wolffd@0: function [sC,old2new,newi] = som_clset(sC,action,par1,par2)
wolffd@0: 
wolffd@0: % SOM_CLSET Create and/or set values in the som_clustering struct.
wolffd@0: %
wolffd@0: %   first argument
wolffd@0: %     sC       (struct) a som_clustering struct
wolffd@0: %     Z        (matrix) size nb-1 x 3, as given by LINKAGE function
wolffd@0: %     base     (vector) size dlen x 1, a partitioning of the data
wolffd@0: %
wolffd@0: %   actions    
wolffd@0: %     'remove'           removes the indicated clusters (par1: vector)
wolffd@0: %     'add'              add a cluster by making a combination of the indicated
wolffd@0: %                        clusters (par1: vector)
wolffd@0: %     %'move'             moves a child cluster (par1: scalar) from a parent to another
wolffd@0: %     %                   (par2: vector 1 x 2)
wolffd@0: %     'merge'            like 'add', followed by removing the indicated clusters (par1: vector)
wolffd@0: %     %'split'            the indicated cluster (par1: scalar) is partitioned into indicated
wolffd@0: %     %                   parts (par2: vector), which are then added, while the indicated cluster
wolffd@0: %     %                   (par1) is removed
wolffd@0: %     'coord'            sets the coordinates of base clusters (par1: matrix nb x *), and 
wolffd@0: %                        recalculates coordinates of the derived clusters (by averaging base cluster
wolffd@0: %                        coordinates)
wolffd@0: %     'color'            sets the colors of base clusters (par1: matrix nb x 3), and recalculates
wolffd@0: %                        colors of the derived clusters (as averages of base cluster colors)
wolffd@0: %                        
wolffd@0: %   sC
wolffd@0: %     .type     (string) 'som_clustering'
wolffd@0: %     .name     (string) Identifier for the clustering.
wolffd@0: %     .nb       (scalar) Number of base clusters in the clustering.
wolffd@0: %     .base     (vector) Size dlen x 1, the basic groups of data 
wolffd@0: %                        forming the base clusters, e.g. as a result 
wolffd@0: %                        of partitive clustering. Allowed values are 
wolffd@0: %                         1:nb   indicating the base cluster
wolffd@0: %                                to which the data belongs to. 
wolffd@0: %                         NaN    indicating that the data has
wolffd@0: %                                been ignored in the clustering                        
wolffd@0: %     .nc       (scalar) Number of clusters in the clustering (nb + derived clusters).
wolffd@0: %     .children (cellarray) size nc x 1, each cell gives the list of indeces
wolffd@0: %                        of child clusters for the cluster
wolffd@0: %     .parent   (vector) size nc x 1, the index of parent of each cluster 
wolffd@0: %                        (or zero if the cluster does not have a parent)
wolffd@0: %     .coord    (matrix) size nc x *, visualization coordinates for each cluster
wolffd@0: %                        By default the coordinates are set so that 
wolffd@0: %                        the base clusters are ordered on a line, and the
wolffd@0: %                        position of each combined cluster is average of 
wolffd@0: %                        the base clusters that constitute it.
wolffd@0: %     .color    (matrix) size nc x 3, color for each cluster. 
wolffd@0: %                        By default the colors are set so that the 
wolffd@0: %                        base clusters are ordered on a line,
wolffd@0: %                        and then colors are assigned from the 'hsv' 
wolffd@0: %                        colormap to the base clusters. The color
wolffd@0: %                        of each combined cluster is average as above.
wolffd@0: %     .cldist   (string) Default cluster distance function.
wolffd@0: 
wolffd@0: inew = []; 
wolffd@0: if isstruct(sC), 
wolffd@0:     % it should be a som_clustering struct
wolffd@0:     old2new = [1:sC.nc];
wolffd@0: elseif size(sC,2)==3, 
wolffd@0:     % assume it is a cluster hierarchy matrix Z 
wolffd@0:     sC = Z2sC(sC); 
wolffd@0:     old2new = [1:sC.nc];
wolffd@0: else
wolffd@0:     % assume it is a partitioning vector
wolffd@0:     base = sC; 
wolffd@0:     u = unique(base(isfinite(base)));
wolffd@0:     old2new = sparse(u,1,1:length(u));
wolffd@0:     base = old2new(base);
wolffd@0:     sC = part2sC(base); 
wolffd@0: end 
wolffd@0: 
wolffd@0: switch action, 
wolffd@0: case 'remove',        
wolffd@0:     for i=1:length(par1),         
wolffd@0:         [sC,o2n] = removecluster(sC,old2new(par1(i)));
wolffd@0:         old2new = o2n(old2new);
wolffd@0:     end 
wolffd@0: case 'add', 
wolffd@0:     [sC,old2new,inew] = addmergedcluster(sC,par1);    
wolffd@0: case 'move',
wolffd@0:     % not implemented yet
wolffd@0: case 'split', 
wolffd@0:     % not implemented yet
wolffd@0: case 'merge', 
wolffd@0:     [sC,old2new,inew] = addmergedcluster(sC,par1);
wolffd@0:     for i=1:length(par1), 
wolffd@0:         [sC,o2n] = removecluster(sC,old2new(par1(i)));
wolffd@0:         old2new = o2n(old2new);
wolffd@0:     end 
wolffd@0: case 'color', 
wolffd@0:     sC.color = derivative_average(sC,par1);
wolffd@0: case 'coord',
wolffd@0:     sC.coord = derivative_average(sC,par1);
wolffd@0: end 
wolffd@0: 
wolffd@0: return;
wolffd@0: 
wolffd@0: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0: %% subfunctions
wolffd@0: 
wolffd@0: function sC = clstruct(nb,nc)
wolffd@0: 
wolffd@0:     sC = struct('type','som_clustering',...
