Mercurial > hg > camir-aes2014

annotate toolboxes/MIRtoolbox1.3.2/somtoolbox/som_clset.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 [sC,old2new,newi] = som_clset(sC,action,par1,par2)
wolffd@0 2
wolffd@0 3 % SOM_CLSET Create and/or set values in the som_clustering struct.
wolffd@0 4 %
wolffd@0 5 % first argument
wolffd@0 6 % sC (struct) a som_clustering struct
wolffd@0 7 % Z (matrix) size nb-1 x 3, as given by LINKAGE function
wolffd@0 8 % base (vector) size dlen x 1, a partitioning of the data
wolffd@0 9 %
wolffd@0 10 % actions
wolffd@0 11 % 'remove' removes the indicated clusters (par1: vector)
wolffd@0 12 % 'add' add a cluster by making a combination of the indicated
wolffd@0 13 % clusters (par1: vector)
wolffd@0 14 % %'move' moves a child cluster (par1: scalar) from a parent to another
wolffd@0 15 % % (par2: vector 1 x 2)
wolffd@0 16 % 'merge' like 'add', followed by removing the indicated clusters (par1: vector)
wolffd@0 17 % %'split' the indicated cluster (par1: scalar) is partitioned into indicated
wolffd@0 18 % % parts (par2: vector), which are then added, while the indicated cluster
wolffd@0 19 % % (par1) is removed
wolffd@0 20 % 'coord' sets the coordinates of base clusters (par1: matrix nb x *), and
wolffd@0 21 % recalculates coordinates of the derived clusters (by averaging base cluster
wolffd@0 22 % coordinates)
wolffd@0 23 % 'color' sets the colors of base clusters (par1: matrix nb x 3), and recalculates
wolffd@0 24 % colors of the derived clusters (as averages of base cluster colors)
wolffd@0 25 %
wolffd@0 26 % sC
wolffd@0 27 % .type (string) 'som_clustering'
wolffd@0 28 % .name (string) Identifier for the clustering.
wolffd@0 29 % .nb (scalar) Number of base clusters in the clustering.
wolffd@0 30 % .base (vector) Size dlen x 1, the basic groups of data
wolffd@0 31 % forming the base clusters, e.g. as a result
wolffd@0 32 % of partitive clustering. Allowed values are
wolffd@0 33 % 1:nb indicating the base cluster
wolffd@0 34 % to which the data belongs to.
wolffd@0 35 % NaN indicating that the data has
wolffd@0 36 % been ignored in the clustering
wolffd@0 37 % .nc (scalar) Number of clusters in the clustering (nb + derived clusters).
wolffd@0 38 % .children (cellarray) size nc x 1, each cell gives the list of indeces
wolffd@0 39 % of child clusters for the cluster
wolffd@0 40 % .parent (vector) size nc x 1, the index of parent of each cluster
wolffd@0 41 % (or zero if the cluster does not have a parent)
wolffd@0 42 % .coord (matrix) size nc x *, visualization coordinates for each cluster
wolffd@0 43 % By default the coordinates are set so that
wolffd@0 44 % the base clusters are ordered on a line, and the
wolffd@0 45 % position of each combined cluster is average of
wolffd@0 46 % the base clusters that constitute it.
wolffd@0 47 % .color (matrix) size nc x 3, color for each cluster.
wolffd@0 48 % By default the colors are set so that the
wolffd@0 49 % base clusters are ordered on a line,
wolffd@0 50 % and then colors are assigned from the 'hsv'
wolffd@0 51 % colormap to the base clusters. The color
wolffd@0 52 % of each combined cluster is average as above.
wolffd@0 53 % .cldist (string) Default cluster distance function.
