机器学习实战第四章——朴素贝叶斯分类(源码解析)

朴素贝叶斯分类

#coding=utf-8
'''
Created on 2016年1月9日

@author: admin
'''


from numpy import *

# 加载数据集函数
def loadDataSet():
    # 定义邮件列表
    postingList = [['my','dog','has','flea',\
                    'problem','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']]
    # 定义标签向量，1——abusive，0——not abusive
    classVec = [0,1,0,1,0,1]
    return postingList,classVec

# 创建词汇列表
def createVocabList(dataSet):
    # 定义词汇集
    vocabSet = set([])
    # 遍历文档
    for document in dataSet:
        # 将每个document合并到vocabSet，|用来联合两个集合
        vocabSet = vocabSet | set(document)
    # 返回词汇集
    return list(vocabSet)

# 把单词转换成向量
def setOfWords2Vec(vocabList,inputSet):
    # 定义要返回的向量
    returnVec = [0] * len(vocabList)
    # 遍历输出集中的单词
    for word in inputSet:
        # 单词在词汇集中
        if word in vocabList:
            # 对应的位置设为1
            returnVec[vocabList.index(word)] = 1
        else:
            print "the word: %s is not in my Vocabulary!" % word
    # 返回向量
    return returnVec

# 把单词转换成向量,用词袋模型，计算词出现的次数
def bagOfWords2VecMN(vocabList,inputSet):
    # 定义要返回的向量
    returnVec = [0] * len(vocabList)
    # 遍历输出集中的单词
    for word in inputSet:
        # 单词在词汇集中
        if word in vocabList:
            # 对应的词出现次数 加1
            returnVec[vocabList.index(word)] += 1
    # 返回向量
    return returnVec

# 条件概率的计算
def trainNB0(trainMatrix,trainCategory):
    # 计算文档的数目
    numTrainDocs = len(trainMatrix)
    # 计算单词的数目
    numWords = len(trainMatrix[0])
    # 计算类别的概率，abusive为1，not abusive为0
    pAbusive =sum(trainCategory) / float(numTrainDocs)
    # 初始化计数器，1行numWords列，p0是not abusive
    # p0Num =zeros(numWords)
    p0Num = ones(numWords)
    # 初始化计数器，p1是abusive
    p1Num = ones(numWords)
    # 初始化分母
    p0Denom = 2.0
    p1Denom = 2.0
    # 遍历文档
    for i in range(numTrainDocs):
        # 计算abusive对应的词汇的数目，trainMatrix为0-1值形成的向量
        if trainCategory[i] == 1:
            # p1Num存储的是每个词出现的次数
            p1Num += trainMatrix[i]
            # p1Denom存储的是词的总数目
            p1Denom += sum(trainMatrix[i])
        # 计算not abusive词汇的数目
        else:
            # 每个词在not abusive下出现的次数
            p0Num += trainMatrix[i]
            # not abusive下的总词数
            p0Denom += sum(trainMatrix[i])
    # 计算abusive下每个词出现的概率
    # p1Vect = p1Num / p1Denom
    p1Vect = log(p1Num / p1Denom)
    # 计算not abusive下每个词出现的概率
    # p0Vect = p0Num / p0Denom
    p0Vect = log(p0Num / p0Denom)
    # 返回词出现的概率和文档为abusive的概率，not abusive的概率为1-pAbusive
    return p0Vect,p1Vect,pAbusive

def classifyNB(vec2Classify,p0Vec,p1Vec,pClass1):
    # 计算abusive的概率
    p1 = sum(vec2Classify * p1Vec) + log(pClass1)
    # 计算not abusive的概率
    p0 = sum(vec2Classify * p0Vec) + log(1.0 - pClass1)
    # 根据概率大小判断属于哪个类
    if p1 > p0:
        return 1
    else:
        return 0

# 测试
def testingNB():
    # 加载数据集
    listOPosts,listClass = loadDataSet()
    # 创建词汇列表
    myVocabList = createVocabList(listOPosts)
    trainMat = []
    for postinDoc in listOPosts:
        trainMat.append(setOfWords2Vec(myVocabList,postinDoc))
    p0V,p1V,pAb = trainNB0(array(trainMat),array(listClass))
    # print p0V,p1V,pAb
    # print trainMat
    testEntry = ['love','my','dalmation']
    thisDoc = array(setOfWords2Vec(myVocabList,testEntry))
    print testEntry,'classified as:',classifyNB(thisDoc, p0V, p1V, pAb)
    testEntry = ['stupid','garbage']
    thisDoc = array(setOfWords2Vec(myVocabList,testEntry))
    print testEntry,'classified as:',classifyNB(thisDoc, p0V, p1V, pAb)

# 文本解析   
# 输入是字符串，输出是单词列表
def textParse(bigString):    
    # 导入正则表达式的包
    import re
    # 用正则表达式分割字符串
    listOfTokens = re.split(r'\W*', bigString)
    # 返回小写单词列表
    return [tok.lower() for tok in listOfTokens if len(tok) > 2] 

# 垃圾邮件测试   
def spamTest():
    # 定义docList文档列表，classList类别列表，fullText所有文档词汇
    docList=[]; classList = []; fullText =[]
    # 遍历email/spam和email/ham下的txt文件
    for i in range(1,26):
        # 定义并读取垃圾邮件文件的词汇分割列表
        wordList = textParse(open('email/spam/%d.txt' % i).read())
        # 将词汇列表加到文档列表中
        docList.append(wordList)
        # 将所有词汇列表汇总到fullText中
        fullText.extend(wordList)
        # 文档类别为1，spam
        classList.append(1)
        # 读取非垃圾邮件的文档
        wordList = textParse(open('email/ham/%d.txt' % i).read())
        # 添加到文档列表中
        docList.append(wordList)
        # 添加到所有词汇列表中
        fullText.extend(wordList)
        # 类别为0，非垃圾邮件
        classList.append(0)
    # 创建词汇列表
    vocabList = createVocabList(docList)
    # 定义训练集的索引和测试集
    trainingSet = range(50); testSet=[]
    # 随机的选择10个作为测试集      
    for i in range(10):
        # 随机索引
        randIndex = int(random.uniform(0,len(trainingSet)))
        # 将随机选择的文档加入到测试集中
        testSet.append(trainingSet[randIndex])
        # 从训练集中删除随机选择的文档
        del(trainingSet[randIndex])
    # 定义训练集的矩阵和类别  
    trainMat=[]; trainClasses = []
    # 遍历训练集，求得先验概率和条件概率
    for docIndex in trainingSet:
        # 将词汇列表变为向量放到trainMat中
        trainMat.append(bagOfWords2VecMN(vocabList, docList[docIndex]))
        # 训练集的类别标签
        trainClasses.append(classList[docIndex])
    # 计算先验概率，条件概率
    p0V,p1V,pSpam = trainNB0(array(trainMat),array(trainClasses))
    # 定义错误计数
    errorCount = 0
    # 对测试集进行分类
    for docIndex in testSet: 
        # 将测试集词汇向量化      
        wordVector = bagOfWords2VecMN(vocabList, docList[docIndex])
        # 对测试数据进行分类
        if classifyNB(array(wordVector),p0V,p1V,pSpam) != classList[docIndex]:
            # 分类不正确，错误计数加1
            errorCount += 1
            print "classification error",docList[docIndex]
    # 输出错误率
    print 'the error rate is: ',float(errorCount)/len(testSet)