annotate reproduce_AES53rd/rerun_figure3.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 % ---
wolffd@0 2 % This script trains similarity measures and evaluates the
wolffd@0 3 % impact of the number of hidden states as displayed in figure 3 of
wolffd@0 4 % % Feature Preprocessing with RBMs for Music Similarity Learning
wolffd@0 5 % Son N. Tran, Daniel Wolff, Tillman Weyde, Artur Garcez, AES53rd
wolffd@0 6 % conference
wolffd@0 7 %
wolffd@0 8 % The output is printed in the console and plotted afterwards
wolffd@0 9 %
wolffd@0 10 % please note that the RBM training is a probabilistic process, and
wolffd@0 11 % thus the papers' results can only be reproduced approximately with
wolffd@0 12 % large numbers iterations of this script, and selection of RBMs according to
wolffd@0 13 % their training set performance.
wolffd@0 14 % ---
wolffd@0 15
wolffd@0 16 % this version reproduces the figure approximately using precomputed RBM
wolffd@0 17 [test(1), train(1)] = rbm_fig3('rbm_h30');
wolffd@0 18 [test(2), train(2)] = rbm_fig3('rbm_h50');
wolffd@0 19 [test(3), train(3)] = rbm_fig3('rbm_h100');
wolffd@0 20 [test(4), train(4)] = rbm_fig3('rbm_h500');
wolffd@0 21 [test(5), train(5)] = rbm_fig3('rbm_h1000');
wolffd@0 22
wolffd@0 23 % optionally, in order to test new RBMs, use the code below
wolffd@0 24 % [test(1), train(1)] = rbm_fig3(30);
wolffd@0 25 % [test(2), train(2)] = rbm_fig3(50);
wolffd@0 26 % [test(3), train(3)] = rbm_fig3(100);
wolffd@0 27 % [test(4), train(4)] = rbm_fig3(500);
wolffd@0 28 % [test(5), train(5)] = rbm_fig3(1000);
wolffd@0 29
wolffd@0 30 hidNum = [30 50 100 500 1000];
wolffd@0 31 hFig = figure;
wolffd@0 32 set(hFig,'Units','centimeters');
wolffd@0 33 set(hFig, 'Position', [10 10 10 6]);
wolffd@0 34 plot(hidNum,train*100,'--rx');
wolffd@0 35 hold on
wolffd@0 36 plot(hidNum,test*100,'-bo');
wolffd@0 37 lg = legend('Training','Test');
wolffd@0 38 set(lg,'Location','SouthEast');
wolffd@0 39 title ('Figure 4: GRADIENT results for different hidNum');