wolffd@0 54
wolffd@0 55 inew = [];
wolffd@0 56 if isstruct(sC),
wolffd@0 57 % it should be a som_clustering struct
wolffd@0 58 old2new = [1:sC.nc];
wolffd@0 59 elseif size(sC,2)==3,
wolffd@0 60 % assume it is a cluster hierarchy matrix Z
wolffd@0 61 sC = Z2sC(sC);
wolffd@0 62 old2new = [1:sC.nc];
wolffd@0 63 else
wolffd@0 64 % assume it is a partitioning vector
wolffd@0 65 base = sC;
wolffd@0 66 u = unique(base(isfinite(base)));
wolffd@0 67 old2new = sparse(u,1,1:length(u));
wolffd@0 68 base = old2new(base);
wolffd@0 69 sC = part2sC(base);
wolffd@0 70 end
wolffd@0 71
wolffd@0 72 switch action,
wolffd@0 73 case 'remove',
wolffd@0 74 for i=1:length(par1),
wolffd@0 75 [sC,o2n] = removecluster(sC,old2new(par1(i)));
wolffd@0 76 old2new = o2n(old2new);
wolffd@0 77 end
wolffd@0 78 case 'add',
wolffd@0 79 [sC,old2new,inew] = addmergedcluster(sC,par1);
wolffd@0 80 case 'move',
wolffd@0 81 % not implemented yet
wolffd@0 82 case 'split',
wolffd@0 83 % not implemented yet
wolffd@0 84 case 'merge',
wolffd@0 85 [sC,old2new,inew] = addmergedcluster(sC,par1);
wolffd@0 86 for i=1:length(par1),
wolffd@0 87 [sC,o2n] = removecluster(sC,old2new(par1(i)));
wolffd@0 88 old2new = o2n(old2new);
wolffd@0 89 end
wolffd@0 90 case 'color',
wolffd@0 91 sC.color = derivative_average(sC,par1);
wolffd@0 92 case 'coord',
wolffd@0 93 sC.coord = derivative_average(sC,par1);
wolffd@0 94 end
wolffd@0 95
wolffd@0 96 return;
wolffd@0 97
wolffd@0 98 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0 99 %% subfunctions
wolffd@0 100
wolffd@0 101 function sC = clstruct(nb,nc)
wolffd@0 102
wolffd@0 103 sC = struct('type','som_clustering',...
wolffd@0 104 'name','','base',[],'nb',nb,'nc',nc,...
wolffd@0 105 'parent',[],'children',[],'coord',[],'color',[],'cldist','centroid');
wolffd@0 106 sC.base = [1:nb];
wolffd@0 107 sC.parent = zeros(nc,1);
wolffd@0 108 sC.children = cell(nc,1); sC.children(:) = {[]};
wolffd@0 109 sC.coord = zeros(nc,2);
wolffd@0 110 sC.color = zeros(nc,3);
wolffd@0 111 return;
wolffd@0 112
wolffd@0 113 function Z = sC2Z(sC,height)
wolffd@0 114
wolffd@0 115 if nargin<2, height = 'level'; end
wolffd@0 116
wolffd@0 117 root = find(sC.parent==0);
wolffd@0 118 order = [root];
wolffd@0 119 ch = sC.children(root);
wolffd@0 120 while any(ch), i = ch(1); order = [ch(1), order]; ch = [ch(2:end), sC.children{i}]; end
wolffd@0 121
wolffd@0 122 he = zeros(sC.nc,1);
wolffd@0 123 if strcmp(height,'level'),
wolffd@0 124 ch = sC.children{root};
wolffd@0 125 while any(ch),
wolffd@0 126 i = ch(1); he(i) = he(sC.parent(i))+1;
wolffd@0 127 ch = [ch(2:end), sC.children{i}];
wolffd@0 128 end
wolffd@0 129 he = max(he)-he;
wolffd@0 130 elseif strcmp(height,'level2'),
