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
|
// Copyright (C) 2003--2004 Ronan Collobert (collober@idiap.ch)
//
// This file is part of Torch 3.1.
//
// 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.
// 3. The name of the author may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``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 AUTHOR 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.
#include "Boosting.h"
#include "BoostingMeasurer.h"
#include "Random.h"
#include "NullXFile.h"
namespace Torch {
static void randw(int *selected_examples, real *ex_weights, int n_examples)
{
real *repartition = (real *)Allocator::sysAlloc(sizeof(real)*(n_examples+1));
repartition[0] = 0;
for(int i = 0; i < n_examples; i++)
repartition[i+1] = repartition[i]+ex_weights[i];
for(int i = 0; i < n_examples; i++)
{
real z = Random::uniform();
int gauche = 0;
int droite = n_examples;
while(gauche+1 != droite)
{
int centre = (gauche+droite)/2;
if(repartition[centre] < z)
gauche = centre;
else
droite = centre;
}
selected_examples[i] = gauche;
// printf("%g < %g < %g\n", repartition[gauche], z, repartition[gauche+1]);
}
free(repartition);
}
Boosting::Boosting(WeightedSumMachine* w_machine_, ClassFormat *class_format_) : Trainer(w_machine_)
{
w_machine = w_machine_;
class_format = class_format_;
n_trainers = w_machine->n_trainers;
weights = w_machine->weights;
}
void Boosting::train(DataSet *data, MeasurerList* measurers)
{
int n_examples = data->n_examples;
int *selected_examples = (int *)Allocator::sysAlloc(n_examples*sizeof(int));
real *ex_weights = (real *)Allocator::sysAlloc(n_examples*sizeof(real));
for(int t = 0; t < n_examples; t++)
ex_weights[t] = 1./((real)n_examples);
NullXFile null_xfile;
BoostingMeasurer *measurer = new BoostingMeasurer(class_format, &null_xfile);
measurer->setDataSet(data);
measurer->setWeights(ex_weights);
message("Boosting: training...");
w_machine->n_trainers_trained = 0;
MeasurerList the_boost_meas;
the_boost_meas.addNode(measurer);
// Initialise le boxon.
for(int i = 0; i < n_trainers; i++)
weights[i] = 0;
for(int i = 0; i < n_trainers; i++)
{
randw(selected_examples, ex_weights, n_examples);
data->pushSubset(selected_examples, n_examples);
w_machine->trainers[i]->machine->reset();
w_machine->trainers[i]->train(data, w_machine->trainers_measurers ? w_machine->trainers_measurers[i] : NULL);
data->popSubset();
// Calcule le nouveau 'beta'...
measurer->setInputs(w_machine->trainers[i]->machine->outputs);
w_machine->trainers[i]->test(&the_boost_meas);
// Ben on vient d'entrainer un truc tu sais...
w_machine->n_trainers_trained = i+1;
// Check if all is classified [cas limite 1]
if(measurer->beta == 0)
{
for(int j = 0; j < n_trainers; j++)
weights[j] = 0;
weights[i] = 1;
warning("Boosting: train stopped. All examples are well classified.");
// On teste quand meme...
if(measurers)
test(measurers);
break;
}
// Regarde si c'est la misere [cas limite 2]
if(measurer->beta >= 1)
{
w_machine->n_trainers_trained = i;
warning("Boosting: train stopped. Model %d too weak.", i);
break;
}
// Si tout va bien... ////////////////////////
// Compute new weights
int *ptr_status = measurer->status;
real mul_pos = exp( 0.5*log(measurer->beta));
real mul_neg = exp(-0.5*log(measurer->beta));
for(int t = 0; t < n_examples; t++)
{
if(ptr_status[t] > 0)
ex_weights[t] *= mul_pos;
else
ex_weights[t] *= mul_neg;
}
// Normalize les poids des exemples
real z = 0;
for(int t = 0; t < n_examples; t++)
z += ex_weights[t];
for(int t = 0; t < n_examples; t++)
ex_weights[t] /= z;
// Refourgue le bon poids a l'autre naze...
weights[i] = -0.5*log(measurer->beta);
// Teste pour voir ce que ca donne...
if(measurers)
test(measurers);
// Fin de si tout va bien... ////////////////
}
// Jarte la misere
free(selected_examples);
free(ex_weights);
delete measurer;
// Vraiment faible mon pote!
if(w_machine->n_trainers_trained == 0)
return;
// Normalize trainers weights...
real z = 0;
for(int i = 0; i < n_trainers; i++)
z += weights[i];
for(int i = 0; i < n_trainers; i++)
weights[i] /= z;
}
Boosting::~Boosting()
{
}
}
|
