Daniel@0: function sC = som_clstruct(Z,varargin)
Daniel@0: 
Daniel@0: %SOM_CLSTRUCT Create a clustering struct or set its field values.
Daniel@0: %
Daniel@0: %  sC = som_clstruct(Z, [argID, value, ...]) 
Daniel@0: %
Daniel@0: %    Z  = linkage(pdist(sM.codebook));
Daniel@0: %    sC = som_clstruct(Z); 
Daniel@0: %    sC = som_clstruct(sC,'coord',som_vis_coords(lattice,msize));
Daniel@0: %    sC = som_clstruct(sC,'color',som_colorcode(sM));
Daniel@0: %    sC = som_clstruct(sC,'base',sC.base(som_bmus(sM,sD)));
Daniel@0: %
Daniel@0: %  Input and output arguments ([]'s are optional): 
Daniel@0: %   Z         (matrix) size clen-1 x 3, where clen is the number of 
Daniel@0: %                      base clusters. This is a clustering matrix 
Daniel@0: %                      similar to that produced by LINKAGE in 
Daniel@0: %                      Statistical Toolbox. See SOM_LINKAGE.
Daniel@0: %             (struct) clustering struct (as produced by this function)
Daniel@0: %   [argID,   (string) See below. Each pair is the fieldname and 
Daniel@0: %    value]   (varies) the value to be given to that field.
Daniel@0: %
Daniel@0: %   sC        (struct) clustering struct
Daniel@0: % 
Daniel@0: %   The clustering struct is based on the assumption that there 
Daniel@0: %   is a base partitioning of the SOM (or data) which is saved in 
Daniel@0: %   the .base field of the struct. Then a hierarchical clustering
Daniel@0: %   is applied to this base partitioning. The results are saved to 
Daniel@0: %   .tree field of the struct. Each cluster (base and combined)
Daniel@0: %   has also three properties: height, coordinate and color, which 
Daniel@0: %   are used in the visualizations. The fields of the struct are:
Daniel@0: %     .type   (string) 'som_clustering'
Daniel@0: %     .name   (string) Identifier for the clustering.
Daniel@0: %     .tree   (matrix) Size clen-1 x 3, as argument Z above. 
Daniel@0: %     .base   (vector) Size dlen x 1, the basic groups of data 
Daniel@0: %                      forming the base clusters, e.g. as a result 
Daniel@0: %                      of partitive clustering. Allowed values are 
Daniel@0: %                       1:clen indicating the base cluster
Daniel@0: %                              to which the data belongs to. 
Daniel@0: %                       NaN    indicating that the data has
Daniel@0: %                              been ignored in the clustering
Daniel@0: %                      By default [1:clen]. 
Daniel@0: %     .height (vector) Size 2*clen-1 x 1, (clustering) height for each 
Daniel@0: %                      cluster. By default 0 for each base cluster and
Daniel@0: %                      .tree(:,3) for the others.
Daniel@0: %     .coord  (matrix) Size 2*clen-1 x *, coordinate for each cluster, 
Daniel@0: %                      By default the coordinates are set so that 
Daniel@0: %                      the base clusters are ordered on a line, and the
Daniel@0: %                      position of each combined cluster is average of 
Daniel@0: %                      the base clusters that constitute it.
Daniel@0: %     .color  (matrix) Size 2*clen-1 x 3, color for each cluster. 
Daniel@0: %                      By default the colors are set so that the 
Daniel@0: %                      base clusters are ordered on a line, like above,
Daniel@0: %                      and then colors are assigned from the 'hsv' 
Daniel@0: %                      colormap to the base clusters. The color
Daniel@0: %                      of each combined cluster is average as above.
Daniel@0: %
Daniel@0: % Height, coord and color can also be specified in alternate forms:
Daniel@0: %   'height' (vector) size 2*clen-1 x 1, if given explicitly
Daniel@0: %                     size clen-1 x 1, specified heights of the 
Daniel@0: %                          combined clusters (the base cluster heights
Daniel@0: %                          are all = 0)
Daniel@0: %                     size 0 x 0, default value is used
Daniel@0: %   'coord'  (matrix) size 2*clen-1 x *, if given explicitly
Daniel@0: %                     size clen x *, to give coordinates for base 
Daniel@0: %                          clusters; the coordinate of combined clusters
Daniel@0: %                          are averaged from these
Daniel@0: %                     size dlen x *, to give coordinates of the 
Daniel@0: %                          original data: the cluster coordinates are
Daniel@0: %                          averaged from these based on base clusters
Daniel@0: %                     size 0 x 0, default value is used
Daniel@0: %   'color'  (matrix) as 'coord'
Daniel@0: %
Daniel@0: % See also  SOM_CLPLOT, SOM_CLVALIDITY, SOM_CLGET, SOM_CLLINKAGE.
Daniel@0: 
Daniel@0: % Copyright (c) 2000 by the SOM toolbox programming team.
