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
|
// Copyright (C) 2002 Ronan Collobert (collober@iro.umontreal.ca)
//
//
// This file is part of Torch. Release II.
// [The Ultimate Machine Learning Library]
//
// Torch is free software; you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation; either version 2 of the License, or
// (at your option) any later version.
//
// Torch is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU General Public License
// along with Torch; if not, write to the Free Software
// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
#include "Linear.h"
namespace Torch {
Linear::Linear(int n_inputs_, int n_outputs_)
{
addROption("weight decay", &weight_decay, 0, "weight decay", true);
n_inputs = n_inputs_;
n_outputs = n_outputs_;
}
int Linear::numberOfParams()
{
return((n_inputs+1)*n_outputs);
}
void Linear::init()
{
GradientMachine::init();
reset();
}
void Linear::reset()
{
real *params_ = (real *)params->ptr;
real bound = 1./sqrt((real)n_inputs);
for(int i = 0; i < n_params; i++)
params_[i] = bounded_uniform(-bound, bound);
}
void Linear::forward(List *inputs)
{
real *ptr_params = (real *)params->ptr;
real *ptr_outputs = (real *)outputs->ptr;
for(int s = 0; s < n_outputs; s++)
{
real out = 0;
List *inputs_ = inputs;
while(inputs_)
{
real *x = (real *)inputs_->ptr;
for(int j = 0; j < inputs_->n; j++)
out += *ptr_params++ * *x++;
inputs_ = inputs_->next;
}
out += *ptr_params++;
*ptr_outputs++ = out;
}
}
void Linear::backward(List *inputs, real *alpha)
{
real *beta_ptr = beta;
for(int k = 0; k < n_inputs; k++)
*beta_ptr++ = 0;
real *alpha_ptr = alpha;
real *params_ptr = (real *)params->ptr;
for(int j = 0; j < n_outputs; j++, alpha_ptr++)
{
beta_ptr = beta;
real z = *alpha_ptr;
for(int k = 0; k < n_inputs; k++)
*beta_ptr++ += z * *params_ptr++;
params_ptr++; // le bias
}
alpha_ptr = alpha;
real *der_params_ptr = (real *)der_params->ptr;
for(int j = 0; j < n_outputs; j++, alpha_ptr++)
{
real z = *alpha_ptr;
List *inputs_ = inputs;
while(inputs_)
{
real *x = (real *)inputs_->ptr;
for(int k = 0; k < inputs_->n; k++)
*der_params_ptr++ = z * *x++;
inputs_ = inputs_->next;
}
*der_params_ptr++ = z;
}
if(weight_decay != 0)
{
params_ptr = (real *)params->ptr;
der_params_ptr = (real *)der_params->ptr;
for(int i = 0; i < n_params; i++)
*der_params_ptr++ += weight_decay * *params_ptr++;
}
}
Linear::~Linear()
{
}
}
|