wolffd@0 131 for i=order, if any(sC.children{i}), he(i) = max(he(sC.children{i}))+1; end, end
wolffd@0 132 else
wolffd@0 133 %he = som_cldist ( between children )
wolffd@0 134 end
wolffd@0 135
wolffd@0 136 Z = zeros(sC.nb-1,3);
wolffd@0 137 i = sC.nb-1;
wolffd@0 138 inds = root;
wolffd@0 139 while i>0,
wolffd@0 140 ch = sC.children{inds(1)}; h = he(inds(1)); inds = [inds(2:end), ch];
wolffd@0 141 if length(ch)>=2,
wolffd@0 142 for k=1:length(ch)-2, Z(i,:) = [i-1, ch(k), h]; i = i - 1; end
wolffd@0 143 Z(i,:) = [ch(end-1) ch(end) h]; i = i - 1;
wolffd@0 144 end
wolffd@0 145 end
wolffd@0 146 return;
wolffd@0 147
wolffd@0 148 function sC = Z2sC(Z)
wolffd@0 149
wolffd@0 150 nb = size(Z,1)+1;
wolffd@0 151 nc = 2*nb-1;
wolffd@0 152 sC = clstruct(nb,nc);
wolffd@0 153 sC.base = [1:nb];
wolffd@0 154 for i=1:nc,
wolffd@0 155 j = find(Z(:,1)==i | Z(:,2)==i);
wolffd@0 156 sC.parent(i) = nb+j;
wolffd@0 157 sC.children{sC.parent(i)}(end+1) = i;
wolffd@0 158 end
wolffd@0 159 % coords and color
wolffd@0 160 order = nc;
wolffd@0 161 nonleaves = 1;
wolffd@0 162 while any(nonleaves),
wolffd@0 163 j = nonleaves(1);
wolffd@0 164 ch = sC.children{order(j)};
wolffd@0 165 if j==1, oleft = []; else oleft = order(1:(j-1)); end
wolffd@0 166 if j==length(order), oright = []; else oright = order((j+1):length(order)); end
wolffd@0 167 order = [oleft, ch, oright];
wolffd@0 168 nonleaves = find(order>nb);
wolffd@0 169 end
wolffd@0 170 [dummy,co] = sort(order);
wolffd@0 171 sC.coord = derivative_average(sC,co');
wolffd@0 172 H = hsv(nb+1);
wolffd@0 173 sC.color = derivative_average(sC,H(co,:));
wolffd@0 174 return;
wolffd@0 175
wolffd@0 176 function sC = part2sC(part)
wolffd@0 177
wolffd@0 178 nb = max(part);
wolffd@0 179 nc = nb+1;
wolffd@0 180 sC = clstruct(nb,nc);
wolffd@0 181 sC.base = part;
wolffd@0 182 sC.parent(1:nb) = nc;
wolffd@0 183 sC.children{nc} = [1:nb];
wolffd@0 184 co = [1:nb]';
wolffd@0 185 sC.coord = derivative_average(sC,co);
wolffd@0 186 H = hsv(nb+1);
wolffd@0 187 sC.color = derivative_average(sC,H(1:nb,:));
wolffd@0 188 return;
wolffd@0 189
wolffd@0 190 function [sC,old2new] = removecluster(sC,ind)
wolffd@0 191
wolffd@0 192 old2new = [1:sC.nc];
wolffd@0 193 parent_ind = sC.parent(ind);
wolffd@0 194 ch = sC.children{ind};
wolffd@0 195 if ~parent_ind,
wolffd@0 196 % trying to remove root cluster - no go
wolffd@0 197 return;
wolffd@0 198 elseif ~any(ch),
wolffd@0 199 % trying to remove a base cluster - no go
wolffd@0 200 return;
wolffd@0 201 else
wolffd@0 202 % ok, proceed
wolffd@0 203 old2new = [1:ind-1 0 ind:sC.nc-1];
wolffd@0 204 % update parent and child fields
wolffd@0 205 sC.parent(ch) = parent_ind;
wolffd@0 206 sC.children{parent_ind} = setdiff([sC.children{parent_ind}, ch],ind);
