Mercurial > hg > camir-aes2014

annotate core/magnatagatune/tests_evals/rbm_subspace/write_mat_results_ISMIR13RBM_windows.m @ 0:e9a9cd732c1e tip

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
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 [out, stats] = write_mat_results_ISMIR13RBM_windows(dirin,fileout)
wolffd@0 2 % combine the test results from the directories supplied,
wolffd@0 3 % and recombine the bins to a coherent dataset
wolffd@0 4
wolffd@0 5 % folders_win1 = {'130507_svm_disttest_subsconv_rbm_largegrid_21_r2385','130507_svm_disttest_subsconv_rbm_largegrid_22_r2385','130507_svm_disttest_subsconv_rbm_largegrid_23_r2385','130507_svm_disttest_subsconv_rbm_largegrid_24_r2385','130508_svm_disttest_subsconv_rbm_largegrid_25_r2391'};
wolffd@0 6 % folders_win2 = {'130511_svm_disttest_subsconv_rbm_largegrid_21_r2391','130511_svm_disttest_subsconv_rbm_largegrid_22_r2391','130511_svm_disttest_subsconv_rbm_largegrid_23_r2391','130511_svm_disttest_subsconv_rbm_largegrid_24_r2391','130513_svm_disttest_subsconv_rbm_largegrid_25_r2391'};
wolffd@0 7 %folders = {folders_win1{:}, folders_win2{:}};
wolffd@0 8
wolffd@0 9 features = [];
wolffd@0 10 show = 1;
wolffd@0 11
wolffd@0 12 if nargin == 0
wolffd@0 13 dirin{1} = './';
wolffd@0 14 end
wolffd@0 15
wolffd@0 16 global comparison;
wolffd@0 17 global comparison_ids;
wolffd@0 18
wolffd@0 19 newout = [];
wolffd@0 20 thisdir = pwd;
wolffd@0 21 % loop through al lthe result directories and
wolffd@0 22 for diri = 1:numel(dirin)
wolffd@0 23
wolffd@0 24 % ---
wolffd@0 25 % go to directory and locate file
wolffd@0 26 % ---
wolffd@0 27 cd(dirin{diri});
wolffd@0 28
wolffd@0 29 u = dir();
wolffd@0 30 u = {u.name};
wolffd@0 31 [idx, strpos] = substrcellfind(u, '_finalresults.mat', 1);
wolffd@0 32
wolffd@0 33 if numel(idx) < 1
wolffd@0 34 error 'This directory contains no valid test data';
wolffd@0 35 end
wolffd@0 36
wolffd@0 37 % just one or more tests in this folder?
wolffd@0 38 if exist('file','var') && isnumeric(file)
wolffd@0 39 cprint(1, 'loading one result file');
wolffd@0 40 file = u{idx(file)};
wolffd@0 41 data = load(file);
wolffd@0 42 sappend(out,data.out);
wolffd@0 43 else
wolffd@0 44 for filei = 1:numel(idx)
wolffd@0 45 cprint(1, 'loading result file %i of %i',filei, numel(idx));
wolffd@0 46 file = u{idx(filei)};
wolffd@0 47 data = load(file);
wolffd@0 48 newout = sappend(newout,data.out);
wolffd@0 49 end
wolffd@0 50 end
wolffd@0 51 % reset act directory
wolffd@0 52 cd(thisdir);
wolffd@0 53 end
wolffd@0 54
wolffd@0 55 % ---
wolffd@0 56 % filter C values!
wolffd@0 57 % ---
wolffd@0 58 allcs = zeros(numel(newout),1);
wolffd@0 59 for i=1:numel(newout)
wolffd@0 60 allcs(i) = newout(i).trainparams.C;
wolffd@0 61 end
wolffd@0 62 valididx = find(allcs == 1); % select C!
