朴素贝叶斯

#朴素：考虑每个特征或者词，出项的可能性与它和其他单词相邻没有关系
#每个特征等权重
from numpy import *

def loadDataSet():
    postingList=[['my', 'dog', 'has', 'flea', 'problems', 'help', 'please'],
                 ['maybe', 'not', 'take', 'him', 'to', 'dog', 'park', 'stupid'],
                 ['my', 'dalmation', 'is', 'so', 'cute', 'I', 'love', 'him'],
                 ['stop', 'posting', 'stupid', 'worthless', 'garbage'],
                 ['mr', 'licks', 'ate', 'my', 'steak', 'how', 'to', 'stop', 'him'],
                 ['quit', 'buying', 'worthless', 'dog', 'food', 'stupid']]
    classVec = [0,1,0,1,0,1]    #1代表侮辱性段落，0代表正常段落
    return postingList,classVec
#创建一个单词的集合
def createVocabList(dataSet):
    vocabSet = set([]) #创建空集合
    for document in dataSet:
        vocabSet |= set(document)
    return list(vocabSet)

#判断文档出现在词汇表中
def setOfWordsVec(vocabSet,inputSet):
    returnVec = [0]*len(vocabSet)
    for word in inputSet:
        if word in vocabSet:
            returnVec[vocabSet.index(word)] = 1
        else: print ("the word: %s is not in the Vocabulary!" % word)
    return returnVec
def trainNB0(trainMatrix,trainCategory):
    numTrainDocs = len(trainMatrix) #有几行话
    numWords = len(trainMatrix[0]) #每行的词汇表的词数
    # print(numTrainDocs)
    # print(numWords)
    pAbusive = sum(trainCategory)/float(numTrainDocs) #p(Ci)
    p0Num = ones(numWords)
    p1Num = ones(numWords)
    # print(p0Num)
    #p0Denom = 2.0; p1Denom = 2.0 #书上是2.0 不知道为什么 p(x1|c1)= (n1 + 1) / (n + N)  看网上的，
    #为了凑成概率和是1，N应该是numWords
    p0Denom = 1.0*numWords; p1Denom = 1.0*numWords
    for i in range(numTrainDocs):
        if trainCategory[i] == 1: 
            #某句话是侮辱性的话
            p1Num += trainMatrix[i] #矩阵相加
            #print(p1Num)
            p1Denom += sum(trainMatrix[i])
            #print(p1Denom)
        else:
            p0Num += trainMatrix[i]
            p0Denom += sum(trainMatrix[i])

    p1Vect = log(p1Num/p1Denom)
    p0Vect = log(p0Num/p0Denom)
    # print(p1Vect)
    # # summ = 0.0
    # # for V in p1Vect:
    # #     summ += V
    # # print(summ)
    # print(p0Vect)
    # print(pAbusive)
    return p0Vect,p1Vect,pAbusive
def classifyNB(vecOClassify,p0Vec,p1Vec,p1Class):
    p1 = sum(vecOClassify*p1Vec) + log(p1Class)
    p0 = sum(vecOClassify*p0Vec) + log(1 - p1Class)
    if p1 > p0:
        return 1
    else: return 0
def main():
    listOPosts,listClasses = loadDataSet()
    myVocabList = createVocabList(listOPosts) #创建一个元素互不重复的集合
    #print (myVocabList)
    trainMat = []
    #print(type(trainMat))
    for postinDoc in listOPosts:
        trainMat.append(setOfWordsVec(myVocabList, postinDoc))
    #print(trainMat)
    p0V,p1V,pC1 = trainNB0(trainMat, listClasses)
    testEntry = ['love','my','dalmation']
    thisDoc = array(setOfWordsVec(myVocabList, testEntry))
    print(testEntry,"classified as: ",classifyNB(thisDoc, p0V, p1V, pC1))
    testEntry = ['stupid','garbage']
    thisDoc = array(setOfWordsVec(myVocabList, testEntry))
    print(testEntry,"classified as: ",classifyNB(thisDoc, p0V, p1V, pC1))
    testEntry = ['stupid','cute','love','help']
    thisDoc = array(setOfWordsVec(myVocabList, testEntry))
    print(testEntry,"classified as: ",classifyNB(thisDoc, p0V, p1V, pC1))
    #print(setOfWordsVec(myVocabList, listOPosts[0]))
    #print(setOfWordsVec(myVocabList, listOPosts[1]))
main()

['love', 'my', 'dalmation'] classified as:  0
['stupid', 'garbage'] classified as:  1
['stupid', 'cute', 'love', 'help'] classified as:  0

经过这一段时间学习，发现ML in Action这本书确实是侧重代码方面，关于数学推导方面还是很少，需要不断查看文档。

链接： 贝叶斯网络、拉普拉斯平滑

链接： Naive Bayes

看完拉普拉斯平滑后，我觉得书上有问题，就是防止概率中有0的时候，分子每项加一，它写的分母 p0Denom = 2.0; p1Denom = 2.0，不明白为什么

看了文章后，

p(x1|c1)= (n1 + 1) / (n + N)
N是所有单词的数目，为了保证概率和为1。
所以我这里用的是NumWords

 

学习致用还是很好，哈哈哈

 

今天做个例子：使用朴素贝叶斯过滤垃圾邮件：

（还是ppt中写公式比较好用）

发现书上的代码和它给的代码库中的有些不一样，我觉得可能是翻译的时候出现了问题。

def textParse(bigString):
    import re
    listOTokens = re.split(r'W*', bigString) #    匹配任意数字和字母，相当于 [a-zA-Z0-9_]
    return [tok.lower() for tok in listOTokens if len(tok) > 2]

def spamTest(): #广告过滤
    docList = [];classList = [];fullText = []
    for i in range(1,26):    
        #wordList = textParse(open('email/spam/%d.txt' % i).read().encode('utf-8').decode('utf-8')) #spam中是垃圾邮件
        wordList = textParse(open('email/spam/%d.txt' % i).read())
        docList.append(wordList)
        fullText.extend(wordList)
        classList.append(1)
        #wordList = textParse(open('email/ham/%d.txt' % i).read().encode('gbk').decode('gbk')) 
        wordList = textParse(open('email/ham/%d.txt' % i).read()) 
        docList.append(wordList)
        fullText.extend(wordList)
        classList.append(0)
    vocabList = createVocabList(docList)
    #print(vocabList)
    # 40个训练集，10个测试集
    trainingSet = list(range(50));testSet = []
    #print(trainingSet)
    #print(type(random.uniform(0, len(trainingSet)))) float
    for i in range(10):
        randIndex = int(random.uniform(0, len(trainingSet)))
        testSet.append(trainingSet[randIndex]) #保证随机性，不管重复,其实不会重复，重复的只是元素索引
        del(trainingSet[randIndex]) 
    trainMax = []; trainClasses = []
    #print(trainingSet)
    for docIndex in trainingSet:
        #print(docIndex)
        trainMax.append(bagOfWordsVec(vocabList, docList[docIndex]))
        trainClasses.append(classList[docIndex])

    p0V,p1V,pSpam = trainNB0(array(trainMax),array(trainClasses))
    errorCount = 0
    for docIndex in testSet:
        wordVector = bagOfWordsVec(vocabList, docList[docIndex])
        if classifyNB(array(wordVector), p0V, p1V, pSpam) != classList[docIndex]:
            errorCount += 1
            print("the error text %s" % docList[docIndex])
    print("error rate: %f " % (float(errorCount)/len(testSet)))
spamTest()