Mercurial > hg > camir-aes2014

annotate toolboxes/MIRtoolbox1.3.2/somtoolbox/som_supervised.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 sM = som_supervised(sData,varargin)
wolffd@0 2
wolffd@0 3 %SOM_SUPERVISED SOM training which utilizes class information.
wolffd@0 4 %
wolffd@0 5 % sM = som_supervised(sData, [ArgID, value,...]))
wolffd@0 6 %
wolffd@0 7 % Input and output arguments ([]'s are optional)
wolffd@0 8 % sData (struct) data struct, the class information is
wolffd@0 9 % taken from the first column of .labels field
wolffd@0 10 % [argID, (string) See below. These are given as
wolffd@0 11 % value] (varies) 'argID', value -pairs.
wolffd@0 12 %
wolffd@0 13 % sMap (struct) map struct
wolffd@0 14 %
wolffd@0 15 % Here are the argument IDs and corresponding values:
wolffd@0 16 % 'munits' (scalar) the preferred number of map units
wolffd@0 17 % 'msize' (vector) map grid size
wolffd@0 18 % 'mask' (vector) BMU search mask, size dim x 1
wolffd@0 19 % 'name' (string) map name
wolffd@0 20 % 'comp_names' (string array / cellstr) component names, size dim x 1
wolffd@0 21 % 'tracking' (scalar) how much to report, default = 1
wolffd@0 22 % The following values are unambiguous and can therefore
wolffd@0 23 % be given without the preceeding argument ID:
wolffd@0 24 % 'algorithm' (string) training algorithm: 'seq' or 'batch'
wolffd@0 25 % 'mapsize' (string) do you want a 'small', 'normal' or 'big' map
wolffd@0 26 % Any explicit settings of munits or msize override this.
wolffd@0 27 % 'topol' (struct) topology struct
wolffd@0 28 % 'som_topol','sTopol' = 'topol'
wolffd@0 29 % 'lattice' (string) map lattice, 'hexa' or 'rect'
wolffd@0 30 % 'shape' (string) map shape, 'sheet', 'cyl' or 'toroid'
wolffd@0 31 % 'neigh' (string) neighborhood function, 'gaussian', 'cutgauss',
wolffd@0 32 % 'ep' or 'bubble'
wolffd@0 33 %
wolffd@0 34 % For more help, try 'type som_supervised', or check out online documentation.
wolffd@0 35 % See also SOM_MAKE, SOM_AUTOLABEL.
wolffd@0 36
wolffd@0 37 %%%%%%%%%%%%% DETAILED DESCRIPTION %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0 38 %
wolffd@0 39 % som_supervised
wolffd@0 40 %
wolffd@0 41 % PURPOSE
wolffd@0 42 %
wolffd@0 43 % Creates, initializes and trains a supervised SOM by taking the
wolffd@0 44 % class-identity into account.
wolffd@0 45 %
wolffd@0 46 % SYNTAX
wolffd@0 47 %
wolffd@0 48 % sMap = som_supervised(sData);
wolffd@0 49 % sMap = som_supervised(...,'argID',value,...)
wolffd@0 50 % sMap = som_make(...,value,...);
wolffd@0 51 %
wolffd@0 52 % DESCRIPTION
wolffd@0 53 %
wolffd@0 54 % Creates, initializes and trains a supervised SOM. It constructs the
wolffd@0 55 % training data by adding 1-of-N -coded matrix to the original data
wolffd@0 56 % based on the class information in the .labels field. The dimension
wolffd@0 57 % of vectors after the process is (the old dimension + number of
wolffd@0 58 % different classes). In each vector, one of the new components has
wolffd@0 59 % value '1' (this depends on the class of the vector), and others '0'.
wolffd@0 60 % Calls SOM_MAKE to construct the map. Then the class of each map unit
wolffd@0 61 % is determined by taking maximum over these added components, and a
wolffd@0 62 % label is give accordingly. Finally, the extra components (the
wolffd@0 63 % 1-of-N -coded ones) are removed.
wolffd@0 64 %
wolffd@0 65 % REFERENCES
wolffd@0 66 %
wolffd@0 67 % Kohonen, T., "Self-Organizing Map", 2nd ed., Springer-Verlag,
wolffd@0 68 % Berlin, 1995, pp. 160-161.
