Mercurial > hg > camir-aes2014

annotate toolboxes/MIRtoolbox1.3.2/somtoolbox/lvq3.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 codebook = lvq3(codebook,data,rlen,alpha,win,epsilon)
wolffd@0 2
wolffd@0 3 %LVQ3 trains codebook with LVQ3 -algorithm
wolffd@0 4 %
wolffd@0 5 % sM = lvq3(sM,D,rlen,alpha,win,epsilon)
wolffd@0 6 %
wolffd@0 7 % sM = lvq3(sM,sD,50*length(sM.codebook),0.05,0.2,0.3);
wolffd@0 8 %
wolffd@0 9 % Input and output arguments:
wolffd@0 10 % sM (struct) map struct, the class information must be
wolffd@0 11 % present on the first column of .labels field
wolffd@0 12 % D (struct) data struct, the class information must
wolffd@0 13 % be present on the first column of .labels field
wolffd@0 14 % rlen (scalar) running length
wolffd@0 15 % alpha (scalar) learning parameter, e.g. 0.05
wolffd@0 16 % win (scalar) window width parameter, e.g. 0.25
wolffd@0 17 % epsilon (scalar) relative learning parameter, e.g. 0.3
wolffd@0 18 %
wolffd@0 19 % sM (struct) map struct, the trained codebook
wolffd@0 20 %
wolffd@0 21 % NOTE: does not take mask into account.
wolffd@0 22 %
wolffd@0 23 % For more help, try 'type lvq3', or check out online documentation.
wolffd@0 24 % See also LVQ1, SOM_SUPERVISED, SOM_SEQTRAIN.
wolffd@0 25
wolffd@0 26 %%%%%%%%%%%%% DETAILED DESCRIPTION %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0 27 %
wolffd@0 28 % lvq3
wolffd@0 29 %
wolffd@0 30 % PURPOSE
wolffd@0 31 %
wolffd@0 32 % Trains codebook with the LVQ3 -algorithm (described below).
wolffd@0 33 %
wolffd@0 34 % SYNTAX
wolffd@0 35 %
wolffd@0 36 % sM = lvq3(sM, data, rlen, alpha, win, epsilon)
wolffd@0 37 %
wolffd@0 38 % DESCRIPTION
wolffd@0 39 %
wolffd@0 40 % Trains codebook with the LVQ3 -algorithm. Codebook contains a number
wolffd@0 41 % of vectors (mi, i=1,2,...,n) and so does data (vectors xj, j=1,2,...k).
wolffd@0 42 % Both vector sets are classified: vectors may have a class (classes are
wolffd@0 43 % set to data- or map -structure's 'labels' -field. For each xj the two
wolffd@0 44 % closest codebookvectors mc1 and mc2 are searched (euclidean distances
wolffd@0 45 % d1 and d2). xj must fall into the zone of window. That happens if:
wolffd@0 46 %
wolffd@0 47 % min(d1/d2, d2/d1) > s, where s = (1-win) / (1+win).
wolffd@0 48 %
wolffd@0 49 % If xj belongs to the same class of one of the mc1 and mc1, codebook
wolffd@0 50 % is updated as follows (let mc1 belong to the same class as xj):
wolffd@0 51 % mc1(t+1) = mc1(t) + alpha * (xj(t) - mc1(t))
wolffd@0 52 % mc2(t+1) = mc2(t) - alpha * (xj(t) - mc2(t))
wolffd@0 53 % If both mc1 and mc2 belong to the same class as xj, codebook is
wolffd@0 54 % updated as follows:
wolffd@0 55 % mc1(t+1) = mc1(t) + epsilon * alpha * (xj(t) - mc1(t))
wolffd@0 56 % mc2(t+1) = mc2(t) + epsilon * alpha * (xj(t) - mc2(t))
wolffd@0 57 % Otherwise updating is not performed.
wolffd@0 58 %
wolffd@0 59 % Argument 'rlen' tells how many times training -sequence is performed.
