-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathCost.h
142 lines (82 loc) · 3.25 KB
/
Cost.h
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
#ifndef COST_H
#define COST_H
#include "Tensor.h"
#include <cmath>
class Cost {
public:
Cost() { }
char type;
virtual double evaluate( Tensor out, Tensor target ) { }
virtual double evaluate( Tensor out, Tensor target, Tensor & delta ) { }
};
class MSE : public Cost {
public:
MSE() { type = 'm'; }
double evaluate( Tensor out, Tensor target ) {
double cost = 0.0;
for (int d = 0; d < out.getDim(); d++)
for (int r = 0; r < out.getRows(); r++)
for (int c = 0; c < out.getCols(); c++)
cost += pow(out(d, r, c) - target(d, r, c), 2);
cost /= (2 * out.getDim() * out.getRows() * out.getCols());
return cost;
}
double evaluate( Tensor out, Tensor target, Tensor & delta ) {
double cost = 0.0;
for (int d = 0; d < out.getDim(); d++)
for (int r = 0; r < out.getRows(); r++)
for (int c = 0; c < out.getCols(); c++) {
delta(d, r, c) = (out(d, r, c) - target(d, r, c)) / (out.getDim() * out.getRows() * out.getCols());
cost += pow(out(d, r, c) - target(d, r, c), 2);
}
cost /= (2 * out.getDim() * out.getRows() * out.getCols());
return cost;
}
};
class CrossEntropy : public Cost {
public:
CrossEntropy() { type = 'c'; }
double evaluate( Tensor out, Tensor target ) {
double cost = 0.0;
for (int d = 0; d < out.getDim(); d++)
for (int r = 0; r < out.getRows(); r++)
for (int c = 0; c < out.getCols(); c++)
cost += target(d, r, c) * log(out(d, r, c)) + (1 - target(d, r, c)) * log(1 - out(d, r, c));
cost /= -(out.getDim() * out.getRows() * out.getCols());
return cost;
}
double evaluate( Tensor out, Tensor target, Tensor & delta ) {
double cost = 0.0;
for (int d = 0; d < out.getDim(); d++)
for (int r = 0; r < out.getRows(); r++)
for (int c = 0; c < out.getCols(); c++) {
delta(d, r, c) = -( (target(d, r, c) - out(d, r, c)) / ( out(d, r, c) * ( 1 - out(d, r, c) ) ) ) / (out.getDim() * out.getRows() * out.getCols());
cost += target(d, r, c) * log(out(d, r, c)) + (1 - target(d, r, c)) * log(1 - out(d, r, c));
}
cost /= -(out.getDim() * out.getRows() * out.getCols());
return cost;
}
};
class KLDivergence : public Cost {
public:
KLDivergence() { type = 'k'; }
double evaluate( Tensor out, Tensor target ) {
double cost = 0.0;
for (int d = 0; d < out.getDim(); d++)
for (int r = 0; r < out.getRows(); r++)
for (int c = 0; c < out.getCols(); c++)
cost += target(d, r, c) * log(target(d, r, c) / out(d, r, c));
return cost;
}
double evaluate( Tensor out, Tensor target, Tensor & delta ) {
double cost = 0.0;
for (int d = 0; d < out.getDim(); d++)
for (int r = 0; r < out.getRows(); r++)
for (int c = 0; c < out.getCols(); c++) {
delta(d, r, c) = -target(d, r, c) / out(d, r, c);
cost += target(d, r, c) * log(target(d, r, c) / out(d, r, c));
}
return cost;
}
};
#endif