forked from wangamama/TransferLearning
-
Notifications
You must be signed in to change notification settings - Fork 0
/
TrAdaboostkmm.py
181 lines (145 loc) · 6.5 KB
/
TrAdaboostkmm.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
import sklearn.svm
from sklearn.datasets import fetch_20newsgroups
from dataQuality.kmm import *
# ala=np.concatenate((trans_A, label_A.reshape(row_A,1)[:,-1:]), axis=1)
# s=np.concatenate((trans_S, label_S.reshape(row_S,1)[:,-1:]), axis=1)
# 初始化权重
# coef = kernel_mean_matching(s,ala,
# kern='rbf', B=10)
# code by chenchiwei
# -*- coding: UTF-8 -*-
import numpy as np
from sklearn import tree
from scipy import sparse
from sklearn import metrics
from sklearn import svm
# H 测试样本分类结果
# TrainS 原训练样本 np数组
# TrainA 辅助训练样本
# LabelS 原训练样本标签
# LabelA 辅助训练样本标签
# Test 测试样本
# N 迭代次数
from KMM import kmmClassification
def tradaboost(trans_S, trans_A, label_S, label_A, test,test_label, N,eliminate=False):
coef = kmmClassification.getBeta(trans_A,test.toarray(),49098)
#排除一些低权重的样本
if eliminate:
percenttile=np.percentile(coef, 7)
indexlist=[]
for index,x in enumerate(coef):
if(x<percenttile):
indexlist.append(index)
del coef[index]
trans_A = np.delete(trans_A, indexlist, axis=0)
label_A = np.delete(label_A, indexlist, axis=0)
print('排除的元素大小',len(indexlist))
# 排除一些低权重的样本
trans_data = sparse.vstack((trans_A, trans_S))
trans_label = np.concatenate((label_A, label_S), axis=0)
row_A = trans_A.shape[0]
row_S = trans_S.shape[0]
row_T = test.shape[0]
#print('目标源的大小',row_S,'辅助源的大小',row_A,'测试集的大小',row_T)
test_data = sparse.vstack((trans_data, test))
weights_A = coef
weights_A = np.asarray(weights_A).reshape(row_A,1)
total=sum(weights_A[:,0])
for j in range(row_A):
weights_A[j,0] = weights_A[j,0]/total
weights_S = np.ones([row_S, 1]) * np.max(weights_A)
# weights_S = np.ones([row_S, 1])/row_S
# weights_A = np.ones([row_A, 1])/row_A
weights = np.concatenate((weights_A, weights_S), axis=0)
bata = 1 / (1 + np.sqrt(2.0 * np.log(row_A/ N)))
# 存储每次迭代的标签和bata值?
bata_T = np.zeros([1, N])
result_label = np.ones([row_A + row_S + row_T, N])
predict = np.zeros([row_T])
# trans_data = np.asarray(trans_data, order='C')
# trans_label = np.asarray(trans_label, order='C')
# test_data = np.asarray(test_data, order='C')
# print(trans_data.shape)
# print(test_data.shape)
accuracy_scorelist=[]
f1_scorelist=[]
recall_scorelist=[]
for i in range(N):
P = calculate_P(weights, trans_label)
result_label[:, i] = train_classify(trans_data, trans_label,
test_data, P)
error_rate = 0.0
for j in range(row_A, row_A + row_S):
error_rate += (weights[j] * abs(result_label[j, i] - trans_label[j]))
error_rate = error_rate / sum(weights[row_A:])
#error_rate = calculate_error_rate(label_S, result_label[row_A:row_A + row_S, i],
# weights[row_A:row_A + row_S, :])
#print ('Error rate:', error_rate)
# if error_rate != 1:
# bata_T[0, i] = error_rate / (1.0 - error_rate)
# if error_rate >= 0.5 and error_rate != 1:
# bata_T[0, i] = 0.45 / (0.51)
# if error_rate == 1:
# bata_T[0, i] = 0.4
if error_rate >= 0.5:
# error_rate = 0.5
error_rate = 0.499;
if error_rate == 0:
# error_rate = 0.000001
# error_rate=0.0001
error_rate = 0.001
bata_T[0, i] = error_rate / (1 - error_rate)
# 调整源域样本权重
for j in range(row_S):
weights[row_A + j] = weights[row_A + j] * np.power(bata_T[0, i],
(-np.abs(result_label[row_A + j, i] - trans_label[row_A+j])))
# 调整辅域样本权重
for j in range(row_A):
weights[j] = weights[j] * np.power(bata, np.abs(result_label[j, i] - trans_label[j]))
##每次迭代完成计算下在测试集合上的误差
# predic_temp=np.zeros([row_T])
# iteration=i+1;
# for j in range(row_T):
# left = np.sum(
# result_label[row_A + row_S + j, 0:iteration] * np.log(1 / bata_T[0, 0:iteration]))
# right = 0.5 * np.sum(np.log(1 / bata_T[0, 0:iteration]))
# # left = np.sum(
# # result_label[row_A + row_S + i, int(np.ceil(iteration / 2)):iteration] * np.log(
# # 1 / bata_T[0, int(np.ceil(iteration / 2)):iteration]))
# # right = 0.5 * np.sum(np.log(1 / bata_T[0, int(np.ceil(iteration / 2)):iteration]))
# if left >= right:
# predic_temp[j] = 1
# else:
# predic_temp[j] = 0
# accuracy_scorelist.append(metrics.accuracy_score(test_label, predic_temp))
# recall_scorelist.append(metrics.recall_score(test_label, predic_temp))
# f1_scorelist.append(metrics.f1_score(test_label, predic_temp))
# print bata_T
for i in range(row_T):
# 跳过训练数据的标签
# left = np.sum(
# result_label[row_A + row_S + i, int(np.ceil(N / 2)):N] * np.log(1 / bata_T[0, int(np.ceil(N / 2)):N]))
# right = 0.5 * np.sum(np.log(1 / bata_T[0, int(np.ceil(N / 2)):N]))
left = np.sum(
result_label[row_A + row_S + i, 0:N] * np.log(1 / bata_T[0, 0:N]))
right = 0.5 * np.sum(np.log(1 / bata_T[0, 0:N]))
if left >= right:
predict[i] = 1
else:
predict[i] = 0
# print left, right, predict[i]
# predict[i]=left-right;
print(accuracy_scorelist)
return predict,accuracy_scorelist,recall_scorelist,f1_scorelist
def calculate_P(weights, label):
total = np.sum(weights)
return np.asarray(weights)/total
from sklearn.linear_model import LogisticRegression
def train_classify(trans_data, trans_label, test_data, P):
clf = LogisticRegression()
clf.fit(trans_data, trans_label, sample_weight=P[:, 0])
return clf.predict(test_data)
# def calculate_error_rate(label_R, label_H, weight):
# total = np.sum(weight)
# #return np.sum((weight[:, 0] / total)* np.abs(label_R - label_H))
# return return_correct_rate(label_R,label_H)