line |
stmt |
bran |
cond |
sub |
pod |
time |
code |
1
|
|
|
|
|
|
|
package AI::NeuralNet::SOM; |
2
|
|
|
|
|
|
|
|
3
|
4
|
|
|
4
|
|
38653
|
use strict; |
|
4
|
|
|
|
|
7
|
|
|
4
|
|
|
|
|
205
|
|
4
|
4
|
|
|
4
|
|
27
|
use warnings; |
|
4
|
|
|
|
|
7
|
|
|
4
|
|
|
|
|
196
|
|
5
|
|
|
|
|
|
|
|
6
|
|
|
|
|
|
|
require Exporter; |
7
|
4
|
|
|
4
|
|
34
|
use base qw(Exporter); |
|
4
|
|
|
|
|
7
|
|
|
4
|
|
|
|
|
377
|
|
8
|
|
|
|
|
|
|
|
9
|
4
|
|
|
4
|
|
4653
|
use Data::Dumper; |
|
4
|
|
|
|
|
44395
|
|
|
4
|
|
|
|
|
12172
|
|
10
|
|
|
|
|
|
|
|
11
|
|
|
|
|
|
|
=pod |
12
|
|
|
|
|
|
|
|
13
|
|
|
|
|
|
|
=head1 NAME |
14
|
|
|
|
|
|
|
|
15
|
|
|
|
|
|
|
AI::NeuralNet::SOM - Perl extension for Kohonen Maps |
16
|
|
|
|
|
|
|
|
17
|
|
|
|
|
|
|
=head1 SYNOPSIS |
18
|
|
|
|
|
|
|
|
19
|
|
|
|
|
|
|
use AI::NeuralNet::SOM::Rect; |
20
|
|
|
|
|
|
|
my $nn = new AI::NeuralNet::SOM::Rect (output_dim => "5x6", |
21
|
|
|
|
|
|
|
input_dim => 3); |
22
|
|
|
|
|
|
|
$nn->initialize; |
23
|
|
|
|
|
|
|
$nn->train (30, |
24
|
|
|
|
|
|
|
[ 3, 2, 4 ], |
25
|
|
|
|
|
|
|
[ -1, -1, -1 ], |
26
|
|
|
|
|
|
|
[ 0, 4, -3]); |
27
|
|
|
|
|
|
|
|
28
|
|
|
|
|
|
|
my @mes = $nn->train (30, ...); # learn about the smallest errors |
29
|
|
|
|
|
|
|
# during training |
30
|
|
|
|
|
|
|
|
31
|
|
|
|
|
|
|
print $nn->as_data; # dump the raw data |
32
|
|
|
|
|
|
|
print $nn->as_string; # prepare a somehow formatted string |
33
|
|
|
|
|
|
|
|
34
|
|
|
|
|
|
|
use AI::NeuralNet::SOM::Torus; |
35
|
|
|
|
|
|
|
# similar to above |
36
|
|
|
|
|
|
|
|
37
|
|
|
|
|
|
|
use AI::NeuralNet::SOM::Hexa; |
38
|
|
|
|
|
|
|
my $nn = new AI::NeuralNet::SOM::Hexa (output_dim => 6, |
39
|
|
|
|
|
|
|
input_dim => 4); |
40
|
|
|
|
|
|
|
$nn->initialize ( [ 0, 0, 0, 0 ] ); # all get this value |
41
|
|
|
|
|
|
|
|
42
|
|
|
|
|
|
|
$nn->value (3, 2, [ 1, 1, 1, 1 ]); # change value for a neuron |
43
|
|
|
|
|
|
|
print $nn->value (3, 2); |
44
|
|
|
|
|
|
|
|
45
|
|
|
|
|
|
|
$nn->label (3, 2, 'Danger'); # add a label to the neuron |
46
|
|
|
|
|
|
|
print $nn->label (3, 2); |
47
|
|
|
|
|
|
|
|
48
|
|
|
|
|
|
|
|
49
|
|
|
|
|
|
|
=head1 DESCRIPTION |
50
|
|
|
|
|
|
|
|
51
|
|
|
|
|
|
|
This package is a stripped down implementation of the Kohonen Maps |
52
|
|
|
|
|
|
|
(self organizing maps). It is B meant as demonstration or for use |
53
|
|
|
|
|
|
|
together with some visualisation software. And while it is not (yet) |
54
|
