Mercurial > hg > camir-aes2014

annotate toolboxes/MIRtoolbox1.3.2/somtoolbox/som_clstruct.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 = som_clstruct(Z,varargin)
wolffd@0 2
wolffd@0 3 %SOM_CLSTRUCT Create a clustering struct or set its field values.
wolffd@0 4 %
wolffd@0 5 % sC = som_clstruct(Z, [argID, value, ...])
wolffd@0 6 %
wolffd@0 7 % Z = linkage(pdist(sM.codebook));
wolffd@0 8 % sC = som_clstruct(Z);
wolffd@0 9 % sC = som_clstruct(sC,'coord',som_vis_coords(lattice,msize));
wolffd@0 10 % sC = som_clstruct(sC,'color',som_colorcode(sM));
wolffd@0 11 % sC = som_clstruct(sC,'base',sC.base(som_bmus(sM,sD)));
wolffd@0 12 %
wolffd@0 13 % Input and output arguments ([]'s are optional):
wolffd@0 14 % Z (matrix) size clen-1 x 3, where clen is the number of
wolffd@0 15 % base clusters. This is a clustering matrix
wolffd@0 16 % similar to that produced by LINKAGE in
wolffd@0 17 % Statistical Toolbox. See SOM_LINKAGE.
wolffd@0 18 % (struct) clustering struct (as produced by this function)
wolffd@0 19 % [argID, (string) See below. Each pair is the fieldname and
wolffd@0 20 % value] (varies) the value to be given to that field.
wolffd@0 21 %
wolffd@0 22 % sC (struct) clustering struct
wolffd@0 23 %
wolffd@0 24 % The clustering struct is based on the assumption that there
wolffd@0 25 % is a base partitioning of the SOM (or data) which is saved in
wolffd@0 26 % the .base field of the struct. Then a hierarchical clustering
wolffd@0 27 % is applied to this base partitioning. The results are saved to
wolffd@0 28 % .tree field of the struct. Each cluster (base and combined)
wolffd@0 29 % has also three properties: height, coordinate and color, which
wolffd@0 30 % are used in the visualizations. The fields of the struct are:
wolffd@0 31 % .type (string) 'som_clustering'
wolffd@0 32 % .name (string) Identifier for the clustering.
wolffd@0 33 % .tree (matrix) Size clen-1 x 3, as argument Z above.
wolffd@0 34 % .base (vector) Size dlen x 1, the basic groups of data
wolffd@0 35 % forming the base clusters, e.g. as a result
wolffd@0 36 % of partitive clustering. Allowed values are
wolffd@0 37 % 1:clen indicating the base cluster
wolffd@0 38 % to which the data belongs to.
wolffd@0 39 % NaN indicating that the data has
wolffd@0 40 % been ignored in the clustering
wolffd@0 41 % By default [1:clen].
wolffd@0 42 % .height (vector) Size 2*clen-1 x 1, (clustering) height for each
wolffd@0 43 % cluster. By default 0 for each base cluster and
wolffd@0 44 % .tree(:,3) for the others.
wolffd@0 45 % .coord (matrix) Size 2*clen-1 x *, coordinate for each cluster,
wolffd@0 46 % By default the coordinates are set so that
wolffd@0 47 % the base clusters are ordered on a line, and the
wolffd@0 48 % position of each combined cluster is average of
wolffd@0 49 % the base clusters that constitute it.
wolffd@0 50 % .color (matrix) Size 2*clen-1 x 3, color for each cluster.
wolffd@0 51 % By default the colors are set so that the
wolffd@0 52 % base clusters are ordered on a line, like above,
wolffd@0 53 % and then colors are assigned from the 'hsv'
wolffd@0 54 % colormap to the base clusters. The color
wolffd@0 55 % of each combined cluster is average as above.
