. 2016-07-28 15:47:35,297: INFO: running train_word2vec_model.py wiki.en.text wiki.en.text.model wiki.en.text.vector
  2. 2016-07-28 15:47:35,302: INFO: collecting all words and their counts
  3. 2016-07-28 15:47:35,370: INFO: PROGRESS: at sentence #0, processed 0 words, keeping 0 word types
  4. 2016-07-28 15:48:05,500: INFO: PROGRESS: at sentence #10000, processed 29336126 words, keeping 434884 word types
  5. 2016-07-28 15:48:39,042: INFO: PROGRESS: at sentence #20000, processed 55594275 words, keeping 628122 word types

训练时打开另一个cmd，进入文件夹，处理中文wiki语料

  1. python process_wiki.py zhwiki-latest-pages-articles.xml.bz2 wiki.zh.text
  1. E:\NLP\word2vec-for-wiki-master>python process_wiki.py zhwiki-latest-pages-articles.xml.bz2 wiki.zh.text
  2. 2016-07-28 16:28:21,686: INFO: Saved 10000 articles
  3. 2016-07-28 16:29:07,536: INFO: Saved 20000 articles

然后就是漫长的等待了···
text8是现成的文本，所以直接进行训练，

  1. python train_word2vec_model.py text8 text8.model text8.vector.

结果如下

  1. 2016-07-28 20:03:42,295: INFO: PROGRESS: at 99.82% examples, 405001 words/s, in_qsize 12, out_qsize 3
  2. 2016-07-28 20:03:42,435: INFO: worker thread finished; awaiting finish of 7 more threads
  3. 2016-07-28 20:03:42,445: INFO: worker thread finished; awaiting finish of 6 more threads
  4. 2016-07-28 20:03:42,445: INFO: worker thread finished; awaiting finish of 5 more threads
  5. 2016-07-28 20:03:42,445: INFO: worker thread finished; awaiting finish of 4 more threads
  6. 2016-07-28 20:03:42,465: INFO: worker thread finished; awaiting finish of 3 more threads
  7. 2016-07-28 20:03:42,495: INFO: worker thread finished; awaiting finish of 2 more threads
  8. 2016-07-28 20:03:42,495: INFO: worker thread finished; awaiting finish of 1 more threads
  9. 2016-07-28 20:03:42,505: INFO: worker thread finished; awaiting finish of 0 more threads
  10. 2016-07-28 20:03:42,505: INFO: training on 85026035 raw words (62532401 effective words) took 154.3s, 405163 effective words/s
  11. 2016-07-28 20:03:42,505: INFO: saving Word2Vec object under text8.model, separately None
  12. 2016-07-28 20:03:42,505: INFO: storing numpy array 'syn0' to text8.model.syn0.npy
  13. 2016-07-28 20:03:43,506: INFO: not storing attribute syn0norm
  14. 2016-07-28 20:03:43,506: INFO: not storing attribute cum_table
  15. 2016-07-28 20:03:43,506: INFO: storing numpy array 'syn1neg' to text8.model.syn1neg.npy
  16. 2016-07-28 20:03:45,225: INFO: storing 71290x400 projection weights into text8.vector.

此时可以进行如下测试：

  1. In [1]: import gensim
  3. In [2]: model = gensim.models.Word2Vec.l
  4. gensim.models.Word2Vec.load                 gensim.models.Word2Vec.log_accuracy
  5. gensim.models.Word2Vec.load_word2vec_format
  7. In [2]: model = gensim.models.Word2Vec.looad('text8.model')
  9. In [3]: model = gensim.models.Word2Vec.load('text8.model')
  11. In [4]: model.mo
  12. model.most_similar        model.most_similar_cosmul
  14. In [4]: model.most_similar('man')
  15. Out[4]:
  16. [('woman', 0.6650575399398804),
  17.  ('girl', 0.5865204334259033),
  18.  ('creature', 0.5350353717803955),
  19.  ('boy', 0.510942816734314),
  20.  ('person', 0.5094308257102966),
  21.  ('men', 0.5073959827423096),
  22.  ('evil', 0.48292240500450134),
  23.  ('totoro', 0.47985178232192993),
  24.  ('god', 0.476554274559021),
  25.  ('vanity', 0.47478240728378296)]
  27. In [5]: model.most_similar('girl')
  28. Out[5]:
  29. [('blonde', 0.7728073596954346),
  30.  ('baby', 0.7689986824989319),
  31.  ('kid', 0.7603048086166382),
  32.  ('woman', 0.7313079833984375),
  33.  ('girls', 0.7117128968238831),
  34.  ('boy', 0.6976305246353149),
  35.  ('joey', 0.6945637464523315),
  36.  ('boys', 0.6894382238388062),
  37.  ('bride', 0.685029149055481),
  38.  ('rabbit', 0.6838369369506836)]

英文wiki的处理时间太长，楼主就没有坚持训练下去了，接下来中文wiki语料在与处理之后还要进行繁简转换，参考的文章
以及网上普遍看好opencc,在github下载并阅读完下面的安装指导后，发现这对于但是在楼主搭建的win10平台上并不是很适用,最后还是参考了Licstar的文章,直接进入下载opencc0.4.2-win32,根据安装说明，解压以后，其实还要添加到path里面去才能直接在cmd中使用，可是输入

  1. opencc -help

看安装是否成功，输出

  1. Open Chinese Convert (OpenCC) Command Line Tool
  2. Version 0.4.2
  4. Author: BYVoid <byvoid@byvoid.com>
  5. Bug Report: http://github.com/BYVoid/OpenCC/issues
  7. Usage:
  8.  opencc [Options]
  10. Options:
  11.  -i [file], --input=[file]   Read original text from [file].
  12.  -o [file], --output=[file]  Write converted text to [file].
  13.  -c [file], --config=[file]  Load configuration of conversion from [file].
  14.  -v, --version               Print version and build information.
  15.  -h, --help                  Print this help.
  17. With no input file, reads standard input and writes converted stream to standard output.
  18. Default configuration(zhs2zht.ini) will be loaded if not set.

说明成功了，就直接输入

  1. opencc -i wiki.zh.text -o wiki.zh.text.jian -c zht2zhs.ini

大约3分钟后，任务完成，如果没有安装jieba，先安装一下

  1. pip install jieba

安装完成以后，就可以直接运行

  1. python separate_words.py wiki.zh.text.jian wiki.zh.text.jian.seq

然后等待完成，输入

  1. python train_word2vec_model.py wiki.zh.text.jian

训练完成后测试即可。

添加新批注
在作者公开此批注前，只有你和作者可见。
回复批注