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']]
# 给出6个文本list
classVec = [0,1,0,1,0,1] # 人为给出标签 1 有侮辱性文字 0 正常
return postingList, classVec
def creatVocabList(dataSet):
VocabSet = set([]) # 创建一个空list，并使用set的数据类型，保证字典中无重复
for word in dataSet:
VocabSet = VocabSet | set(word) # 操作符‘|’表示两个集合的并集
return list(VocabSet) # 返回list的数据类型
def setOfWords2Vec(vocabList, inputSet): #词集模型
# input为输入要对比的样本 比如postingList[0]
returnVec = [0]*len(vocabList) # 创建所有为0的向量
for word in inputSet:
if word in vocabList:
returnVec[vocabList.index(word)] = 1 #vocabList.index(word) 为word在vocabList中的索引值 如果在vocabList中有这个单词，相应位置设为1
else:
print("the word %s is not in VocabList" % word)
return returnVec
# 输出文档向量，向量的元素为0或者1，分别表示字典中的单词在输入文档中是否没出现还是出现。
# 训练算法：计算概率
from numpy import *
def trainNB(trainMatrix, trainLabel): # 输入训练数据矩阵 和 训练标签
numTrainDocs = len(trainMatrix) # 训练样本个数
numWords = len(trainMatrix[0]) # 字典的长度 相当于属性个数
pAbusive = sum(trainLabel)/float(numTrainDocs ) # P(Vj) V={1,0} 含有侮辱性文档的概率，由于是二分类，另外一个 可以用1减去即可。
p0Num = ones(numWords) # 第0类文档中每个词条出现的总数量
p1Num = ones(numWords)
p0Denom = 2.0 # 第0类文档所有词条的数量
p1Denom = 2.0
for i in range(numTrainDocs): # 遍历所有的文档 对每篇训练文档：
if (trainLabel[i] == 1): # 对每个类别
p1Num += trainMatrix[i] # 增加该词条的计算值
p1Denom += sum(trainMatrix[i]) # 增加所有词条的计数值
else:
p0Num += trainMatrix[i]
p0Denom += sum(trainMatrix[i])
p1Vect = log(p1Num/p1Denom) # 对每个类别，对每个词条，计算条件概率
p0Vect = log(p0Num/p0Denom) # 对每个元素除以该类别中的总词数，（用一个数组除以浮点数即可实现）
return p0Vect,p1Vect,pAbusive # 返回两个向量 一个概率
'''
data, label = loadDataSet()
vocab = creatVocabList(data)
#print(vocab)
trainMat = []
for i in data:
trainMat.append(setOfWords2Vec(vocab, i))
p0v, p1v, pAb = trainNB(trainMat, label)
#print(p0v)
#print(p1v)
#print(pAb)
'''
#测试算法 : 朴素贝叶斯分类器
# p(v_j)*(多个P(ai|vj)相乘)
def classifyNB(Vect2classify, p0Vec, p1Vec, pAb): # 朴素贝叶斯分类器
# 输入Vect2classify 为要分类的文档,已经变为向量形式
p1 = sum(Vect2classify*p1Vec) + log(pAb) # log(ab)=log(a)+log(b)
p0 = sum(Vect2classify*p0Vec) + log(1-pAb) # 计算V_NB 括号里面的想成为对应元素相乘
if p1>p0: # 分类
return 1
else:
return 0
def testingNB():
data, label = loadDataSet()
Vocab = creatVocabList(data)
trainMat = []
for i in data:
trainMat.append(setOfWords2Vec(Vocab, i))
p0Vec, p1Vec, pAb = trainNB(trainMat, label)
test1 = ['love','my','dalmation']
test2 = ['stupid','garbage','love','dog']
testDoc1 = array(setOfWords2Vec(Vocab, test1)) # 为什么要用array数组？？ 注（a）
testDoc2 = array(setOfWords2Vec(Vocab, test2))
print(test1,'classified as: ', classifyNB(testDoc1, p0Vec, p1Vec, pAb))
print(test2,'classified as: ', classifyNB(testDoc2, p0Vec, p1Vec, pAb))
testingNB()
from numpy import *
def creatVocabList(dataSet):
def bagOfWords2Vec(vocabList, inputSet):
def trainNB(): # 上面已经解读
def classifyNB():
def textSeg(emailText):
def spamTest(): # 垃圾邮件测试
DocList = [] #
labelList = [] # 标签list
FullList = [] #
for i in range(1, 26): # 每个文件夹下25个txt文件
emailText = open('email/spam/%d.txt' % i).read()
wordList = textSeg(emailText)
DocList.append(wordList) # append extend 区别，决策树里面有详细讲解
FullList.extend(wordList)
labelList.append(1) # 人为标记
emailText = open('email/ham/%d.txt' % i).read()
wordList = textSeg(emailText)
DocList.append(wordList) # append extend 区别，决策树里面有详细讲解
FullList.extend(wordList)
labelList.append(0) # 人为标记
Vocab = creatVocabList(DocList) # 创建字典
# 随机抽取训练测试样本，总共50个样本 选取10个测试
traingSetIndex = list(range(50))
testSetIndex = []
for i in range(10):
index = int(random.uniform(0,50)) # 随机生成0-50的数
testSetIndex.append(traingSetIndex[index]) # 选出来的放入 testSetIndex
del(traingSetIndex[index]) # 还要删除traingSetIndex相应的位置索引
trainMat = []
trainLabel = []
for i in traingSetIndex:
trainMat.append(bagOfWords2Vec(Vocab, DocList[i]))
trainLabel.append(labelList[i])
p0Vec, p1Vec, pAb = trainNB(trainMat, trainLabel)
errorcount = 0
for i in testSetIndex: # 测试
test = bagOfWords2Vec(Vocab, DocList(i))
if classifyNB(test, p0Vec, p1Vec, pAb) != labelList[i]:
errorcount += 1
print('the error is: ', float(errorcount)/len(testSetIndex))
From：贝叶斯学习 -- matlab、python代码分析（4）