|
|
|
|
|
|
optimized for speed, some consideration has been given that it is not |
55
|
|
|
|
|
|
|
overly slow. |
56
|
|
|
|
|
|
|
|
57
|
|
|
|
|
|
|
Particular emphasis has been given that the package plays nicely with |
58
|
|
|
|
|
|
|
others. So no use of files, no arcane dependencies, etc. |
59
|
|
|
|
|
|
|
|
60
|
|
|
|
|
|
|
=head2 Scenario |
61
|
|
|
|
|
|
|
|
62
|
|
|
|
|
|
|
The basic idea is that the neural network consists of a 2-dimensional |
63
|
|
|
|
|
|
|
array of N-dimensional vectors. When the training is started these |
64
|
|
|
|
|
|
|
vectors may be completely random, but over time the network learns |
65
|
|
|
|
|
|
|
from the sample data, which is a set of N-dimensional vectors. |
66
|
|
|
|
|
|
|
|
67
|
|
|
|
|
|
|
Slowly, the vectors in the network will try to approximate the sample |
68
|
|
|
|
|
|
|
vectors fed in. If in the sample vectors there were clusters, then |
69
|
|
|
|
|
|
|
these clusters will be neighbourhoods within the rectangle (or |
70
|
|
|
|
|
|
|
whatever topology you are using). |
71
|
|
|
|
|
|
|
|
72
|
|
|
|
|
|
|
Technically, you have reduced your dimension from N to 2. |
73
|
|
|
|
|
|
|
|
74
|
|
|
|
|
|
|
=head1 INTERFACE |
75
|
|
|
|
|
|
|
|
76
|
|
|
|
|
|
|
=head2 Constructor |
77
|
|
|
|
|
|
|
|
78
|
|
|
|
|
|
|
The constructor takes arguments: |
79
|
|
|
|
|
|
|
|
80
|
|
|
|
|
|
|
=over |
81
|
|
|
|
|
|
|
|
82
|
|
|
|
|
|
|
=item C : (mandatory, no default) |
83
|
|
|
|
|
|
|
|
84
|
|
|
|
|
|
|
A positive integer specifying the dimension of the sample vectors (and hence that of the vectors in |
85
|
|
|
|
|
|
|
the grid). |
86
|
|
|
|
|
|
|
|
87
|
|
|
|
|
|
|
=item C: (optional, default C<0.1>) |
88
|
|
|
|
|
|
|
|
89
|
|
|
|
|
|
|
This is a magic number which controls how strongly the vectors in the grid can be influenced. Stronger |
90
|
|
|
|
|
|
|
movement can mean faster learning if the clusters are very pronounced. If not, then the movement is |
91
|
|
|
|
|
|
|
like noise and the convergence is not good. To mediate that effect, the learning rate is reduced |
92
|
|
|
|
|
|
|
over the iterations. |
93
|
|
|
|
|
|
|
|
94
|
|
|
|
|
|
|
=item C: (optional, defaults to radius) |
95
|
|
|
|
|
|
|
|
96
|
|
|
|
|
|
|
A non-negative number representing the start value for the learning radius. Practically, the value |
97
|
|
|
|
|
|
|
should be chosen in such a way to cover a larger part of the map. During the learning process this |
