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
|
/**
\file lbfgs_method.hpp
\brief large bfgs method, used to perform large scale optimization
\author Junhua Gu
*/
#ifndef LBFGS_METHOD
#define LBFGS_METHOD
#define OPT_HEADER
#include <core/optimizer.hpp>
//#include <blitz/array.h>
#include <limits>
#include <cstdlib>
#include <core/opt_traits.hpp>
#include "../linmin/linmin.hpp"
#include <math/num_diff.hpp>
#include <cassert>
#include <cmath>
#include <ctime>
#include <vector>
#include <algorithm>
#include "lbfgs.h"
#include "lbfgs.cpp"
/*
*
*/
#include <iostream>
using std::cerr;
using std::endl;
namespace opt_utilities
{
template<typename rT,typename pT>
lbfgsfloatval_t lbfgs_adapter(
void *instance,
const lbfgsfloatval_t *x,
lbfgsfloatval_t *g,
const int n,
const lbfgsfloatval_t step
)
{
pT px;
resize(px,n);
for(int i=0;i<n;++i)
{
set_element(px,i,x[i]);
}
lbfgsfloatval_t result=((func_obj<rT,pT>*)instance)->eval(px);
pT grad(gradient(*static_cast<func_obj<rT,pT>*>(instance),px));
for(int i=0;i<n;++i)
{
g[i]=get_element(grad,i);
}
return result;
}
template <typename rT,typename pT>
class lbfgs_method
:public opt_method<rT,pT>
{
public:
typedef pT array1d_type;
typedef rT T;
private:
func_obj<rT,pT>* p_fo;
optimizer<rT,pT>* p_optimizer;
//typedef blitz::Array<rT,2> array2d_type;
private:
array1d_type start_point;
array1d_type end_point;
private:
rT threshold;
private:
rT func(const pT& x)
{
assert(p_fo!=0);
return p_fo->eval(x);
}
const char* do_get_type_name()const
{
return "large scale bfgs";
}
public:
lbfgs_method()
:threshold(1e-4)
{}
virtual ~lbfgs_method()
{
};
lbfgs_method(const lbfgs_method<rT,pT>& rhs)
:p_fo(rhs.p_fo),p_optimizer(rhs.p_optimizer),
start_point(rhs.start_point),
end_point(rhs.end_point),
threshold(rhs.threshold)
{
}
lbfgs_method<rT,pT>& operator=(const lbfgs_method<rT,pT>& rhs)
{
threshold=rhs.threshold;
p_fo=rhs.p_fo;
p_optimizer=rhs.p_optimizer;
opt_eq(start_point,rhs.start_point);
opt_eq(end_point,rhs.end_point);
}
opt_method<rT,pT>* do_clone()const
{
return new lbfgs_method<rT,pT>(*this);
}
void do_set_start_point(const array1d_type& p)
{
start_point.resize(get_size(p));
opt_eq(start_point,p);
}
array1d_type do_get_start_point()const
{
return start_point;
}
void do_set_precision(rT t)
{
threshold=t;
}
rT do_get_precision()const
{
return threshold;
}
void do_set_optimizer(optimizer<rT,pT>& o)
{
p_optimizer=&o;
p_fo=p_optimizer->ptr_func_obj();
}
pT do_optimize()
{
lbfgs_parameter_t param;
lbfgs_parameter_init(¶m);
param.ftol=threshold;
std::vector<lbfgsfloatval_t> buffer(get_size(start_point));
for(int i=0;i<buffer.size();++i)
{
buffer[i]=get_element(start_point,i);
}
lbfgsfloatval_t fx;
lbfgs(get_size(start_point),&buffer[0],&fx,
lbfgs_adapter<rT,pT>,0,p_fo,¶m);
for(int i=0;i<buffer.size();++i)
{
set_element(start_point,i,buffer[i]);
}
return start_point;
}
};
}
#endif
//EOF
|