wolffd@0 63 newout = newout(valididx);
wolffd@0 64
wolffd@0 65 % ---
wolffd@0 66 % filter according to training parameter deltafun_params(windows)
wolffd@0 67 % ---
wolffd@0 68 wind = cell(numel(newout),1);
wolffd@0 69 for i=1:numel(newout)
wolffd@0 70 wind{i} = cell2str(newout(i).trainparams.deltafun_params);
wolffd@0 71 end
wolffd@0 72 windu = unique(wind);
wolffd@0 73
wolffd@0 74 % save the combined out
wolffd@0 75 newout2 = newout;
wolffd@0 76
wolffd@0 77 % do the loop as in write_mat_results_ISMIR13RBM
wolffd@0 78 for wi=1:numel(windu)
wolffd@0 79 newout = newout2(strcellfind(wind, windu{wi},1));
wolffd@0 80
wolffd@0 81 % bundle all datasets
wolffd@0 82 fout = sameparamsubset(newout, 'dataset','');
wolffd@0 83 out = [];
wolffd@0 84 for ci=1:numel(fout)
wolffd@0 85 filteredout = fout{ci};
wolffd@0 86
wolffd@0 87 ok_test = zeros(2, numel(filteredout));
wolffd@0 88 ok_train = zeros(2, numel(filteredout));
wolffd@0 89 ok_config = [];
wolffd@0 90
wolffd@0 91
wolffd@0 92 tmpout = filteredout(1);
wolffd@0 93 % cycle over all test sets and get new means
wolffd@0 94 for i=1:numel(filteredout)
wolffd@0 95 ok_test(:,i) = filteredout(i).mean_ok_test;
wolffd@0 96 ok_train(:,i) = filteredout(i).mean_ok_train;
wolffd@0 97 end
wolffd@0 98
wolffd@0 99 % save the stuff
wolffd@0 100 tmpout.mean_ok_test = mean(ok_test,2);
wolffd@0 101 tmpout.var_ok_test = var(ok_test,0,2);
wolffd@0 102 tmpout.mean_ok_train = mean(ok_train,2);
wolffd@0 103 tmpout.var_ok_train = var(ok_train,0,2);
wolffd@0 104
wolffd@0 105 tmpout.ok_test = ok_test;
wolffd@0 106 tmpout.ok_train = ok_train;
wolffd@0 107
wolffd@0 108 % put it in output structure
wolffd@0 109 out = sappend(out,tmpout);
wolffd@0 110 end
wolffd@0 111
wolffd@0 112 % % ---
wolffd@0 113 % % show results
wolffd@0 114 % % ---
wolffd@0 115 % if numel([out.mean_ok_test]) > 1 && show
wolffd@0 116 %
wolffd@0 117 % % plot means % plot std = sqrt(var) % plot training results
wolffd@0 118 % figure;
wolffd@0 119 % boxplot([out.mean_ok_test], sqrt([out.var_ok_test]), [out.mean_ok_train]);
wolffd@0 120 % title (sprintf('Performance for all configs'));
wolffd@0 121 % end
wolffd@0 122
wolffd@0 123 % ---
wolffd@0 124 % write max. test/train success
wolffd@0 125 % ---
wolffd@0 126 mean_ok_train = [out.mean_ok_train];
wolffd@0 127 [val, idx] = max(mean_ok_train(1,:));
wolffd@0 128 if show
wolffd@0 129 fprintf(' --- Maximal train set success for %s: nr. %d, %3.2f percent, test result %3.2f percent. --- \n', windu{wi}, idx, val * 100,out(idx).mean_ok_test(1,:)*100)
wolffd@0 130 end
wolffd@0 131
wolffd@0 132 end
wolffd@0 133
wolffd@0 134 end
wolffd@0 135
wolffd@0 136 function [out, param_hash,idx] = sameparamsubset(in, ignoret,ignoref)
wolffd@0 137 % ---
wolffd@0 138 % build index of all existing configurations
wolffd@0 139 % ---
wolffd@0 140 param_hash = cell(numel(in),1);
wolffd@0 141 for i=1:numel(in)
wolffd@0 142 params = struct('trainparams',in(i).trainparams, ...
wolffd@0 143 'fparams',in(i).fparams);
wolffd@0 144
wolffd@0 145 % remove the dataset param
wolffd@0 146 if ~isempty(ignoret)
wolffd@0 147 params.trainparams = rmfield(params.trainparams,ignoret);
wolffd@0 148 end
wolffd@0 149 if ~isempty(ignoref)
wolffd@0 150 params.fparams = rmfield(params.fparams,ignoref);
wolffd@0 151 end
wolffd@0 152
wolffd@0 153 phash = hash(xml_format(params),'md5');
wolffd@0 154 param_hash{i} = phash;
wolffd@0 155 end
wolffd@0 156
wolffd@0 157 % ---
wolffd@0 158 % recombine the data for different datasets!
wolffd@0 159 % ---
wolffd@0 160 cvals = unique(param_hash);
wolffd@0 161
wolffd@0 162 out = {};
wolffd@0 163 for ci=1:numel(cvals)
wolffd@0 164 idx{ci} = strcellfind(param_hash,cvals(ci),1);
wolffd@0 165 out{ci} = in(idx{ci});
wolffd@0 166 end
wolffd@0 167
wolffd@0 168 end
wolffd@0 169
wolffd@0 170 function boxplot(mean, std, train);
wolffd@0 171
wolffd@0 172 bar([train; mean]', 1.5);
wolffd@0 173 hold on;
wolffd@0 174 errorbar(1:size(mean,2), mean(1,:), std(1,:),'.');
wolffd@0 175 % plot(train,'rO');
wolffd@0 176 colormap(spring);
wolffd@0 177 axis([0 size(mean,2)+1 max(0, min(min([train mean] - 0.1))) max(max([train mean] + 0.1))]);
wolffd@0 178 end