98
|
|
|
|
|
|
|
value will be narrowed down, so that the learning radius impacts less and less neurons. |
99
|
|
|
|
|
|
|
|
100
|
|
|
|
|
|
|
B: Do not choose C<1> as the C function is used on this value. |
101
|
|
|
|
|
|
|
|
102
|
|
|
|
|
|
|
=back |
103
|
|
|
|
|
|
|
|
104
|
|
|
|
|
|
|
Subclasses will (re)define some of these parameters and add others: |
105
|
|
|
|
|
|
|
|
106
|
|
|
|
|
|
|
Example: |
107
|
|
|
|
|
|
|
|
108
|
|
|
|
|
|
|
my $nn = new AI::NeuralNet::SOM::Rect (output_dim => "5x6", |
109
|
|
|
|
|
|
|
input_dim => 3); |
110
|
|
|
|
|
|
|
|
111
|
|
|
|
|
|
|
=cut |
112
|
|
|
|
|
|
|
|
113
|
0
|
|
|
0
|
0
|
0
|
sub new { die; } |
114
|
|
|
|
|
|
|
|
115
|
|
|
|
|
|
|
=pod |
116
|
|
|
|
|
|
|
|
117
|
|
|
|
|
|
|
=head2 Methods |
118
|
|
|
|
|
|
|
|
119
|
|
|
|
|
|
|
=over |
120
|
|
|
|
|
|
|
|
121
|
|
|
|
|
|
|
=item I |
122
|
|
|
|
|
|
|
|
123
|
|
|
|
|
|
|
I<$nn>->initialize |
124
|
|
|
|
|
|
|
|
125
|
|
|
|
|
|
|
You need to initialize all vectors in the map before training. There are several options |
126
|
|
|
|
|
|
|
how this is done: |
127
|
|
|
|
|
|
|
|
128
|
|
|
|
|
|
|
=over |
129
|
|
|
|
|
|
|
|
130
|
|
|
|
|
|
|
=item providing data vectors |
131
|
|
|
|
|
|
|
|
132
|
|
|
|
|
|
|
If you provide a list of vectors, these will be used in turn to seed the neurons. If the list is |
133
|
|
|
|
|
|
|
shorter than the number of neurons, the list will be started over. That way it is trivial to |
134
|
|
|
|
|
|
|
zero everything: |
135
|
|
|
|
|
|
|
|
136
|
|
|
|
|
|
|
$nn->initialize ( [ 0, 0, 0 ] ); |
137
|
|
|
|
|
|
|
|
138
|
|
|
|
|
|
|
=item providing no data |
139
|
|
|
|
|
|
|
|
140
|
|
|
|
|
|
|
Then all vectors will get randomized values (in the range [ -0.5 .. 0.5 ]). |
141
|
|
|
|
|
|
|
|
142
|
|
|
|
|
|
|
=item using eigenvectors (see L) |
143
|
|
|
|
|
|
|
|
144
|
|
|
|
|
|
|
=back |
145
|
|
|
|
|
|
|
|
146
|
|
|
|
|
|
|
=item I |
147
|
|
|
|
|
|
|
|
148
|
|
|
|
|
|
|
I<$nn>->train ( I<$epochs>, I<@vectors> ) |
149
|
|
|
|
|
|
|
|
150
|
|
|
|
|
|
|
I<@mes> = I<$nn>->train ( I<$epochs>, I<@vectors> ) |
151
|
|
|
|
|
|
|
|
152
|
|
|
|
|
|
|
The training uses the list of sample vectors to make the network learn. Each vector is simply a |
153
|
|
|
|
|
|
|
reference to an array of values. |
154
|
|
|
|
|
|
|
|
155
|
|
|
|
|
|
|
The C parameter controls how many vectors are processed. The vectors are B used in |
156
|
|
|
|
|
|
|
