Mercurial > hg > camir-aes2014

annotate toolboxes/MIRtoolbox1.3.2/somtoolbox/som_sompaktrain.m @ 0:e9a9cd732c1e tip

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
first hg version after svn
author wolffd
date Tue, 10 Feb 2015 15:05:51 +0000
parents
children
rev   line source
wolffd@0 1 function [sMap, sTrain] = som_sompaktrain(sMap, D, varargin)
wolffd@0 2
wolffd@0 3 %SOM_SOMPAKTRAIN Use SOM_PAK to train the Self-Organizing Map.
wolffd@0 4 %
wolffd@0 5 % [sM,sT] = som_sompaktrain(sM, D, [[argID,] value, ...])
wolffd@0 6 %
wolffd@0 7 % sM = som_sompaktrain(sM,D);
wolffd@0 8 % sM = som_sompaktrain(sM,sD,'alpha_type','inv');
wolffd@0 9 % [M,sT] = som_sompaktrain(M,D,'bubble','trainlen',10,'inv','hexa');
wolffd@0 10 %
wolffd@0 11 % Input and output arguments ([]'s are optional):
wolffd@0 12 % sM (struct) map struct, the trained and updated map is returned
wolffd@0 13 % (matrix) codebook matrix of a self-organizing map
wolffd@0 14 % size munits x dim or msize(1) x ... x msize(k) x dim
wolffd@0 15 % The trained map codebook is returned.
wolffd@0 16 % D (struct) training data; data struct
wolffd@0 17 % (matrix) training data, size dlen x dim
wolffd@0 18 % (string) name of data file
wolffd@0 19 % [argID, (string) See below. The values which are unambiguous can
wolffd@0 20 % value] (varies) be given without the preceeding argID.
wolffd@0 21 %
wolffd@0 22 % sT (struct) learning parameters used during the training
wolffd@0 23 %
wolffd@0 24 % Here are the valid argument IDs and corresponding values. The values which
wolffd@0 25 % are unambiguous (marked with '*') can be given without the preceeding argID.
wolffd@0 26 % 'msize' (vector) map size
wolffd@0 27 % 'radius_ini' (scalar) neighborhood radius
wolffd@0 28 % 'radius' = 'radius_ini'
wolffd@0 29 % 'alpha_ini' (scalar) initial learning rate
wolffd@0 30 % 'alpha' = 'alpha_ini'
wolffd@0 31 % 'trainlen' (scalar) training length
wolffd@0 32 % 'seed' (scalar) seed for random number generator
wolffd@0 33 % 'snapfile' (string) base name for snapshot files
wolffd@0 34 % 'snapinterval' (scalar) snapshot interval
wolffd@0 35 % 'tlen_type' *(string) is the given trainlen 'samples' or 'epochs'
wolffd@0 36 % 'train' *(struct) train struct, parameters for training
wolffd@0 37 % 'sTrain','som_train' = 'train'
wolffd@0 38 % 'alpha_type' *(string) learning rate function, 'inv' or 'linear'
wolffd@0 39 % 'neigh' *(string) neighborhood function, 'gaussian' or 'bubble'
wolffd@0 40 % 'topol' *(struct) topology struct
wolffd@0 41 % 'som_topol','sTopol' = 'topol'
wolffd@0 42 % 'lattice' *(string) map lattice, 'hexa' or 'rect'
wolffd@0 43 %
wolffd@0 44 % For more help, try 'type som_sompaktrain' or check out online documentation.
wolffd@0 45 % See also SOM_MAKE, SOM_SEQTRAIN, SOM_BATCHTRAIN, SOM_TRAIN_STRUCT.
wolffd@0 46
wolffd@0 47 %%%%%%%%%%%%% DETAILED DESCRIPTION %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0 48 %
wolffd@0 49 % som_sompaktrain
wolffd@0 50 %
wolffd@0 51 % PURPOSE
wolffd@0 52 %
wolffd@0 53 % Use SOM_PAK to train the Self-Organizing Map.
