前两章我们要求分类器作出艰难的抉择，不过分类器有时候会产生错误，这时会产生错误结果，这是可以要求分类器给出一个最优的类别猜测结果，同事给出这个猜测的概率估计值。
本章会给出一些使用概率论进行分类的方法。首先从一个最简单的概率分类器开始，然后给出一些假设来学习朴素贝叶斯分类器。我们称之为“朴素”，是因为整个形式化过程只做最原始、最简单的假设。
朴素贝叶斯
优点：对于输入数据的准备方式较为敏感
缺点：对于输入数据的准备方式较为敏感
适用数据类型：标称型数据
在文档分类中，整个文档就是实例，而某些元素则构成特征。
# -*- coding:utf-8 -*-
#4-1 词表到向量的转化函数
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]
return postingList,classVec#词条切分后的集合，类别标签
def createVocabList(dataSet):
vocabSet = set([])
for document in dataSet:
vocabSet = vocabSet | set(document) #合集
return list(vocabSet)
def setOfwords2Vec(vocabList, inputSet):
returnVec = [0]*len(vocabList) #创建一个其中所有元素都为0的向量
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
#bayes-1.py
import bayes
listOPosts,listClasses = bayes.loadDataSet()
myVocabList = bayes.createVocabList(listOPosts)
>>> myVocabList
['cute', 'love', 'help', 'garbage', 'quit', 'I', 'problems', 'is', 'park', 'stop', 'flea', 'dalmation', 'licks', 'food', 'not', 'him', 'buying', 'posting', 'has', 'worthless', 'ate', 'to', 'maybe', 'please', 'dog', 'how', 'stupid', 'so', 'take', 'mr', 'steak', 'my']
>>> bayes.setOfwords2Vec(myVocabList, listOPosts[0])
[0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 1]
>>> bayes.setOfwords2Vec(myVocabList, listOPosts[3])
[0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0]
从词向量计算概率
#4-2 朴素贝叶斯分类器训练函数
import numpy as np
def trainNB0(trainMatrix, trainCategory):
m = len(trainMatrix) #numTrainDocs
n = len(trainMatrix[0]) #numWords
p0Num = np.zeros(n); p1Num = np.zeros(n)
p0Denom = 0.0; p1Denom = 0.0#初始化概率
for i in range(m):
if trainCategory[i] == 1:
p1Num += trainMatrix[i]
p1Denom += sum(trainMatrix[i])
else:
p0Num += trainMatrix[i]
p0Denom += sum(trainMatrix[i])
#此次因为缩进错误debug了很久...
p1Vect = p1Num/p1Denom
p0Vect = p0Num/p0Denom
pAbusive = sum(trainCategory)/float(m)
return p0Vect,p1Vect,pAbusive
#bayes-1.py
import bayes
from numpy import *
reload(bayes)
listOPosts,listClasses = bayes.loadDataSet()
myVocabList = bayes.createVocabList(listOPosts)
trainMat=[]
for postinDoc in listOPosts:
trainMat.append(bayes.setOfwords2Vec(myVocabList, postinDoc))
p0V,p1V,pAb = bayes.trainNB0(trainMat,listClasses)
>>> pAb
0.5
>>> p0V
array([ 0.04166667, 0.04166667, 0.04166667, 0. , 0. ,
0.04166667, 0.04166667, 0.04166667, 0. , 0.04166667,
0.04166667, 0.04166667, 0.04166667, 0. , 0. ,
0.08333333, 0. , 0. , 0.04166667, 0. ,
0.04166667, 0.04166667, 0. , 0.04166667, 0.04166667,
0.04166667, 0. , 0.04166667, 0. , 0.04166667,
0.04166667, 0.125 ])
>>> p1V
array([ 0. , 0. , 0. , 0.05263158, 0.05263158,
0. , 0. , 0. , 0.05263158, 0.05263158,
0. , 0. , 0. , 0.05263158, 0.05263158,
0.05263158, 0.05263158, 0.05263158, 0. , 0.10526316,
0. , 0.05263158, 0.05263158, 0. , 0.10526316,
0. , 0.15789474, 0. , 0.05263158, 0. ,
0. , 0. ])
>>>
利用贝叶斯分类器对文档进行分类时，要计算多个概率的乘积以获得文档属于某个类别的概率。
如果其中一个概率值为0，那么最后乘积也是0。
为了降低这种影响，可以将所有词的出现数改为1，并将分母改为2
另外可以遇到下溢出。由于太多很小的树相乘，可能四舍五入得到0。可以改成log
p0Num = np.ones(n); p1Num = np.ones(n)
p0Denom = 2.0; p1Denom = 2.0
p1Vect = log(p1Num/p1Denom)
p0Vect = log(p0Num/p0Denom)
开始使用numpy向量处理功能
def classifyNB(vec2Classify, p0Vec, p1Vec, pClass1):#要分类的向量
p1 = sum(vec2Classify * p1Vec) + log(pClass1)#向量相乘