wolffd@0 69 % Kohonen, T., Mäkivasara, K., Saramäki, T., "Phonetic Maps -
wolffd@0 70 % Insightful Representation of Phonological Features For
wolffd@0 71 % Speech Recognition", In proceedings of International
wolffd@0 72 % Conference on Pattern Recognition (ICPR), Montreal, Canada,
wolffd@0 73 % 1984, pp. 182-185.
wolffd@0 74 %
wolffd@0 75 % REQUIRED INPUT ARGUMENTS
wolffd@0 76 %
wolffd@0 77 % sData The data to use in the training.
wolffd@0 78 % (struct) A data struct. '.comp_names' as well as '.name'
wolffd@0 79 % is copied to the map. The class information is
wolffd@0 80 % taken from the first column of '.labels' field.
wolffd@0 81 %
wolffd@0 82 % OPTIONAL INPUT ARGUMENTS
wolffd@0 83 %
wolffd@0 84 % argID (string) Argument identifier string (see below).
wolffd@0 85 % value (varies) Value for the argument (see below).
wolffd@0 86 %
wolffd@0 87 % The optional arguments can be given as 'argID',value -pairs. If an
wolffd@0 88 % argument is given value multiple times, the last one is used.
wolffd@0 89 % Here are the argument IDs and corresponding values:
wolffd@0 90 % 'munits' (scalar) the preferred number of map units - this may
wolffd@0 91 % change a bit, depending on the properties of the data
wolffd@0 92 % 'msize' (vector) map grid size
wolffd@0 93 % 'mask' (vector) BMU search mask, size dim x 1
wolffd@0 94 % 'name' (string) map name
wolffd@0 95 % 'comp_names' (string array / cellstr) component names, size dim x 1
wolffd@0 96 % 'tracking' (scalar) how much to report, default = 1. This parameter
wolffd@0 97 % is also passed to the training functions.
wolffd@0 98 % The following values are unambiguous and can therefore
wolffd@0 99 % be given without the preceeding argument ID:
wolffd@0 100 % 'algorithm' (string) training algorithm: 'seq' or 'batch' (default)
wolffd@0 101 % 'mapsize' (string) do you want a 'small', 'normal' or 'big' map
wolffd@0 102 % Any explicit settings of munits or msize (or topol)
wolffd@0 103 % override this.
wolffd@0 104 % 'topol' (struct) topology struct
wolffd@0 105 % 'som_topol','sTopol' = 'topol'
wolffd@0 106 % 'lattice' (string) map lattice, 'hexa' or 'rect'
wolffd@0 107 % 'shape' (string) map shape, 'sheet', 'cyl' or 'toroid'
wolffd@0 108 % 'neigh' (string) neighborhood function, 'gaussian', 'cutgauss',
wolffd@0 109 % 'ep' or 'bubble'
wolffd@0 110 %
wolffd@0 111 % OUTPUT ARGUMENTS
wolffd@0 112 %
wolffd@0 113 % sMap (struct) SOM -map struct
wolffd@0 114 %
wolffd@0 115 % EXAMPLES
wolffd@0 116 %
wolffd@0 117 % To simply train a map with default parameters:
wolffd@0 118 %
wolffd@0 119 % sMap = som_supervised(sData);
wolffd@0 120 %
wolffd@0 121 % With the optional arguments, the initialization and training can be
wolffd@0 122 % influenced. To change map size, use 'msize', 'munits' or 'mapsize'
wolffd@0 123 % arguments:
wolffd@0 124 %
wolffd@0 125 % sMap = som_supervised(D,'mapsize','big'); or
wolffd@0 126 % sMap = som_supervised(D,'big');
wolffd@0 127 % sMap = som_supervised(D,'munits', 100);
wolffd@0 128 % sMap = som_supervised(D,'msize', [20 10]);
wolffd@0 129 %
wolffd@0 130 % Argument 'algorithm' can be used to switch between 'seq' and 'batch'
wolffd@0 131 % algorithms. 'batch' is the default, so to use 'seq' algorithm:
wolffd@0 132 %
wolffd@0 133 % sMap = som_supervised(D,'algorithm','seq'); or
wolffd@0 134 % sMap = som_supervised(D,'seq');
wolffd@0 135 %
wolffd@0 136 % The 'tracking' argument can be used to control the amout of reporting
wolffd@0 137 % during training. The argument is used in this function, and it is
wolffd@0 138 % passed to the training functions. To make the function work silently
wolffd@0 139 % set it to 0.
wolffd@0 140 %
wolffd@0 141 % sMap = som_supervised(D,'tracking',0);
wolffd@0 142 %
wolffd@0 143 % SEE ALSO
wolffd@0 144 %
wolffd@0 145 % som_make Create, initialize and train Self-Organizing map.