sequence, but picked randomly from the list. For this reason it is wise to run several epochs, |
157
|
|
|
|
|
|
|
not just one. But within one epoch B vectors are visited exactly once. |
158
|
|
|
|
|
|
|
|
159
|
|
|
|
|
|
|
Example: |
160
|
|
|
|
|
|
|
|
161
|
|
|
|
|
|
|
$nn->train (30, |
162
|
|
|
|
|
|
|
[ 3, 2, 4 ], |
163
|
|
|
|
|
|
|
[ -1, -1, -1 ], |
164
|
|
|
|
|
|
|
[ 0, 4, -3]); |
165
|
|
|
|
|
|
|
|
166
|
|
|
|
|
|
|
=cut |
167
|
|
|
|
|
|
|
|
168
|
|
|
|
|
|
|
sub train { |
169
|
47
|
|
|
47
|
1
|
2911
|
my $self = shift; |
170
|
47
|
|
50
|
|
|
142
|
my $epochs = shift || 1; |
171
|
47
|
50
|
|
|
|
129
|
die "no data to learn" unless @_; |
172
|
|
|
|
|
|
|
|
173
|
47
|
|
|
|
|
279
|
$self->{LAMBDA} = $epochs / log ($self->{_Sigma0}); # educated guess? |
174
|
|
|
|
|
|
|
|
175
|
47
|
|
|
|
|
107
|
my @mes = (); # this will contain the errors during the epochs |
176
|
47
|
|
|
|
|
116
|
for my $epoch (1..$epochs) { |
177
|
3060
|
|
|
|
|
5211
|
$self->{T} = $epoch; |
178
|
3060
|
|
|
|
|
9605
|
my $sigma = $self->{_Sigma0} * exp ( - $self->{T} / $self->{LAMBDA} ); # compute current radius |
179
|
3060
|
|
|
|
|
8016
|
my $l = $self->{_L0} * exp ( - $self->{T} / $epochs ); # current learning rate |
180
|
|
|
|
|
|
|
|
181
|
3060
|
|
|
|
|
6264
|
my @veggies = @_; # make a local copy, that will be destroyed in the loop |
182
|
3060
|
|
|
|
|
6900
|
while (@veggies) { |
183
|
8780
|
|
|
|
|
22556
|
my $sample = splice @veggies, int (rand (scalar @veggies) ), 1; # find (and take out) |
184
|
|
|
|
|
|
|
|
185
|
8780
|
|
|
|
|
29744
|
my @bmu = $self->bmu ($sample); # find the best matching unit |
186
|
8780
|
100
|
|
|
|
22235
|
push @mes, $bmu[2] if wantarray; |
187
|
8780
|
|
|
|
|
30356
|
my $neighbors = $self->neighbors ($sigma, @bmu); # find its neighbors |
188
|
8780
|
|
|
|
|
14777
|
map { _adjust ($self, $l, $sigma, $_, $sample) } @$neighbors; # bend them like Beckham |
|
57992
|
|
|
|
|
107416
|
|
189
|
|
|
|
|
|
|
} |
190
|
|
|
|
|
|
|
} |
191
|
47
|
|
|
|
|
238
|
return @mes; |
192
|
|
|
|
|
|
|
} |
193
|
|
|
|
|
|
|
|
194
|
|
|
|
|
|
|
sub _adjust { # http://www.ai-junkie.com/ann/som/som4.html |
195
|
57992
|
|
|
57992
|
|
71494
|
my $self = shift; |
196
|
57992
|
|
|
|
|
62607
|
my $l = shift; # the learning rate |
197
|
57992
|
|
|
|
|
60432
|
my $sigma = shift; # the current radius |
198
|
57992
|
|
|
|
|
78081
|
my $unit = shift; # which unit to change |
199
|
57992
|
|
|
|
|
80791
|
