8.1 读书《机器学习实战》

## 4.5 使用Python进行文本分类 代码错误及修正

　　原代码4-2中条件概率分母有误, 如P(cute=1|ci=0)应为1/3.

def trainNB0(trainMatrix, trainCategory):
    numTrainDocs = len(trainMatrix)
    numWords = len(trainMatrix[0])
    pAbusive = sum(trainCategory)/float(numTrainDocs)
    p0Num = ones(numWords)
    p1Num = ones(numWords)
    p0Denom = 2.0
    p1Denom = 2.0
    for i in range(numTrainDocs):
        if trainCategory[i] == 1:
            p1Num += trainMatrix[i]
            p1Denom += 1 #条件概率分母修正
        else:
            p0Num += trainMatrix[i]
            p0Denom += 1 #条件概率分母修正
    p1Vect = p1Num/p1Denom #求log放在后面了
    p0Vect = p0Num/p0Denom #求log放在后面了
    return p0Vect, p1Vect, pAbusive

　　原代码4-3中计算p1和p0时只考虑了所有P(wi=1|ci)分量，而忽略了P(wi=0|ci)分量, 而P(wi=0|ci) = 1-P(wi=1|ci).

def classifyNB(vec2Classify, p0Vect, p1Vect, pClass1):
    oneVect = ones(len(p0Vect))  #制造一个等维度的全1向量
    p1VectInv = oneVect - p1Vect #制造P(wi=0|ci=1)向量
    p0VectInv = oneVect - p0Vect #制造P(wi=0|ci=0)向量
    p1Vect = log(p1Vect); p0Vect = log(p0Vect)
    p1VectInv = log(p1VectInv); p0VectInv = log(p0VectInv) #全部取对数
    vec2ClassifyInv = oneVect - vec2Classify #制造用于取出各个P(wi=0|ci)的向量
    p1 = sum(vec2Classify*p1Vect) + sum(vec2ClassifyInv*p1VectInv) + log(pClass1) 
    p0 = sum(vec2Classify*p0Vect) + sum(vec2ClassifyInv*p0VectInv) + log(1.0-pClass1)
    if p1 > p0:
        return 1
    else:
        return 0

 

## 5-5回归系数丢掉了w0项，应在训练集和测试集分别添加X0=1.0列.

 

注：本书配合CMU课程食用效果更佳：http://www.cs.cmu.edu/~ninamf/courses/601sp15/index.html