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
|
const char *help = "\
SVMTorch Multi III (c) Trebolloc & Co 2002\n\
\n\
This program will train several SVMs, for classification with\n\
more than two classes, using a one-against-all approach.\n\
It uses a gaussian kernel (default) or a polynomial kernel.\n";
#include <torch/MatDataSet.h>
#include <torch/OneHotClassFormat.h>
#include <torch/ClassMeasurer.h>
#include <torch/MSEMeasurer.h>
#include <torch/QCTrainer.h>
#include <torch/CmdLine.h>
#include <torch/Random.h>
#include <torch/SVMClassification.h>
#include <torch/DiskXFile.h>
#include <torch/ClassFormatDataSet.h>
using namespace Torch;
int main(int argc, char **argv)
{
char *file;
real c_cst, stdv;
real accuracy, cache_size;
int iter_shrink;
int the_seed;
int max_load;
char *dir_name;
char *model_file;
bool binary_mode;
int degree;
real a_cst, b_cst;
int n_classes;
Allocator *allocator = new Allocator;
//=================== The command-line ==========================
// Construct the command line
CmdLine cmd;
// Put the help line at the beginning
cmd.info(help);
// Train mode
cmd.addText("\nArguments:");
cmd.addSCmdArg("file", &file, "the train file");
cmd.addSCmdArg("model", &model_file, "the model file");
cmd.addICmdArg("# classes", &n_classes, "the number of classes", true);
cmd.addText("\nModel Options:");
cmd.addRCmdOption("-c", &c_cst, 100., "trade off cst between error/margin");
cmd.addRCmdOption("-std", &stdv, 10., "the std parameter in the gaussian kernel [exp(-|x-y|^2/std^2)]", true);
cmd.addICmdOption("-degree", °ree, -1, "if positive, use a polynomial kernel [(a xy + b)^d] with the specified degree", true);
cmd.addRCmdOption("-a", &a_cst, 1., "constant a in the polynomial kernel", true);
cmd.addRCmdOption("-b", &b_cst, 1., "constant b in the polynomial kernel", true);
cmd.addText("\nLearning Options:");
cmd.addRCmdOption("-e", &accuracy, 0.01, "end accuracy");
cmd.addRCmdOption("-m", &cache_size, 50., "cache size in Mo");
cmd.addICmdOption("-h", &iter_shrink, 100, "minimal number of iterations before shrinking");
cmd.addText("\nMisc Options:");
cmd.addICmdOption("-seed", &the_seed, -1, "the random seed");
cmd.addICmdOption("-load", &max_load, -1, "max number of examples to load for train");
cmd.addSCmdOption("-dir", &dir_name, ".", "directory to save measures");
cmd.addBCmdOption("-bin", &binary_mode, false, "binary mode for files");
// Test mode
cmd.addMasterSwitch("--test");
cmd.addText("\nArguments:");
cmd.addSCmdArg("model", &model_file, "the model file");
cmd.addSCmdArg("file", &file, "the test file");
cmd.addText("\nMisc Options:");
cmd.addICmdOption("-load", &max_load, -1, "max number of examples to load for test");
cmd.addSCmdOption("-dir", &dir_name, ".", "directory to save measures");
cmd.addBCmdOption("-bin", &binary_mode, false, "binary mode for files");
// Read the command line
int mode = cmd.read(argc, argv);
DiskXFile *model = NULL;
if(mode == 1)
{
model = new(allocator) DiskXFile(model_file, "r");
cmd.loadXFile(model);
}
// If the user didn't give any random seed,
// generate a random random seed...
if(mode == 0)
{
if(the_seed == -1)
Random::seed();
else
Random::manualSeed((long)the_seed);
}
cmd.setWorkingDirectory(dir_name);
//=================== Create the SVMs... =========================
Kernel *kernel = NULL;
if(degree > 0)
kernel = new(allocator) PolynomialKernel(degree, a_cst, b_cst);
else
kernel = new(allocator) GaussianKernel(1./(stdv*stdv));
SVM **svms = (SVM **)allocator->alloc(sizeof(SVM *)*n_classes);
for(int i = 0; i < n_classes; i++)
{
svms[i] = new(allocator) SVMClassification(kernel);
if(mode == 0)
{
svms[i]->setROption("C", c_cst);
svms[i]->setROption("cache size", cache_size);
}
}
//=================== DataSets & Measurers... ===================
// Create the training dataset
MatDataSet *mat_data = new(allocator) MatDataSet(file, -1, 1, false, max_load, binary_mode);
// Reload the model in test mode
if(mode == 1)
{
for(int i = 0; i < n_classes; i++)
svms[i]->loadXFile(model);
}
//=================== Let's go... ===============================
// Train
if(mode == 0)
{
DiskXFile model_(model_file, "w");
cmd.saveXFile(&model_);
for(int i = 0; i < n_classes; i++)
{
message("Training class %d against the others", i);
QCTrainer trainer(svms[i]);
trainer.setROption("end accuracy", accuracy);
trainer.setIOption("iter shrink", iter_shrink);
Sequence class_labels(n_classes, 1);
for(int j = 0; j < n_classes; j++)
{
if(j == i)
class_labels.frames[j][0] = 1;
else
class_labels.frames[j][0] = -1;
}
ClassFormatDataSet data(mat_data, &class_labels);
trainer.train(&data, NULL);
message("%d SV with %d at bounds", svms[i]->n_support_vectors, svms[i]->n_support_vectors_bound);
svms[i]->saveXFile(&model_);
}
}
// Test
if(mode == 1)
{
OneHotClassFormat class_format(n_classes);
int n_errors = 0;
real *buffer = (real *)allocator->alloc(sizeof(real)*n_classes);
for(int t = 0; t < mat_data->n_examples; t++)
{
mat_data->setExample(t);
for(int i = 0; i < n_classes; i++)
{
svms[i]->forward(mat_data->inputs);
buffer[i] = svms[i]->outputs->frames[0][0];
}
int the_class = (int)mat_data->targets->frames[0][0];
if(the_class != class_format.getClass(buffer))
n_errors++;
}
message("%g%% of missclassification. (%d errors)", ((real)n_errors)/((real)mat_data->n_examples)*100., n_errors);
}
delete allocator;
return(0);
}
|
