annotate core/tools/machine_learning/display_mahalanobis_metric.m @ 0:e9a9cd732c1e tip

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 display_mahalanobis_metric(A, labels)
wolffd@0 2 % display a mala matrix and its stats
wolffd@0 3
wolffd@0 4 if nargin < 2
wolffd@0 5 labels = num2cell(1:size(A,1));
wolffd@0 6
wolffd@0 7 elseif ~iscell(labels)
wolffd@0 8
wolffd@0 9 features = labels;
wolffd@0 10 labels = features.labels;
wolffd@0 11 end
wolffd@0 12
wolffd@0 13
wolffd@0 14
wolffd@0 15 figure;
wolffd@0 16
wolffd@0 17 % plot matrix
wolffd@0 18 imagesc(A);
wolffd@0 19 axis xy;
wolffd@0 20
wolffd@0 21 % set labels
wolffd@0 22 set(gca,'YTick', 1:numel(labels), ...
wolffd@0 23 'YTickLabel', labels);
wolffd@0 24 set(gca,'XTick',1:numel(labels), ...
wolffd@0 25 'XTickLabel', labels);
wolffd@0 26
wolffd@0 27 % ---
wolffd@0 28 % approximate parameter weights:
wolffd@0 29 % diagonal and sum(abs(row))
wolffd@0 30 % TODO: make nshow dependend on percentile
wolffd@0 31 % ---
wolffd@0 32
wolffd@0 33 nshow = min(numel(labels), 50);
wolffd@0 34 figure;
wolffd@0 35
wolffd@0 36 % get diagonal values of the Matrix
wolffd@0 37 diagw = abs(diag(A));
wolffd@0 38
wolffd@0 39 % ---
wolffd@0 40 % weight with feature values if possible
wolffd@0 41 % ---
wolffd@0 42 if exist('features','var')
wolffd@0 43
wolffd@0 44 diagw = diagw.* mean(features.vector(),2);
wolffd@0 45 end
wolffd@0 46
wolffd@0 47
wolffd@0 48 [diagw, idx] = sort(diagw, 'descend');
wolffd@0 49
wolffd@0 50 % normalise
wolffd@0 51 alld = sum(diagw);
wolffd@0 52
wolffd@0 53 % plot
wolffd@0 54 bar(diagw(1:nshow)./ alld);
wolffd@0 55 set(gca,'XTick',1:nshow, ...
wolffd@0 56 'XTickLabel', labels(idx(1:nshow)));
wolffd@0 57
wolffd@0 58 ylabel ('relevance factor');
wolffd@0 59
wolffd@0 60 if exist('features','var')
wolffd@0 61 xlabel 'normalised weight'
wolffd@0 62 else
wolffd@0 63 xlabel 'matrix factors'
wolffd@0 64 end
wolffd@0 65
wolffd@0 66 end