机器学习-决策树模型-西瓜书代码(C4.5)-预剪枝修正
以下代码是本人在学习西瓜书时花费两个礼拜根据原理进行原创，若需转载请咨询本人，谢谢！
自我研究模拟代码
附上离散类别运行截图
c45_config.py
"""
Filename: kdd_config
Author: kdd_zyx
Description: 机器学习 - C4.5(连续值) - 剪枝
Start: 18.10.1
End:
"""
import time as t
import numpy as np
import pandas as pd
import random as r
import cProfile as cPf
text_tnum = 791 # 抽取数据量
kddt_tnum = 1 # KDD测试量
text_pret = 0.225 # 测试集百分比(计算需要排除属性集)
let = 0 # 小于
got = 1 # 大于
start = 1 # 数据上标
end = -1 # 结尾下标
def divi(x, y):
# 除法异常捕捉
if y == 0:
return 0
else:
return x / y
def log2(x):
# log函数异常捕捉
if x == 0:
return 0
else:
return np.log2(x)
def tw_copy(old_arrs):
# 避免deepcopy
new_arrs = []
for arr in old_arrs:
new_arrs.append(arr.copy())
return new_arrs
def flat(datas):
re_datas = []
for data in datas:
if data == datas[end]:
re_datas.append(data)
return re_datas
else:
re_datas.append(float(data))
def radf(text_addr, text_splt):
Data = pd.read_csv('D:\\BaiduNetdiskDownload\\dc\\train.csv')
Data = Data.values
dataset = np.delete(Data, 0, axis=1)
dataset[0] = np.arange(6001)
# with open(text_addr, 'r', enco
# with open(text_addr, 'r', encoding='UTF-8') as f:
# dataset = [eachLine.replace('\n', '').split(text_splt) for eachLine in f]
# f.close()
return dataset.tolist()
def writ(NUM, result):
try:
with open('kdd{}.txt'.format(NUM), 'a+', encoding='UTF-8') as f:
f.write(result)
f.close()
except Exception as e:
print('Error：', e)
writ(result)
# 引入训练集、验证集、测试集
def Lead_dataset():
try:
dataset = radf('c45_csv.txt', ',')
fin_dataset = []
pro_dataset = dataset[0]
txt_dataset = dataset[start : text_tnum]
r.shuffle(txt_dataset)
fin_dataset.append(pro_dataset)
for data in txt_dataset:
fin_dataset.append(data)
kdd_dataset = dataset[text_tnum:]
for data in kdd_dataset:
fin_dataset.append(data)
return fin_dataset
except Exception as e:
print('Error：', e)
c45.py
"""
Filename: kdd_check_one
Author: kdd_zyx
Description: 机器学习 - C4.5(连续值) - 预剪枝
Datas：kdd - 随机划分
Start: 18.10.1
End: 18.10.7
"""
from c45_config import *
class Decision_tree(object):
# 初始化决策树对象
def __init__(self):
self.tra_dataset = []
self.pro_dataset = []
self.end_dataset = []
self.tex_dataset = []
# 引入类别集
def Lead_attrset(self):
end_data = []
for data in self.tra_dataset:
end_data.append(data[end])
return list(set(end_data))
# 构建参数集
def Main_data(self):
dataset = Lead_dataset()
self.tra_num = len(dataset)
if text_pret != 0:
self.tex_num = int(len(dataset[start: ]) * text_pret + 1)
self.tra_dataset = dataset[start : -self.tex_num]
self.tex_dataset = dataset[-self.tex_num: ]
self.pro_dataset = dataset[0]
self.end_dataset = self.Lead_attrset()
else:
pass
# 构建决策树模型
def TreeCreate(self):
self.C45_Tree = []
self.Main_data()
self.TreeGenerate([], self.tra_dataset, self.pro_dataset)
self.Tree_gui()
self.Tree_verify(True)
# 构建决策树枝叶
def TreeGenerate(self, limb, tra_dataset, pro_dataset):
print(len(tra_dataset))
# situation one
condition_one = True
for data in tra_dataset:
if data[end] != tra_dataset[0][end]:
condition_one = False
break
if condition_one:
limb.append(data[end])
self.C45_Tree.append(limb)
return
# situation two
condition_two = True
for data in tra_dataset:
if data[:end] != tra_dataset[0][:end]:
condition_two = False
break
if condition_two or not pro_dataset:
Pk = self.Calc_Max(tra_dataset)
limb.append(self.end_dataset[Pk.index(max(Pk))])
self.C45_Tree.append(limb)
return
# 预剪枝前
Pk = self.Calc_Max(tra_dataset)
limb.append(self.end_dataset[Pk.index(max(Pk))])
self.C45_Tree.append(limb.copy())
accury_font = self.Tree_verify()
Font_Tree = tw_copy(self.C45_Tree)
del limb[end]
del self.C45_Tree[end]
# situation three
tra_data, tra_pro, tra_Ta = self.Calc_InfoGR(tra_dataset, pro_dataset)
limb.append(tra_pro)
limb.append(str(round(tra_Ta, 2)))
comp_copy = {let: '>', got: '<'}
# 预剪枝后
for data in tra_data:
limb.append(comp_copy[tra_data.index(data)])
Pk = self.Calc_Max(data)
limb.append(self.end_dataset[Pk.index(max(Pk))])
self.C45_Tree.append(limb.copy())
del limb[-2: ]
accury_back = self.Tree_verify()
del self.C45_Tree[-2: ]
# 判断剪枝结果
if accury_font >= accury_back:
self.C45_Tree = Font_Tree
return
# 进入非剪枝递归
for data in tra_data:
limb.append(comp_copy[tra_data.index(data)])
self.TreeGenerate(limb.copy(), data, pro_dataset)
del limb[end]
# 模拟树图形结构
def Tree_gui(self):
Go_Tr = '~~~~~'
for num in self.C45_Tree[0]:
print(Go_Tr + str(num), end='')
print()
for tree in self.C45_Tree[start: ]:
num_TF = False
for num in range(len(tree)):
try:
if tree[num] != self.C45_Tree[self.C45_Tree.index(tree) - 1][num] or num_TF:
num_TF = True
print(Go_Tr + str(tree[num]), end='')
else:
print(' ' * len(Go_Tr) + ' ' * len(tree[num]), end='')
except:
print(Go_Tr + str(tree[num]), end='')
print()
print()
# 验证模型
def Tree_verify(self, Tree = False):
verify_num = 0
for data in self.tex_dataset:
for tree in self.C45_Tree:
data_verify = True
for tree_num in range(2, len(tree[ :end]), 3):
pro_num = self.pro_dataset.index(tree[tree_num - 2])
if tree[tree_num] == '>' and float(tree[tree_num - 1]) < data[pro_num]:
data_verify = False
break
elif tree[tree_num] == '<' and float(tree[tree_num - 1]) > data[pro_num]:
data_verify = False
break
if data_verify and tree[end] == data[end]:
verify_num += 1
elif Tree and data_verify and tree[end] != data[end] and data[end] == 'No':
writ(1, tree[end] + '\n')
accury = round(divi(verify_num, self.tex_num - kddt_tnum), 3)
if Tree == True:
print('Training dataset number:', self.tra_num - self.tex_num - 1)
print('Texting dataset number:', self.tex_num - kddt_tnum)
print('The correct number:', verify_num)
print('Model node number:', len(self.C45_Tree))
print('Accuracy:', accury)
return accury
# 信息增益率
def Calc_InfoGR(self, tra_dataset, pro_dataset):
All_ta = []
All_dt = []
All_gain = []
All_ratio = []
for pro in pro_dataset[:end]:
pro_gain, pro_dt, pro_ta = self.Calc_InfoG(tra_dataset, pro)
# 增益率计算
Iv = 0.0
for dt in pro_dt:
iv_num = len(dt)
value = divi(iv_num, len(tra_dataset))
Iv -= (value * log2(value))
pro_ratio = divi(pro_gain, Iv)
All_ta.append(pro_ta)
All_dt.append(pro_dt)
All_gain.append(pro_gain)
All_ratio.append(pro_ratio)
Max_num = 0
Max_ratio = 0.0
All_gain = np.array(All_gain)
aver_gain = All_gain.mean()
for num in range(len(All_gain)):
if All_ratio[num] > Max_ratio and All_gain[num] > aver_gain:
Max_num = num
high_gain = All_gain[Max_num]
Max_ratio = All_ratio[Max_num]
elif All_ratio[num] == Max_ratio and All_gain[num] > high_gain and All_gain[num] > aver_gain:
Max_num = num
high_gain = All_gain[Max_num]
Max_ratio = All_ratio[Max_num]
return All_dt[Max_num], pro_dataset[Max_num], All_ta[Max_num]
# 信息增益
def Calc_InfoG(self, tra_dataset, pro):
Et_Pk = self.Calc_Max(tra_dataset)
Ent_D = self.Calc_Info(Pk=Et_Pk)
pro_indx = self.pro_dataset.index(pro)
tra_dataset = sorted(tra_dataset, key=lambda data:data[pro_indx])
Pk = [[0] * len(self.end_dataset), Et_Pk]
max_num = 0
max_gain_d = 0.0
for data in tra_dataset:
Pk[let][self.end_dataset.index(data[end])] += 1
Pk[got][self.end_dataset.index(data[end])] -= 1
gain_d = Ent_D
num = sum(Pk[let])
for pk in Pk:
ent_d = self.Calc_Info(Pk=pk)
gain_d -= divi(ent_d * sum(pk), len(tra_dataset))
if gain_d > max_gain_d:
max_num = num
max_gain_d = gain_d
max_dt = [tra_dataset[ :max_num], tra_dataset[max_num: ]]
max_ta = divi(max_dt[let][-1][pro_indx] + max_dt[got][0][pro_indx], 2)
return max_gain_d, max_dt, max_ta
# 信息熵
def Calc_Info(self, tra_dataset = None, Pk = None):
if not Pk:
data_num = len(tra_dataset)
Pk = self.Calc_Max(tra_dataset)
else:
data_num = sum(Pk)
Ent_D = 0.0
for pk_value in Pk:
value = divi(pk_value, data_num)
Ent_D -= (value * log2(value))
return Ent_D
# 信息标记分类
def Calc_Max(self, tra_dataset):
Pk = [0] * len(self.end_dataset)
datas = []
for data in tra_dataset:
datas.append(data[end])
for num in range(len(Pk)):
pk_num = datas.count(self.end_dataset[num])
Pk[num] = pk_num
return Pk
if __name__ == '__main__':
start_time = t.time()
tree = Decision_tree()
tree.TreeCreate()
end_time = t.time()
print('Run time:', end_time - start_time)