wolffd@0: function display_mahalanobis_metric(A, labels)
wolffd@0: % display a mala matrix and its stats
wolffd@0: 
wolffd@0: if nargin < 2
wolffd@0:     labels = num2cell(1:size(A,1));
wolffd@0:     
wolffd@0: elseif ~iscell(labels)
wolffd@0:     
wolffd@0:     features = labels;
wolffd@0:     labels = features.labels;
wolffd@0: end
wolffd@0: 
wolffd@0: 
wolffd@0: 
wolffd@0: figure;
wolffd@0: 
wolffd@0: % plot matrix
wolffd@0: imagesc(A);
wolffd@0: axis xy;
wolffd@0: 
wolffd@0: % set labels
wolffd@0: set(gca,'YTick', 1:numel(labels), ...
wolffd@0:     'YTickLabel', labels);
wolffd@0: set(gca,'XTick',1:numel(labels), ...
wolffd@0:     'XTickLabel', labels);
wolffd@0: 
wolffd@0: % ---
wolffd@0: % approximate parameter weights: 
wolffd@0: %   diagonal and sum(abs(row))
wolffd@0: % TODO: make nshow dependend on percentile
wolffd@0: % ---
wolffd@0: 
wolffd@0: nshow = min(numel(labels), 50);
wolffd@0: figure;
wolffd@0: 
wolffd@0: % get diagonal values of the Matrix
wolffd@0: diagw = abs(diag(A));
wolffd@0: 
wolffd@0: % ---
wolffd@0: % weight with feature values if possible
wolffd@0: % ---
wolffd@0: if exist('features','var')
wolffd@0:     
wolffd@0:     diagw = diagw.* mean(features.vector(),2);
wolffd@0: end
wolffd@0:     
wolffd@0: 
wolffd@0: [diagw, idx] = sort(diagw, 'descend');
wolffd@0: 
wolffd@0: % normalise
wolffd@0: alld = sum(diagw);
wolffd@0: 
wolffd@0: % plot
wolffd@0: bar(diagw(1:nshow)./ alld);
wolffd@0: set(gca,'XTick',1:nshow, ...
wolffd@0:     'XTickLabel', labels(idx(1:nshow)));
wolffd@0: 
wolffd@0: ylabel ('relevance factor');
wolffd@0: 
wolffd@0: if exist('features','var')
wolffd@0:     xlabel 'normalised weight'
wolffd@0: else
wolffd@0:     xlabel 'matrix factors'
wolffd@0: end
wolffd@0: 
wolffd@0: end