wolffd@0 54 %
wolffd@0 55 % SYNTAX
wolffd@0 56 %
wolffd@0 57 % sM = som_sompaktrain(sM,D);
wolffd@0 58 % sM = som_sompaktrain(sM,sD);
wolffd@0 59 % sM = som_sompaktrain(...,'argID',value,...);
wolffd@0 60 % sM = som_sompaktrain(...,value,...);
wolffd@0 61 % [sM,sT] = som_sompaktrain(M,D,...);
wolffd@0 62 %
wolffd@0 63 % DESCRIPTION
wolffd@0 64 %
wolffd@0 65 % Trains the given SOM (sM or M above) with the given training data (sD or
wolffd@0 66 % D) using SOM_PAK. If no optional arguments (argID, value) are
wolffd@0 67 % given, a default training is done, the parameters are obtained from
wolffd@0 68 % SOM_TRAIN_STRUCT function. Using optional arguments the training
wolffd@0 69 % parameters can be specified. Returns the trained and updated SOM and a
wolffd@0 70 % train struct which contains information on the training.
wolffd@0 71 %
wolffd@0 72 % Notice that the SOM_PAK program 'vsom' must be in the search path of your
wolffd@0 73 % shell. Alternatively, you can set a variable 'SOM_PAKDIR' in the Matlab
wolffd@0 74 % workspace to tell the som_sompaktrain where to find the 'vsom' program.
wolffd@0 75 %
wolffd@0 76 % Notice also that many of the training parameters are much more limited in
wolffd@0 77 % values than when using SOM Toolbox function for training:
wolffd@0 78 % - the map shape is always 'sheet'
wolffd@0 79 % - only initial value for neighborhood radius can be given
wolffd@0 80 % - neighborhood function can only be 'bubble' or 'gaussian'
wolffd@0 81 % - only initial value for learning rate can be given
wolffd@0 82 % - learning rate can only be 'linear' or 'inv'
wolffd@0 83 % - mask cannot be used: all variables are always used in BMU search
wolffd@0 84 % Any parameters not confirming to these restrictions will be converted
wolffd@0 85 % so that they do before training. On the other hand, there are some
wolffd@0 86 % additional options that are not present in the SOM Toolbox:
wolffd@0 87 % - random seed
wolffd@0 88 % - snapshot file and interval
wolffd@0 89 %
wolffd@0 90 % REQUIRED INPUT ARGUMENTS
wolffd@0 91 %
wolffd@0 92 % sM The map to be trained.
wolffd@0 93 % (struct) map struct
wolffd@0 94 % (matrix) codebook matrix (field .data of map struct)
wolffd@0 95 % Size is either [munits dim], in which case the map grid
wolffd@0 96 % dimensions (msize) should be specified with optional arguments,
wolffd@0 97 % or [msize(1) ... msize(k) dim] in which case the map
wolffd@0 98 % grid dimensions are taken from the size of the matrix.
wolffd@0 99 % Lattice, by default, is 'rect' and shape 'sheet'.
wolffd@0 100 % D Training data.
wolffd@0 101 % (struct) data struct
wolffd@0 102 % (matrix) data matrix, size [dlen dim]
wolffd@0 103 % (string) name of data file
wolffd@0 104 %
wolffd@0 105 % OPTIONAL INPUT ARGUMENTS
wolffd@0 106 %
wolffd@0 107 % argID (string) Argument identifier string (see below).
wolffd@0 108 % value (varies) Value for the argument (see below).
wolffd@0 109 %
wolffd@0 110 % The optional arguments can be given as 'argID',value -pairs. If an
wolffd@0 111 % argument is given value multiple times, the last one is
wolffd@0 112 % used. The valid IDs and corresponding values are listed below. The values
wolffd@0 113 % which are unambiguous (marked with '*') can be given without the
wolffd@0 114 % preceeding argID.