wolffd@0 146 % som_autolabel Label SOM/data set based on another SOM/data set.
wolffd@0 147
wolffd@0 148 % Contributed to SOM Toolbox vs2, Feb 2nd, 2000 by Juha Parhankangas
wolffd@0 149 % Copyright (c) by Juha Parhankangas
wolffd@0 150 % http://www.cis.hut.fi/projects/somtoolbox/
wolffd@0 151
wolffd@0 152 % Juha Parhankangas 050100
wolffd@0 153
wolffd@0 154 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0 155
wolffd@0 156 D0 = sData.data;
wolffd@0 157 [c,n,classlabels] = class2num(sData.labels(:,1));
wolffd@0 158
wolffd@0 159 %%%%%%%% Checking arguments %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0 160
wolffd@0 161 if ~isstruct(sData)
wolffd@0 162 error('Argument ''sData'' must be a ''som_data'' -struct.');
wolffd@0 163 else
wolffd@0 164 data_name = sData.name;
wolffd@0 165 comp_names = sData.comp_names;
wolffd@0 166 comp_norm = sData.comp_norm;
wolffd@0 167 end
wolffd@0 168
wolffd@0 169 [dlen,dim] = size(sData.data);
wolffd@0 170
wolffd@0 171 % defaults
wolffd@0 172
wolffd@0 173 mapsize = '';
wolffd@0 174 sM = som_map_struct(dim+n);
wolffd@0 175 sTopol = sM.topol;
wolffd@0 176 munits = prod(sTopol.msize); % should be zero
wolffd@0 177 mask = sM.mask;
wolffd@0 178 name = sM.name;
wolffd@0 179 neigh = sM.neigh;
wolffd@0 180 tracking = 1;
wolffd@0 181 algorithm = 'batch';
wolffd@0 182
wolffd@0 183 %%%% changes to defaults (checking varargin) %%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0 184
wolffd@0 185 i=1;
wolffd@0 186 while i <= length(varargin)
wolffd@0 187 argok = 1;
wolffd@0 188 if ischar(varargin{i})
wolffd@0 189 switch varargin{i},
wolffd@0 190 % argument IDs
wolffd@0 191 case 'mask',
wolffd@0 192 i=i+1;
wolffd@0 193 mask = varargin{i};
wolffd@0 194 case 'munits',
wolffd@0 195 i=i+1;
wolffd@0 196 munits = varargin{i};
wolffd@0 197 case 'msize',
wolffd@0 198 i=i+1;
wolffd@0 199 sTopol.msize = varargin{i};
wolffd@0 200 munits = prod(sTopol.msize);
wolffd@0 201 case 'mapsize',
wolffd@0 202 i=i+1;
wolffd@0 203 mapsize = varargin{i};
wolffd@0 204 case 'name',
wolffd@0 205 i=i+1;
wolffd@0 206 name = varargin{i};
wolffd@0 207 case 'comp_names',
wolffd@0 208 i=i+1;
wolffd@0 209 comp_names = varargin{i};
wolffd@0 210 case 'lattice',
wolffd@0 211 i=i+1;
wolffd@0 212 sTopol.lattice = varargin{i};
wolffd@0 213 case 'shape',
wolffd@0 214 i=i+1;
wolffd@0 215 sTopol.shape = varargin{i};
wolffd@0 216 case {'topol','som_topol','sTopol'},
wolffd@0 217 i=i+1;
wolffd@0 218 sTopol = varargin{i};
wolffd@0 219 munits = prod(sTopol.msize);
wolffd@0 220 case 'neigh',
wolffd@0 221 i=i+1;
wolffd@0 222 neigh = varargin{i};
wolffd@0 223 case 'tracking',
wolffd@0 224 i=i+1;
wolffd@0 225 tracking = varargin{i};
wolffd@0 226 case 'algorithm',
wolffd@0 227 i=i+1;
wolffd@0 228 algorithm = varargin{i};
wolffd@0 229 % unambiguous values
wolffd@0 230 case {'hexa','rect'},
wolffd@0 231 sTopol.lattice = varargin{i};
wolffd@0 232 case {'sheet','cyl','toroid'},
wolffd@0 233 sTopol.shape = varargin{i};
wolffd@0 234 case {'gaussian','cutgauss','ep','bubble'},
wolffd@0 235 neigh = varargin{i};
wolffd@0 236 case {'seq','batch'},
wolffd@0 237 algorithm = varargin{i};
wolffd@0 238 case {'small','normal','big'},
wolffd@0 239 mapsize = varargin{i};
wolffd@0 240 otherwise argok=0;
wolffd@0 241 end
wolffd@0 242 elseif isstruct(varargin{i}) & isfield(varargin{i},'type'),
wolffd@0 243 switch varargin{i}(1).type,
wolffd@0 244 case 'som_topol',
wolffd@0 245 sTopol = varargin{i};
wolffd@0 246 otherwise argok=0;
wolffd@0 247 end
wolffd@0 248 else
wolffd@0 249 argok = 0;
wolffd@0 250 end
wolffd@0 251 if ~argok,
wolffd@0 252 disp(['(som_supervised) Ignoring invalid argument #' num2str(i+1)]);
wolffd@0 253 end
wolffd@0 254 i = i+1;
wolffd@0 255 end
wolffd@0 256
wolffd@0 257 %%%%%%%% Action %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0 258
wolffd@0 259
wolffd@0 260
wolffd@0 261 % constructing the training data by adding 1-of-N -coded matrix to the
wolffd@0 262 % original data.
