机器学习实战笔记(Python实现)-03-朴素贝叶斯

 1 from numpy import *
 2 
 3 def loadDataSet():
 4     ‘‘‘
 5     postingList: 进行词条切分后的文档集合
 6     classVec:类别标签    
 7     ‘‘‘
 8     postingList=[[‘my‘, ‘dog‘, ‘has‘, ‘flea‘, ‘problems‘, ‘help‘, ‘please‘],
 9                  [‘maybe‘, ‘not‘, ‘take‘, ‘him‘, ‘to‘, ‘dog‘, ‘park‘, ‘stupid‘],
10                  [‘my‘, ‘dalmation‘, ‘is‘, ‘so‘, ‘cute‘, ‘I‘, ‘love‘, ‘him‘],
11                  [‘stop‘, ‘posting‘, ‘stupid‘, ‘worthless‘, ‘garbage‘],
12                  [‘mr‘, ‘licks‘, ‘ate‘, ‘my‘, ‘steak‘, ‘how‘, ‘to‘, ‘stop‘, ‘him‘],
13                  [‘quit‘, ‘buying‘, ‘worthless‘, ‘dog‘, ‘food‘, ‘stupid‘]]
14     classVec = [0,1,0,1,0,1]    #1代表侮辱性文字，0代表正常言论
15     return postingList,classVec
16 
17 def createVocabList(dataSet):
18     vocabSet = set([])#使用set创建不重复词表库
19     for document in dataSet:
20         vocabSet = vocabSet | set(document) #创建两个集合的并集
21     return list(vocabSet)
22 
23 def setOfWords2Vec(vocabList, inputSet):
24     returnVec = [0]*len(vocabList)#创建一个所包含元素都为0的向量
25     #遍历文档中的所有单词，如果出现了词汇表中的单词，则将输出的文档向量中的对应值设为1
26     for word in inputSet:
27         if word in vocabList:
28             returnVec[vocabList.index(word)] = 1
29         else: print("the word: %s is not in my Vocabulary!" % word)
30     return returnVec
31 ‘‘‘
32 我们将每个词的出现与否作为一个特征，这可以被描述为词集模型(set-of-words model)。
33 如果一个词在文档中出现不止一次，这可能意味着包含该词是否出现在文档中所不能表达的某种信息,
34 这种方法被称为词袋模型(bag-of-words model)。
35 在词袋中，每个单词可以出现多次，而在词集中，每个词只能出现一次。
36 为适应词袋模型，需要对函数setOfWords2Vec稍加修改，修改后的函数称为bagOfWords2VecMN
37 ‘‘‘
38 def bagOfWords2VecMN(vocabList, inputSet):
39     returnVec = [0]*len(vocabList)
40     for word in inputSet:
41         if word in vocabList:
42             returnVec[vocabList.index(word)] += 1
43     return returnVec

 1 def trainNB0(trainMatrix,trainCategory):
 2     ‘‘‘
 3     朴素贝叶斯分类器训练函数(此处仅处理两类分类问题)
 4     trainMatrix:文档矩阵
 5     trainCategory:每篇文档类别标签
 6     ‘‘‘
 7     numTrainDocs = len(trainMatrix)
 8     numWords = len(trainMatrix[0])
 9     pAbusive = sum(trainCategory)/float(numTrainDocs)
10     #初始化所有词出现数为1，并将分母初始化为2，避免某一个概率值为0
11     p0Num = ones(numWords); p1Num = ones(numWords)#
12     p0Denom = 2.0; p1Denom = 2.0 #
13     for i in range(numTrainDocs):
14         if trainCategory[i] == 1:
15             p1Num += trainMatrix[i]
16             p1Denom += sum(trainMatrix[i])
17         else:
18             p0Num += trainMatrix[i]
19             p0Denom += sum(trainMatrix[i])
20     #将结果取自然对数，避免下溢出，即太多很小的数相乘造成的影响
21     p1Vect = log(p1Num/p1Denom)#change to log()
22     p0Vect = log(p0Num/p0Denom)#change to log()
23     return p0Vect,p1Vect,pAbusive

 1 def classifyNB(vec2Classify, p0Vec, p1Vec, pClass1):
 2     ‘‘‘
 3     分类函数
 4     vec2Classify:要分类的向量
 5     p0Vec, p1Vec, pClass1:分别对应trainNB0计算得到的3个概率
 6     ‘‘‘
 7     p1 = sum(vec2Classify * p1Vec) + log(pClass1)
 8     p0 = sum(vec2Classify * p0Vec) + log(1.0 - pClass1)
 9     if p1 > p0:
10         return 1
11     else: 
12         return 0

 1 def testingNB():
 2     listOPosts,listClasses = loadDataSet()
 3     myVocabList = createVocabList(listOPosts)
 4     trainMat=[]
 5     for postinDoc in listOPosts:
 6         trainMat.append(setOfWords2Vec(myVocabList, postinDoc))
 7     #训练模型，注意此处使用array
 8     p0V,p1V,pAb = trainNB0(array(trainMat),array(listClasses))
 9     testEntry = [‘love‘, ‘my‘, ‘dalmation‘]
10     thisDoc = array(setOfWords2Vec(myVocabList, testEntry))
11     print(testEntry,‘classified as: ‘,classifyNB(thisDoc,p0V,p1V,pAb))
12     testEntry = [‘stupid‘, ‘garbage‘]
13     thisDoc = array(setOfWords2Vec(myVocabList, testEntry))
14     print(testEntry,‘classified as: ‘,classifyNB(thisDoc,p0V,p1V,pAb))