wolffd@0 115 %
wolffd@0 116 % 'msize' (vector) map grid dimensions. Default is the one
wolffd@0 117 % in sM (field sM.topol.msize) or
wolffd@0 118 % 'si = size(sM); msize = si(1:end-1);'
wolffd@0 119 % if only a codebook matrix was given.
wolffd@0 120 % 'radius_ini' (scalar) initial neighborhood radius
wolffd@0 121 % 'radius' (scalar) = 'radius_ini'
wolffd@0 122 % 'alpha_ini' (vector) initial learning rate
wolffd@0 123 % 'alpha' (scalar) = 'alpha_ini'
wolffd@0 124 % 'trainlen' (scalar) training length (see also 'tlen_type')
wolffd@0 125 % 'seed' (scalar) seed for random number generator
wolffd@0 126 % 'snapfile' (string) base name for snapshot files
wolffd@0 127 % 'snapinterval' (scalar) snapshot interval
wolffd@0 128 % 'tlen_type' *(string) is the trainlen argument given in 'epochs' or
wolffd@0 129 % in 'samples'. Default is 'epochs'.
wolffd@0 130 % 'train' *(struct) train struct, parameters for training.
wolffd@0 131 % Default parameters, unless specified,
wolffd@0 132 % are acquired using SOM_TRAIN_STRUCT (this
wolffd@0 133 % also applies for 'trainlen', 'alpha_type',
wolffd@0 134 % 'alpha_ini', 'radius_ini' and 'radius_fin').
wolffd@0 135 % 'sTrain', 'som_topol' (struct) = 'train'
wolffd@0 136 % 'neigh' *(string) The used neighborhood function. Default is
wolffd@0 137 % the one in sM (field '.neigh') or 'gaussian'
wolffd@0 138 % if only a codebook matrix was given. The other
wolffd@0 139 % possible value is 'bubble'.
wolffd@0 140 % 'topol' *(struct) topology of the map. Default is the one
wolffd@0 141 % in sM (field '.topol').
wolffd@0 142 % 'sTopol', 'som_topol' (struct) = 'topol'
wolffd@0 143 % 'alpha_type' *(string) learning rate function, 'inv' or 'linear'
wolffd@0 144 % 'lattice' *(string) map lattice. Default is the one in sM
wolffd@0 145 % (field sM.topol.lattice) or 'rect'
wolffd@0 146 % if only a codebook matrix was given.
wolffd@0 147 %
wolffd@0 148 % OUTPUT ARGUMENTS
wolffd@0 149 %
wolffd@0 150 % sM the trained map
wolffd@0 151 % (struct) if a map struct was given as input argument, a
wolffd@0 152 % map struct is also returned. The current training
wolffd@0 153 % is added to the training history (sM.trainhist).
wolffd@0 154 % The 'neigh' and 'mask' fields of the map struct
wolffd@0 155 % are updated to match those of the training.
wolffd@0 156 % (matrix) if a matrix was given as input argument, a matrix
wolffd@0 157 % is also returned with the same size as the input
wolffd@0 158 % argument.
wolffd@0 159 % sT (struct) train struct; information of the accomplished training
wolffd@0 160 %
wolffd@0 161 % EXAMPLES
wolffd@0 162 %
wolffd@0 163 % Simplest case:
wolffd@0 164 % sM = som_sompaktrain(sM,D);
wolffd@0 165 % sM = som_sompaktrain(sM,sD);
wolffd@0 166 %
wolffd@0 167 % The change training parameters, the optional arguments 'train',
wolffd@0 168 % 'neigh','mask','trainlen','radius','radius_ini', 'alpha',
wolffd@0 169 % 'alpha_type' and 'alpha_ini' are used.
