Python探索之pLSA实现代码

# -*- coding: utf-8 -*-
import math
import random
import jieba
import codecs
import datetime

class pLSA_lph():
  def __init__(self, ntopic = 5):
    self.n_doc = 0
    self.n_word = 0
    self.n_topic = ntopic
    self.corpus = None
    self.p_z_dw = None
    self.p_w_z = None
    self.p_z_d = None
    self.likelihood = 0
    self.vocab = None
    self.stop_words = [u'，', u'。', u'、', u'（', u'）', u'·', u'！', u' ', u'：', u'“', u'”', u'\n']
  # 每行和为1的正实数，概率分布;
  def _rand_mat(self, sizex, sizey):
    ret = []
    for i in xrange(sizex):
      ret.append([])
      for _ in xrange(sizey):
        ret[-1].append(random.random())
      norm = sum(ret[-1])
      for j in xrange(sizey):
        ret[-1][j] /= norm
    return ret
  #从文本中计算词频稀疏矩阵，这里存储模型仿照LDA
  def loadCorpus(self, fn):
    # 中文分词
    f = open(fn, 'r')
    text = f.readlines()
    text = r' '.join(text)
    seg_generator = jieba.cut(text)
    seg_list = [i for i in seg_generator if i not in self.stop_words]
    seg_list = r' '.join(seg_list)
    # 切割统计所有出现的词纳入词典
    seglist = seg_list.split(" ")
    self.vocab = []
    for word in seglist:
      if (word != u' ' and word not in self.vocab):
        self.vocab.append(word)
    self.n_word =len(self.vocab)
    CountMatrix = []
    f.seek(0, 0)
    # 统计每个文档中出现的词频
    for line in f:
      # 置零
      count = [0 for i in range(len(self.vocab))]
      text = line.strip()
      # 但还是要先分词
      seg_generator = jieba.cut(text)
      seg_list = [i for i in seg_generator if i not in self.stop_words]
      seg_list = r' '.join(seg_list)
      seglist = seg_list.split(" ")
      # 查询词典中的词出现的词频
      for word in seglist:
        if word in self.vocab:
          count[self.vocab.index(word)] += 1
      CountMatrix.append(count)
    f.close()
    self.corpus = CountMatrix
    self.n_doc = len(CountMatrix)
    #初始化
    self.p_z_d = self._rand_mat(self.n_topic, self.n_doc)
    self.p_w_z = self._rand_mat(self.n_word, self.n_topic)
    self.p_z_dw =[]
    for k in range(self.n_topic):
      self.p_z_dw.append(self._rand_mat(self.n_doc, self.n_word))

  def _e_step(self):
    for k in range(self.n_topic):
      for d in range(self.n_doc):
        for j in range(self.n_word):
          _d_wz_zd = 0
          for kk in range(self.n_topic):
            _d_wz_zd += self.p_w_z[j][kk]*self.p_z_d[kk][d]
          if _d_wz_zd <= 0:
            _d_wz_zd = 1e-6
          self.p_z_dw[k][d][j] = self.p_w_z[j][k]*self.p_z_d[k][d]/_d_wz_zd
  def _m_step(self):
    print "updating Pn(Wj|Zk)...\r"
    for j in range(self.n_word):
      for k in range(self.n_topic):
        _d_dw_zdw = 0
        for d in range(self.n_doc):
          _d_dw_zdw += self.corpus[d][j]*self.p_z_dw[k][d][j]
        _d_dw_zdw_sum = 0
        for jj in range(self.n_word):
          _d_dw_zdw_i = 0
          for d in range(self.n_doc):
            _d_dw_zdw_i += self.corpus[d][jj]*self.p_z_dw[k][d][jj]
          _d_dw_zdw_sum += _d_dw_zdw_i
        if _d_dw_zdw_sum <= 0:
          _d_dw_zdw_sum = 1e-6
        self.p_w_z[j][k] = _d_dw_zdw/_d_dw_zdw_sum
    print "updating Pn(Zk|Di)...\r"
    for k in range(self.n_topic):
      for d in range(self.n_doc):
        _d_dw_zdw = 0
        for j in range(self.n_word):
          _d_dw_zdw += self.corpus[d][j]*self.p_z_dw[k][d][j]
        _d_dw_zdw_sum = 0
        for kk in range(self.n_topic):
          _d_dw_zdw_i = 0
          for j in range(self.n_word):
            _d_dw_zdw_i += self.corpus[d][j]*self.p_z_dw[kk][d][j]
          _d_dw_zdw_sum += _d_dw_zdw_i
        if _d_dw_zdw_sum <= 0:
          _d_dw_zdw_sum = 1e-6
        self.p_z_d[k][d] = _d_dw_zdw/_d_dw_zdw_sum
  #计算最大似然值
  def _cal_max_likelihood(self):
    self.likelihood = 0
    for d in range(self.n_doc):
      for j in range(self.n_word):
        _dP_wjdi = 0
        for k in range(self.n_topic):
          _dP_wjdi += self.p_w_z[j][k]*self.p_z_d[k][d]
        _dP_wjdi = 1.0/self.n_doc*_dP_wjdi
        self.likelihood += self.corpus[d][j]*math.log(_dP_wjdi)
  #迭代训练
  def train(self, n_iter = 100, d_delta = 1e-6,log_fn = "log.log"):
    itr = 0
    delta =10e9
    _likelihood = 0
    f = open(log_fn, 'w')
    while itr < n_iter and delta > d_delta:
      _likelihood = self.likelihood
      self._e_step()
      self._m_step()
      self._cal_max_likelihood()
      itr += 1
      delta = abs(self.likelihood - _likelihood)
      t1 = datetime.datetime.now().strftime('%Y-%m-%d-%y %H:%M:%S');
      f.write("%s iteration %d, max-likelihood = %.6f\n"%(t1, itr, self.likelihood))
      print "%s iteration %d, max-likelihood = %.6f"%(t1, itr, self.likelihood)
    f.close()