p0 = sum(vec2Classify * p0Vec) + log(1.0 - pClass1)
if p1 > p0:
return 1
else:
return 0
def testingNB():
listOPosts,listClasses = loadDataSet()
myVocabList = createVocabList(listOPosts)
trainMat=[]
for postinDoc in listOPosts:
trainMat.append(setOfWords2Vec(myVocabList,postinDoc))
p0V,p1V,pAb = trainNB0(array(trainMat),array(listClasses))
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)
#bayes-1.py
import bayes
from numpy import *
reload(bayes)
listOPosts,listClasses = bayes.loadDataSet()
myVocabList = bayes.createVocabList(listOPosts)
trainMat=[]
for postinDoc in listOPosts:
trainMat.append(bayes.setOfWords2Vec(myVocabList, postinDoc))
p0V,p1V,pAb = bayes.trainNB0(trainMat,listClasses)
bayes.testingNB()
测试这两句话
['love', 'my', 'dalmation'] classified as: 0
['stupid', 'garbage'] classified as: 1
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，这种方法被称为词袋模型。
4-4朴素贝叶斯词袋模型
def bagOfWords2VecMN(vocabList, inputSet):
returnVec = [0]*len(vocavList)
for word in inputSet:
if word in inputSet:
returnVec[vocabList.index(word)] += 1
return returnVec
下面开始了解贝叶斯的一个著名应用：电子邮件垃圾过滤。
首先准备数据，切分文本
>>> mySent='This book is the best book on Python or M.L. I have ever laid eyes upon.'
>>> mySent.split()
['This', 'book', 'is', 'the', 'best', 'book', 'on', 'Python', 'or', 'M.L.', 'I', 'have', 'ever', 'laid', 'eyes', 'upon.']
但是标点符号也被当成了词的一部分，可以用正则表达式来切分句子，其中分隔符是除单词、数字外的任意字符串。
>>> import re
>>> regEx = re.compile('\\W*')
>>> listOfTokens = regEx.split(mySent)
>>> listOfTokens
['This', 'book', 'is', 'the', 'best', 'book', 'on', 'Python', 'or', 'M', 'L', 'I', 'have', 'ever', 'laid', 'eyes', 'upon', '']
接着我们把里面的空字符串去掉
[tok for tok in listOfTokens if len(tok) > 0]
['This', 'book', 'is', 'the', 'best', 'book', 'on', 'Python', 'or', 'M', 'L', 'I', 'have', 'ever', 'laid', 'eyes', 'upon']
然后把所有字幕改成小写
>>> [tok.lower() for tok in listOfTokens if len(tok) > 0]
['this', 'book', 'is', 'the', 'best', 'book', 'on', 'python', 'or', 'm', 'l', 'i', 'have', 'ever', 'laid', 'eyes', 'upon']
现在来看数据集中一封完整的电子邮件的实际处理结果。
>>> emailText = open('E:/上学/机器学习实战/4.朴素贝叶斯/email/ham/6.txt').read()
>>> listOfTokens=regEx.split(emailText)
>>> listOfTokens
['Hello', 'Since', 'you', 'are', 'an', 'owner', 'of', 'at', 'least', 'one', 'Google', 'Groups', 'group', 'that', 'uses', 'the', 'customized', 'welcome', 'message', 'pages', 'or', 'files', 'we', 'are', 'writing', 'to', 'inform', 'you', 'that', 'we', 'will', 'no', 'longer', 'be', 'supporting', 'these', 'features', 'starting', 'February', '2011', 'We', 'made', 'this', 'decision', 'so', 'that', 'we', 'can', 'focus', 'on', 'improving', 'the', 'core', 'functionalities', 'of', 'Google', 'Groups', 'mailing', 'lists', 'and', 'forum', 'discussions', 'Instead', 'of', 'these', 'features', 'we', 'encourage', 'you', 'to', 'use', 'products', 'that', 'are', 'designed', 'specifically', 'for', 'file', 'storage', 'and', 'page', 'creation', 'such', 'as', 'Google', 'Docs', 'and', 'Google', 'Sites', 'For', 'example', 'you', 'can', 'easily', 'create', 'your', 'pages', 'on', 'Google', 'Sites', 'and', 'share', 'the', 'site', 'http', 'www', 'google', 'com', 'support', 'sites', 'bin', 'answer', 'py', 'hl', 'en', 'answer', '174623', 'with', 'the', 'members', 