机器学习贝叶斯-垃圾邮件识别

---

一、何为贝叶斯

贝叶斯基本公式:P(A|B)=P(B|A)*P(A)/P(B)
假设事件 B={一封邮件}，事件A={这封邮件是垃圾邮件}，P(A|B)表示这封邮件是垃圾邮件的概率。可想而知，我们无法直接求得这个概率，此时借助贝叶斯就可以对该事件进行一次逆推

• P(A|B)称为后验概率(posterior),这是我们需要结合先验概率和证据计算之后才能知道的。

• P(B|A)称为似然(likelihood),在事件A发生的情况下，事件B(或evidence)的概率有多大

• P(A)称为先验概率(prior), 事件A发生的概率有多大

• P(B)称为证据(evidence)，即无论事件如何，事件B(或evidence)的可能性有多大

二、项目实战-垃圾分类

1. 导包

import random
import numpy as np
import re

2. 代码示例

def textParse(input_text):
    '''
    读取文本数据，进行正则切分，将切分后的单词进行小写化
    :param input_text: 邮件文本
    :return: 切分后的单词列表
    '''
    list_world = re.split(r'\W+', input_text)
    return [world.lower() for world in list_world if len(world) > 2]


def createVocabList(doclist):
    '''
    创建语料库
    :param doclist: 传入单词列表
    :return: 返回无重复的语料库
    '''
    vocablist = set()
    for doc in doclist:
        vocablist = vocablist | set(doc)

    return list(vocablist)


def setOfWordVec(vocablist, inputSet):
    '''
    文本向量化
    :param vocablist: 语料库
    :param inputSet: 单篇邮件的单词列表
    :return: 统计单词出现次数的向量
    '''
    returnVec = [0] * len(vocablist)
    for word in inputSet:
        if word in vocablist:
            returnVec[vocablist.index(word)] += 1
    return returnVec


def trainNB(trainMat, trainClass):
    '''
    训练模型
    :param trainMat: 训练集单词矩阵
    :param trainClass: 训练集对应类别(1代表垃圾邮件，0代表正常邮件)
    :return: 垃圾邮件似然概率，正常文件似然概率，垃圾文件先验概率
    '''
    numTrainDocs = len(trainMat)
    numberWorld = len(trainMat[0])
    p1 = sum(trainClass) / numTrainDocs
    p0Num = np.ones((numberWorld))  # 拉普拉斯平滑
    p1Num = np.ones((numberWorld))
    p0Demo = 2  # 一般情况下分为几类就写几
    p1Demo = 2

    for i in range(numTrainDocs):
        if trainClass[i] == 1:
            p1Num += trainMat[i]
            p1Demo += sum(trainMat[i])
        else:
            p0Num += trainMat[i]
            p0Demo += sum(trainMat[i])

    p1Vec = np.log(p1Num / p1Demo)
    p0Vec = np.log(p0Num / p0Demo)
    return p1Vec, p0Vec, p1


def classifyNB(wordVec, p1Vec, p0Vec, p1):
    '''
    贝叶斯计算，分类
    :param wordVec: 测试集文本向量
    :param p1Vec: 垃圾邮件似然概率
    :param p0Vec: 正常邮件似然概率
    :param p1: 垃圾邮件先验概率
    :return: 分类结果
    '''
    p1 = np.log(p1) + sum(wordVec * p1Vec)
    p0 = np.log(1 - p1) + sum(wordVec * p0Vec)
    if p1 > p0:
        return 1
    else:
        return 0


def spam():
    '''
    主函数
    :return:
    '''
    doclist = []
    classlist = []
    for i in range(1, 26):
        wordlist = textParse(open("data/ham/{0}.txt".format(i), 'r').read())
        doclist.append(wordlist)
        classlist.append(1)  # 垃圾邮件标签

        wordlist = textParse(open("data/spam/{0}.txt".format(i), 'r').read())
        doclist.append(wordlist)
        classlist.append(0)  # 正常邮件标签
    vocablist = createVocabList(doclist)  # 创建语料表
    trainset = list(range(50))
    testset = []
    for i in range(10):
        randIndex = int(random.uniform(0, len(trainset)))
        testset.append(trainset[randIndex])
        del trainset[randIndex]

    trainMat = []
    trainClass = []
    for docIndex in trainset:
        trainMat.append(setOfWordVec(vocablist, doclist[docIndex]))
        trainClass.append(classlist[docIndex])

    p1Vec, p0Vec, p1 = trainNB(np.array(trainMat), np.array(trainClass))
    errorCount = 0
    for docIndex in testset:
        wordVec = setOfWordVec(vocablist, doclist[docIndex])
        p = classifyNB(np.array(wordVec), p1Vec, p0Vec, p1)
        if p != classlist[docIndex]:
            errorCount += 1
    return errorCount

3. 测试代码

if __name__ == "__main__":
    errorCount = 0
    for i in range(10):
        print(f"第{i+1}轮分类错误：{spam()}个...")

4.测试结果

注意：需要数据请私聊作者！！！