my ($x, $y, $d) = @$unit; # it contains the distance |
200
|
57992
|
|
|
|
|
58590
|
my $v = shift; # the vector which makes the impact |
201
|
|
|
|
|
|
|
|
202
|
57992
|
|
|
|
|
90497
|
my $w = $self->{map}->[$x]->[$y]; # find the data behind the unit |
203
|
57992
|
|
|
|
|
126059
|
my $theta = exp ( - ($d ** 2) / (2 * $sigma ** 2)); # gaussian impact (using distance and current radius) |
204
|
|
|
|
|
|
|
|
205
|
57992
|
|
|
|
|
109832
|
foreach my $i (0 .. $#$w) { # adjusting values |
206
|
173976
|
|
|
|
|
480692
|
$w->[$i] = $w->[$i] + $theta * $l * ( $v->[$i] - $w->[$i] ); |
207
|
|
|
|
|
|
|
} |
208
|
|
|
|
|
|
|
} |
209
|
|
|
|
|
|
|
|
210
|
|
|
|
|
|
|
=pod |
211
|
|
|
|
|
|
|
|
212
|
|
|
|
|
|
|
=item I |
213
|
|
|
|
|
|
|
|
214
|
|
|
|
|
|
|
(I<$x>, I<$y>, I<$distance>) = I<$nn>->bmu (I<$vector>) |
215
|
|
|
|
|
|
|
|
216
|
|
|
|
|
|
|
This method finds the I, i.e. that neuron which is closest to the vector passed |
217
|
|
|
|
|
|
|
in. The method returns the coordinates and the actual distance. |
218
|
|
|
|
|
|
|
|
219
|
|
|
|
|
|
|
=cut |
220
|
|
|
|
|
|
|
|
221
|
0
|
|
|
0
|
1
|
0
|
sub bmu { die; } |
222
|
|
|
|
|
|
|
|
223
|
|
|
|
|
|
|
=pod |
224
|
|
|
|
|
|
|
|
225
|
|
|
|
|
|
|
=item I |
226
|
|
|
|
|
|
|
|
227
|
|
|
|
|
|
|
I<$me> = I<$nn>->mean_error (I<@vectors>) |
228
|
|
|
|
|
|
|
|
229
|
|
|
|
|
|
|
This method takes a number of vectors and produces the I, i.e. the average I |
230
|
|
|
|
|
|
|
which the SOM makes when finding the Cs for the vectors. At least one vector must be passed in. |
231
|
|
|
|
|
|
|
|
232
|
|
|
|
|
|
|
Obviously, the longer you let your SOM be trained, the smaller the error should become. |
233
|
|
|
|
|
|
|
|
234
|
|
|
|
|
|
|
=cut |
235
|
|
|
|
|
|
|
|
236
|
|
|
|
|
|
|
sub mean_error { |
237
|
81
|
|
|
81
|
1
|
352
|
my $self = shift; |
238
|
81
|
|
|
|
|
111
|
my $error = 0; |
239
|
243
|
|
|
|
|
360
|
map { $error += $_ } # then add them all up |
|
243
|
|
|
|
|
745
|
|
240
|
81
|
|
|
|
|
180
|
map { ( $self->bmu($_) )[2] } # then find the distance |
241
|
|
|
|
|
|
|
@_; # take all data vectors |
242
|
81
|
|
|
|
|
491
|
return ($error / scalar @_); # return the mean value |
243
|
|
|
|
|
|
|
} |
244
|
|
|
|
|
|
|
|
245
|
|
|
|
|
|
|
=pod |
246
|
|
|
|
|
|
|
|
247
|
|
|
|
|
|
|
=item I |
248
|
|
|
|
|
|
|
|
249
|
|
|
|
|
|
|
I<$ns> = I<$nn>->neighbors (I<$sigma>, I<$x>, I<$y>) |
250
|
|
|
|
|
|
|
|
251
|
|
|
|
|
|
|
