-
Notifications
You must be signed in to change notification settings - Fork 2
/
BCSH.py
134 lines (105 loc) · 4.07 KB
/
BCSH.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
# -*- coding: utf-8 -*-
"""
@author: yingwenjie
"""
import scipy
import numpy as np
from utils import *
def BCSH1(weight, bits=64, iters=10, lambd=0.1):
"""
文档哈希码B的取值等于{0, 1} 词表中所有词的概率连乘，包括出现、不出现的概率
"""
X = weight.transpose()
X[X > 0] = 1 ##tfidf 0,1化
X1 = X.copy()
X0 = 1 - X
X1 = normalize(X1)
X0 = normalize(X0)
[n, m] = X.shape
B = np.random.randint(0, 2, (bits, m))
B[B == 0] = -1
for i in range(iters):
tempB1 = B.copy()
tempB0 = -B
tempB1[tempB1 < 0] = 0
tempB0[tempB0 < 0] = 0
PX1_B1 = (np.dot(tempB1, X.transpose()) + 1) / (m + 2)
PX1_B0 = (np.dot(tempB0, X.transpose()) + 1) / (m + 2)
PB1 = np.sum(tempB1, 1, keepdims=True) / m
PX0_B1 = 1 - PX1_B1
PX0_B0 = 1 - PX1_B0
PB0 = 1 - PB1
logPX1_B1 = np.log2(PX1_B1)
logPX1_B0 = np.log2(PX1_B0)
logPX0_B1 = np.log2(PX0_B1)
logPX0_B0 = np.log2(PX0_B0)
logPB1 = np.dot(logPX1_B1, X1) + np.dot(logPX0_B1, X0)
logPB0 = np.dot(logPX1_B0, X1) + np.dot(logPX0_B0, X0)
tmp = logPB1 - logPB0
tmp[tmp > 32] = 32
PXB1 = np.power(2, tmp)
PXB1 = PXB1 / (1 + PXB1)
Fx = PXB1 * 2 -1
Y = Update(B, bits)
old_B = B.copy()
for i in range(bits):
for j in range(m):
if((np.power((1 - Fx[i, j]), 2) + lambd * np.power((1 - Y[i, j]), 2)) <= (np.power((-1 - Fx[i, j]), 2) + lambd * np.power((-1 - Y[i, j]), 2))):
B[i, j] = 1
else:
B[i, j] = -1
updateB = sum(sum(B != old_B))
print('update-------------------')
print(updateB)
scipy.io.savemat('./argfile/arg.mat', {'B': B, 'logPX1_B1': logPX1_B1, 'logPX1_B0': logPX1_B0, 'logPX0_B1': logPX0_B1, 'logPX0_B0': logPX0_B0})
return B
def BCSH2(weight, bits=64, iters=10, lambd=0.1):
"""
文档哈希码B的取值等于{0, 1} 文档中出现词的概率连乘
"""
X = weight.transpose()
X = normalize(X)
[n, m] = X.shape
print(n, m)
B = np.random.randint(0, 2, (bits, m))
B[B == 0] = -1
for it in range(iters):
tempB1 = B.copy()
tempB0 = -B
tempB1[tempB1 < 0] = 0
tempB0[tempB0 < 0] = 0
PX1_B1 = np.dot(tempB1, X.transpose())
ALL1 = np.sum(PX1_B1, 1, keepdims=True)
for r in range(bits):
PX1_B1[r, :] = (PX1_B1[r, :] + 1) / (ALL1[r, 0] + n)
PX1_B0 = np.dot(tempB0,X.transpose())
ALL0 = np.sum(PX1_B0, 1, keepdims=True)
for r in range(bits):
PX1_B0[r, :] = (PX1_B0[r, :] + 1) / (ALL0[r, 0] + n)
PB1 = np.sum(tempB1, 1, keepdims=True) / m
logPX1_B1 = np.log2(PX1_B1)
logPX1_B0 = np.log2(PX1_B0)
logPB1 = np.dot(logPX1_B1, X)
logPB0 = np.dot(logPX1_B0, X)
tmp = (logPB1 - logPB0) ### 规范化很重要，特征进行规范化
tmp[tmp > 32] = 32
PXB1 = np.power(2, tmp)
PXB1 = PXB1 / (1 + PXB1)
Fx = PXB1 * 2 -1
Y = Update(B, bits)
old_B = B.copy()
for i in range(bits):
for j in range(m):
if((np.power((1 - Fx[i, j]), 2) + lambd * np.power((1 - Y[i, j]), 2)) <= (np.power((-1 - Fx[i, j]), 2) + lambd * np.power((-1 - Y[i, j]), 2))):
B[i, j] = 1
else:
B[i, j] = -1
updateB = sum(sum(B != old_B))
print('update-----------------' + str(lambd))
print(updateB)
loss1 = np.trace(np.dot((B - Fx), (B - Fx).transpose()))
loss2 = np.trace(np.dot((B - Y), (B - Y).transpose()))
Loss = loss1 + lambd * loss2
print('Loss=' + str(Loss))
scipy.io.savemat('./argfile/arg.mat', {'B': B, 'logPX1_B1': logPX1_B1, 'logPX1_B0': logPX1_B0})
return B