wolffd@0:                 'name','','base',[],'nb',nb,'nc',nc,...
wolffd@0:                 'parent',[],'children',[],'coord',[],'color',[],'cldist','centroid');
wolffd@0:     sC.base = [1:nb]; 
wolffd@0:     sC.parent = zeros(nc,1);
wolffd@0:     sC.children = cell(nc,1); sC.children(:) = {[]}; 
wolffd@0:     sC.coord = zeros(nc,2);
wolffd@0:     sC.color = zeros(nc,3);
wolffd@0:     return;
wolffd@0: 
wolffd@0: function Z = sC2Z(sC,height)
wolffd@0: 
wolffd@0:     if nargin<2, height = 'level'; end
wolffd@0: 
wolffd@0:     root   = find(sC.parent==0); 
wolffd@0:     order  = [root]; 
wolffd@0:     ch     = sC.children(root); 
wolffd@0:     while any(ch), i = ch(1); order = [ch(1), order]; ch = [ch(2:end), sC.children{i}]; end 
wolffd@0: 
wolffd@0:     he = zeros(sC.nc,1); 
wolffd@0:     if strcmp(height,'level'), 
wolffd@0:         ch = sC.children{root}; 
wolffd@0:         while any(ch),
wolffd@0:             i = ch(1); he(i) = he(sC.parent(i))+1; 
wolffd@0:             ch = [ch(2:end), sC.children{i}]; 
wolffd@0:         end 
wolffd@0:         he = max(he)-he; 
wolffd@0:     elseif strcmp(height,'level2'), 
wolffd@0:         for i=order, if any(sC.children{i}), he(i) = max(he(sC.children{i}))+1; end, end
wolffd@0:     else
wolffd@0:         %he = som_cldist ( between children )
wolffd@0:     end 
wolffd@0:     
wolffd@0:     Z = zeros(sC.nb-1,3);    
wolffd@0:     i = sC.nb-1; 
wolffd@0:     inds = root; 
wolffd@0:     while i>0, 
wolffd@0:         ch = sC.children{inds(1)}; h = he(inds(1)); inds = [inds(2:end), ch]; 
wolffd@0:         if length(ch)>=2,
wolffd@0:             for k=1:length(ch)-2, Z(i,:) = [i-1, ch(k), h]; i = i - 1; end
wolffd@0:             Z(i,:) = [ch(end-1) ch(end) h]; i = i - 1;             
wolffd@0:         end 
wolffd@0:     end 
wolffd@0:     return;
wolffd@0: 
wolffd@0: function sC = Z2sC(Z)
wolffd@0: 
wolffd@0:     nb        = size(Z,1)+1;
wolffd@0:     nc        = 2*nb-1;
wolffd@0:     sC        = clstruct(nb,nc);
wolffd@0:     sC.base   = [1:nb];
wolffd@0:     for i=1:nc, 
wolffd@0:         j = find(Z(:,1)==i | Z(:,2)==i); 
wolffd@0:         sC.parent(i) = nb+j;
wolffd@0:         sC.children{sC.parent(i)}(end+1) = i; 
wolffd@0:     end 
wolffd@0:     % coords and color
wolffd@0:     order = nc; 
wolffd@0:     nonleaves = 1; 
wolffd@0:     while any(nonleaves), 
wolffd@0:         j = nonleaves(1); 
wolffd@0:         ch = sC.children{order(j)};
wolffd@0:         if j==1, oleft = []; else oleft = order(1:(j-1)); end
wolffd@0:         if j==length(order), oright = []; else oright = order((j+1):length(order)); end
wolffd@0:         order = [oleft, ch, oright];
wolffd@0:         nonleaves = find(order>nb); 
wolffd@0:     end
wolffd@0:     [dummy,co] = sort(order);     
wolffd@0:     sC.coord   = derivative_average(sC,co');
wolffd@0:     H          = hsv(nb+1);
wolffd@0:     sC.color   = derivative_average(sC,H(co,:));    
wolffd@0:     return;
wolffd@0:     
wolffd@0: function sC = part2sC(part)
wolffd@0: 
wolffd@0:     nb      = max(part); 
wolffd@0:     nc      = nb+1; 
wolffd@0:     sC      = clstruct(nb,nc);
wolffd@0:     sC.base = part; 
wolffd@0:     sC.parent(1:nb) = nc; 
wolffd@0:     sC.children{nc} = [1:nb]; 
wolffd@0:     co       = [1:nb]'; 
wolffd@0:     sC.coord = derivative_average(sC,co);
wolffd@0:     H        = hsv(nb+1);
wolffd@0:     sC.color = derivative_average(sC,H(1:nb,:));
wolffd@0:     return;
wolffd@0: 
wolffd@0: function [sC,old2new] = removecluster(sC,ind)
wolffd@0:   
wolffd@0:     old2new = [1:sC.nc]; 
wolffd@0:     parent_ind = sC.parent(ind);