Finds all neighbors of (X, Y) with a distance smaller than SIGMA. Returns a list reference of (X, Y, |
252
|
|
|
|
|
|
|
distance) triples. |
253
|
|
|
|
|
|
|
|
254
|
|
|
|
|
|
|
=cut |
255
|
|
|
|
|
|
|
|
256
|
0
|
|
|
0
|
1
|
0
|
sub neighbors { die; } |
257
|
|
|
|
|
|
|
|
258
|
|
|
|
|
|
|
=pod |
259
|
|
|
|
|
|
|
|
260
|
|
|
|
|
|
|
=item I (read-only) |
261
|
|
|
|
|
|
|
|
262
|
|
|
|
|
|
|
I<$dim> = I<$nn>->output_dim |
263
|
|
|
|
|
|
|
|
264
|
|
|
|
|
|
|
Returns the output dimensions of the map as passed in at constructor time. |
265
|
|
|
|
|
|
|
|
266
|
|
|
|
|
|
|
=cut |
267
|
|
|
|
|
|
|
|
268
|
|
|
|
|
|
|
sub output_dim { |
269
|
2
|
|
|
2
|
1
|
5
|
my $self = shift; |
270
|
2
|
|
|
|
|
9
|
return $self->{output_dim}; |
271
|
|
|
|
|
|
|
} |
272
|
|
|
|
|
|
|
|
273
|
|
|
|
|
|
|
=pod |
274
|
|
|
|
|
|
|
|
275
|
|
|
|
|
|
|
=item I (read-only) |
276
|
|
|
|
|
|
|
|
277
|
|
|
|
|
|
|
I<$radius> = I<$nn>->radius |
278
|
|
|
|
|
|
|
|
279
|
|
|
|
|
|
|
Returns the I of the map. Different topologies interpret this differently. |
280
|
|
|
|
|
|
|
|
281
|
|
|
|
|
|
|
=item I |
282
|
|
|
|
|
|
|
|
283
|
|
|
|
|
|
|
I<$m> = I<$nn>->map |
284
|
|
|
|
|
|
|
|
285
|
|
|
|
|
|
|
This method returns a reference to the map data. See the appropriate subclass of the data |
286
|
|
|
|
|
|
|
representation. |
287
|
|
|
|
|
|
|
|
288
|
|
|
|
|
|
|
=cut |
289
|
|
|
|
|
|
|
|
290
|
|
|
|
|
|
|
sub map { |
291
|
6
|
|
|
6
|
1
|
2586
|
my $self = shift; |
292
|
6
|
|
|
|
|
31
|
return $self->{map}; |
293
|
|
|
|
|
|
|
} |
294
|
|
|
|
|
|
|
|
295
|
|
|
|
|
|
|
=pod |
296
|
|
|
|
|
|
|
|
297
|
|
|
|
|
|
|
=item I |
298
|
|
|
|
|
|
|
|
299
|
|
|
|
|
|
|
I<$val> = I<$nn>->value (I<$x>, I<$y>) |
300
|
|
|
|
|
|
|
|
301
|
|
|
|
|
|
|
I<$nn>->value (I<$x>, I<$y>, I<$val>) |
302
|
|
|
|
|
|
|
|
303
|
|
|
|
|
|
|
Set or get the current vector value for a particular neuron. The neuron is addressed via its |
304
|
|
|
|
|
|
|
coordinates. |
305
|
|
|
|
|
|
|
|
306
|
|
|
|
|
|
|
=cut |
307
|
|
|
|
|
|
|
|
308
|
|
|
|
|
|
|
sub value { |
309
|
45
|
|
|
45
|
1
|
14904
|
my $self = shift; |
310
|
45
|
|
|
|
|
82
|
my ($x, $y) = (shift, shift); |
311
|
45
|
|
|
|
|
52
|
my $v = shift; |
312
|
45
|
100
|
|
|
|
216
|
return defined $v ? $self->{map}->[$x]->[$y] = $v : $self->{map}->[$x]->[$y]; |
313
|
|
|
|
|
|
|
} |
314
|
|
|
|
|
|
|
|
315
|
|
|
|
|
|
|
=pod |
316
|
|
|
|
|
|
|
|
317
|
|
|
|
|
|
|
=item I |
318
|
|
|
|
|
