wolffd@0: % ---
wolffd@0: % The DistMeasureMahal class states a wrapper for 
wolffd@0: % special Mahalanobis similarity distance, and is compatible with the 
wolffd@0: % DistMeasure class
wolffd@0: % ---
wolffd@0: classdef DistMeasureNNet <  handle
wolffd@0:     
wolffd@0:     properties (SetAccess = private)
wolffd@0:         
wolffd@0:         net;
wolffd@0:         
wolffd@0:         featX;
wolffd@0:         
wolffd@0:         ids;
wolffd@0:         ret_ids;
wolffd@0:     end
wolffd@0: 
wolffd@0:     methods
wolffd@0:         
wolffd@0:     % ---
wolffd@0:     % constructor
wolffd@0:     % ---
wolffd@0:     function m = DistMeasureNNet(clips, net, featX)
wolffd@0: 
wolffd@0:         if size(featX, 2) ~= numel(clips) 
wolffd@0:             error 'wrong input format'
wolffd@0:         end
wolffd@0: 
wolffd@0:         % fill index and generate matrix;
wolffd@0:         m.ids = [clips.id];
wolffd@0: 
wolffd@0:         % reverse index
wolffd@0:         m.ret_ids = sparse(numel(m.ids),1);
wolffd@0:         m.ret_ids(m.ids) = 1:numel(m.ids);
wolffd@0: 
wolffd@0:         % ---
wolffd@0:         % save neural net and lazy-copy features
wolffd@0:         % ---
wolffd@0:         m.net = net;
wolffd@0:         
wolffd@0:         m.featX = featX;
wolffd@0:         
wolffd@0:     end
wolffd@0:     
wolffd@0: 
wolffd@0:     % ---
wolffd@0:     % this function returns the
wolffd@0:     % similarity of two clip indices
wolffd@0:     % --- 
wolffd@0:     function out = mat(m, idxa, idxb)
wolffd@0:     
wolffd@0:         if nargin == 1
wolffd@0:             idxa = 1:numel(m.ids);
wolffd@0:             idxb = 1:numel(m.ids);
wolffd@0:         end
wolffd@0:         
wolffd@0:         % cycle through all index combinations
wolffd@0:         out = zeros(numel(idxa), numel(idxb));
wolffd@0:         for i = 1:numel(idxa)
wolffd@0:            for j = 1:numel(idxb)
wolffd@0:                
wolffd@0:                 % calculate distance vector
wolffd@0:                 deltas = m.featX(:,idxa(i)) - m.featX(:,idxb(j));
wolffd@0:                 
wolffd@0:                 % return distance from net
wolffd@0:                 out(i,j) = m.net.calcValue(deltas);
wolffd@0:            end
wolffd@0:         end
wolffd@0:         
wolffd@0:     end
wolffd@0:   
wolffd@0:     % ---
wolffd@0:     % returns the distance for the two input clips
wolffd@0:     % ---
wolffd@0:     function out = distance(m, clipa, clipb)
wolffd@0:         posa = m.get_clip_pos(clipa);
wolffd@0:         posb = m.get_clip_pos(clipb);
wolffd@0:         
wolffd@0:         out = m.mat(posa, posb);
wolffd@0:     end
wolffd@0:     
wolffd@0:     % ---
wolffd@0:     % returns a list of n (default = 10) clips most 
wolffd@0:     % similar to the  input
wolffd@0:     % ---
wolffd@0:     function [clips, dist] = get_nearest(m, clip, n)
wolffd@0:     % list = get_nearest(m, clip, n)
wolffd@0:     % 
wolffd@0:     % returns a list of n (default = 10) clips most 
wolffd@0:     % similar to the  input
wolffd@0: 
wolffd@0:         % default number of results
wolffd@0:         if nargin == 2 
wolffd@0: 
wolffd@0:             n = 10;
wolffd@0:         end
wolffd@0:         
wolffd@0:         % return all clips in case n = 0
wolffd@0:         if n == 0; n = numel(m.ids); end
wolffd@0: 
wolffd@0:         % get clip positions
wolffd@0:         pos = m.get_clip_pos(clip);
wolffd@0: 
wolffd@0:         % sort according to distance
wolffd@0:         [sc, idx] = sort( m.mat(pos, 1:numel(m.ids)), 'ascend');
wolffd@0: 
wolffd@0:         % we only output relevant data
wolffd@0:         idx = idx(sc < inf);
wolffd@0: 
wolffd@0:         if numel(idx) > 0
wolffd@0:             % create clips form best ids
wolffd@0:             clips = MTTClip( m.ids( idx(1:min(n, end))));
wolffd@0:             dist = m.mat(pos, idx(1:min(n, end)));
wolffd@0: 
wolffd@0:         else
wolffd@0:             clips = [];
wolffd@0:             dist = [];
wolffd@0:         end
wolffd@0:     end
wolffd@0:     
wolffd@0:     
wolffd@0: 
wolffd@0:     function [clips, dist] = present_nearest(m, clip, n)
wolffd@0:     % plays and shows the n best hits for a given clip    
wolffd@0: 
wolffd@0:         % default number of results
wolffd@0:         if nargin == 2 
wolffd@0: 
wolffd@0:             n = 3;
wolffd@0:         end
wolffd@0: 
wolffd@0:         % get best list
wolffd@0:         [clips, dist] = get_nearest(m, clip, n);
wolffd@0: 
wolffd@0:         clip.audio_features_basicsm.visualise();
wolffd@0:         for i = 1:numel(clips)
wolffd@0:             fprintf('\n\n\n- Rank %d, distance: %1.4f \n\n',i, dist(i));
wolffd@0: 
wolffd@0:             clips(i).audio_features_basicsm.visualise();
wolffd@0:             h = gcf();
wolffd@0:             t = clips(i).play(20);
wolffd@0:             pause(t);
wolffd@0:             close(h);
wolffd@0:         end
wolffd@0:     end
wolffd@0: 
wolffd@0:     function a = visualise(m)
wolffd@0: 
wolffd@0:         figure;
wolffd@0: 
wolffd@0:         % plot data
wolffd@0: 
wolffd@0:         imagesc(m.mat);
wolffd@0: 
wolffd@0:         a = gca;
wolffd@0:         set(a,'YTick',[1:numel(m.ids)], 'YTickLabel',m.ids);
wolffd@0:         set(a,'XTick',[1:numel(m.ids)], 'XTickLabel', m.ids);
wolffd@0: 
wolffd@0:         axis xy;
wolffd@0:         colormap(hot);
wolffd@0:     end
wolffd@0: 
wolffd@0:     % end methods
wolffd@0:     end
wolffd@0:     
wolffd@0:     % ---
wolffd@0:     % private methods
wolffd@0:     % ---
wolffd@0:     methods(Access = private)
wolffd@0:         
wolffd@0:     function out = get_clip_pos(m, clip)
wolffd@0:         % returns position in mat for given clip
wolffd@0: 
wolffd@0:         out = m.ret_ids(clip.id);
wolffd@0:     end
wolffd@0:         
wolffd@0:     end
wolffd@0:     
wolffd@0: end