机器学习实战之决策树

目前代码都是基于书本(机器学习实战)的，主要整理一下决策树的代码思路：

首先要明确一点：决策树中叶子节点一定是属于同一类的(因此这里如果有无法分类的特征时采用多数表决法)。

决策树的算法是ID3算法，核心是信息熵，也就是通过公式计算，不断找到信息熵最高的特征值并根据该特征对数据进行分类，信息熵也叫信息增益， 信息增益越大, 表示新集合越趋向有序 。因此这里侧重于处理数据，如何选取最优的数据集划分特征（即如何计算信息增益）？如何将一个特征值的列抽出来？ 最后如何建树？  

from math import log
import operator

def calcShannonEnt(dataSet):#计算单个特征
    numEntries = len(dataSet)
    labelCounts = {}
    for featVec in dataSet:
        currentLabel = featVec[-1]
        if currentLabel not in labelCounts.keys():
            labelCounts[currentLabel] = 0
        labelCounts[currentLabel] += 1
    shannonEnt = 0.0
    for key in labelCounts:
        prob = float(labelCounts[key])/numEntries
        shannonEnt -= prob * log(prob,2)#总的熵
    return shannonEnt
# myDat = [[1,1,'yes'],[1,1,'yes'],[1,0,'middle'],[0,1,'no'],[0,1,'no']]
myDat=[[1, 1, 'yes'],
      [1, 1, 'yes'],
      [1, 0, 'middle'],
      [0, 1, 'no'],
      [0, 1, 'no']]
labels =['no surfacing','flippers']
# print(calcShannonEnt(myDat))

def splitDataSet(dataSet,axis,value):# axis=0  3是指定值
    retDataSet = []
    for featVec in dataSet:
        if featVec[axis] == value:
            reducedFeatVec = featVec[:axis]
            reducedFeatVec.extend(featVec[axis+1:])
            retDataSet.append(reducedFeatVec)
    # print(axis)
    # print(retDataSet)
    # print()
    return retDataSet

# print(splitDataSet(myDat,0,1))
# print(splitDataSet(myDat,0,0))

def chooseBestFeatureToSplit(dataSet):
    numFeatures = len(dataSet[0])-1 #征数（总特列数）
    baseEntropy = calcShannonEnt(dataSet)#整个数据集的原始香农熵
    bestInfoGain = 0.0; bestFeature = -1
    for i in range(numFeatures):#特征个数
        featList = [example[i] for example in dataSet] #一个类别的是否放在一个列表里  一个特征的抽出来
        # print(featList)
        uniqueVals = set(featList)#状态数？？
        newEntropy = 0.0
        for value in uniqueVals:#用过的特征值不会再用
            subDataSet = splitDataSet(dataSet,i,value)# i表示特征数（第几个特征）
            prob = len(subDataSet)/float(len(dataSet))
            newEntropy += prob * calcShannonEnt(subDataSet) #表示抽出哪一列 即去掉那一列
        infoGain = baseEntropy - newEntropy
        if(infoGain > bestInfoGain):#找出最高的香农熵
            bestInfoGain = infoGain
            bestFeature = i
    return bestFeature

#print(chooseBestFeatureToSplit(myDat))#以哪个特征来分最好

#类标签仍不唯一时 采用多数表决
def majority(classList):
    classCount={}
    for vote in classList:
        if vote not in classCount.keys():
            classCount[vote]=0
        classCount[vote]+=1
    sortedClassCount = sorted(classCount.iteritems(),key=operator.itemgetter(1),reverse=True)#降序
    return sortedClassCount[0][0]

def createTree(dataSet,labels):
    # print(dataSet)
    classList = [example[-1] for example in dataSet]
    # print(classList)
    # print()
    if classList.count(classList[0]) == len(classList):
        return classList[0]#类别完全相同
    if len(dataSet[0]) == 1:
        return majority(classList)
    print(dataSet)
    print()
    bestFeat = chooseBestFeatureToSplit(dataSet)

    bestFeatLabel = labels[bestFeat]
    myTree = {bestFeatLabel:{}}
    del(labels[bestFeat])
    featValues = [example[bestFeat] for example in dataSet]
    uniqueVals = set(featValues)
    for value in uniqueVals:
        subLabels = labels[:]
        myTree[bestFeatLabel][value] = createTree(splitDataSet(dataSet,bestFeat,value),subLabels)
        return myTree
# print()
myTree = createTree(myDat,labels)
# print(myTree)
# print(myTree)

转载于:https://my.oschina.net/u/3719898/blog/1579758