Daniel@0: % Contributed to SOM Toolbox on XXX by Juha Vesanto
Daniel@0: % http://www.cis.hut.fi/projects/somtoolbox/
Daniel@0: 
Daniel@0: % Version 2.0beta juuso 180800
Daniel@0: 
Daniel@0: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Daniel@0: 
Daniel@0: if isstruct(Z), 
Daniel@0:   base = Z.base; 
Daniel@0:   color = Z.color; 
Daniel@0:   coord = Z.coord; 
Daniel@0:   height = Z.height; 
Daniel@0:   name = Z.name; 
Daniel@0:   Z = Z.tree; 
Daniel@0: else
Daniel@0:   base  = []; 
Daniel@0:   color = []; 
Daniel@0:   coord = []; 
Daniel@0:   height = []; 
Daniel@0:   name = ''; 
Daniel@0: end    
Daniel@0: clen  = size(Z,1)+1; 
Daniel@0: 
Daniel@0: i=1; 
Daniel@0: while i<=length(varargin), 
Daniel@0:   argok = 1; 
Daniel@0:   if ischar(varargin{i}), 
Daniel@0:     switch varargin{i}, 
Daniel@0:      case 'tree',   i=i+1; Z = varargin{i}; clen = size(Z,1)+1;
Daniel@0:      case 'base',   i=i+1; base = varargin{i}; 
Daniel@0:      case 'color',  i=i+1; color = varargin{i}; 
Daniel@0:      case 'coord',  i=i+1; coord = varargin{i}; 
Daniel@0:      case 'height', i=i+1; height = varargin{i}; 
Daniel@0:      case 'name',   i=i+1; name = varargin{i}; 
Daniel@0:      otherwise argok=0; 
Daniel@0:     end
Daniel@0:   else argok = 0; 
Daniel@0:   end
Daniel@0:   if ~argok, disp(['(som_clstruct) Ignoring invalid argument #' num2str(i+1)]); end
Daniel@0:   i = i+1; 
Daniel@0: end
Daniel@0: 
Daniel@0: if isempty(base), 
Daniel@0:   dlen = clen; 
Daniel@0:   base = 1:dlen; 
Daniel@0: else
Daniel@0:   dlen = length(base); 
Daniel@0:   if any(base)>clen | any(base)<1, error('Incorrect base partition vector.'); end
Daniel@0: end
Daniel@0: 
Daniel@0: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Daniel@0: %% analysis of hierarchy
Daniel@0: 
Daniel@0: % order of base clusters
Daniel@0: order = 2*clen-1; 
Daniel@0: nonleaves = 1; 
Daniel@0: while any(nonleaves), 
Daniel@0:   j = nonleaves(1); 
Daniel@0:   ch = Z(order(j)-clen,1:2);
Daniel@0:   if j==1, oleft = []; else oleft = order(1:(j-1)); end
Daniel@0:   if j==length(order), oright = []; else oright = order((j+1):length(order)); end
Daniel@0:   order = [oleft, ch, oright];
Daniel@0:   nonleaves = find(order>clen); 
Daniel@0: end
Daniel@0: 
Daniel@0: % base cluster indeces for each non-base cluster
Daniel@0: basecl = cell(clen-1,1); 
Daniel@0: for i=1:clen-1, 
Daniel@0:   c1 = Z(i,1); if c1>clen, c1 = basecl{c1-clen}; end
Daniel@0:   c2 = Z(i,2); if c2>clen, c2 = basecl{c2-clen}; end
Daniel@0:   basecl{i} = [c1 c2];   
Daniel@0: end
Daniel@0: 
Daniel@0: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Daniel@0: %% set coordinates, color and height and make the struct
Daniel@0: 
Daniel@0: % coordinates
Daniel@0: if size(coord,1)==2*clen-1, % this is ok already
Daniel@0: else
Daniel@0:   if size(coord,1)==0, % the default    
Daniel@0:     [dummy,coord] = sort(order); 
Daniel@0:     coord = coord'; 
Daniel@0:   elseif size(coord,1)==dlen & dlen>clen, % coordinates given for original data
Daniel@0:     codata = coord; 
Daniel@0:     coord = zeros(clen,size(coord,2)); 
Daniel@0:     for i=1:clen, coord(i,:) = mean(codata(find(base==i),:),1); end  
Daniel@0:   end
Daniel@0:   if size(coord,1)==clen, % average from base clusters
Daniel@0:     coord = [coord; zeros(clen-1,size(coord,2))]; 
Daniel@0:     for i=1:clen-1, coord(i+clen,:) = mean(coord(basecl{i},:),1); end
Daniel@0:   else
Daniel@0:     error('Incorrect coordinate matrix.'); 
Daniel@0:   end
Daniel@0: end
Daniel@0: 
Daniel@0: % color
Daniel@0: if size(color,1)==2*clen-1, % this is ok already
Daniel@0: else
Daniel@0:   if size(color,1)==0, % the default
Daniel@0:     color(order,:) = hsv(length(order)); 
Daniel@0:   elseif size(color,1)==dlen & dlen>clen, % colors given for original data
Daniel@0:     codata = color; 
Daniel@0:     color = zeros(clen,3); 
Daniel@0:     for i=1:clen, color(i,:) = mean(codata(find(base==i),:),1); end  
Daniel@0:   end
Daniel@0:   if size(color,1)==clen, % average from base clusters
Daniel@0:     color = [color; zeros(clen-1,3)]; 
Daniel@0:     for i=1:clen-1, color(i+clen,:) = mean(color(basecl{i},:),1); end
Daniel@0:   else
Daniel@0:     error('Incorrect color matrix.'); 
Daniel@0:   end
Daniel@0: end
Daniel@0: 
Daniel@0: % height 
Daniel@0: if isempty(height), 
Daniel@0:   height = [zeros(clen,1); Z(:,3)]; 
Daniel@0: elseif length(height)==clen-1, 
Daniel@0:   if size(height,2)==clen-1, height = height'; end
Daniel@0:   height = [zeros(clen,1); height]; 
Daniel@0: elseif length(height)~=2*clen-1, 
Daniel@0:   error('Incorrect height vector.'); 
Daniel@0: end
Daniel@0: 
Daniel@0: % make the struct
Daniel@0: sC = struct('type','som_clustering',...
Daniel@0: 	    'name',name,'base',base,'tree',Z,...
Daniel@0: 	    'color',color,'coord',coord,'height',height); 
Daniel@0: return; 
Daniel@0: 
Daniel@0: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%  
Daniel@0: