File: mlp.html

package info (click to toggle)
torch 2-1
  • links: PTS
  • area: main
  • in suites: woody
  • size: 5,488 kB
  • ctags: 3,217
  • sloc: cpp: 14,272; makefile: 201
file content (240 lines) | stat: -rw-r--r-- 12,919 bytes parent folder | download 
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
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
 
<!doctype html public "-//w3c//dtd html 4.0 transitional//en">
<html>
<head>
   <meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">
   <meta name="GENERATOR" content="Mozilla/4.77 [en] (X11; U; SunOS 5.7 sun4u) [Netscape]">
</head>
<body text="#000000" bgcolor="#FFFFFF" link="#0000EE" vlink="#551A8B" alink="#FF0000">
<tt><font color="#CC6600">const char *help = "\</font></tt>
<br><tt><font color="#CC6600">TorchMLP\n\</font></tt>
<br><tt><font color="#CC6600">\n\</font></tt>
<br><tt><font color="#CC6600">This program will train a MLP with tanh outputs
for\n\</font></tt>
<br><tt><font color="#CC6600">classification and linear outputs for regression\n";</font></tt><tt><font color="#FF9900"></font></tt>
<p><tt><font color="#009900">#include "ConnectedMachine.h"</font></tt>
<br><tt><font color="#009900">#include "Linear.h"</font></tt>
<br><tt><font color="#009900">#include "FileDataSet.h"</font></tt>
<br><tt><font color="#009900">#include "MseCriterion.h"</font></tt>
<br><tt><font color="#009900">#include "Tanh.h"</font></tt>
<br><tt><font color="#009900">#include "MseMeasurer.h"</font></tt>
<br><tt><font color="#009900">#include "ClassMeasurer.h"</font></tt>
<br><tt><font color="#009900">#include "TwoClassFormat.h"</font></tt>
<br><tt><font color="#009900">#include "OneHotClassFormat.h"</font></tt>
<br><tt><font color="#009900">#include "StochasticGradient.h"</font></tt>
<br><tt><font color="#009900">#include "GMTrainer.h"</font></tt>
<br><tt><font color="#009900">#include "CmdLine.h"</font></tt><tt></tt>
<p><tt>int main(int argc, char **argv)</tt>
<br><tt>{</tt>
<br><tt>&nbsp; char *model_file, *test_model_file;</tt>
<br><tt>&nbsp; char *valid_file;</tt>
<br><tt>&nbsp; char *file;</tt><tt></tt>
<p><tt>&nbsp; int n_inputs;</tt>
<br><tt>&nbsp; int n_targets;</tt>
<br><tt>&nbsp; int n_hu;</tt><tt></tt>
<p><tt>&nbsp; int max_load;</tt>
<br><tt>&nbsp; real accuracy;</tt>
<br><tt>&nbsp; real learning_rate;</tt>
<br><tt>&nbsp; real decay;</tt>
<br><tt>&nbsp; int max_iter;</tt>
<br><tt>&nbsp; bool regression;</tt>
<br><tt>&nbsp; int k_fold;</tt>
<br><tt>&nbsp; int the_seed;</tt><tt></tt>
<p><tt>&nbsp; <font color="#FF0000">//=================== The command-line
==========================</font></tt><tt><font color="#FF0000"></font></tt>
<p><tt>&nbsp; <font color="#CC6600">// Construct the command line</font></tt>
<br><tt>&nbsp; CmdLine cmd;</tt><tt></tt>
<p><tt>&nbsp; <font color="#CC6600">// Put the help line at the beginning</font></tt>
<br><tt>&nbsp; cmd.info(help);</tt><tt></tt>
<p><tt>&nbsp; <font color="#CC6600">// Ask for arguments</font></tt>
<br><tt>&nbsp; cmd.addText("\nArguments:");</tt>
<br><tt>&nbsp; cmd.addSCmdArg("file", &amp;file, "the train or test file");</tt>
<br><tt>&nbsp; cmd.addICmdArg("n_inputs", &amp;n_inputs, "input dimension
of the data");</tt>
<br><tt>&nbsp; cmd.addICmdArg("n_targets", &amp;n_targets, "output dimension
of the data");</tt><tt></tt>
<p><tt>&nbsp; <font color="#CC6600">// Propose some options</font></tt>
<br><tt>&nbsp; cmd.addText("\nModel Options:");</tt>
<br><tt>&nbsp; cmd.addICmdOption("-nhu", &amp;n_hu, 25, "number of hidden
units");</tt>
<br><tt>&nbsp; cmd.addBCmdOption("-rm", &amp;regression, false, "regression
mode");</tt><tt></tt>
<p><tt>&nbsp; cmd.addText("\nLearning Options:");</tt>
<br><tt>&nbsp; cmd.addICmdOption("-iter", &amp;max_iter, 25, "max number
of iterations");</tt>
<br><tt>&nbsp; cmd.addRCmdOption("-lr", &amp;learning_rate, 0.01, "learning
rate");</tt>
<br><tt>&nbsp; cmd.addRCmdOption("-e", &amp;accuracy, 0.00001, "end accuracy");</tt>
<br><tt>&nbsp; cmd.addRCmdOption("-lrd", &amp;decay, 0, "learning rate
decay");</tt><tt></tt>
<p><tt>&nbsp; cmd.addText("\nMisc Options:");</tt>
<br><tt>&nbsp; cmd.addICmdOption("-seed", &amp;the_seed, -1, "the random
seed");</tt>
<br><tt>&nbsp; cmd.addICmdOption("-Kfold", &amp;k_fold, -1, "number of
subsets for K-fold cross-validation");</tt>
<br><tt>&nbsp; cmd.addICmdOption("-load", &amp;max_load, -1, "max number
of examples to load");</tt>
<br><tt>&nbsp; cmd.addSCmdOption("-valid", &amp;valid_file, "", "validation
file, if you want it");</tt>
<br><tt>&nbsp; cmd.addSCmdOption("-sm", &amp;model_file, "", "file to save
the model");</tt>
<br><tt>&nbsp; cmd.addSCmdOption("-test", &amp;test_model_file, "", "model
file to test");</tt><tt></tt>
<p><tt>&nbsp; <font color="#CC6600">// Read the command line</font></tt>
<br><tt>&nbsp; cmd.read(argc, argv);</tt><tt></tt>
<p><tt>&nbsp; <font color="#CC6600">// If the user didn't give any random
seed,</font></tt>
<br><tt><font color="#CC6600">&nbsp; // generate a random random seed...</font></tt>
<br><tt>&nbsp; if(the_seed == -1)</tt>
<br><tt>&nbsp;&nbsp;&nbsp; seed();</tt>
<br><tt>&nbsp; else</tt>
<br><tt>&nbsp;&nbsp;&nbsp; manual_seed((long)the_seed);</tt><tt></tt>
<p><tt>&nbsp; <font color="#FF0000">//=================== Create the MLP...
=========================</font></tt>
<br><tt>&nbsp; ConnectedMachine MLP;</tt><tt></tt>
<p><tt>&nbsp; <font color="#CC6600">// Create the layers of the MLP</font></tt>
<br><tt>&nbsp; Linear hidden_linear(n_inputs, n_hu);</tt>
<br><tt>&nbsp; Tanh hidden_nlinear(n_hu);</tt>
<br><tt>&nbsp; Linear output_linear(n_hu, n_targets);</tt>
<br><tt>&nbsp; Tanh output_nlinear(n_targets);</tt><tt></tt>
<p><tt>&nbsp; <font color="#CC6600">// Initialize the layers</font></tt>
<br><tt>&nbsp; hidden_linear.init();</tt>
<br><tt>&nbsp; hidden_nlinear.init();</tt>
<br><tt>&nbsp; output_linear.init();</tt>
<br><tt>&nbsp; output_nlinear.init();</tt><tt></tt>
<p><tt>&nbsp; <font color="#CC6600">// Add the layers (Full Connected Layers)
to the MLP</font></tt>
<br><tt>&nbsp; MLP.addFCL(&amp;hidden_linear);</tt>
<br><tt>&nbsp; MLP.addFCL(&amp;hidden_nlinear);</tt>
<br><tt>&nbsp; MLP.addFCL(&amp;output_linear);</tt><tt></tt>
<p><tt>&nbsp; <font color="#CC6600">// If regression, don't add the tanh
output layer</font></tt>
<br><tt>&nbsp; if(!regression)</tt>
<br><tt>&nbsp;&nbsp;&nbsp; MLP.addFCL(&amp;output_nlinear);</tt><tt></tt>
<p><tt>&nbsp; <font color="#CC6600">// Initialize the MLP</font></tt>
<br><tt>&nbsp; MLP.init();</tt>
<br><tt></tt>&nbsp;<tt></tt>
<p><tt>&nbsp; <font color="#FF0000">//=================== DataSets &amp;
Measurers... ===================</font></tt><tt><font color="#FF0000"></font></tt>
<p><tt>&nbsp; <font color="#CC6600">// Create the training dataset (normalize
inputs)</font></tt>
<br><tt>&nbsp; FileDataSet data(file, n_inputs, n_targets, false, max_load);</tt>
<br><tt>&nbsp; data.setBOption("normalize inputs", true);</tt>
<br><tt>&nbsp; data.init();</tt><tt></tt>
<p><tt>&nbsp; <font color="#CC6600">// The list of measurers...</font></tt>
<br><tt>&nbsp; List *measurers = NULL;</tt><tt></tt>
<p><tt>&nbsp; <font color="#CC6600">// The class format</font></tt>
<br><tt>&nbsp; ClassFormat *class_format = NULL;</tt>
<br><tt>&nbsp; if(!regression)</tt>
<br><tt>&nbsp; {</tt>
<br><tt>&nbsp;&nbsp;&nbsp; if(n_targets == 1)</tt>
<br><tt>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; class_format = new TwoClassFormat(&amp;data);</tt>
<br><tt>&nbsp;&nbsp;&nbsp; else</tt>
<br><tt>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; class_format = new OneHotClassFormat(&amp;data);</tt>
<br><tt>&nbsp; }</tt><tt></tt>
<p><tt>&nbsp; <font color="#CC6600">// The validation set...</font></tt>
<br><tt>&nbsp; FileDataSet *valid_data = NULL;</tt>
<br><tt>&nbsp; MseMeasurer *valid_mse_meas = NULL;</tt>
<br><tt>&nbsp; ClassMeasurer *valid_class_meas = NULL;</tt><tt></tt>
<p><tt>&nbsp; <font color="#CC6600">// Create a validation set, if any</font></tt>
<br><tt>&nbsp; if(strcmp(valid_file, ""))</tt>
<br><tt>&nbsp; {</tt>
<br><tt>&nbsp;&nbsp;&nbsp; <font color="#CC6600">// Load the validation
set and normalize it with the</font></tt>
<br><tt><font color="#CC6600">&nbsp;&nbsp;&nbsp; // values in the train
dataset</font></tt>
<br><tt>&nbsp;&nbsp;&nbsp; valid_data = new FileDataSet(valid_file, n_inputs,
n_targets);</tt>
<br><tt>&nbsp;&nbsp;&nbsp; valid_data->init();</tt>
<br><tt>&nbsp;&nbsp;&nbsp; valid_data->normalizeUsingDataSet(&amp;data);</tt><tt></tt>
<p><tt>&nbsp;&nbsp;&nbsp; <font color="#CC6600">// Create a MSE measurer
and an error class measurer</font></tt>
<br><tt><font color="#CC6600">&nbsp;&nbsp;&nbsp; // on the validation dataset
(if we are not in regression)</font></tt>
<br><tt>&nbsp;&nbsp;&nbsp; valid_mse_meas = new MseMeasurer(MLP.outputs,
valid_data, "the_valid_mse");</tt>
<br><tt>&nbsp;&nbsp;&nbsp; valid_mse_meas->init();</tt>
<br><tt>&nbsp;&nbsp;&nbsp; addToList(&amp;measurers, 1, valid_mse_meas);</tt><tt></tt>
<p><tt>&nbsp;&nbsp;&nbsp; if(!regression)</tt>
<br><tt>&nbsp;&nbsp;&nbsp; {</tt>
<br><tt>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; valid_class_meas = new ClassMeasurer(MLP.outputs,
valid_data, class_format, "the_valid_class_err");</tt>
<br><tt>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; valid_class_meas->init();</tt>
<br><tt>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; addToList(&amp;measurers, 1, valid_class_meas);</tt>
<br><tt>&nbsp;&nbsp;&nbsp; }</tt>
<br><tt>&nbsp; }</tt><tt></tt>
<p><tt>&nbsp; <font color="#CC6600">// Measurers on the training dataset</font></tt>
<br><tt>&nbsp; MseMeasurer *mse_meas = new MseMeasurer(MLP.outputs, &amp;data,
"the_mse");</tt>
<br><tt>&nbsp; mse_meas->init();</tt>
<br><tt>&nbsp; addToList(&amp;measurers, 1, mse_meas);</tt><tt></tt>
<p><tt>&nbsp; ClassMeasurer *class_meas = NULL;</tt>
<br><tt>&nbsp; if(!regression)</tt>
<br><tt>&nbsp; {</tt>
<br><tt>&nbsp;&nbsp;&nbsp; class_meas = new ClassMeasurer(MLP.outputs,
&amp;data, class_format, "the_class_err");</tt>
<br><tt>&nbsp;&nbsp;&nbsp; class_meas->init();</tt>
<br><tt>&nbsp;&nbsp;&nbsp; addToList(&amp;measurers, 1, class_meas);</tt>
<br><tt>&nbsp; }</tt><tt></tt>
<p><tt>&nbsp; <font color="#FF0000">//=================== The Trainer ===============================</font></tt>
<br><tt>&nbsp;</tt>
<br><tt>&nbsp; <font color="#CC6600">// The criterion for the GMTrainer
(MSE criterion)</font></tt>
<br><tt>&nbsp; MseCriterion mse(n_targets);</tt>
<br><tt>&nbsp; mse.init();</tt><tt></tt>
<p><tt>&nbsp; // The optimizer for the GMTrainer</tt>
<br><tt>&nbsp; StochasticGradient opt;</tt>
<br><tt>&nbsp; opt.setIOption("max iter", max_iter);</tt>
<br><tt>&nbsp; opt.setROption("end accuracy", accuracy);</tt>
<br><tt>&nbsp; opt.setROption("learning rate", learning_rate);</tt>
<br><tt>&nbsp; opt.setROption("learning rate decay", decay);</tt><tt></tt>
<p><tt>&nbsp; <font color="#CC6600">// The Gradient Machine Trainer</font></tt>
<br><tt>&nbsp; GMTrainer trainer(&amp;MLP, &amp;data, &amp;mse, &amp;opt);</tt><tt></tt>
<p><tt>&nbsp; <font color="#FF0000">//=================== Let's go... ===============================</font></tt><tt></tt>
<p><tt>&nbsp; <font color="#CC6600">// Print the number of parameter of
the MLP (just for fun)</font></tt>
<br><tt>&nbsp; message("Number of parameters: %d", MLP.n_params);</tt><tt></tt>
<p><tt>&nbsp; <font color="#CC6600">// If the user provides a previously
trained model,</font></tt>
<br><tt><font color="#CC6600">&nbsp; // test it...</font></tt>
<br><tt>&nbsp; if( strcmp(test_model_file, "") )</tt>
<br><tt>&nbsp; {</tt>
<br><tt>&nbsp;&nbsp;&nbsp; trainer.load(test_model_file);</tt>
<br><tt>&nbsp;&nbsp;&nbsp; trainer.test(measurers);</tt>
<br><tt>&nbsp; }</tt><tt></tt>
<p><tt>&nbsp; <font color="#CC6600">// ...else...</font></tt>
<br><tt>&nbsp; else</tt>
<br><tt>&nbsp; {</tt>
<br><tt>&nbsp;&nbsp;&nbsp; <font color="#CC6600">// If the user provides
a number for the K-fold validation,</font></tt>
<br><tt><font color="#CC6600">&nbsp;&nbsp;&nbsp; // do a K-fold validation</font></tt>
<br><tt>&nbsp;&nbsp;&nbsp; if(k_fold > 0)</tt>
<br><tt>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; trainer.crossValidate(k_fold, NULL,
measurers);</tt><tt></tt>
<p><tt>&nbsp;&nbsp;&nbsp; <font color="#CC6600">// Else, train the model</font></tt>
<br><tt>&nbsp;&nbsp;&nbsp; else</tt>
<br><tt>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; trainer.train(measurers);</tt><tt></tt>
<p><tt>&nbsp;&nbsp;&nbsp; <font color="#CC6600">// Save the model if the
user provides a name for that</font></tt>
<br><tt>&nbsp;&nbsp;&nbsp; if( strcmp(model_file, "") )</tt>
<br><tt>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; trainer.save(model_file);</tt>
<br><tt>&nbsp; }</tt><tt></tt>
<p><tt>&nbsp; <font color="#FF0000">//=================== Quit... ===================================</font></tt>
<br><tt>&nbsp; if(strcmp(valid_file, ""))</tt>
<br><tt>&nbsp; {</tt>
<br><tt>&nbsp;&nbsp;&nbsp; delete valid_data;</tt>
<br><tt>&nbsp;&nbsp;&nbsp; delete valid_mse_meas;</tt>
<br><tt>&nbsp;&nbsp;&nbsp; if(!regression)</tt>
<br><tt>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; delete valid_class_meas;</tt>
<br><tt>&nbsp; }</tt><tt></tt>
<p><tt>&nbsp; delete mse_meas;</tt>
<br><tt>&nbsp; if(!regression)</tt>
<br><tt>&nbsp; {</tt>
<br><tt>&nbsp;&nbsp;&nbsp; delete class_meas;</tt>
<br><tt>&nbsp;&nbsp;&nbsp; delete class_format;</tt>
<br><tt>&nbsp; }</tt><tt></tt>
<p><tt>&nbsp; freeList(&amp;measurers);</tt><tt></tt>
<p><tt>&nbsp; return(0);</tt>
<br><tt>}</tt>
<br>&nbsp;
</body>
</html>