在上一篇文章中，我們已經構建了決策樹，接下來可以使用它用於實際的資料分類。在執行資料分類時，需要決策時以及標籤向量。程式比較測試資料和決策樹上的數值，遞迴執行直到進入葉子節點。

這篇文章主要使用決策樹分類器就行分類，資料集採用UCI資料庫中的紅酒，白酒資料，主要特徵包括12個，主要有非揮發性酸,揮發性酸度, 檸檬酸, 殘糖含量,氯化物, 遊離二氧化硫, 總二氧化硫,密度, pH,硫酸鹽,酒精, 質量等特徵。

下面是具體程式碼的實現：

#coding :utf-8
'''
2017.6.26 author :Erin 
          function: "decesion tree" ID3
          
'''
import numpy as np
import pandas as pd
from math import log
import operator  
import random
def load_data():
   
    red = [line.strip().split(';') for line in open('e:/a/winequality-red.csv')]
    white = [line.strip().split(';') for line in open('e:/a/winequality-white.csv')]
    data=red+white
    random.shuffle(data)  #打亂data
    x_train=data[:800]
    x_test=data[800:]
    
    features=['fixed','volatile','citric','residual','chlorides','free','total','density','pH','sulphates','alcohol','quality']
    return x_train,x_test,features

def cal_entropy(dataSet):
 
    
    numEntries = len(dataSet)
    labelCounts = {}
    for featVec in dataSet:
        label = featVec[-1]
        if label not in labelCounts.keys():
            labelCounts[label] = 0
        labelCounts[label] += 1
    entropy = 0.0
    for key in labelCounts.keys():
        p_i = float(labelCounts[key]/numEntries)
        entropy -= p_i * log(p_i,2)#log(x,10)表示以10 為底的對數
    return entropy

def split_data(data,feature_index,value):
    '''
    劃分資料集
    feature_index：用於劃分特徵的列數，例如“年齡”
    value:劃分後的屬性值：例如“青少年”
    '''
    data_split=[]#劃分後的資料集
    for feature in data:
        if feature[feature_index]==value:
            reFeature=feature[:feature_index]
            reFeature.extend(feature[feature_index+1:])
            data_split.append(reFeature)
    return data_split
def choose_best_to_split(data):
    
    '''
    根據每個特徵的資訊增益，選擇最大的劃分資料集的索引特徵
    '''
    
    count_feature=len(data[0])-1#特徵個數4
    #print(count_feature)#4
    entropy=cal_entropy(data)#原資料總的資訊熵
    #print(entropy)#0.9402859586706309
    
    max_info_gain=0.0#資訊增益最大
    split_fea_index = -1#資訊增益最大，對應的索引號

    for i in range(count_feature):
        
        feature_list=[fe_index[i] for fe_index in data]#獲取該列所有特徵值
        #######################################

       # print(feature_list)
        unqval=set(feature_list)#去除重複
        Pro_entropy=0.0#特徵的熵
        for value in unqval:#遍歷改特徵下的所有屬性
            sub_data=split_data(data,i,value)
            pro=len(sub_data)/float(len(data))
            Pro_entropy+=pro*cal_entropy(sub_data)
            #print(Pro_entropy)
            
        info_gain=entropy-Pro_entropy
        if(info_gain>max_info_gain):
            max_info_gain=info_gain
            split_fea_index=i
    return split_fea_index
        
        
##################################################
def most_occur_label(labels):
    #sorted_label_count[0][0]  次數最多的類標籤
    label_count={}
    for label in labels:
        if label not in label_count.keys():
            label_count[label]=0
        else:
            label_count[label]+=1
        sorted_label_count = sorted(label_count.items(),key = operator.itemgetter(1),reverse = True)
    return sorted_label_count[0][0]
def build_decesion_tree(dataSet,featnames):
    '''
    字典的鍵存放節點資訊，分支及葉子節點存放值
    '''
    featname = featnames[:]              ################
    classlist = [featvec[-1] for featvec in dataSet]  #此節點的分類情況
    if classlist.count(classlist[0]) == len(classlist):  #全部屬於一類
        return classlist[0]
    if len(dataSet[0]) == 1:         #分完了,沒有屬性了
        return Vote(classlist)       #少數服從多數
    # 選擇一個最優特徵進行劃分
    bestFeat = choose_best_to_split(dataSet)
    bestFeatname = featname[bestFeat]
    del(featname[bestFeat])     #防止下標不準
    DecisionTree = {bestFeatname:{}}
    # 建立分支,先找出所有屬性值,即分支數
    allvalue = [vec[bestFeat] for vec in dataSet]
    specvalue = sorted(list(set(allvalue)))  #使有一定順序
    for v in specvalue:
        copyfeatname = featname[:]
        DecisionTree[bestFeatname][v] =  build_decesion_tree(split_data(dataSet,bestFeat,v),copyfeatname)
    return DecisionTree

def classify(Tree, featnames, X):
    classLabel=''
    root = list(Tree.keys())[0]
    firstDict = Tree[root]
    featindex = featnames.index(root)  #根節點的屬性下標
    #classLabel='0'
    for key in firstDict.keys():   #根屬性的取值,取哪個就走往哪顆子樹
        if X[featindex] == key:
            if type(firstDict[key]) == type({}):
                classLabel = classify(firstDict[key],featnames,X)
            else:
                classLabel = firstDict[key]
    return classLabel



            
    
if __name__ == '__main__':
    x_train,x_test,features=load_data()
    split_fea_index=choose_best_to_split(x_train)
    newtree=build_decesion_tree(x_train,features)
    #print(newtree)
    #classLabel=classify(newtree, features, ['7.4','0.66','0','1.8','0.075','13','40','0.9978','3.51','0.56','9.4','5'] )
    #print(classLabel)
    
    count=0
    for test in x_test:
        label=classify(newtree, features,test)
        
        if(label==test[-1]):
            count=count+1
    acucy=float(count/len(x_test))
    print(acucy)