wolffd@0 170 % sM = som_sompaktrain(sM,D,'bubble','trainlen',10,'radius_ini',3);
wolffd@0 171 %
wolffd@0 172 % Another way to specify training parameters is to create a train struct:
wolffd@0 173 % sTrain = som_train_struct(sM,'dlen',size(D,1),'algorithm','seq');
wolffd@0 174 % sTrain = som_set(sTrain,'neigh','gaussian');
wolffd@0 175 % sM = som_sompaktrain(sM,D,sTrain);
wolffd@0 176 %
wolffd@0 177 % You don't necessarily have to use the map struct, but you can operate
wolffd@0 178 % directly with codebook matrices. However, in this case you have to
wolffd@0 179 % specify the topology of the map in the optional arguments. The
wolffd@0 180 % following commads are identical (M is originally a 200 x dim sized matrix):
wolffd@0 181 % M = som_sompaktrain(M,D,'msize',[20 10],'lattice','hexa');
wolffd@0 182 %
wolffd@0 183 % M = som_sompaktrain(M,D,'msize',[20 10],'hexa');
wolffd@0 184 %
wolffd@0 185 % sT= som_set('som_topol','msize',[20 10],'lattice','hexa');
wolffd@0 186 % M = som_sompaktrain(M,D,sT);
wolffd@0 187 %
wolffd@0 188 % M = reshape(M,[20 10 dim]);
wolffd@0 189 % M = som_sompaktrain(M,D,'hexa');
wolffd@0 190 %
wolffd@0 191 % The som_sompaktrain also returns a train struct with information on the
wolffd@0 192 % accomplished training. This is the same one as is added to the end of the
wolffd@0 193 % trainhist field of map struct, in case a map struct is given.
wolffd@0 194 % [M,sTrain] = som_sompaktrain(M,D,'msize',[20 10]);
wolffd@0 195 %
wolffd@0 196 % [sM,sTrain] = som_sompaktrain(sM,D); % sM.trainhist(end)==sTrain
wolffd@0 197 %
wolffd@0 198 % SEE ALSO
wolffd@0 199 %
wolffd@0 200 % som_make Initialize and train a SOM using default parameters.
wolffd@0 201 % som_seqtrain Train SOM with sequential algorithm.
wolffd@0 202 % som_batchtrain Train SOM with batch algorithm.
wolffd@0 203 % som_train_struct Determine default training parameters.
wolffd@0 204
wolffd@0 205 % Copyright (c) 1999-2000 by the SOM toolbox programming team.
wolffd@0 206 % http://www.cis.hut.fi/projects/somtoolbox/
wolffd@0 207
wolffd@0 208 % Version 2.0beta juuso 151199
wolffd@0 209
wolffd@0 210 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0 211 %% Check arguments
wolffd@0 212
wolffd@0 213 error(nargchk(2, Inf, nargin)); % check the number of input arguments
wolffd@0 214
wolffd@0 215 % map
wolffd@0 216 struct_mode = isstruct(sMap);
wolffd@0 217 if struct_mode,
wolffd@0 218 sTopol = sMap.topol;
wolffd@0 219 else
wolffd@0 220 orig_size = size(sMap);
wolffd@0 221 if ndims(sMap) > 2,
wolffd@0 222 si = size(sMap); dim = si(end); msize = si(1:end-1);
wolffd@0 223 M = reshape(sMap,[prod(msize) dim]);
wolffd@0 224 else
wolffd@0 225 msize = [orig_size(1) 1];
wolffd@0 226 dim = orig_size(2);
wolffd@0 227 end
wolffd@0 228 sMap = som_map_struct(dim,'msize',msize);
wolffd@0 229 sTopol = sMap.topol;
wolffd@0 230 end
wolffd@0 231 [munits dim] = size(sMap.codebook);
wolffd@0 232
wolffd@0 233 % data
wolffd@0 234 givendatafile = '';
wolffd@0 235 if ischar(D),
wolffd@0 236 data_name = D;
wolffd@0 237 givendatafile = D;
wolffd@0 238 D = [];
wolffd@0 239 dlen = NaN;
wolffd@0 240 else
wolffd@0 241 if isstruct(D),
wolffd@0 242 data_name = D.name;
wolffd@0 243 D = D.data;
wolffd@0 244 else
wolffd@0 245 data_name = inputname(2);
wolffd@0 246 end
wolffd@0 247 D = D(find(sum(isnan(D),2) < dim),:); % remove empty vectors from the data
wolffd@0 248 [dlen ddim] = size(D); % check input dimension
wolffd@0 249 if ddim ~= dim, error('Map and data dimensions must agree.'); end
wolffd@0 250 end
wolffd@0 251
wolffd@0 252 % varargin
wolffd@0 253 sTrain = som_set('som_train','algorithm','seq',...
