-
Notifications
You must be signed in to change notification settings - Fork 0
/
tree_improve.py
211 lines (179 loc) · 5.84 KB
/
tree_improve.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
import csv
import sys
import numpy as np
import copy
import collections
from collections import Counter
import random
import math
attr=["fixed acidity", "volatile acidity", "citric acid", "residual sugar", "chlorides", "free sulfur dioxide", "total sulfur dioxide","density", "pH", "sulphates", "alcohol"]
training=[]
# Implement your decision tree below
class DecisionTree():
def get_gini(self, llist):
n=llist.size
count={}
for curlabel in llist:
if curlabel not in count.keys():
count[curlabel]=0
count[curlabel]=count[curlabel]+1
ss=0
for i in range (0, len(llist)):
ss=ss+llist[i]
gini_impurity=0
if (ss!=0):
for j in range (0, len(llist)):
if(i!=j):
gini_impurity=gini_impurity+float(llist[i]/ss)*float(llist[j]/ss);
return gini_impurity
def get_id(self, j, llist):
for i in range(0, len(llist)):
if j==llist[i]:
return i
return -1
def ig_calculation(self, data, attr_list, t_attr):
b_id=-1
min_entr=1000
labels=[]
for i in range(0, len(data)):
if data[i][t_attr] not in labels:
labels.append(data[i][t_attr])
for i in range(0, len(attr_list)):
values=[]
checked=[]
for j in range(0, len(data)):
if data[j][i] not in checked:
checked.append(data[j][i])
values.append(data[j][i])
if ((1.0*len(labels)/len(values))<0.008):
continue;
valueLabels=np.zeros((len(values), len(labels)))
for k in range(0, len(data)):
rid=self.get_id(data[k][i], values)
if (rid!=-1):
cid=self.get_id(data[k][t_attr], labels)
if ((rid!=-1) and (cid!=-1)):
valueLabels[rid,cid]=valueLabels[rid,cid]+1
entrs=np.zeros((1,valueLabels.shape[0]))
weightings=np.zeros((1,valueLabels.shape[0]))
for m in range(0, valueLabels.shape[0]):
entrs[0,m]=self.get_gini(valueLabels[m])
ss=0
for k in range(0, valueLabels.shape[0]):
ss=ss+sum(valueLabels[k])
for l in range(0, valueLabels.shape[0]):
weightings[0, l]=float(sum(valueLabels[l])/ss)
sum_entr=0
for m in range(0, valueLabels.shape[0]):
sum_entr=sum_entr+weightings[0,m]*entrs[0,m]
print "Gini impurity of",attr[attr_list[i]],"= %.4f" % sum_entr
if (sum_entr<min_entr):
b_id=i
min_entr=sum_entr
attribIndex=attr_list.pop(b_id)
toStop=False
return attribIndex,toStop;
tree={}
defaultLabel="NULL"
def buildtree(self, data, attr_list, t_attr, ts_length):
classCounts=Counter([instance[t_attr] for instance in data])
default=classCounts.most_common(1)[0][0]
labels=[]
for i in range(0, len(data)):
labels.append(data[i][t_attr])
if ((len(data)*1.0)/ts_length<=0.02):
return default
if (data is None) or len(attr_list)<=0:
return default
if (labels.count(labels[0])==len(labels)):
return labels[0]
toStop=False
if (len(attr_list)>-1):
b_id_attr, toStop=self.ig_calculation(data, attr_list, t_attr)
print "Best attribute:", attr[b_id_attr]
else:
random.shuffle(attr_list)
b_id_attr=attr_list.pop(-1)
tree={b_id_attr:{}}
unique=[]
checked=[]
for i in range(0, len(data)):
if data[i][b_id_attr] not in checked:
checked.append(data[i][b_id_attr])
unique.append(data[i][b_id_attr])
print "#b=",len(unique)
for value in unique:
dataSubset=[]
for i in range(0, len(data)):
if (data[i][b_id_attr]==value):
dataSubset.append(data[i])
subTree=self.buildtree(dataSubset, attr_list, t_attr, ts_length)
tree[b_id_attr][value]=subTree
return tree
def cf(self, tree, instance, max, defaultLabel=None):
if (tree is None):
return None
if (not isinstance (tree, dict)):
return tree
root=tree.keys()[0]
subTrees=tree.values()[0]
branch=len(tree.values()[0])
if (branch>max[0]):max[0]=branch
if instance[root] not in subTrees:
return None
return self.cf(subTrees[instance[root]], instance, max)
def classify(self, test_instance):
max=[0]
result=self.cf(self.tree, test_instance, max)
if result is None:
result=self.defaultLabel
return result
def implement():
with open("hw4-data.csv") as ff:
data=[tuple(line) for line in csv.reader(ff, delimiter=",")]
#print data
ss=0
for j in range (0,10):
training=[x for i, x in enumerate(data) if i % 10!=j]
test=[x for i, x in enumerate(data) if i % 10==j]
tree=DecisionTree()
attribs=range(0,(len(training[0])-1))
t_attrib=len(training[0])-1
exattrs=[]
classCounts=Counter([instance[t_attrib] for instance in training])
tree.defaultLabel=classCounts.most_common(1)[0][0]
labels=[]
for k in range(0, len(training)):
if training[k][t_attrib] not in labels:
labels.append(data[k][t_attrib])
for k in range(0, len(attribs)):
values=[]
checked=[]
for l in range(0, len(training)):
if training[l][k] not in checked:
checked.append(data[l][k])
values.append(data[l][k])
if ( (1.0*len(labels)/len(values))<0.001):
exattrs.append(attribs[k])
notexattrs=[]
for i in range(0,len(attribs)):
if(attribs[i] not in exattrs):
notexattrs.append(attribs[i])
tree.tree=tree.buildtree(training, notexattrs, t_attrib, len(training))
results=[]
for instance in test:
result=tree.classify(instance[:-1])
results.append(result==instance[-1])
# Accuracy
accuracy=float(results.count(True))/float(len(results))
ss += accuracy
print "cv=",j," accuracy: %.4f" % accuracy
print "\n"
accuracy=ss/10
print "The accuracy is %.4f" % accuracy
# Writing results to a file (DO NOT CHANGE)
f=open("result.txt", "w")
f.write("accuracy: %.4f" % accuracy)
f.close()
if __name__=="__main__":
implement()