|
|
|
319
|
|
|
|
|
|
|
I<$label> = I<$nn>->label (I<$x>, I<$y>) |
320
|
|
|
|
|
|
|
|
321
|
|
|
|
|
|
|
I<$nn>->label (I<$x>, I<$y>, I<$label>) |
322
|
|
|
|
|
|
|
|
323
|
|
|
|
|
|
|
Set or get the label for a particular neuron. The neuron is addressed via its coordinates. |
324
|
|
|
|
|
|
|
The label can be anything, it is just attached to the position. |
325
|
|
|
|
|
|
|
|
326
|
|
|
|
|
|
|
=cut |
327
|
|
|
|
|
|
|
|
328
|
|
|
|
|
|
|
sub label { |
329
|
3
|
|
|
3
|
1
|
869
|
my $self = shift; |
330
|
3
|
|
|
|
|
5
|
my ($x, $y) = (shift, shift); |
331
|
3
|
|
|
|
|
4
|
my $l = shift; |
332
|
3
|
100
|
|
|
|
18
|
return defined $l ? $self->{labels}->[$x]->[$y] = $l : $self->{labels}->[$x]->[$y]; |
333
|
|
|
|
|
|
|
} |
334
|
|
|
|
|
|
|
|
335
|
|
|
|
|
|
|
=pod |
336
|
|
|
|
|
|
|
|
337
|
|
|
|
|
|
|
=item I |
338
|
|
|
|
|
|
|
|
339
|
|
|
|
|
|
|
print I<$nn>->as_string |
340
|
|
|
|
|
|
|
|
341
|
|
|
|
|
|
|
This methods creates a pretty-print version of the current vectors. |
342
|
|
|
|
|
|
|
|
343
|
|
|
|
|
|
|
=cut |
344
|
|
|
|
|
|
|
|
345
|
0
|
|
|
0
|
1
|
|
sub as_string { die; } |
346
|
|
|
|
|
|
|
|
347
|
|
|
|
|
|
|
=pod |
348
|
|
|
|
|
|
|
|
349
|
|
|
|
|
|
|
=item I |
350
|
|
|
|
|
|
|
|
351
|
|
|
|
|
|
|
print I<$nn>->as_data |
352
|
|
|
|
|
|
|
|
353
|
|
|
|
|
|
|
This methods creates a string containing the raw vector data, row by |
354
|
|
|
|
|
|
|
row. This can be fed into gnuplot, for instance. |
355
|
|
|
|
|
|
|
|
356
|
|
|
|
|
|
|
=cut |
357
|
|
|
|
|
|
|
|
358
|
0
|
|
|
0
|
1
|
|
sub as_data { die; } |
359
|
|
|
|
|
|
|
|
360
|
|
|
|
|
|
|
=pod |
361
|
|
|
|
|
|
|
|
362
|
|
|
|
|
|
|
=back |
363
|
|
|
|
|
|
|
|
364
|
|
|
|
|
|
|
=head1 HOWTOs |
365
|
|
|
|
|
|
|
|
366
|
|
|
|
|
|
|
=over |
367
|
|
|
|
|
|
|
|
368
|
|
|
|
|
|
|
=item I |
369
|
|
|
|
|
|
|
|
370
|
|
|
|
|
|
|
See the example script in the directory C provided in the |
371
|
|
|
|
|
|
|
distribution. It uses L (for speed and scalability, but the |
372
|
|
|
|
|
|
|
results are not as good as I had thought). |
373
|
|
|
|
|
|
|
|
374
|
|
|
|
|
|
|
=item I |
375
|
|
|
|
|
|
|
|
376
|
|
|
|
|
|
|
See the example script in the directory C. It uses |
377
|
|
|
|
|
|
|
C to directly dump the data structure onto disk. Storage and |
378
|
|
|
|
|
|
|
retrieval is quite fast. |
379
|
|
|
|
|
|
|
|
380
|
|
|
|
|
|
|
=back |
381
|
|
|
|
|
|
|
|
382
|
|
|
|
|
|
|
=head1 FAQs |
383
|
|
|
|
|
|
|
|
384
|
|
|