wolffd@0 60 %
wolffd@0 61 % Argument 'alpha' is recommended to be smaller than 0.1 and argument
wolffd@0 62 % 'epsilon' should be between 0.1 and 0.5.
wolffd@0 63 %
wolffd@0 64 % NOTE: does not take mask into account.
wolffd@0 65 %
wolffd@0 66 % REFERENCES
wolffd@0 67 %
wolffd@0 68 % Kohonen, T., "Self-Organizing Map", 2nd ed., Springer-Verlag,
wolffd@0 69 % Berlin, 1995, pp. 181-182.
wolffd@0 70 %
wolffd@0 71 % See also LVQ_PAK from http://www.cis.hut.fi/research/som_lvq_pak.shtml
wolffd@0 72 %
wolffd@0 73 % REQUIRED INPUT ARGUMENTS
wolffd@0 74 %
wolffd@0 75 % sM The data to be trained.
wolffd@0 76 % (struct) A map struct.
wolffd@0 77 %
wolffd@0 78 % data The data to use in training.
wolffd@0 79 % (struct) A data struct.
wolffd@0 80 %
wolffd@0 81 % rlen (integer) Running length of LVQ3 -algorithm.
wolffd@0 82 %
wolffd@0 83 % alpha (float) Learning rate used in training, e.g. 0.05
wolffd@0 84 %
wolffd@0 85 % win (float) Window length, e.g. 0.25
wolffd@0 86 %
wolffd@0 87 % epsilon (float) Relative learning parameter, e.g. 0.3
wolffd@0 88 %
wolffd@0 89 % OUTPUT ARGUMENTS
wolffd@0 90 %
wolffd@0 91 % sM Trained data.
wolffd@0 92 % (struct) A map struct.
wolffd@0 93 %
wolffd@0 94 % EXAMPLE
wolffd@0 95 %
wolffd@0 96 % lab = unique(sD.labels(:,1)); % different classes
wolffd@0 97 % mu = length(lab)*5; % 5 prototypes for each
wolffd@0 98 % sM = som_randinit(sD,'msize',[mu 1]); % initial prototypes
wolffd@0 99 % sM.labels = [lab;lab;lab;lab;lab]; % their classes
wolffd@0 100 % sM = lvq1(sM,sD,50*mu,0.05); % use LVQ1 to adjust
wolffd@0 101 % % the prototypes
wolffd@0 102 % sM = lvq3(sM,sD,50*mu,0.05,0.2,0.3); % then use LVQ3
wolffd@0 103 %
wolffd@0 104 % SEE ALSO
wolffd@0 105 %
wolffd@0 106 % lvq1 Use LVQ1 algorithm for training.
wolffd@0 107 % som_supervised Train SOM using supervised training.
wolffd@0 108 % som_seqtrain Train SOM with sequential algorithm.