'of', 'your', 'group', 'You', 'can', 'also', 'store', 'your', 'files', 'on', 'the', 'site', 'by', 'attaching', 'files', 'to', 'pages', 'http', 'www', 'google', 'com', 'support', 'sites', 'bin', 'answer', 'py', 'hl', 'en', 'answer', '90563', 'on', 'the', 'site', 'If', 'you', 're', 'just', 'looking', 'for', 'a', 'place', 'to', 'upload', 'your', 'files', 'so', 'that', 'your', 'group', 'members', 'can', 'download', 'them', 'we', 'suggest', 'you', 'try', 'Google', 'Docs', 'You', 'can', 'upload', 'files', 'http', 'docs', 'google', 'com', 'support', 'bin', 'answer', 'py', 'hl', 'en', 'answer', '50092', 'and', 'share', 'access', 'with', 'either', 'a', 'group', 'http', 'docs', 'google', 'com', 'support', 'bin', 'answer', 'py', 'hl', 'en', 'answer', '66343', 'or', 'an', 'individual', 'http', 'docs', 'google', 'com', 'support', 'bin', 'answer', 'py', 'hl', 'en', 'answer', '86152', 'assigning', 'either', 'edit', 'or', 'download', 'only', 'access', 'to', 'the', 'files', 'you', 'have', 'received', 'this', 'mandatory', 'email', 'service', 'announcement', 'to', 'update', 'you', 'about', 'important', 'changes', 'to', 'Google', 'Groups', '']
文本解析是个相当复杂的过程，接下来将构建一个极其简单的函数。
def textParse(bigString):
import re
listOfTokens = re.split(r'\W*',bigString)
return [tok.lower() for tok in listOfTokens if len(tok) > 2]
def spamTest():
docList=[]; classList = []; fullText = []
for i in range(1,26):
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())
docList.append(wordList)
fullText.extend(wordList)
classList.append(0)
vocabList = createVocabList(docList)
trainingSet = range(50); testSet=[]
for i in range(10):#随机选择10个作为训练集
randIndex = int(random.uniform(0,len(trainingSet)))
testSet.append(trainingSet[randIndex])
del(trainingSet[randIndex])
trainMat=[]; trainClasses = []
for docIndex in trainingSet:
trainMat.append(bagOfWords2VecMN(vocabList,docList[docIndex]))
trainClasses.append(classList[docIndex])
p0V,p1V,pSpam = trainNB0(array(trainMat),array(trainClasses))
errorCount = 0
for docIndex in testSet:
wordVector = bagOfWords2VecMN(vocabList,docList[docIndex])
if classifyNB(array(wordVector),p0V,p1V,pSpam) !=classList[docIndex]:
errorCount += 1
print "classification error",docList[docIndex]
print 'the error rate is: ',float(errorCount)/len(testSet)
因为是随机选择10封电子邮件，所以每次都有差别。如果发现错误，函数会输出错分文档的词表，这样就可以分析是那篇文档发生了错误。
bayes.spamTest()
classification error ['yeah', 'ready', 'may', 'not', 'here', 'because', 'jar', 'jar', 'has', 'plane', 'tickets', 'germany', 'for']
the error rate is: 0.1
bayes.spamTest()
the error rate is: 0.0
bayes.spamTest()
classification error ['benoit', 'mandelbrot', '1924', '2010', 'benoit', 'mandelbrot', '1924', '2010', 'wilmott', 'team', 'benoit', 'mandelbrot', 'the', 'mathematician', 'the', 'father', 'fractal', 'mathematics', 'and', 'advocate', 'more', 'sophisticated', 'modelling', 'quantitative', 'finance', 'died', '14th', 'october', '2010', 'aged', 'wilmott', 'magazine', 'has', 'often', 'featured', 'mandelbrot', 'his', 'ideas', 'and', 'the', 'work', 'others', 'inspired', 'his', 'fundamental', 'insights', 'you', 'must', 'logged', 'view', 'these', 'articles', 'from', 'past', 'issues', 'wilmott', 'magazine']