|
|
|
|
=over |
385
|
|
|
|
|
|
|
|
386
|
|
|
|
|
|
|
=item I |
387
|
|
|
|
|
|
|
|
388
|
|
|
|
|
|
|
There is most likely something wrong with the C you |
389
|
|
|
|
|
|
|
specified and your vectors should be having. |
390
|
|
|
|
|
|
|
|
391
|
|
|
|
|
|
|
=back |
392
|
|
|
|
|
|
|
|
393
|
|
|
|
|
|
|
=head1 TODOs |
394
|
|
|
|
|
|
|
|
395
|
|
|
|
|
|
|
=over |
396
|
|
|
|
|
|
|
|
397
|
|
|
|
|
|
|
=item maybe implement the SOM on top of PDL? |
398
|
|
|
|
|
|
|
|
399
|
|
|
|
|
|
|
=item provide a ::SOM::Compat to have compatibility with the original AI::NeuralNet::SOM? |
400
|
|
|
|
|
|
|
|
401
|
|
|
|
|
|
|
=item implement different window forms (bubble/gaussian), linear/random |
402
|
|
|
|
|
|
|
|
403
|
|
|
|
|
|
|
=item implement the format mentioned in the original AI::NeuralNet::SOM |
404
|
|
|
|
|
|
|
|
405
|
|
|
|
|
|
|
=item add methods as_html to individual topologies |
406
|
|
|
|
|
|
|
|
407
|
|
|
|
|
|
|
=item add iterators through vector lists for I and I |
408
|
|
|
|
|
|
|
|
409
|
|
|
|
|
|
|
=back |
410
|
|
|
|
|
|
|
|
411
|
|
|
|
|
|
|
=head1 SUPPORT |
412
|
|
|
|
|
|
|
|
413
|
|
|
|
|
|
|
Bugs should always be submitted via the CPAN bug tracker |
414
|
|
|
|
|
|
|
L |
415
|
|
|
|
|
|
|
|
416
|
|
|
|
|
|
|
=head1 SEE ALSO |
417
|
|
|
|
|
|
|
|
418
|
|
|
|
|
|
|
Explanation of the algorithm: |
419
|
|
|
|
|
|
|
|
420
|
|
|
|
|
|
|
L |
421
|
|
|
|
|
|
|
|
422
|
|
|
|
|
|
|
Old version of AI::NeuralNet::SOM from Alexander Voischev: |
423
|
|
|
|
|
|
|
|
424
|
|
|
|
|
|
|
L |
425
|
|
|
|
|
|
|
|
426
|
|
|
|
|
|
|
Subclasses: |
427
|
|
|
|
|
|
|
|
428
|
|
|
|
|
|
|
L |
429
|
|
|
|
|
|
|
L |
430
|
|
|
|
|
|
|
L |
431
|
|
|
|
|
|
|
|
432
|
|
|
|
|
|
|
|
433
|
|
|
|
|
|
|
=head1 AUTHOR |
434
|
|
|
|
|
|
|
|
435
|
|
|
|
|
|
|
Robert Barta, Erho@devc.atE |
436
|
|
|
|
|
|
|
|
437
|
|
|
|
|
|
|
=head1 COPYRIGHT AND LICENSE |
438
|
|
|
|
|
|
|
|
439
|
|
|
|
|
|
|
Copyright (C) 200[78] by Robert Barta |
440
|
|
|
|
|
|
|
|
441
|
|
|
|
|
|
|
This library is free software; you can redistribute it and/or modify |
442
|
|
|
|
|
|
|
it under the same terms as Perl itself, either Perl version 5.8.8 or, |
443
|
|
|
|
|
|
|
at your option, any later version of Perl 5 you may have available. |
444
|
|
|
|
|
|
|
|
445
|
|
|
|
|
|
|
=cut |
446
|
|
|
|
|
|
|
|
447
|
|
|
|
|
|
|
our $VERSION = '0.07'; |
448
|
|
|
|
|
|
|
|
449
|
|
|
|
|
|
|
1; |
450
|
|
|
|
|
|
|
|
451
|
|
|
|
|
|
|
__END__ |