wolffd@0 263
wolffd@0 264 [dlen,dim] = size(D0);
wolffd@0 265
wolffd@0 266 Dc = zeros(dlen,n);
wolffd@0 267
wolffd@0 268 for i=1:dlen
wolffd@0 269 if c(i)
wolffd@0 270 Dc(i,c(i)) = 1;
wolffd@0 271 end
wolffd@0 272 end
wolffd@0 273
wolffd@0 274 D = [D0, Dc];
wolffd@0 275
wolffd@0 276 % initialization and training
wolffd@0 277
wolffd@0 278 sD = som_data_struct(D,...
wolffd@0 279 'name',data_name);
wolffd@0 280
wolffd@0 281 sM = som_make(sD,...
wolffd@0 282 'mask',mask,...
wolffd@0 283 'munits',munits,...
wolffd@0 284 'name',data_name,...
wolffd@0 285 'tracking',tracking,...
wolffd@0 286 'algorithm',algorithm,...
wolffd@0 287 'mapsize',mapsize,...
wolffd@0 288 'topol',sTopol,...
wolffd@0 289 'neigh',neigh);
wolffd@0 290
wolffd@0 291 % add labels
wolffd@0 292
wolffd@0 293 for i=1:prod(sM.topol.msize),
wolffd@0 294 [dummy,class] = max(sM.codebook(i,dim+[1:n]));
wolffd@0 295 sM.labels{i} = classlabels{class};
wolffd@0 296 end
wolffd@0 297
wolffd@0 298 %sD.labels = sData.labels;
wolffd@0 299 %sM = som_autolabel(sM,sD,'vote');
wolffd@0 300
wolffd@0 301 % remove extra components and modify map -struct
wolffd@0 302
wolffd@0 303 sM.codebook = sM.codebook(:,1:dim);
wolffd@0 304 sM.mask = sM.mask(1:dim);
wolffd@0 305 sM.comp_names = sData.comp_names;
wolffd@0 306 sM.comp_norm = sData.comp_norm;
wolffd@0 307
wolffd@0 308 % remove extras from sM.trainhist
wolffd@0 309
wolffd@0 310 for i=1:length(sM.trainhist)
wolffd@0 311 if sM.trainhist(i).mask
wolffd@0 312 sM.trainhist(i).mask = sM.trainhist(i).mask(1:dim);
wolffd@0 313 end
wolffd@0 314 end
wolffd@0 315
wolffd@0 316 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0 317
wolffd@0 318 function [numbers, n, names] = class2num(class)
wolffd@0 319
wolffd@0 320 names = {};
wolffd@0 321 numbers = zeros(length(class),1);
wolffd@0 322
wolffd@0 323 for i=1:length(class)
wolffd@0 324 if ~isempty(class{i}) & ~any(strcmp(class{i},names))
wolffd@0 325 names=cat(1,names,class(i));
wolffd@0 326 end
wolffd@0 327 end
wolffd@0 328
wolffd@0 329 n=length(names);
wolffd@0 330
wolffd@0 331 tmp_numbers = (1:n)';
wolffd@0 332
wolffd@0 333 for i=1:length(class)
wolffd@0 334 if ~isempty(class{i})
wolffd@0 335 numbers(i,1) = find(strcmp(class{i},names));
wolffd@0 336 end
wolffd@0 337 end