/
FirstOrderMinimizer.h
224 lines (205 loc) · 6.52 KB
/
FirstOrderMinimizer.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
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
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
/*
* Copyright (c) The Shogun Machine Learning Toolbox
* Written (w) 2015 Wu Lin
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice, this
* list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
* ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
* SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
* The views and conclusions contained in the software and documentation are those
* of the authors and should not be interpreted as representing official policies,
* either expressed or implied, of the Shogun Development Team.
*
*/
#ifndef FIRSTORDERMINIMIZER_H
#define FIRSTORDERMINIMIZER_H
#include <shogun/lib/config.h>
#include <shogun/optimization/FirstOrderCostFunction.h>
#include <shogun/optimization/MinimizerContext.h>
#include <shogun/optimization/Penalty.h>
namespace shogun
{
/** @brief The first order minimizer base class.
*
* This class gives the interface of a first-order gradient-based unconstrained minimizer
*
* This kind of minimizers will find optimal target variables based on gradient information wrt target variables.
* For example, the gradient descend method is a minimizer.
*
* A minimizer requires the following objects as input:
* a supported cost function object (eg, FirstOrderCostFunction )
* a penalty object if regularization is enabled (eg, Penalty )
* a context object to restore mutable variables if deserialization is actived (eg, CMinimizerContext )
*
*/
class FirstOrderMinimizer
{
public:
/** Default constructor */
FirstOrderMinimizer()
{
init();
}
/** Constructor
* @param fun cost function (user have to manully delete the pointer)
*/
FirstOrderMinimizer(FirstOrderCostFunction *fun)
{
init();
set_cost_function(fun);
}
/** Destructor */
virtual ~FirstOrderMinimizer()
{}
/** Do minimization and get the optimal value
*
* @return optimal value
*/
virtual float64_t minimize()=0;
/** Does minimizer support batch update?
*
* @return whether minimizer supports batch update
*/
virtual bool supports_batch_update() const=0;
/** Set cost function used in the minimizer
*
* @param fun the cost function
*/
virtual void set_cost_function(FirstOrderCostFunction *fun)
{
REQUIRE(fun,"The cost function must be not NULL\n");
m_fun=fun;
}
/** Return a context object which stores mutable variables
* Usually it is used in serialization.
*
* @return a context object
*/
virtual CMinimizerContext* save_to_context()
{
CMinimizerContext* result=new CMinimizerContext();
update_context(result);
return result;
}
/** Load the given context object to restores mutable variables
* Usually it is used in deserialization.
*
* @param context, a context object
*/
virtual void load_from_context(CMinimizerContext* context)
{
REQUIRE(context,"Context must set\n");
if(m_penalty_type)
m_penalty_type->load_from_context(context);
}
/** Set the weight of penalty
*
* @param penalty_weight the weight of penalty, which is positive
*/
virtual void set_penalty_weight(float64_t penalty_weight)
{
REQUIRE(penalty_weight>0,"The weight of penalty must be positive\n");
m_penalty_weight=penalty_weight;
}
/** Set the type of penalty
* For example, L2 penalty
*
* @param penalty_type the type of penalty. If NULL is given, regularization is not enabled.
*/
virtual void set_penalty_type(Penalty* penalty_type)
{
m_penalty_type=penalty_type;
}
protected:
/** Update a context object to store mutable variables
*
* @param context, a context object
*/
virtual void update_context(CMinimizerContext* context)
{
REQUIRE(context,"Context must set\n");
if(m_penalty_type)
m_penalty_type->update_context(context);
}
/** Get the penalty given target variables
* For L2 penalty,
* the target variable is \f$w\f$
* and
* the value of penalty is \f$\lambda \frac{w^t w}{2}\f$,
* where \lambda is the weight of penalty
*
*
* @param var the variable used in regularization
*/
virtual float64_t get_penalty(SGVector<float64_t> var)
{
float64_t penalty=0.0;
if(m_penalty_type)
{
REQUIRE(m_penalty_weight>0,"The weight of penalty must be set first\n");
for(index_t idx=0; idx<var.vlen; idx++)
penalty+=m_penalty_weight*m_penalty_type->get_penalty(var[idx]);
}
return penalty;
}
/** Add gradient of the penalty wrt target variables to unpenalized gradient
* For least sqaure with L2 penalty,
* \f[
* L2f(w)=f(w) + L2(w) \f]
* where \f$ f(w)=\sum_i{(y_i-w^T x_i)^2}\f$ is the least sqaure cost function
* and \f$L2(w)=\lambda \frac{w^t w}{2}\f$ is the L2 penalty
*
* Target variables is \f$w\f$
* Unpenalized gradient is \f$\frac{\partial f(w) }{\partial w}\f$
* Gradient of the penalty wrt target variables is \f$\frac{\partial L2(w) }{\partial w}\f$
*
* @param gradient, unpenalized gradient wrt its target variable
* @param var the target variable
*/
virtual void update_gradient(SGVector<float64_t> gradient, SGVector<float64_t> var)
{
if(m_penalty_type)
{
REQUIRE(m_penalty_weight>0,"The weight of penalty must be set first\n");
for(index_t idx=0; idx<var.vlen; idx++)
{
float64_t grad=gradient[idx];
float64_t variable=var[idx];
gradient[idx]+=m_penalty_weight*m_penalty_type->get_penalty_gradient(variable,grad);
}
}
}
/* Cost function */
FirstOrderCostFunction *m_fun;
/* the type of penalty*/
Penalty* m_penalty_type;
/* the weight of penalty*/
float64_t m_penalty_weight;
private:
/* init */
void init()
{
m_fun=NULL;
m_penalty_type=NULL;
m_penalty_weight=0;
}
};
}
#endif