wolffd@0 254 'neigh',sMap.neigh,...
wolffd@0 255 'mask',ones(dim,1),...
wolffd@0 256 'data_name',data_name);
wolffd@0 257 tlen_type = 'epochs';
wolffd@0 258 random_seed = 0;
wolffd@0 259 snapshotname = '';
wolffd@0 260 snapshotinterval = 0;
wolffd@0 261
wolffd@0 262 i=1;
wolffd@0 263 while i<=length(varargin),
wolffd@0 264 argok = 1;
wolffd@0 265 if ischar(varargin{i}),
wolffd@0 266 switch varargin{i},
wolffd@0 267 % argument IDs
wolffd@0 268 case 'msize', i=i+1; sTopol.msize = varargin{i};
wolffd@0 269 case 'lattice', i=i+1; sTopol.lattice = varargin{i};
wolffd@0 270 case 'neigh', i=i+1; sTrain.neigh = varargin{i};
wolffd@0 271 case 'trainlen', i=i+1; sTrain.trainlen = varargin{i};
wolffd@0 272 case 'tlen_type', i=i+1; tlen_type = varargin{i};
wolffd@0 273 case 'radius_ini', i=i+1; sTrain.radius_ini = varargin{i};
wolffd@0 274 case 'radius', i=i+1; sTrain.radius_ini = varargin{i}(1);
wolffd@0 275 case 'alpha_type', i=i+1; sTrain.alpha_type = varargin{i};
wolffd@0 276 case 'alpha_ini', i=i+1; sTrain.alpha_ini = varargin{i};
wolffd@0 277 case 'alpha', i=i+1; sTrain.alpha_ini = varargin{i}(1);
wolffd@0 278 case 'seed', i=i+1; random_seed = varargin{i};
wolffd@0 279 case 'snapshotname',i=i+1; snapshotname = varargin{i};
wolffd@0 280 case 'snapshotinterval',i=i+1; snapshotinterval = varargin{i};
wolffd@0 281 case {'sTrain','train','som_train'}, i=i+1; sTrain = varargin{i};
wolffd@0 282 case {'topol','sTopol','som_topol'},
wolffd@0 283 i=i+1;
wolffd@0 284 sTopol = varargin{i};
wolffd@0 285 if prod(sTopol.msize) ~= munits,
wolffd@0 286 error('Given map grid size does not match the codebook size.');
wolffd@0 287 end
wolffd@0 288 % unambiguous values
wolffd@0 289 case {'inv','linear'}, sTrain.alpha_type = varargin{i};
wolffd@0 290 case {'hexa','rect'}, sTopol.lattice = varargin{i};
wolffd@0 291 case {'gaussian','bubble'}, sTrain.neigh = varargin{i};
wolffd@0 292 case {'epochs','samples'}, tlen_type = varargin{i};
wolffd@0 293 otherwise argok=0;
wolffd@0 294 end
wolffd@0 295 elseif isstruct(varargin{i}) & isfield(varargin{i},'type'),
wolffd@0 296 switch varargin{i}(1).type,
wolffd@0 297 case 'som_topol',
wolffd@0 298 sTopol = varargin{i};
wolffd@0 299 if prod(sTopol.msize) ~= munits,
wolffd@0 300 error('Given map grid size does not match the codebook size.');
wolffd@0 301 end
wolffd@0 302 case 'som_train', sTrain = varargin{i};
wolffd@0 303 otherwise argok=0;
wolffd@0 304 end
wolffd@0 305 else
wolffd@0 306 argok = 0;
wolffd@0 307 end
wolffd@0 308 if ~argok,
wolffd@0 309 disp(['(som_sompaktrain) Ignoring invalid argument #' num2str(i+2)]);
wolffd@0 310 end
wolffd@0 311 i = i+1;
wolffd@0 312 end
wolffd@0 313
wolffd@0 314 % check topology
wolffd@0 315 if struct_mode,
wolffd@0 316 if ~strcmp(sTopol.lattice,sMap.topol.lattice) | ...
