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
|
/*********************************************************************
MLDemos: A User-Friendly visualization toolkit for machine learning
Copyright (C) 2010 Basilio Noris
Contact: mldemos@b4silio.com
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
This library 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
Library General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, write to the Free
Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
*********************************************************************/
#include "public.h"
#include "basicMath.h"
#include "regressorGB.h"
RegressorGB::RegressorGB()
: boostIters(0), boostLossType(0), boostTreeDepths(0), gbt(0)
{
}
RegressorGB::~RegressorGB()
{
DEL(gbt);
}
void RegressorGB::Train(std::vector< fvec > samples, ivec labels)
{
u32 sampleCnt = samples.size();
if(!sampleCnt) return;
dim = samples[0].size();
if(outputDim != -1 && outputDim < dim -1)
{
// we need to swap the current last dimension with the desired output
FOR(i, samples.size())
{
float val = samples[i][dim-1];
samples[i][dim-1] = samples[i][outputDim];
samples[i][outputDim] = val;
}
}
DEL(gbt);
dim = samples[0].size()-1;
// convert the samples into CvMat* format
CvMat *trainSamples = cvCreateMat(sampleCnt, dim, CV_32FC1);
CvMat *trainOutputs = cvCreateMat(sampleCnt, 1, CV_32FC1);
const CvMat *sampleIdx = 0;//cvCreateMat(samples[0].size(), 1, CV_32FC1);//dim;//samples[0].size();// should be dim
const CvMat *varIdx = 0;// 0 if the sused ubset if not set
const CvMat *varType = 0;
const CvMat *missingDataMask = 0;
int tflag = CV_ROW_SAMPLE;
//loss_function_type – Type of the loss function used for training (see Training the GBT model). It must be one of the following types: CvGBTrees::SQUARED_LOSS, CvGBTrees::ABSOLUTE_LOSS, CvGBTrees::HUBER_LOSS, CvGBTrees::DEVIANCE_LOSS. The first three types are used for regression problems, and the last one for classification.
int activationFunction;
switch(boostLossType==1)
{
case 1:
activationFunction = CvGBTrees::SQUARED_LOSS;
break;
case 2:
activationFunction = CvGBTrees::ABSOLUTE_LOSS;
break;
case 3:
activationFunction = CvGBTrees::HUBER_LOSS;
break;
}
//weak_count – Count of boosting algorithm iterations. weak_count*K is the total count of trees in the GBT model, where K is the output classes count (equal to one in case of a regression).
int weak_count = boostIters;
//shrinkage – Regularization parameter (see Training the GBT model).
float shrinkage = 0.1f;
//subsample_portion – Portion of the whole training set used for each algorithm iteration. Subset is generated randomly. For more information see http://www.salfordsystems.com/doc/StochasticBoostingSS.pdf.
float subsample_portion = 0.5f;
//max_depth – Maximal depth of each decision tree in the ensemble (see CvDTree).
int max_depth = boostTreeDepths;
//use_surrogates – If true, surrogate splits are built (see CvDTree).
bool use_surrogates = false;
CvGBTreesParams params= CvGBTreesParams(activationFunction, weak_count, shrinkage, subsample_portion, max_depth, use_surrogates);
u32 *perm = randPerm(sampleCnt);
FOR(i, sampleCnt)
{
FOR(j, dim) cvSetReal2D(trainSamples, i, j, samples[perm[i]][j]);
cvSet1D(trainOutputs, i, cvScalar(samples[perm[i]][dim]));
}
delete [] perm;
gbt = new CvGBTrees();
gbt->train(trainSamples, tflag,trainOutputs,varIdx,sampleIdx,varType,missingDataMask,params);
cvReleaseMat(&trainSamples);
cvReleaseMat(&trainOutputs);
}
fvec RegressorGB::Test( const fvec &sample)
{
fvec res;
res.resize(2);
if(!gbt) return res;
float *_input = new float[dim];
if(outputDim != -1 & outputDim < sample.size())
{
fvec newSample = sample;
newSample[outputDim] = sample[sample.size()-1];
newSample[sample.size()-1] = sample[outputDim];
FOR(d, min(dim,(u32)sample.size())) _input[d] = newSample[d];
for(int d=min(dim,(u32)sample.size()); d<dim; d++) _input[d] = 0;
}
else
{
FOR(d, min(dim,(u32)sample.size())) _input[d] = sample[d];
for(int d=min(dim,(u32)sample.size()); d<dim; d++) _input[d] = 0;
}
CvMat input = cvMat(1,dim,CV_32FC1, _input);
float output;
output = gbt->predict(&input);
res[0] = output;
res[1] = 0;
delete [] _input;
return res;
}
void RegressorGB::SetParams(int boostIters, int boostLossType, int boostTreeDepths)
{
this->boostIters = boostIters;
this->boostLossType = boostLossType;
this->boostTreeDepths = boostTreeDepths;
}
const char *RegressorGB::GetInfoString()
{
char *text = new char[1024];
sprintf(text, "Gradient Boosting Tree\n");
sprintf(text, "\n");
return text;
}
|