wolffd@0 56 %
wolffd@0 57 % Height, coord and color can also be specified in alternate forms:
wolffd@0 58 % 'height' (vector) size 2*clen-1 x 1, if given explicitly
wolffd@0 59 % size clen-1 x 1, specified heights of the
wolffd@0 60 % combined clusters (the base cluster heights
wolffd@0 61 % are all = 0)
wolffd@0 62 % size 0 x 0, default value is used
wolffd@0 63 % 'coord' (matrix) size 2*clen-1 x *, if given explicitly
wolffd@0 64 % size clen x *, to give coordinates for base
wolffd@0 65 % clusters; the coordinate of combined clusters
wolffd@0 66 % are averaged from these
wolffd@0 67 % size dlen x *, to give coordinates of the
wolffd@0 68 % original data: the cluster coordinates are
wolffd@0 69 % averaged from these based on base clusters
wolffd@0 70 % size 0 x 0, default value is used
wolffd@0 71 % 'color' (matrix) as 'coord'
wolffd@0 72 %
wolffd@0 73 % See also SOM_CLPLOT, SOM_CLVALIDITY, SOM_CLGET, SOM_CLLINKAGE.
wolffd@0 74
wolffd@0 75 % Copyright (c) 2000 by the SOM toolbox programming team.
wolffd@0 76 % Contributed to SOM Toolbox on XXX by Juha Vesanto
wolffd@0 77 % http://www.cis.hut.fi/projects/somtoolbox/
wolffd@0 78
wolffd@0 79 % Version 2.0beta juuso 180800
wolffd@0 80
wolffd@0 81 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0 82
wolffd@0 83 if isstruct(Z),
wolffd@0 84 base = Z.base;
wolffd@0 85 color = Z.color;
wolffd@0 86 coord = Z.coord;
wolffd@0 87 height = Z.height;
wolffd@0 88 name = Z.name;
wolffd@0 89 Z = Z.tree;
wolffd@0 90 else
wolffd@0 91 base = [];
wolffd@0 92 color = [];
wolffd@0 93 coord = [];
wolffd@0 94 height = [];
wolffd@0 95 name = '';
wolffd@0 96 end
wolffd@0 97 clen = size(Z,1)+1;
wolffd@0 98
wolffd@0 99 i=1;
wolffd@0 100 while i<=length(varargin),
wolffd@0 101 argok = 1;
wolffd@0 102 if ischar(varargin{i}),
wolffd@0 103 switch varargin{i},
wolffd@0 104 case 'tree', i=i+1; Z = varargin{i}; clen = size(Z,1)+1;
wolffd@0 105 case 'base', i=i+1; base = varargin{i};
wolffd@0 106 case 'color', i=i+1; color = varargin{i};
wolffd@0 107 case 'coord', i=i+1; coord = varargin{i};
wolffd@0 108 case 'height', i=i+1; height = varargin{i};
wolffd@0 109 case 'name', i=i+1; name = varargin{i};
wolffd@0 110 otherwise argok=0;
wolffd@0 111 end
wolffd@0 112 else argok = 0;
wolffd@0 113 end
wolffd@0 114 if ~argok, disp(['(som_clstruct) Ignoring invalid argument #' num2str(i+1)]); end
wolffd@0 115 i = i+1;
wolffd@0 116 end
wolffd@0 117
wolffd@0 118 if isempty(base),
wolffd@0 119 dlen = clen;
wolffd@0 120 base = 1:dlen;
wolffd@0 121 else
wolffd@0 122 dlen = length(base);
wolffd@0 123 if any(base)>clen | any(base)<1, error('Incorrect base partition vector.'); end
wolffd@0 124 end
wolffd@0 125
wolffd@0 126 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0 127 %% analysis of hierarchy
wolffd@0 128
wolffd@0 129 % order of base clusters
wolffd@0 130 order = 2*clen-1;
wolffd@0 131 nonleaves = 1;
wolffd@0 132 while any(nonleaves),
wolffd@0 133 j = nonleaves(1);
wolffd@0 134 ch = Z(order(j)-clen,1:2);