wolffd@0 317 ~strcmp(sTopol.shape,sMap.topol.shape) | ...
wolffd@0 318 any(sTopol.msize ~= sMap.topol.msize),
wolffd@0 319 warning('Changing the original map topology.');
wolffd@0 320 end
wolffd@0 321 end
wolffd@0 322 sMap.topol = sTopol;
wolffd@0 323
wolffd@0 324 % complement the training struct
wolffd@0 325 if ~isnan(dlen),
wolffd@0 326 sTrain = som_train_struct(sTrain,sMap,'dlen',dlen);
wolffd@0 327 else
wolffd@0 328 sTrain = som_train_struct(sTrain,sMap);
wolffd@0 329 end
wolffd@0 330 if isempty(sTrain.mask), sTrain.mask = ones(dim,1); end
wolffd@0 331
wolffd@0 332 % training length
wolffd@0 333 if strcmp(tlen_type,'epochs'),
wolffd@0 334 if isnan(dlen),
wolffd@0 335 error('Training length given as epochs, but data length is not known.\n');
wolffd@0 336 else
wolffd@0 337 rlen = sTrain.trainlen*dlen;
wolffd@0 338 end
wolffd@0 339 else
wolffd@0 340 rlen = sTrain.trainlen;
wolffd@0 341 sTrain.trainlen = sTrain.trainlen/dlen;
wolffd@0 342 end
wolffd@0 343
wolffd@0 344 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0 345 %% check arguments
wolffd@0 346
wolffd@0 347 % mask
wolffd@0 348 if any(sTrain.mask~=1),
wolffd@0 349 sTrain.mask = ones(dim,1);
wolffd@0 350 fprintf(1,'Ignoring given mask.\n');
wolffd@0 351 end
wolffd@0 352
wolffd@0 353 % learning rate
wolffd@0 354 if strcmp(sTrain.alpha_type,'power'),
wolffd@0 355 sTrain.alpha_type = 'inv';
wolffd@0 356 fprintf(1,'Using ''inv'' learning rate type instead of ''power''\n');
wolffd@0 357 end
wolffd@0 358
wolffd@0 359 % neighborhood
wolffd@0 360 if any(strcmp(sTrain.neigh,{'cutgauss','ep'})),
wolffd@0 361 fprintf(1,'Using ''gaussian'' neighborhood function instead of %s.\n',sTrain.neigh);
wolffd@0 362 sTrain.neigh = 'gaussian';
wolffd@0 363 end
wolffd@0 364
wolffd@0 365 % map shape
wolffd@0 366 if ~strcmp(sMap.topol.shape,'sheet'),
wolffd@0 367 fprintf(1,'Using ''sheet'' map shape of %s.\n',sMap.topol.shape);
wolffd@0 368 sMap.topol.shape = 'sheet';
wolffd@0 369 end
wolffd@0 370
wolffd@0 371 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0 372 %% Action
wolffd@0 373
wolffd@0 374 % write files
wolffd@0 375 if ~isempty(givendatafile),
wolffd@0 376 temp_din = givendatafile;
wolffd@0 377 else
wolffd@0 378 temp_din = tempname;
wolffd@0 379 som_write_data(D, temp_din, 'x')
wolffd@0 380 end
wolffd@0 381 temp_cin = tempname;
wolffd@0 382 som_write_cod(sMap, temp_cin)
wolffd@0 383 temp_cout = tempname;
wolffd@0 384
