[res_ordering] = ViewPCStruct(run_structures,viewparams)
'ViewPCStruct': Parallel coordinate visualization of the generations at which most
frequent edges appearing in the structures learned by an EDA. The
vertical axis represents the generation at which edges (shown in the
horizontal axis) has been learned. A line between two points means that
both edges appear in the same structure learned at the same generation.
INPUT:
run_structures: Contain the data structures with all the structures
learned by the probability models in every run and generation (see program ReadStructures.m for details.
viewparams{1} = fs; % fs: Font size for the images
viewparams{2} : Matrices with edges that will be shown. One row for each
edge. If viewparams{2}== [], the algorithm finds a subset of edges
according viewparams{3}
viewparams{3} = const_edg : Minimal number of times that an edge has to appear in (all) the structures learned
to be selected for visualization. Since the clarity of the parallel coordinate
visualization depend on the number of variables, this is an important parameter.
viewparams{4} = min_edg : Minimal number of edges in the substructures selected (min_edg>0)
viewparams{5} : Method used to order the variables before displaying them
using parallel coordinates. Ordering may help to reduce cluttering, improving
visualization. viewparams{5} = 'none' if the current given ordering is used.
viewparams{5} = 'random' for random order of variables
Ordering methods can be implemented by the user. Currently implemented is
'ClusterUsingCorr' which clusters togethers variables with strong
correlation using affinity propagation.
viewparams{6} = distance. Distance used to cluster edges from their
appearance in the structures (distances used by matlab command pdist (ex. 'correlation', 'euclidean',etc.) can
be used (see help pdist).
OUTPUT:
res_ordering{1}: Generations from which the edges have been extracted
res_ordering{2}: Executions from which the edges have been extracted
res_ordering{3}: List of the edges
Last version 8/26/2008. Roberto Santana (roberto.santana@ehu.es)0001 function[res_ordering] = ViewPCStruct(run_structures,viewparams) 0002 % [res_ordering] = ViewPCStruct(run_structures,viewparams) 0003 % 'ViewPCStruct': Parallel coordinate visualization of the generations at which most 0004 % frequent edges appearing in the structures learned by an EDA. The 0005 % vertical axis represents the generation at which edges (shown in the 0006 % horizontal axis) has been learned. A line between two points means that 0007 % both edges appear in the same structure learned at the same generation. 0008 % INPUT: 0009 % run_structures: Contain the data structures with all the structures 0010 % learned by the probability models in every run and generation (see program ReadStructures.m for details. 0011 % viewparams{1} = fs; % fs: Font size for the images 0012 % viewparams{2} : Matrices with edges that will be shown. One row for each 0013 % edge. If viewparams{2}== [], the algorithm finds a subset of edges 0014 % according viewparams{3} 0015 % viewparams{3} = const_edg : Minimal number of times that an edge has to appear in (all) the structures learned 0016 % to be selected for visualization. Since the clarity of the parallel coordinate 0017 % visualization depend on the number of variables, this is an important parameter. 0018 % 0019 % viewparams{4} = min_edg : Minimal number of edges in the substructures selected (min_edg>0) 0020 % viewparams{5} : Method used to order the variables before displaying them 0021 % using parallel coordinates. Ordering may help to reduce cluttering, improving 0022 % visualization. viewparams{5} = 'none' if the current given ordering is used. 0023 % viewparams{5} = 'random' for random order of variables 0024 % Ordering methods can be implemented by the user. Currently implemented is 0025 % 'ClusterUsingCorr' which clusters togethers variables with strong 0026 % correlation using affinity propagation. 0027 % viewparams{6} = distance. Distance used to cluster edges from their 0028 % appearance in the structures (distances used by matlab command pdist (ex. 'correlation', 'euclidean',etc.) can 0029 % be used (see help pdist). 0030 % OUTPUT: 0031 % res_ordering{1}: Generations from which the edges have been extracted 0032 % res_ordering{2}: Executions from which the edges have been extracted 0033 % res_ordering{3}: List of the edges 0034 % 0035 % Last version 8/26/2008. Roberto Santana (roberto.santana@ehu.es) 0036 0037 indexmatrix = run_structures{1}; 0038 0039 fs = viewparams{1}; 0040 substruct = viewparams{2}; 0041 const_edg = viewparams{3}; 0042 min_edg = viewparams{4}; 0043 edge_ordering_method = viewparams{5}; 0044 distance = viewparams{6}; 0045 0046 0047 0048 results{1} = []; 0049 0050 if isempty(substruct) 0051 [RepEdges] = SelectEdgesToShow(run_structures,const_edg); % Edges are selected according to const_edg value 0052 if(isempty(RepEdges)) 0053 disp('No edges were selected. Reduce the number of times required to an edge to appear.') %No edges were found 0054 return 0055 end 0056 else 0057 for i=1:size(substruct,1) 0058 RepEdges(1,i) = indexmatrix(substruct(i,1),substruct(i,2)); % Indices of the edges in indexmatrix 0059 end 0060 end 0061 0062 0063 0064 [AllRepVectors] = ExtractSubstructures(run_structures,RepEdges,min_edg); % The substructures to be shown are extracted 0065 0066 nedges = size(RepEdges,2); 0067 ordering = [1:nedges]; 0068 0069 if(strcmp(edge_ordering_method,'random') == 1) 0070 ordering = randperm(nedges); 0071 elseif(strcmp(edge_ordering_method,'none') ~= 1) 0072 res_ordering = eval([edge_ordering_method,'(AllRepVectors,distance)';]); %The method for ordering the variables is invoked 0073 ordering = res_ordering{1}; % The new ordering of the variables should be the first output of the method 0074 end, 0075 0076 0077 ShowParallelCoord(fs,AllRepVectors(:,ordering)); % Parallel coordinates are shown in the new ordering 0078 0079 res_ordering{1} = ordering; 0080 0081 for i=1:nedges 0082 [row,col] = find(indexmatrix==RepEdges(ordering(i))); 0083 res_ordering{3}(i,:) = [row(1),col(1)]; 0084 end 0085 0086 return 0087 0088 0089 0090 0091 0092 0093 0094 0095 0096