wolffd@0 135 if j==1, oleft = []; else oleft = order(1:(j-1)); end
wolffd@0 136 if j==length(order), oright = []; else oright = order((j+1):length(order)); end
wolffd@0 137 order = [oleft, ch, oright];
wolffd@0 138 nonleaves = find(order>clen);
wolffd@0 139 end
wolffd@0 140
wolffd@0 141 % base cluster indeces for each non-base cluster
wolffd@0 142 basecl = cell(clen-1,1);
wolffd@0 143 for i=1:clen-1,
wolffd@0 144 c1 = Z(i,1); if c1>clen, c1 = basecl{c1-clen}; end
wolffd@0 145 c2 = Z(i,2); if c2>clen, c2 = basecl{c2-clen}; end
wolffd@0 146 basecl{i} = [c1 c2];
wolffd@0 147 end
wolffd@0 148
wolffd@0 149 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0 150 %% set coordinates, color and height and make the struct
wolffd@0 151
wolffd@0 152 % coordinates
wolffd@0 153 if size(coord,1)==2*clen-1, % this is ok already
wolffd@0 154 else
wolffd@0 155 if size(coord,1)==0, % the default
wolffd@0 156 [dummy,coord] = sort(order);
wolffd@0 157 coord = coord';
wolffd@0 158 elseif size(coord,1)==dlen & dlen>clen, % coordinates given for original data
wolffd@0 159 codata = coord;
wolffd@0 160 coord = zeros(clen,size(coord,2));
wolffd@0 161 for i=1:clen, coord(i,:) = mean(codata(find(base==i),:),1); end
wolffd@0 162 end
wolffd@0 163 if size(coord,1)==clen, % average from base clusters
wolffd@0 164 coord = [coord; zeros(clen-1,size(coord,2))];
wolffd@0 165 for i=1:clen-1, coord(i+clen,:) = mean(coord(basecl{i},:),1); end
wolffd@0 166 else
wolffd@0 167 error('Incorrect coordinate matrix.');
wolffd@0 168 end
wolffd@0 169 end
wolffd@0 170
wolffd@0 171 % color
wolffd@0 172 if size(color,1)==2*clen-1, % this is ok already
wolffd@0 173 else
wolffd@0 174 if size(color,1)==0, % the default
wolffd@0 175 color(order,:) = hsv(length(order));
wolffd@0 176 elseif size(color,1)==dlen & dlen>clen, % colors given for original data
wolffd@0 177 codata = color;
wolffd@0 178 color = zeros(clen,3);
wolffd@0 179 for i=1:clen, color(i,:) = mean(codata(find(base==i),:),1); end
wolffd@0 180 end
wolffd@0 181 if size(color,1)==clen, % average from base clusters
wolffd@0 182 color = [color; zeros(clen-1,3)];
wolffd@0 183 for i=1:clen-1, color(i+clen,:) = mean(color(basecl{i},:),1); end
wolffd@0 184 else
wolffd@0 185 error('Incorrect color matrix.');
wolffd@0 186 end
wolffd@0 187 end
wolffd@0 188
wolffd@0 189 % height
wolffd@0 190 if isempty(height),
wolffd@0 191 height = [zeros(clen,1); Z(:,3)];
wolffd@0 192 elseif length(height)==clen-1,
wolffd@0 193 if size(height,2)==clen-1, height = height'; end
wolffd@0 194 height = [zeros(clen,1); height];
wolffd@0 195 elseif length(height)~=2*clen-1,
wolffd@0 196 error('Incorrect height vector.');
wolffd@0 197 end
wolffd@0 198
wolffd@0 199 % make the struct
wolffd@0 200 sC = struct('type','som_clustering',...
wolffd@0 201 'name',name,'base',base,'tree',Z,...
wolffd@0 202 'color',color,'coord',coord,'height',height);
wolffd@0 203 return;
wolffd@0 204
wolffd@0 205 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0 206