决策树（ID3）

参考自“机器学习实战”

import numpy as np
from collections import *
from math import log
def createDataset():
    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):

    label_count = Counter([data[-1] for data in dataset])
    print(label_count)
    probs = [p[1]/len(dataset) for p in label_count.items()]
    shannonEnt = sum([(-p)*log(p,2) for p in probs])
    return shannonEnt

def splitDataSet(dataSet,index,value):
    retDataset = [data for data in dataset for i,v in enumerate(data) if i==index and v==value]

def chooseBestFeatureToSplit(dataSet):
    base_entropy = calcShannonEnt(dataset)
    base_info_gain =0
    best_feature = -1

    for i in range(len(dataset[0]-1)):
        feature_count = Counter(data[i] for data in dataset)
        new_entropy = sum(feature[1]/float(len(dataset)) * calcShannonEnt(splitDataSet(dataset,i,feature[0])) \
                            for feature in feature_count.items() )

        info_gain = base_entropy - new_entropy
        print('now. {%d} feature info is {.3f}'.format(i,info_gain))
        if info_gain > base_info_gain:
            base_info_gain = info_gain
            best_feature = i

    return best_feature

def majorityCnt(classList):
    major_label = Counter(classList).most_common(1)[0]
    return major_label


def createTree(dataSet,labels):
    classList = [e[-1] for e in dataset]






if __name__ == "__main__":
    dataset,label = createDataset()
    calcShannonEnt(dataset)



    # d = [1, 1, 'yes']
    # for i,v in enumerate(d):
    #     print(i,v)

然后，构建树的过程，是一个递归的过程，递归终止的条件：

• 1）遍历完所有划分数据集的属性；
• 2）某个分支下所有实例都具有相同的分类。

递归的流程：

• 寻找划分数据集的最好特征
• 划分数据集
• 创建分支节点
  • for 每个分支节点
    • 递归，并增加返回结果到分支节点中