the error rate is: 0.1
下面使用朴素贝叶斯分类器从个人广告中获取区域倾向
本人现在使用的是Anaconda，先安装feedparser包，然后构建一个类似于spamTest()的函数
def calcMostFreq(vocabList,fullText):#计算单词出现的频率
import operator
freqDict = {}
for token in vocabList:
freqDict[token]=fullText.count(token)
sortedFreq = sorted(freqDict.iteritems(),key=operator.itemgetter(1),reverse=True)#排序
return sortedFreq[:30]#选出最多的30个
def localWords(feed1,feed0):
import feedparser
docList=[];classList = []; fullText = []
minLen = min(len(feed1['entries']),len(feed0['entries']))
for i in range(minLen):
wordList = textParse(feed1['entries'][i]['summary'])
docList.append(wordList)
fullText.extend(wordList)
classList.append(1)
wordList = textParse(feed0['entries'][i]['summary'])
docList.append(wordList)
fullText.extend(wordList)
classList.append(0)
vocabList = createVocabList(docList)
top30Words = calcMostFreq(vocabList,fullText)#去掉出现次数最高的那些词
for pairW in top30Words:
if pairW[0] in vocabList: vocabList.remove(pairW[0])
trainingSet = range(2*minLen); testSet=[]
for i in range(20):
randIndex = int(random.uniform(0,len(trainingSet)))
testSet.append(trainingSet[randIndex])
del(trainingSet[randIndex])
trainMat=[]; trainClasses = []
for docIndex in trainingSet:
trainMat.append(bagOfWords2VecMN(vocabList,docList[docIndex]))
trainClasses.append(classList[docIndex])
p0V,p1V,pSpam = trainNB0(array(trainMat),array(trainClasses))
errorCount = 0
for docIndex in testSet:
wordVector = bagOfWords2VecMN(vocabList, docList[docIndex])
if classifyNB(array(wordVector),p0V,p1V,pSpam) != classList[docIndex]:
errorCount += 1
print 'the error rate is:',float(errorCount)/len(testSet)
return vocabList,p0V,p1V
在iPython中调试
import bayes
ny=feedparser.parse('http://newyork.craigslist.org/stp/index.rss')
sf=feedparser.parse('http://sfbay.craigslist.org/stp/index.rss')
vocabList,pSF,pNY=bayes.localWords(ny,sf)
the error rate is: 0.5
import bayes
ny=feedparser.parse('http://newyork.craigslist.org/stp/index.rss')
sf=feedparser.parse('http://sfbay.craigslist.org/stp/index.rss')
vocabList,pSF,pNY=bayes.localWords(ny,sf)
the error rate is: 0.55
为了得到错误率的精确估计，应该多次进行上述实验，最后取平均值
下面开始显示地域相关的用词
def getTopWords(ny,sf):
import operator
vocabList,p0V,p1V=localWords(ny,sf)
topNY=[];topSF=[]
for i in range(len(p0V)):
if p0V[i] > -4.5 : topSF.append((vocabList[i],p0V[i]))
if p1V[i] > -4.5 : topNY.append((vocabList[i],p1V[i]))
sortedSF = sorted(topSF,key=lambda pair:pair[1],reverse=True)
print 'SF**SF**SF**SF**SF**SF**SF**SF**SF**SF**SF**SF**SF**SF**SF**SF**'
for item in sortedSF:
print item[0]
sortedNY = sorted(topNY,key=lambda pair:pair[1], reverse=True)
print "NY**NY**NY**NY**NY**NY**NY**NY**NY**NY**NY**NY**NY**NY**NY**NY**"
for item in sortedNY:
print item[0]
和之前返回排名最高的X个单词不同，这里可以返回大于某个阈值的所有词，这些元组会按照它们的条件概率进行排序。
import bayes
ny=feedparser.parse('http://newyork.craigslist.org/stp/index.rss')
sf=feedparser.parse('http://sfbay.craigslist.org/stp/index.rss')
bayes.getTopWords(ny,sf)
the error rate is: 0.4
SF**SF**SF**SF**SF**SF**SF**SF**SF**SF**SF**SF**SF**SF**SF**SF**
meet
all
61514
great
open
any
movie
about
area
NY**NY**NY**NY**NY**NY**NY**NY**NY**NY**NY**NY**NY**NY**NY**NY**
show
contact
info
very
guy
massage
talk
nyc
first
need
only
off
网友评论