wolffd@0:     ch = sC.children{ind};
wolffd@0:     if ~parent_ind, 
wolffd@0:         % trying to remove root cluster - no go
wolffd@0:         return; 
wolffd@0:     elseif ~any(ch), 
wolffd@0:         % trying to remove a base cluster - no go
wolffd@0:         return;
wolffd@0:     else
wolffd@0:         % ok, proceed
wolffd@0:         old2new = [1:ind-1 0 ind:sC.nc-1];
wolffd@0:         % update parent and child fields
wolffd@0:         sC.parent(ch) = parent_ind;
wolffd@0:         sC.children{parent_ind} = setdiff([sC.children{parent_ind}, ch],ind);
wolffd@0:         % remove old cluster
wolffd@0:         j = [1:ind-1, ind+1:sC.nc]; 
wolffd@0:         sC.parent   = sC.parent(j);
wolffd@0:         sC.children = sC.children(j);
wolffd@0:         sC.color    = sC.color(j,:);
wolffd@0:         sC.coord    = sC.coord(j,:);
wolffd@0:         sC.nc       = sC.nc-1; 
wolffd@0:         % update old indeces to new indices
wolffd@0:         sC.parent = old2new(sC.parent);
wolffd@0:         for i=1:sC.nc, sC.children{i} = old2new(sC.children{i}); end
wolffd@0:     end     
wolffd@0:     return;
wolffd@0: 
wolffd@0: function [sC,old2new,inew] = addmergedcluster(sC,inds)
wolffd@0: 
wolffd@0:     old2new    = [1:sC.nc]; 
wolffd@0:     inew       = 0; 
wolffd@0:     p_inds     = sC.parent(inds); 
wolffd@0:     if ~all(p_inds(1)==p_inds),  
wolffd@0:         % clusters are not siblings - no go
wolffd@0:         return;
wolffd@0:     end
wolffd@0:     parent_ind = p_inds(1); 
wolffd@0:     if isempty(setdiff(sC.children{parent_ind},inds)),  
wolffd@0:         % such a merged cluster exists already
wolffd@0:         return;     
wolffd@0:     else
wolffd@0:         % ok, proceed
wolffd@0:         inew = parent_ind;
wolffd@0:         old2new = [1:inew-1,inew+1:sC.nc+1];
wolffd@0:         % add the new cluster (=copy of parent_ind) 
wolffd@0:         j = [1:inew,inew:sC.nc];
wolffd@0:         sC.parent   = sC.parent(j);
wolffd@0:         sC.children = sC.children(j);
wolffd@0:         sC.color    = sC.color(j,:);
wolffd@0:         sC.coord    = sC.coord(j,:);
wolffd@0:         sC.nc       = sC.nc+1;
wolffd@0:         % update old indeces to new indices
wolffd@0:         sC.parent = old2new(sC.parent);
wolffd@0:         for i=1:sC.nc, sC.children{i} = old2new(sC.children{i}); end
wolffd@0:         inds = old2new(inds);
wolffd@0:         parent_ind = old2new(parent_ind);
wolffd@0:         % update parent, child, color and coord fields
wolffd@0:         sC.parent(inds)         = inew; 
wolffd@0:         sC.parent(inew)         = parent_ind;
wolffd@0:         sC.children{inew}       = inds; 
wolffd@0:         sC.children{parent_ind} = [setdiff(sC.children{parent_ind}, inds), inew];
wolffd@0:         b = baseind(sC,inew); 
wolffd@0:         sC.color(inew,:)        = mean(sC.color(b,:));
wolffd@0:         sC.coord(inew,:)        = mean(sC.coord(b,:));
wolffd@0:     end    
wolffd@0:     return;
wolffd@0:     
wolffd@0: function C = derivative_average(sC,Cbase)
wolffd@0: 
wolffd@0:     [n dim] = size(Cbase);
wolffd@0:     if n ~= sC.nb, error('Color / Coord matrix should have nb rows'); end
wolffd@0:     C = zeros(sC.nc,dim);     
wolffd@0:     for i=1:sC.nc, C(i,:) = mean(Cbase(baseind(sC,i),:)); end   
wolffd@0:     return;
wolffd@0:     
wolffd@0: function bi = baseind(sC,ind)
wolffd@0: 
wolffd@0:     bi = [ind]; 
wolffd@0:     i = 1; 
wolffd@0:     while i<=length(bi), bi = [bi, sC.children{bi(i)}]; end 
wolffd@0:     bi = bi(bi<=sC.nb);
wolffd@0:     return;
wolffd@0:   
wolffd@0: 
wolffd@0:       
wolffd@0: