-
Notifications
You must be signed in to change notification settings - Fork 0
/
C45.py
263 lines (223 loc) · 9.01 KB
/
C45.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
# coding: utf-8
import sys
from collections import defaultdict
from operator import itemgetter
import numpy as np
from utils import (
read_data,
cal_set_info,
get_disc_val,
binary_sp,
cal_gain_ratio,
check_purity,
check_accurcy,
get_err_sum,
get_cls_from_data,
fcv
)
class TreeNode(object):
def __init__(self,dataset):
self.cls = 0.0 # class of the leaf node
self.childNode = {} # type also TreeNode, 属性值: TreeNode
self.dataset = dataset
self.attr_type = 0 # 1 for categorical, 0 for numerical
self.attr_index = -1 # 分裂数据集所使用属性的序号
self.demark = 0.0 # 如果是连续属性,分界点
class DecisionTree(object):
def __init__(self, dataset, attrset, disc_type):
self.dataset = dataset
self.attrset = attrset
self.disc_type = disc_type
self.root = TreeNode(dataset)
self.construct_tree()
def __construct_tree(self, cur_node, attr_list):
'''
递归构建决策树
'''
data = cur_node.dataset
data_classified = {}
max_gain_ratio, index = sys.float_info.min, -1
num_border = 0.0
for idx in attr_list:
if idx in self.disc_type: #离散属性
gain_r = self.disc_gain_rt(idx, data)
else: #数值属性
gain_r, num_border = self.num_gain_rt(idx,data)
if gain_r > max_gain_ratio:
max_gain_ratio = gain_r
index = idx
if index == -1: #所有属性都不能满足条件
cur_node.cls = get_cls_from_data(data)
return
cur_node.attr_index = index
if index in self.disc_type: #离散属性
cur_node.attr_type = 1
#对数据进行分类
for val in self.disc_type[index]:
data_classified[val] = []
for d in data:
data_classified[d[index]].append(d)
else: #连续属性
cur_node.attr_type = 0
cur_node.demark = num_border
data_classified[0] = []
data_classified[1] = []
for d in data:
if d[index] < num_border:
data_classified[0].append(d)
else:
data_classified[1].append(d)
if len(attr_list) == 1: #下一次递归属性集为空
for k, v in data_classified.items():
child_node = TreeNode(v)
#属性值对应的数据集为空,则使用当前节点的数据集判断节点对应的分类
if len(v) == 0:
child_node.cls = get_cls_from_data(data)
else:
child_node.cls = get_cls_from_data(v)
cur_node.childNode[k] = child_node
else:
sub_attr = list(attr_list)
sub_attr.remove(index)
for k, v in data_classified.items():
child_node = TreeNode(v)
if len(v) == 0:
child_node.cls = get_cls_from_data(data)
elif check_purity(v) == 1:
child_node.cls = v[0][-1] #随便取一个sample的标签
else:
self.__construct_tree(child_node, sub_attr) #对子节点进行递归
cur_node.childNode[k] = child_node
def construct_tree(self):
'''
决策树递归构建entrance
'''
#init_attr_list = range(len(self.dataset[0]))
self.__construct_tree(self.root, self.attrset)
def disc_gain_rt(self, index, data):
'''
计算一个属性的信息增益
'''
statisc_dict = {}
index_val = self.disc_type[index]
total_info = cal_set_info(data)
for val in index_val:
statisc_dict[val] = {}
for d in data:
if d[-1] in statisc_dict[d[index]]:
statisc_dict[d[index]][d[-1]] += 1
else:
statisc_dict[d[index]][d[-1]] = 1
'''
statisc_dict结构:
{
attr_value1:{yes:num1, no:num2},
...
attr_value2:{yes:num1, no:num2},
}
'''
info_gain, info_measure = cal_gain_ratio(statisc_dict, data)
return -1 if info_measure == -1 else info_gain / info_measure
def num_gain_rt(self, index, data):
'''
连续数值属性计算信息增益,先根据第index列排序,选取标签改变时对应的index列属性值,
作为分界点,分别计算出每个分界点对应的信息增益,返回最大增益及其对应的分界点
'''
ctgs = set()
sorted_data = sorted(data, key=itemgetter(index))
cls = sorted_data[0][-1]
#只选取便签改变时对应的属性值
for d in sorted_data:
if d[-1] != cls:
cls = d[-1]
ctgs.add(d[index])
max_gain, border, gain_ratio = sys.float_info.min, 0.0, -1.0
for ctg in ctgs:
statisc_dict = {}
info_gain = 0.0
'''
结构为
{
'left': {yes: num1, no:num2}
'right': {yes:num1, no:num2}
}
'''
statisc_dict['left'], statisc_dict['right'] = binary_sp(data, ctg, index)
info_gain, info_measure = cal_gain_ratio(statisc_dict, data)
if info_measure == -1:
continue
if info_gain > max_gain:
max_gain, border, gain_ratio = info_gain, ctg, info_gain / info_measure
return gain_ratio, border
def classify(self,dataset):
'''对给定的一个数据集中的数据进行分类'''
predict_cls = []
for d in dataset:
predict_cls.append(self.__classify_data(d, self.root))
return predict_cls
def __classify_data(self, data, cur_node):
'''递归分类过程'''
if len(cur_node.childNode) == 0:
return cur_node.cls
else:
criteria_val = data[cur_node.attr_index]
if cur_node.attr_type == 1: #离散属性
next_node = cur_node.childNode[criteria_val]
return self.__classify_data(data, next_node)
else:
if criteria_val < cur_node.demark:
next_node = cur_node.childNode[0]
else:
next_node = cur_node.childNode[1]
return self.__classify_data(data, next_node)
def leaf_err_sum(self, cur_node, err_set):
'''
悲观剪枝,用于计算一个当前节点子树的错误率
err_num: 当一个叶子节点数据集为空时,错误节点数目就是父节点的错误节点数
'''
if len(cur_node.childNode) == 0: #叶子节点
if len(cur_node.dataset) == 0:
err_set.append(0)
else:
err_sum = get_err_sum(cur_node.cls, cur_node.dataset)
err_set.append(err_sum)
else: # 内部节点
for _, c in cur_node.childNode.items():
if len(c.childNode) == 0 and len(c.dataset) == 0:
self.leaf_err_sum(c, err_set)
else:
self.leaf_err_sum(c, err_set)
def __prun_tree(self, cur_node):
'''剪枝'''
if len(cur_node.childNode) == 0: #叶子节点直接跳过
return
else:
cur_node.cls = get_cls_from_data(cur_node.dataset)
cur_err_sum = get_err_sum(cur_node.cls, cur_node.dataset) + 0.5
leaf_err_set = []
self.leaf_err_sum(cur_node, leaf_err_set)
leaf_e_sum = sum(leaf_err_set) + 0.5 * len(leaf_err_set)
leaf_err_ratio = leaf_e_sum / len(cur_node.dataset)
std_dev = np.sqrt(leaf_err_ratio * (1 - leaf_err_ratio))
if leaf_e_sum + std_dev > cur_err_sum:
print leaf_e_sum + std_dev, cur_err_sum, " prun!!!!"
cur_node.childNode = {}
cur_node.cls = get_cls_from_data(cur_node.dataset)
else:
for _, c in cur_node.childNode.items():
self.__prun_tree(c)
def prun_tree(self):
self.__prun_tree(self.root)
if __name__ == '__main__':
#dataset = read_data("test.txt")
#dataset = read_data("breast-cancer-assignment5.txt")
dataset = read_data("german-assignment5.txt")
attr_set = range(len(dataset[0]))
DiscType = get_disc_val(dataset)
decisin_tree = DecisionTree(dataset[1:],attr_set, DiscType)
#decisin_tree.prun_tree()
res_cls = decisin_tree.classify(dataset[1:])
#res_cls = decisin_tree.classify(dataset[1:])
#print res_cls
acc = check_accurcy(dataset[1:], res_cls)
print acc