决策树（decision tree）简介

基本概念：

决策树用来解决分类问题，一棵树包含一个根结点，多个内部结点和多个叶子结点。未知数据从树的根结点开始判断，一直到叶子结点得到分类结果。本文主要关注决策树的构造方法。决策树算法是一种根据数据属性的泛化能力大小来进行分类的方法。其中属性的泛化能力大小由信息增益和信息增益率来评估。使用信息增益方法评估称为ID3算法，使用信息增益率评估称为C4.5。

具体步骤：

划分数据集是将信息增益或信息增益率最高的特征挑选出来。
1.信息熵定义：

2.信息增益定义：

属性的信息增益越大表明属性的泛化能力越强。
3.增益率定义：

C4.5决策树算法中不使用信息增益来衡量最优划分，而是使用信息增益率。
ID3的具体的代码如下：

from math import log
import time


def createDataSet():
    dataSet = [[1, 1, 'yes'],
               [1, 1, 'yes'],
               [1, 0, 'no'],
               [0, 1, 'no'],
               [0, 1, 'no']]
    labels = ['no surfaceing', 'flippers']
    return dataSet, labels


# 计算信息熵
def calcShannonEnt(dataSet):
    numEntries = len(dataSet)
    labelCounts = {}
    for feaVec in dataSet:
        currentLabel = feaVec[-1]
        if currentLabel not in labelCounts:
            labelCounts[currentLabel] = 0
        labelCounts[currentLabel] += 1
    shannonEnt = 0.0
    for key in labelCounts:
        prob = float(labelCounts[key]) / numEntries
        shannonEnt -= prob * log(prob, 2)
    return shannonEnt

#以指定特征将数据进行划分
def splitDataSet(dataSet, axis, value):#传入待划分的数据集、划分数据集的特征以及需要返回的特征的值
    retDataSet = []
    for featVec in dataSet:
        if featVec[axis] == value:
            reducedFeatVec = featVec[:axis]
            reducedFeatVec.extend(featVec[axis + 1:])
            retDataSet.append(reducedFeatVec)
    return retDataSet

#选择最优特征作为节点(选取每个特征划分数据集，从中选取信息增益最大的作为最优划分)
def chooseBestFeatureToSplit(dataSet):
    numFeatures = len(dataSet[0]) - 1  # 因为数据集的最后一项是标签
    baseEntropy = calcShannonEnt(dataSet)
    bestInfoGain = 0.0
    bestFeature = -1
    for i in range(numFeatures):
        featList = [example[i] for example in dataSet]
        uniqueVals = set(featList)
        newEntropy = 0.0
        for value in uniqueVals:
            subDataSet = splitDataSet(dataSet, i, value)
            prob = len(subDataSet) / float(len(dataSet))
            newEntropy += prob * calcShannonEnt(subDataSet)
        infoGain = baseEntropy - newEntropy
        if infoGain > bestInfoGain:
            bestInfoGain = infoGain
            bestFeature = i
    return bestFeature


# 因为我们递归构建决策树是根据属性的消耗进行计算的，所以可能会存在最后属性用完了，但是分类
# 还是没有算完，这时候就会采用多数表决的方式计算节点分类
def majorityCnt(classList):
    classCount = {}
    for vote in classList:
        if vote not in classCount.keys():
            classCount[vote] = 0
        classCount[vote] += 1
    return max(classCount)

#创建决策树
def createTree(dataSet, labels):
    classList = [example[-1] for example in dataSet]
    if classList.count(classList[0]) == len(classList):  # 类别相同则停止划分
        return classList[0]
    if len(dataSet[0]) == 1:  # 所有特征已经用完
        return majorityCnt(classList)
    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[:]  # 为了不改变原始列表的内容复制了一下
        myTree[bestFeatLabel][value] = createTree(splitDataSet(dataSet,
                                                               bestFeat, value), subLabels)
    return myTree


def main():
    data, label = createDataSet()
    t1 = time.clock()
    myTree = createTree(data, label)
    t2 = time.clock()
    print(myTree)
    print('execute for ', t2 - t1)


if __name__ == '__main__':
    main()

ID3与C4.5代码的主要区别在于选择数据集的划分方法，C4.5的划分方法如下：

def chooseBestFeatureToSplit(dataSet):
    numFeatures = len(dataSet[0]) - 1  # 因为数据集的最后一项是标签
    baseEntropy = calcShannonEnt(dataSet)
    bestInfoGain = 0.0
    bestFeature = -1
    for i in range(numFeatures):
        featList = [example[i] for example in dataSet]
        uniqueVals = set(featList)
        newEntropy = 0.0
        splitInfo = 0.0;
        for value in uniqueVals:
            subDataSet = splitDataSet(dataSet, i, value)
            prob = len(subDataSet) / float(len(dataSet))
            newEntropy += prob * calcShannonEnt(subDataSet)
            splitInfo -= prob * log(prob, 2);
            infoGain = (baseEntropy - newEntropy) / splitInfo;
        if infoGain > bestInfoGain:
            bestInfoGain = infoGain
            bestFeature = i
    return bestFeature