1 def textParse(bigString):#
2     ‘‘‘
3     文本切分
4     输入文本字符串，输出词表
5     ‘‘‘
6     import re
7     listOfTokens = re.split(r‘\W*‘, bigString)
8     return [tok.lower() for tok in listOfTokens if len(tok) > 2] 
9     

 1 def spamTest():
 2     ‘‘‘
 3     垃圾邮件测试函数
 4     ‘‘‘
 5     docList=[]; classList = []; fullText =[]
 6     for i in range(1,26):
 7         #读取垃圾邮件
 8         wordList = textParse(open(‘email/spam/%d.txt‘ % i,‘r‘,encoding= ‘utf-8‘).read())
 9         docList.append(wordList)
10         fullText.extend(wordList)
11         #设置垃圾邮件类标签为1
12         classList.append(1)        
13         wordList = textParse(open(‘email/ham/%d.txt‘ % i,‘r‘,encoding= ‘utf-8‘).read())
14         docList.append(wordList)
15         fullText.extend(wordList)
16         classList.append(0)
17     vocabList = createVocabList(docList)#生成次表库
18     trainingSet = list(range(50))
19     testSet=[]           #
20     #随机选10组做测试集
21     for i in range(10):
22         randIndex = int(random.uniform(0,len(trainingSet)))
23         testSet.append(trainingSet[randIndex])
24         del(trainingSet[randIndex])  
25     trainMat=[]; trainClasses = []
26     for docIndex in trainingSet:#生成训练矩阵及标签
27         trainMat.append(bagOfWords2VecMN(vocabList, docList[docIndex]))
28         trainClasses.append(classList[docIndex])
29     p0V,p1V,pSpam = trainNB0(array(trainMat),array(trainClasses))
30     errorCount = 0
31     #测试并计算错误率
32     for docIndex in testSet:
33         wordVector = bagOfWords2VecMN(vocabList, docList[docIndex])
34         if classifyNB(array(wordVector),p0V,p1V,pSpam) != classList[docIndex]:
35             errorCount += 1
36             print("classification error",docList[docIndex])
37     print(‘the error rate is: ‘,float(errorCount)/len(testSet))
38     #return vocabList,fullText

 1 ‘‘‘
 2 函数localWords()与程序清单中的spamTest()函数几乎相同，区别在于这里访问的是
 3 RSS源而不是文件。然后调用函数calcMostFreq()来获得排序最高的30个单词并随后将它们移除
 4 ‘‘‘
 5 def localWords(feed1,feed0):
 6     import feedparser
 7     docList=[]; classList = []; fullText =[]
 8     minLen = min(len(feed1[‘entries‘]),len(feed0[‘entries‘]))
 9     for i in range(minLen):
10         wordList = textParse(feed1[‘entries‘][i][‘summary‘])
11         docList.append(wordList)
12         fullText.extend(wordList)
13         classList.append(1) #NY is class 1
14         wordList = textParse(feed0[‘entries‘][i][‘summary‘])
15         docList.append(wordList)
16         fullText.extend(wordList)
17         classList.append(0)
18     vocabList = createVocabList(docList)#create vocabulary
19     top30Words = calcMostFreq(vocabList,fullText)   #remove top 30 words
20     for pairW in top30Words:
21         if pairW[0] in vocabList: vocabList.remove(pairW[0])
22     trainingSet = list(range(2*minLen)); testSet=[]           #create test set
23     for i in range(10):
24         randIndex = int(random.uniform(0,len(trainingSet)))
25         testSet.append(trainingSet[randIndex])
26         del(trainingSet[randIndex])  
27     trainMat=[]; trainClasses = []
28     for docIndex in trainingSet:#train the classifier (get probs) trainNB0
29         trainMat.append(bagOfWords2VecMN(vocabList, docList[docIndex]))
30         trainClasses.append(classList[docIndex])
31     p0V,p1V,pSpam = trainNB0(array(trainMat),array(trainClasses))
32     errorCount = 0
33     for docIndex in testSet:        #classify the remaining items
34         wordVector = bagOfWords2VecMN(vocabList, docList[docIndex])
35         if classifyNB(array(wordVector),p0V,p1V,pSpam) != classList[docIndex]:
36             errorCount += 1
37     print(‘the error rate is: ‘,float(errorCount)/len(testSet))
38     return vocabList,p0V,p1V
39 
40 def calcMostFreq(vocabList,fullText):
41     ‘‘‘
42     返回前30个高频词
43     ‘‘‘
44     import operator
45     freqDict = {}
46     for token in vocabList:
47         freqDict[token]=fullText.count(token)
48     sortedFreq = sorted(freqDict.items(), key=operator.itemgetter(1), reverse=True) 
49     return sortedFreq[:30]
50 
51 if __name__== "__main__":  
52     #导入RSS数据源
53     import operator
54     ny=feedparser.parse(‘http://newyork.craigslist.org/stp/index.rss‘)
55     sf=feedparser.parse(‘http://sfbay.craigslist.org/stp/index.rss‘)
56     localWords(ny,sf)