【机器学习算法】之决策树算法的实现

为了加深对机器学习算法的理解，以及熟悉python,pandas,scikit-learn。现在自己实现一下主要的机器学习算法，程序记录如下：

决策树类的实现程序：

import numpy as np  
import pandas as pd  
import random as rd  
import re  
from sklearn import tree  
from sklearn import preprocessing  
from sklearn import cross_validation  
from math import log

class dec_tree():
     def __init__(self, name):
         self.train_file = name
         self.model={}
         self.feature_name =[]
     def train(self):
         data_set,fea_name=load_data_set(self.train_file)
         mytree=build_tree(data_set,fea_name)
         self.model = mytree
     def test(self,x):
         data_set,fea_name=load_data_set(self.train_file)
         self.feature_name=fea_name
         res = tree_test(self.model,x,self.feature_name)
         return res 









def gen_entro(data_set):
    set_len=len(data_set)
    data_dic={}
    for num in data_set:
        if(data_dic.get(num[-1],0)==0):
            data_dic[num[-1]]=1
        else:
            data_dic[num[-1]]+=1
    h=0
    for num in data_dic.keys():
       curr_val= data_dic[num] 
       h+= -float(curr_val/set_len)*log(float(curr_val)/set_len,2)
    return h

def cut_data_set(data_set,fea_idx,value):
    new_data_set=[]
    for data in data_set:
        if(data[fea_idx]==value):
            curr_list=[]
            if(fea_idx==0):
                curr_list=data[(fea_idx+1):]
            else:
                curr_list=data[:fea_idx]+data[(fea_idx+1):]
            new_data_set.append(curr_list)
    return new_data_set

def chose_best_fea(data_set):
    data_set=np.array(data_set)
    fea_len = len(data_set[0])-1
    curr_max_gain=0;curr_max_idx=0
    for fea_idx in range(fea_len):
        curr_list = np.unique(data_set[:,fea_idx])
        info_gain = 0        
        for val in curr_list:
            new_data_set=cut_data_set(data_set,fea_idx,val)
            prob = len(new_data_set)/len(data_set)
            entro = gen_entro(new_data_set)
            info_gain+=-prob*entro
        curr_gain = gen_entro(data_set)-info_gain  
        if(curr_gain>curr_max_gain):
            curr_max_idx = fea_idx
            curr_max_gain = curr_gain
    return curr_max_idx

def build_tree(data_set,label):
    tree={}
    data_set=np.array(data_set)
    if(len(data_set[0])==1):
        major_label = gen_major_label(data_set)       
        return  major_label
    label_set = set(data_set[:,-1])
    if(len(label_set)==1):
        return data_set[0][-1]
    fea_idx = chose_best_fea(data_set)
    curr_list = np.unique(data_set[:,fea_idx])
    label_name = label[fea_idx]
    del label[fea_idx]
    for val in curr_list:
       new_data_set=cut_data_set(data_set,fea_idx,val)     
       key=tuple([label_name,val])            
       tree[key]=build_tree(new_data_set,label)
    return tree


def gen_major_label(data_set):
    data_set = np.array(data_set)
    label_array = data_set[:,-1]
    label_dic={}
    for num in data_set:
        if(label_dic.get(num[-1],0)==0):
            label_dic[num[-1]]=1
        else:
            label_dic[num[-1]]+=1
    curr_max=0 ;curr_max_label=0
    for key in label_dic.keys():
        if(label_dic[key]>curr_max):
            curr_max = label_dic[key]
            curr_max_label = key
    return curr_max_label

def load_data_set(file_name):   
    df = pd.read_csv(file_name,header=None)
    df['label']=pd.factorize(df[4])[0]
    del df[4] 
    #data_set = df.values
    data_set=[[1,1,1],[1,1,1],[1,0,0],[0,1,0],[0,1,0]]
    fea_name = ['fea0','fea1','fea2','fea3']

    return data_set,fea_name

def tree_test(model,x,label_name):
    fea_len = len(x)
    curr_model=model
    for i in range(fea_len):
        curr_fea=x[i]
        curr_key=tuple([label_name[i],curr_fea])
        curr_res = curr_model[curr_key]
        if(type(curr_res).__name__!='dict'):
            return curr_res
        else:
            curr_model=curr_res

测试程序：

import tree
model = tree.dec_tree('flower.csv')
model.train()
x=[1,1,1]
res=model.test(x)
print res