wolffd@0: % ---
wolffd@0: % This script trains similarity measures and evaluates the 
wolffd@0: % impact of the number of hidden states as displayed in figure 3 of
wolffd@0: % % Feature Preprocessing with RBMs for Music Similarity Learning
wolffd@0: % Son N. Tran, Daniel Wolff, Tillman Weyde, Artur Garcez, AES53rd
wolffd@0: % conference 
wolffd@0: %
wolffd@0: % The output is printed in the console and plotted afterwards
wolffd@0: % 
wolffd@0: % please note that the RBM training is a probabilistic process, and 
wolffd@0: % thus the papers' results can only be reproduced approximately with 
wolffd@0: % large numbers iterations of this script, and selection of RBMs according to
wolffd@0: % their training set performance.
wolffd@0: % ---
wolffd@0: 
wolffd@0: % this version reproduces the figure approximately using precomputed RBM
wolffd@0: [test(1), train(1)] = rbm_fig3('rbm_h30');
wolffd@0: [test(2), train(2)] = rbm_fig3('rbm_h50');
wolffd@0: [test(3), train(3)] = rbm_fig3('rbm_h100');
wolffd@0: [test(4), train(4)] = rbm_fig3('rbm_h500');
wolffd@0: [test(5), train(5)] = rbm_fig3('rbm_h1000');
wolffd@0: 
wolffd@0: % optionally, in order to test new RBMs, use the code below
wolffd@0: % [test(1), train(1)] = rbm_fig3(30);
wolffd@0: % [test(2), train(2)] = rbm_fig3(50);
wolffd@0: % [test(3), train(3)] = rbm_fig3(100);
wolffd@0: % [test(4), train(4)] = rbm_fig3(500);
wolffd@0: % [test(5), train(5)] = rbm_fig3(1000);
wolffd@0: 
wolffd@0: hidNum = [30 50 100 500 1000];
wolffd@0: hFig = figure;
wolffd@0: set(hFig,'Units','centimeters');
wolffd@0: set(hFig, 'Position', [10 10 10 6]);
wolffd@0: plot(hidNum,train*100,'--rx');
wolffd@0: hold on
wolffd@0: plot(hidNum,test*100,'-bo');
wolffd@0: lg = legend('Training','Test');
wolffd@0: set(lg,'Location','SouthEast');
wolffd@0: title ('Figure 4: GRADIENT results for different hidNum');