  def printVocabulary(self):
    print "vocabulary:"
    for word in self.vocab:
      print word,
    print
  def saveVocabulary(self, fn):
    f = codecs.open(fn, 'w', 'utf-8')
    for word in self.vocab:
      f.write("%s\n"%word)
    f.close()
  def printWordOfTopic(self):
    for k in range(self.n_topic):
      print "Topic %d"%k,
      for j in range(self.n_word):
        print self.p_w_z[j][k],
      print
  def saveWordOfTopic(self,fn):
    f = open(fn, 'w')
    for j in range(self.n_word):
      f.write(", w%d"%j)
    f.write("\n")
    for k in range(self.n_topic):
      f.write("topic %d"%k)
      for j in range(self.n_word):
        f.write(", %.6f"%self.p_w_z[j][k])
      f.write("\n")
    f.close()
  def printTopicOfDoc(self):
    for d in range(self.n_doc):
      print "Doc %d"%d,
      for k in range(self.n_topic):
        print self.p_z_d[k][d],
      print
  def saveTopicOfDoc(self, fn):
    f = open(fn, 'w')
    for k in range(self.n_topic):
      f.write(", z%d" % k)
    f.write("\n")
    for d in range(self.n_doc):
      f.write("doc %d" % d)
      for k in range(self.n_topic):
        f.write(", %.6f" % self.p_z_d[k][d])
      f.write("\n")
    f.close()

<span style="font-family: Arial, Helvetica, sans-serif; background-color: rgb(255, 255, 255);">新华社北京11月26日电中共中央总书记、国家主席习近平26日向古巴共产党中央委员会第一书记、国务委员会主席兼部长会议主席劳尔·卡斯特罗致唁电，代表中国党、政府、人民并以个人名义，对菲德尔·卡斯特罗同志逝世表示最沉痛的哀悼，向其家属致以最诚挚的慰问。全文如下：惊悉古巴革命领导人菲德尔·卡斯特罗同志不幸逝世，我谨代表中国共产党、政府、人民，并以我个人的名义，向你并通过你向古巴共产党、政府、人民，对菲德尔·卡斯特罗同志的逝世表示最沉痛的哀悼，向其家属致以最诚挚的慰问。菲德尔·卡斯特罗同志是古巴共产党和古巴社会主义事业的缔造者，是古巴人民的伟大领袖。他把毕生精力献给了古巴人民争取民族解放、维护国家主权、建设社会主义的壮丽事业，为古巴人民建立了不朽的历史功勋，也为世界社会主义发展建立了不朽的历史功勋。菲德尔·卡斯特罗同志是我们这个时代的伟大人物，历史和人民将记住他。我多次同菲德尔·卡斯特罗同志见面，促膝畅谈，他的真知灼见令我深受启发，他的音容笑貌犹在眼前。我深深怀念他，中国人民深深怀念他。菲德尔·卡斯特罗同志生前致力于中古友好，密切关注和高度评价中国发展进程，在他亲自关心和支持下，古巴成为第一个同新中国建交的拉美国家。建交56年来，中古关系长足发展，各领域务实合作成果丰硕，两国人民友谊与日俱增，这都与菲德尔·卡斯特罗同志的关怀和心血密不可分。菲德尔·卡斯特罗同志的逝世是古巴和拉美人民的重大损失，不仅使古巴和拉美人民失去了一位优秀儿子，也使中国人民失去了一位亲密的同志和真诚的朋友。他的光辉形象和伟大业绩将永载史册。我相信，在主席同志坚强领导下，古巴党、政府、人民必将继承菲德尔·卡斯特罗同志的遗志，化悲痛为力量，在社会主义建设事业中不断取得新的成绩。中古两党、两国、两国人民友谊必将得到巩固和发展。伟大的菲德尔·卡斯特罗同志永垂不朽！（完）</span> 

if __name__=="__main__":

  _plsa = pLSA_lph(2)
  _plsa.loadCorpus(u"C:\\Users\Administrator\Desktop\\zhongwen.txt")
  _plsa.train()
  _plsa.printTopicOfDoc()
  _plsa.printWordOfTopic()
  _plsa.saveTopicOfDoc(u"C:\\Users\Administrator\Desktop\\topic_doc.txt")
  _plsa.saveWordOfTopic(u"C:\\Users\Administrator\Desktop\\word_topic.txt")