决策树剪枝中的损失函数的实现

#!/usr/bin/python
#-*-coding:utf-8 -*-
#决策树的剪枝算法
import ID3alogorithem as id3
from math import log
#  myTree数据类型 {'no surfacing': {0: 'no', 1: {'flippers': {0: 'no', 1: 'yes'}}}}
#计算每个结点的经验熵
def eachNodeEntropy():
    dataSet,labels=id3.createDataSET()
    myTree=id3.createTree(dataSet,labels)
    i,j=0,0
    label=""
    node=(i,j,label)
    #存储所有节点个数
    nodeList=[]
    return 0
#得到所有叶节点
# def getNodeList(dataSet):
#     firstDict=dataSet[0]


#找到当前节点的路径
#思考一部分，写一部分，不要想一下子写完
#{0: {'no surfacing': 0}, 1: {'no surfacing': 1, 'flippers': 0}, 2: {'no surfacing': 1, 'flippers': 1}}
#
def getNodePath(dataSet,path,i,allPaths):
    firstFeature=dataSet.keys()[0]
    secondDict=dataSet[firstFeature]
    #只在最开始，初始化paths
    if firstFeature=="no surfacing":
        paths={}
        i=0
        allPaths={}
    else :
        paths=path
    # print(str(dataSet)+"dataSet")
    for key in secondDict.keys():
        if type(secondDict[key]).__name__=="dict":
            path[firstFeature]=key
            getNodePath(secondDict[key],path,i,allPaths)
        else:
            # 保存当前的 特征以及特征值
            path[firstFeature]=key
            paths=dict(paths.items()+path.items())
            # print(paths)
            allPaths[i]=paths
            i=i+1
            path={}
    # print(allPath)
    return allPaths

def getfeatureIndex(labels,feature):
    i=-1
    # print(labels)
    for label in labels:
        i+=1
        if label==feature:
            # print(i)
            return i

# 由路径找到当前的叶节点的样本点
def searchSample(path,dataSet,labels):
    feature=[]
    feaValue=[]
    for item,value in path.items():
        # print(item+"-----"+str(value))
        feature.append(item)
        feaValue.append(value)
    # print(str(feature)+"feaValue  "+str(feaValue))
    i=0
    while i<len(feature):
        fea=getfeatureIndex(labels,feature[i])
        # print(str(feature[i])+" dataSet"+str(dataSet)+"----"+str(feaValue[i]))
        dataSet=id3.splitDataSet(dataSet,fea,feaValue[i])
        #没有把根数据破坏掉
        nowData=[]
        before=labels[:fea]
        after=labels[fea+1:]
        nowData.extend(before)
        nowData.extend(after)
        labels=nowData
        i+=1
        # print(i)
    return dataSet

#得到损失函数的值
def getLossMethod(paremeter):
    dataSet,labels=id3.createDataSET()
    #在这里labels 被改变了
    allPaths=getNodePath(id3.createTree(dataSet,labels),{},0,{})
    # print(allPaths[0])
    dataSet,labels2=id3.createDataSET()
    print(allPaths)
    CT=0.0
    method=0.0
    for key,path in allPaths.items():
        print(str(path)+" path")
        nowdata=searchSample(path,dataSet,labels2)
          #叶节点的样本点
        print(nowdata)
        #得到损失函数值
        CT+=getCT(nowdata)
    method=CT+paremeter*list(allPaths.keys())[-1]
    print(str(method)+" methodValues")


#得到C（T）的值
def getCT(dataSet):
    numData=len(dataSet)
    MarkList=[data[-1] for data in dataSet]
    uniqueList=set(MarkList)
    ratio=0.0
    for value in uniqueList:
        numTK=0
        for data in dataSet:
            if value==data[-1]:
                numTK+=1
        ratio+=numTK*(log(numTK/numData,2))
    print("ratio "+str(ratio))
    return ratio
getLossMethod(0.3)