wolffd@0 109
wolffd@0 110 % Contributed to SOM Toolbox vs2, February 2nd, 2000 by Juha Parhankangas
wolffd@0 111 % Copyright (c) by Juha Parhankangas
wolffd@0 112 % http://www.cis.hut.fi/projects/somtoolbox/
wolffd@0 113
wolffd@0 114 % Juha Parhankangas 310100 juuso 020200
wolffd@0 115
wolffd@0 116 NOTFOUND = 1;
wolffd@0 117
wolffd@0 118 cod = codebook.codebook;
wolffd@0 119 dat = data.data;
wolffd@0 120
wolffd@0 121 c_class = codebook.labels(:,1);
wolffd@0 122 d_class = data.labels(:,1);
wolffd@0 123
wolffd@0 124 s = (1-win)/(1+win);
wolffd@0 125
wolffd@0 126 x = size(dat,1);
wolffd@0 127 y = size(cod,2);
wolffd@0 128
wolffd@0 129 c_class=class2num(c_class);
wolffd@0 130 d_class=class2num(d_class);
wolffd@0 131
wolffd@0 132 ONES=ones(size(cod,1),1);
wolffd@0 133
wolffd@0 134 for t=1:rlen
wolffd@0 135 fprintf('\rTraining round: %d/%d',t,rlen);
wolffd@0 136 tmp = NaN*ones(x,y);
wolffd@0 137
wolffd@0 138 for j=1:x
wolffd@0 139 flag = 0;
wolffd@0 140 mj = 0;
wolffd@0 141 mi = 0;
wolffd@0 142 no_NaN=find(~isnan(dat(j,:)));
wolffd@0 143 di=sqrt(sum([cod(:,no_NaN) - ONES*dat(j,no_NaN)].^2,2));
wolffd@0 144 [foo, ind1] = min(di);
wolffd@0 145 di(ind1)=Inf;
wolffd@0 146 [foo,ind2] = min(di);
wolffd@0 147
wolffd@0 148 %ind2=ind2+1;
wolffd@0 149
wolffd@0 150 if d_class(j) & d_class(j)==c_class(ind1)
wolffd@0 151 mj = ind1;
wolffd@0 152 mi = ind2;
wolffd@0 153 if d_class(j)==c_class(ind2)
wolffd@0 154 flag = 1;
wolffd@0 155 end
wolffd@0 156 elseif d_class(j) & d_class(j)==c_class(ind2)
wolffd@0 157 mj = ind2;
wolffd@0 158 mi = ind1;
wolffd@0 159 if d_class(j)==c_class(ind1)
wolffd@0 160 flag = 1;
wolffd@0 161 end
wolffd@0 162 end
wolffd@0 163
wolffd@0 164 if mj & mi
wolffd@0 165 if flag
wolffd@0 166 tmp([mj mi],:) = cod([mj mi],:) + epsilon*alpha*...
wolffd@0 167 (dat([j j],:) - cod([mj mi],:));
wolffd@0 168 else
wolffd@0 169 tmp(mj,:) = cod(mj,:) + alpha * (dat(j,:)-cod(mj,:));
wolffd@0 170 tmp(mi,:) = cod(mi,:) - alpha * (dat(j,:)-cod(mj,:));
wolffd@0 171 end
wolffd@0 172 end
wolffd@0 173 end
wolffd@0 174 inds = find(~isnan(sum(tmp,2)));
wolffd@0 175 cod(inds,:) = tmp(inds,:);
wolffd@0 176 end
wolffd@0 177 fprintf(1,'\n');
wolffd@0 178
wolffd@0 179 sTrain = som_set('som_train','algorithm','lvq3',...
wolffd@0 180 'data_name',data.name,...
wolffd@0 181 'neigh','',...
wolffd@0 182 'mask',ones(y,1),...
wolffd@0 183 'radius_ini',NaN,...
wolffd@0 184 'radius_fin',NaN,...
wolffd@0 185 'alpha_ini',alpha,...
wolffd@0 186 'alpha_type','constant',...
wolffd@0 187 'trainlen',rlen,...
wolffd@0 188 'time',datestr(now,0));
wolffd@0 189 codebook.trainhist(end+1) = sTrain;
wolffd@0 190
wolffd@0 191 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0 192
wolffd@0 193 function nos = class2num(class)
wolffd@0 194
wolffd@0 195 names = {};
wolffd@0 196 nos = zeros(length(class),1);
wolffd@0 197
wolffd@0 198 for i=1:length(class)
wolffd@0 199 if ~isempty(class{i}) & ~any(strcmp(class{i},names))
wolffd@0 200 names=cat(1,names,class(i));
wolffd@0 201 end
wolffd@0 202 end
wolffd@0 203
wolffd@0 204 tmp_nos = (1:length(names))';
wolffd@0 205
wolffd@0 206 for i=1:length(class)
wolffd@0 207 if ~isempty(class{i})
wolffd@0 208 nos(i,1) = find(strcmp(class{i},names));
wolffd@0 209 end
wolffd@0 210 end
wolffd@0 211
wolffd@0 212
wolffd@0 213
wolffd@0 214
wolffd@0 215