机器学习实战 第三章：决策树

from math import log

def createDataSet():             %创建数据集
    dataSet = [[1, 1, 'yes'],
               [1, 1, 'yes'],
               [1, 0, 'no'],
               [0, 1, 'no'],
               [0, 1, 'no']]
    labels = ['no surfacing','flippers']
    return dataSet, labels

def calcShannonEnt(dataSet):         %计算香农熵
    numEntries = len(dataSet)        %numEntries =5
    labelCounts = {}                 %空的字典
    for featVec in dataSet: 
        currentLabel = featVec[-1]    %比如第一次循环，currentLabel = 'yes'
        if currentLabel not in labelCounts.keys(): 
        labelCounts[currentLabel] = 0
        labelCounts[currentLabel] += 1
    shannonEnt = 0.0
    for key in labelCounts:           
        prob = float(labelCounts[key])/numEntries        %以第一次循环为例，prob= float(2)/5 
        shannonEnt -= prob * log(prob,2)                 %求和
    return shannonEnt

在python命令提示符下输入以下内容：

reload(trees)

myDat,labels = trees.createDataSet()

改为了

import trees
myDat,labels = trees.createDataSet()
print(myDat)

myDat[0][-1]='maybe'   %第0行，最后一个元素改为'maybe'

def splitDataSet(dataSet, axis, value):    %划分数据集，输入数据集、分类特征、特征值
    retDataSet = []
    for featVec in dataSet:      %比如axis==0 value==1
        if featVec[axis] == value:      %   dataSet第一个元素满足条件       reducedFeatVec = featVec[:axis]     %reducedFeatVec =featVec[:0]=[]
       reducedFeatVec.extend(featVec[axis+1:])   %reducedFeatVec.extend(featVec[0+1:])=[1,'yes'] 
       retDataSet.append(reducedFeatVec)
    return retDataSet           

def chooseBestFeatureToSplit(dataSet):
    numFeatures = len(dataSet[0]) - 1      #dateset的最后一列作为标签
    baseEntropy = calcShannonEnt(dataSet)    #计算原始香农熵
    bestInfoGain = 0.0; bestFeature = -1
    for i in range(numFeatures):        #numFeatures=2,i=0,1
        featList = [example[i] for example in dataSet]    #数据集中i=0,1 列的特征值featList=[1, 1, 1, 0, 0]，[1, 1, 0, 1, 1]
        uniqueVals = set(featList)       #得到独立不重复的值uniqueVals ={0，1}
        newEntropy = 0.0
        for value in uniqueVals: #valiue=0,1俩次循环
            subDataSet = splitDataSet(dataSet, i, value)       #划分数据集
            prob = len(subDataSet)/float(len(dataSet))          #计算概率
            newEntropy += prob * calcShannonEnt(subDataSet)      #这里的prob算是每个香农熵的权值，求和
        infoGain = baseEntropy - newEntropy      
        if (infoGain > bestInfoGain):       
            bestInfoGain = infoGain         
            bestFeature = i                                     #熵减少越多，信息增益越大
    return bestFeature

def majorityCnt(classList):     #多数投票
    classCount={}
    for vote in classList:
        if vote not in classCount.keys():     #类标签计数
            classCount[vote] = 0
        classCount[vote] += 1
    sortedClassCount = sorted(classCount.iteritems(), key=operator.itemgetter(1), reverse=True)  #排序
    return sortedClassCount[0][0]