wolffd@0 385 % check if the environment variable 'SOM_PAKDIR' has been defined
wolffd@0 386 if any(strcmp('SOM_PAKDIR', evalin('base', 'who')))
wolffd@0 387 som_pak_dir = evalin('base', 'SOM_PAKDIR');
wolffd@0 388 else
wolffd@0 389 som_pak_dir = '';
wolffd@0 390 end
wolffd@0 391 if ~isempty(som_pak_dir) & ~strncmp(som_pak_dir(end), '/', 1)
wolffd@0 392 som_pak_dir(end + 1) = '/';
wolffd@0 393 end
wolffd@0 394
wolffd@0 395 aini = sTrain.alpha_ini;
wolffd@0 396 atype = sTrain.alpha_type;
wolffd@0 397 if strcmp(atype,'inv'), atype = 'inverse_t'; end
wolffd@0 398 rad = sTrain.radius_ini;
wolffd@0 399 str = [som_pak_dir 'vsom ' ...
wolffd@0 400 sprintf('-cin %s -din %s -cout %s', temp_cin, temp_din, temp_cout) ...
wolffd@0 401 sprintf(' -rlen %d -alpha %g -alpha_type %s', rlen, aini, atype) ...
wolffd@0 402 sprintf(' -radius %g -rand %g ',rad,random_seed)];
wolffd@0 403 if ~isempty(snapshotname) & snapinterval>0,
wolffd@0 404 str = [str, sprintf(' -snapfile %s -snapinterval %d',snapshotname,snapshotinterval)];
wolffd@0 405 end
wolffd@0 406
wolffd@0 407 fprintf(1,'Execute: %s\n',str);
wolffd@0 408 if isunix,
wolffd@0 409 [status,w] = unix(str);
wolffd@0 410 if status, fprintf(1,'Execution failed.\n'); end
wolffd@0 411 if ~isempty(w), fprintf(1,'%s\n',w); end
wolffd@0 412 else
wolffd@0 413 [status,w] = dos(str);
wolffd@0 414 if status, fprintf(1,'Execution failed.\n'); end
wolffd@0 415 if ~isempty(w), fprintf(1,'%s\n',w); end
wolffd@0 416 end
wolffd@0 417
wolffd@0 418 sMap_temp = som_read_cod(temp_cout);
wolffd@0 419 M = sMap_temp.codebook;
wolffd@0 420
wolffd@0 421 if isunix
wolffd@0 422 unix(['/bin/rm -f ' temp_din ' ' temp_cin ' ' temp_cout]);
wolffd@0 423 else
wolffd@0 424 dos(['del ' temp_din ' ' temp_cin ' ' temp_cout]);
wolffd@0 425 end
wolffd@0 426
wolffd@0 427 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0 428 %% Build / clean up the return arguments
wolffd@0 429
wolffd@0 430 % update structures
wolffd@0 431 sTrain = som_set(sTrain,'time',datestr(now,0));
wolffd@0 432 if struct_mode,
wolffd@0 433 sMap = som_set(sMap,'codebook',M,'mask',sTrain.mask,'neigh',sTrain.neigh);
wolffd@0 434 tl = length(sMap.trainhist);
wolffd@0 435 sMap.trainhist(tl+1) = sTrain;
wolffd@0 436 else
wolffd@0 437 sMap = reshape(M,orig_size);
wolffd@0 438 end
wolffd@0 439
wolffd@0 440 return;
wolffd@0 441
wolffd@0 442 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0 443