wolffd@0 207 % remove old cluster
wolffd@0 208 j = [1:ind-1, ind+1:sC.nc];
wolffd@0 209 sC.parent = sC.parent(j);
wolffd@0 210 sC.children = sC.children(j);
wolffd@0 211 sC.color = sC.color(j,:);
wolffd@0 212 sC.coord = sC.coord(j,:);
wolffd@0 213 sC.nc = sC.nc-1;
wolffd@0 214 % update old indeces to new indices
wolffd@0 215 sC.parent = old2new(sC.parent);
wolffd@0 216 for i=1:sC.nc, sC.children{i} = old2new(sC.children{i}); end
wolffd@0 217 end
wolffd@0 218 return;
wolffd@0 219
wolffd@0 220 function [sC,old2new,inew] = addmergedcluster(sC,inds)
wolffd@0 221
wolffd@0 222 old2new = [1:sC.nc];
wolffd@0 223 inew = 0;
wolffd@0 224 p_inds = sC.parent(inds);
wolffd@0 225 if ~all(p_inds(1)==p_inds),
wolffd@0 226 % clusters are not siblings - no go
wolffd@0 227 return;
wolffd@0 228 end
wolffd@0 229 parent_ind = p_inds(1);
wolffd@0 230 if isempty(setdiff(sC.children{parent_ind},inds)),
wolffd@0 231 % such a merged cluster exists already
wolffd@0 232 return;
wolffd@0 233 else
wolffd@0 234 % ok, proceed
wolffd@0 235 inew = parent_ind;
wolffd@0 236 old2new = [1:inew-1,inew+1:sC.nc+1];
wolffd@0 237 % add the new cluster (=copy of parent_ind)
wolffd@0 238 j = [1:inew,inew:sC.nc];
wolffd@0 239 sC.parent = sC.parent(j);
wolffd@0 240 sC.children = sC.children(j);
wolffd@0 241 sC.color = sC.color(j,:);
wolffd@0 242 sC.coord = sC.coord(j,:);
wolffd@0 243 sC.nc = sC.nc+1;
wolffd@0 244 % update old indeces to new indices
wolffd@0 245 sC.parent = old2new(sC.parent);
wolffd@0 246 for i=1:sC.nc, sC.children{i} = old2new(sC.children{i}); end
wolffd@0 247 inds = old2new(inds);
wolffd@0 248 parent_ind = old2new(parent_ind);
wolffd@0 249 % update parent, child, color and coord fields
wolffd@0 250 sC.parent(inds) = inew;
wolffd@0 251 sC.parent(inew) = parent_ind;
wolffd@0 252 sC.children{inew} = inds;
wolffd@0 253 sC.children{parent_ind} = [setdiff(sC.children{parent_ind}, inds), inew];
wolffd@0 254 b = baseind(sC,inew);
wolffd@0 255 sC.color(inew,:) = mean(sC.color(b,:));
wolffd@0 256 sC.coord(inew,:) = mean(sC.coord(b,:));
wolffd@0 257 end
wolffd@0 258 return;
wolffd@0 259
wolffd@0 260 function C = derivative_average(sC,Cbase)
wolffd@0 261
wolffd@0 262 [n dim] = size(Cbase);
wolffd@0 263 if n ~= sC.nb, error('Color / Coord matrix should have nb rows'); end
wolffd@0 264 C = zeros(sC.nc,dim);
wolffd@0 265 for i=1:sC.nc, C(i,:) = mean(Cbase(baseind(sC,i),:)); end
wolffd@0 266 return;
wolffd@0 267
wolffd@0 268 function bi = baseind(sC,ind)
wolffd@0 269
wolffd@0 270 bi = [ind];
wolffd@0 271 i = 1;
wolffd@0 272 while i<=length(bi), bi = [bi, sC.children{bi(i)}]; end
wolffd@0 273 bi = bi(bi<=sC.nb);
wolffd@0 274 return;
wolffd@0 275
wolffd@0 276
wolffd@0 277
wolffd@0 278