朴素贝叶斯

朴素贝叶斯

常用于文本分类的朴素贝叶斯算法基于贝叶斯假设而来。和逻辑回归是概率判别式模型不同，朴素贝叶斯算法是概率生成式模型，不需要直接计算生成的概率值。朴素贝叶斯算法对小规模数据处理效果较好，但是若样本属性有关联时分类效果不好。
其最优化目标是是使得后验概率最大，通过贝叶斯假设使得联合概率可以分开乘积（否则则要通过极大似然方法估计参数，再转化成最优化的极值问题求解），这也就大大简化了计算，算法因此也十分简洁。
代码实现如下：

import numpy as np

#载入文本
def loadDataSet():
	postingList=[['my', 'dog', 'has', 'flea', 'problems', 'help', 'please'],				#切分的词条
				['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]   																#类别标签向量，1代表侮辱性词汇，0代表不是
	return postingList,classVec																#返回实验样本切分的词条和类别标签向量

def createVocabList(dataSet):
    VocabList = set([])   #创建不重复的空集
    for document in dataSet:
        VocabList = VocabList | set(document)    #取并集
    return list(VocabList)

#词集模型
def setOfWords2Vec(VocabList,inputSet):
    retVec = [0] * len(VocabList)
    for word in inputSet:
        if word in VocabList:
            retVec[VocabList.index(word)] = 1
    return retVec

# #词袋模型
# def setOfWords2Vec(VocabList,inputSet):
#     retVec = [0] * len(VocabList)
#     for word in inputSet:
#         if word in VocabList:
#             retVec[VocabList.index(word)] += 1
#     return retVec

#朴素贝叶斯分类器训练函数
def trainNB0(trainMatrix,trainCategory):
    numDocs = len(trainMatrix)
    numwords = len(trainMatrix[0])
    p0Num = np.zeros(numwords)
    p1Num = np.zeros(numwords)
    p0Demon = 0.0
    p1Demon = 0.0
    pc1 = sum(trainCategory)/float(numDocs)
    for i in range(numDocs):
        if trainCategory[i] == 1:
            p1Num += trainMatrix[i]
            p1Demon += sum(trainMatrix[i])
        else:
            p0Num += trainMatrix[i]
            p0Demon += sum(trainMatrix[i])
    p1Vect = p1Num / p1Demon
    p0Vect = p0Num / p0Demon
    return p0Vect,p1Vect,pc1

#朴素贝叶斯分类函数
def classifyNB(vec2Classify,p0Vect,p1Vect,pc1):    #vec2Classify为文本经过向量化后的向量
	#这里转变成log方便数学计算
    p1 = sum(vec2Classify * p1Vect) + np.log(pc1)
    p0 = sum(vec2Classify * p0Vect) + np.log(1.0-pc1)
    if p1 > p0:
        return 1
    else:
        return 0

def testingNB():
    list0posts, listClasses = loadDataSet()
    myVocabulist = createVocabList(list0posts)
    trainmat = []
    for postinDoc in list0posts:
        trainmat.append(setOfWords2Vec(myVocabulist,postinDoc))
    p0V,p1V,pcV = trainNB0(np.array(trainmat),np.array(listClasses))
    testEntry = ['love','my','dalmation']
    testVec = np.array(setOfWords2Vec(myVocabulist, testEntry))
    if classifyNB(testVec, p0V, p1V, pcV) == 1:
        print(testEntry, '属于侮辱类')
    else:
        print(testEntry, '属于非侮辱类')
    testEntry = ['stupid','garbage']
    testVec = np.array(setOfWords2Vec(myVocabulist,testEntry))
    if classifyNB(testVec,p0V,p1V,pcV) == 1:
        print(testEntry,'属于侮辱类')
    else:
        print(testEntry,'属于非侮辱类')

if __name__ == '__main__':
    testingNB()

运行结果：

参考：
1、《机器学习实战》