贝叶斯-垃圾邮件
开始一点点写贝叶斯过滤垃圾邮件
之前写的，没注意把文件名对应上，可能会有些不清楚导包
代码都会上传到github
首先写代码之前好好看一下西瓜书和机器学习实战里面关于贝叶斯理论的介绍$p(c|x) = \frac{p(x|c)p(c)}{p(x)}$
在这里有：后验概率=先验概率*调查因子
分别解释一下，先验概率是指x发生前，对c的判断
后验概率是指x发生后，对c的重新评估
那么剩下的$\frac{p(x|c)}{p(x)}$就是调查因子，这里指的是，使预估概率更接近真实概率
有了以上，结合看机器学习实战的4.2节应该会好点
首先进行文本分类，这里我创建一个wordVec.py文件，代码见下
# 这个程序的目的是创建一个n*m列表，然后得到set集合，再用给定的序列对其判定，输出一个序列
# 本来这个序列是未排序的，所以多次输出结果会不同，但是如果排序之后再输出
# 那么当我们知道某个位置是什么的情况下，就可对这个序列进行判定
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']]
# sum = 0
# for i in postingList:
# sum += len(i)
# print(sum)
classVec = [0, 1, 0, 1, 0, 1]
return postingList, classVec
# 用一个set把数据集的内容全部打包
def createVocabList(dataSet):
vocabSet = set([])
for document in dataSet:
vocabSet = vocabSet | set(document)
return sorted(list(vocabSet))
# 对某一指定序列进行向量化输出
def setOfWords2Vec(vocabList, inputSet):
returnVec = [0] * len(vocabList)
for word in inputSet:
if word in vocabList:
returnVec[vocabList.index(word)] = 1
else:
print("the word:%s is not in my vocabulary!" % word)
return returnVec
if __name__ == "__main__":
listOposts, listClasses = loadDataSet()
result = createVocabList(listOposts)
print(result)
print(len(result))
print(setOfWords2Vec(result,listOposts[0]))
这里比较容易理解，创建了一个6行的数据集，然后createVocabList函数将其变成了一个set集合(一开始也不太熟set，这里我多练习了几次),然后用setOfwords2Vec函数，将某一行数据转换成只有01的向量，这里我们要对输入和输出注意一下
这时候来到了4.5.2 我们要从词向量计算概率
从整体来看，也就是对每一行的词出现的次数进行统计，然后再汇总，再除以类别数，具体来看就是侮辱性的是3行，这3行对应总共词汇是19正常的也是3行(这3行对应24，总共43个词，但是set会去掉一些重复的，所以set之后是32个词)，然后stupid这个词出现了3次，3/19≈0.15789474，也就是书上说的分类为侮辱的依据
那么我们可以在写代码的过程中多次调用print函数来查看
这里我创建了classifier_bayes.py文件，代码如下
from math import log
import numpy as np
import wordVec
# 传入的trainMatrix是6*32的矩阵，其中6行中每一列32个元素，出现过的位置为1，未出现则为0
# trainCategory是[0,1,0,1,0,1]
def trainNBO(trainMatrix, trainCtegory):
numTrainDocs = len(trainMatrix) # 6行
# print(numTrainDocs)
numWords = len(trainMatrix[0]) # 每行32个元素
# print(numWords)
pAbusive = sum(trainCtegory) / numTrainDocs # 3
# print(sum(trainCtegory))
# print(pAbusive)
p0Num = np.zeros(numWords) # 32个0
# p0Num = np.ones(numWords)
# print(pONum)
p1Num = np.zeros(numWords)
# p1Num = np.ones(numWords)
p0Denom = 0
# p0Denom = 2
p1Denom = 0
# p1Denom = 2
for i in range(numTrainDocs):
if trainCtegory[i] == 1:
p1Num += trainMatrix[i]
print("p0Num:", p1Num)
p1Denom += sum(trainMatrix[i])
print("p1Denom:", p1Denom)
else:
p0Num += trainMatrix[i]
print("p0Num:",p0Num)
p0Denom += sum(trainMatrix[i])
print("p0Denom:",p0Denom)
p1Vect = p1Num / p1Denom
p0Vect = p0Num / p0Denom
return p0Vect, p1Vect, pAbusive
if __name__ == "__main__":
listPosts, listClasses = wordVec.loadDataSet()
# print(listPosts)
'''
[['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']]
'''
myVocabList = wordVec.createVocabList(listPosts)
# print(myVocabList)
'''
['not', 'to', 'ate', 'I', 'so', 'cute', 'help',
'is', 'dog', 'worthless', 'posting', 'quit', 'him', 'love', 'food',
'garbage', 'please', 'my', 'mr', 'take', 'maybe',
'has', 'stupid', 'steak', 'stop', 'buying', 'licks', 'problems',
'park', 'how', 'flea', 'dalmation']
'''
# print(len(myVocabList)) # 32
trainMat = []
for postinDoc in listPosts:
# print(postinDoc)
'''
['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']
'''
trainMat.append(wordVec.setOfWords2Vec(myVocabList,postinDoc))
# print(trainMat)
'''
[[0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 1, 0, 1],
[0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 1],
[0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0],
[0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0],
[1, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0],
[0, 1, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1]]
'''
p0V,p1V,pAb = trainNBO(trainMat,listClasses)
print("p0V",p0V)
print("p1V",p1V)
print("pAb",pAb)
现在我们已经对侮辱(1)和非侮辱(0)的概率p0V和p1V分别进行了计算
如果我们像进行类别判断，不必把$\frac{p(x|c)p(c)}{p(x)}$所有求出来，我们只需要求$p(x|c)p(c)$，并且我们是假设独立的，那么$p(x|c)$就变成了$p(x_1|c)p(x_2|c)...*p(x_n|c)$
，并且考虑到如果其中有某一项为0，那么整个概率为0，采用取对数法进行计算
进行修改，并创建bayes.py文件，代码见下
import numpy as np
import wordVec
# 需要的时候释放一些注释
def trainNBO(trainMatrix, trainCtegory):
numTrainDocs = len(trainMatrix)
numWords = len(trainMatrix[0])
pAbusive = sum(trainCtegory) / numTrainDocs
p0Num = np.ones(numWords)
p1Num = np.ones(numWords)
p0Denom = 2
p1Denom = 2
for i in range(numTrainDocs):
if trainCtegory[i] == 1:
p1Num += trainMatrix[i]
p1Denom += sum(trainMatrix[i])
else:
p0Num += trainMatrix[i]
p0Denom += sum(trainMatrix[i])
p1Vect = np.log(p1Num / p1Denom)
p0Vect = np.log(p0Num / p0Denom)
# print(p0Vect)
# print(p1Vect)
return p0Vect, p1Vect, pAbusive
def classifyNB(vec2Classify, p0Vec, p1Vec, pClass1):
p1 = sum(vec2Classify * p1Vec) + np.log(pClass1)
# print(vec2Classify * p1Vec)
# print("p1:",p1)
p0 = sum(vec2Classify * p0Vec) + np.log(1 - pClass1)
# print(vec2Classify * p0Vec)
# print("p0:",p0)
if p1 > p0:
return 1
else:
return 0
def testingNB():
listPosts, listClasses = wordVec.loadDataSet()
myVocabList = wordVec.createVocabList(listPosts)
trainMat = []
for postinDoc in listPosts:
trainMat.append(wordVec.setOfWords2Vec(myVocabList, postinDoc))
p0V, p1V, pAb = trainNBO(trainMat, listClasses)
print(p0V)
print(p1V)
print(pAb)
testEntry = ['love', 'my', 'dalmation']
thisDoc = np.array(wordVec.setOfWords2Vec(myVocabList, testEntry))
print("%s classified as %d" % (testEntry, classifyNB(thisDoc, p0V, p1V, pAb)))
testEntry = ['stupid', 'garbage']
thisDoc = np.array(wordVec.setOfWords2Vec(myVocabList, testEntry))
print("%s classified as %d" % (testEntry, classifyNB(thisDoc, p0V, p1V, pAb)))
if __name__ == "__main__":
testingNB()
最后，用上述写好的代码，来进行一个垃圾邮件的测试
代码见下
# 现在要对邮件进行检测
import random
import bayes
import wordVec
import numpy as np
# 使用正则表达式来获取其中内容
def textParse(listString):
import re
listTokens = re.split(r'\W+',listString)
return [tok.lower() for tok in listTokens if len(tok) > 2]
def spamTest():
# 里面装每个文件的内容，也即最后列表中有50个列表
docList = []
# 标记
classList = []
# 最后装到一个列表里面
fullText = []
for i in range(1,26):
# 首先打开一个spam文件
wordList = textParse(open('email/spam/%d.txt'%i).read())
docList.append(wordList)
# print(docList)
# print(i)
fullText.extend(wordList)
classList.append(1)
# 然后打开一个ham文件
# 加上encoding='windows-1252'不会报错
wordList = textParse(open('email/ham/%d.txt'%i,encoding='windows-1252').read())
docList.append(wordList)
# print(docList)
fullText.extend(wordList)
classList.append(0)
# print(classList)
# print(fullText)
# print(docList)
# print(len(docList))
# 现在docList有50个列表，classList有25个0和25个1交替，fullText是汇总的
vocabList = wordVec.createVocabList(docList) # 去掉重复的
# print(vocabList)
trainingSet = list(range(50))
print("trainingSet:",trainingSet)
testSet = []
# 从50行中随机获得10条样本作为测试集
for i in range(10):
randIndex = int(random.uniform(0,len(trainingSet)))
testSet.append(trainingSet[randIndex])
print("testSet",testSet)
del(trainingSet[randIndex])
print(len(trainingSet))
# print(testSet)
trainMat = []
trainClasses = []
for docIndex in trainingSet:
# print(docIndex)
trainMat.append(wordVec.setOfWords2Vec(vocabList,docList[docIndex]))
trainClasses.append(classList[docIndex])
# print(trainMat)
p0V,p1V,pSpam = bayes.trainNBO(np.array(trainMat),np.array(trainClasses))
errorCount = 0
for docIndex in testSet:
wordVector = wordVec.setOfWords2Vec(vocabList,docList[docIndex])
if bayes.classifyNB(np.array(wordVector),p0V,p1V,pSpam) != classList[docIndex]:
errorCount += 1
print(errorCount)
print("分类错误的测试集：", docList[docIndex])
print(classList)
print(errorCount)
print(len(testSet))
print("the error rate is :",errorCount/len(testSet))
spamTest()
这里可能试几次错误率都是0，多run几次试试
