wolffd@0: function [out, stats] = write_mat_results_ISMIR13RBM_singletraining(dirin,fileout)
wolffd@0: % combine the test results from the directories supplied,
wolffd@0: % group them according to dataset parameter values
wolffd@0: % combine the test results from the directories supplied,
wolffd@0: % group them according to dataset parameter values
wolffd@0: 
wolffd@0: features = [];
wolffd@0: show = 1;
wolffd@0: 
wolffd@0: if nargin == 0
wolffd@0:     dirin{1} = './';
wolffd@0: end
wolffd@0: 
wolffd@0: global comparison;
wolffd@0: global comparison_ids;
wolffd@0:      
wolffd@0: newout = [];
wolffd@0: thisdir = pwd;
wolffd@0: % loop through al lthe result directories and 
wolffd@0: for diri = 1:numel(dirin)
wolffd@0: 
wolffd@0:     % ---
wolffd@0:     % go to directory and locate file
wolffd@0:     % ---
wolffd@0:     cd(dirin{diri});
wolffd@0: 
wolffd@0:     u = dir();
wolffd@0:     u = {u.name};
wolffd@0:     [idx, strpos] = substrcellfind(u, '_finalresults.mat', 1);
wolffd@0:     
wolffd@0:     if numel(idx) < 1
wolffd@0:         error 'This directory contains no valid test data';
wolffd@0:     end
wolffd@0:     
wolffd@0:     % just one or more tests in this folder?
wolffd@0:     if exist('file','var') && isnumeric(file)
wolffd@0:         cprint(1, 'loading one result file');
wolffd@0:         file = u{idx(file)};
wolffd@0:         data = load(file);
wolffd@0:         sappend(out,data.out);
wolffd@0:     else
wolffd@0:         for filei = 1:numel(idx)
wolffd@0:             cprint(1, 'loading result file %i of %i',filei, numel(idx));
wolffd@0:             file = u{idx(filei)};
wolffd@0:             data = load(file);
wolffd@0:             newout = sappend(newout,data.out);
wolffd@0:         end
wolffd@0:     end
wolffd@0:     % reset act directory
wolffd@0:     cd(thisdir);
wolffd@0: end
wolffd@0: 
wolffd@0: % ---
wolffd@0: % filter according to training parameter C
wolffd@0: %
wolffd@0: % NOTE :if we don't filter by C, we get strong overfitting with training
wolffd@0: % success > 96 % and test set performance aorund 65 %
wolffd@0: % ---
wolffd@0: cs = zeros(numel(newout),1);
wolffd@0: for i=1:numel(newout)
wolffd@0:     cs(i) = newout(i).trainparams.C;
wolffd@0: end
wolffd@0: cvals = unique(cs);
wolffd@0: 
wolffd@0: for ci=1:numel(cvals)
wolffd@0:     valididx = find(cs == cvals(ci));
wolffd@0:     filteredout = newout(valididx);
wolffd@0:     
wolffd@0:     % ---
wolffd@0:     % get parameter statistics
wolffd@0:     % ---
wolffd@0:     stats = test_generic_display_param_influence(filteredout, show);
wolffd@0: 
wolffd@0:     % get maximal values for each test set bin
wolffd@0:     % ---
wolffd@0:     % trainparams.dataset contains sets which have each only one bin of the
wolffd@0:     % ismir testsets 
wolffd@0:     % ---
wolffd@0:     max_idx = [stats.trainparams.dataset.mean_ok_train.max_idx];   
wolffd@0:     ok_test = zeros(2, numel(max_idx));
wolffd@0:     ok_train =  zeros(2, numel(max_idx));
wolffd@0:     ok_config = [];
wolffd@0:     % cycle over all test sets and save best result
wolffd@0:     for i=1:numel(max_idx)
wolffd@0:         ok_test(:,i) = filteredout(max_idx(i)).mean_ok_test;
wolffd@0:         ok_train(:,i) = filteredout(max_idx(i)).mean_ok_train;
wolffd@0:         ok_config = sappend(ok_config,struct('trainparams',filteredout(max_idx(i)).trainparams, ...
wolffd@0:                                                 'fparams',filteredout(max_idx(i)).fparams));
wolffd@0:     end
wolffd@0:     % save the stuff
wolffd@0:     out(ci).mean_ok_test = mean(ok_test,2);
wolffd@0:     out(ci).var_ok_test = var(ok_test,0,2);
wolffd@0:     out(ci).mean_ok_train = mean(ok_train,2);
wolffd@0:     out(ci).var_ok_train = var(ok_train,0,2);
wolffd@0:     out(ci).trainparams.C = cvals(ci);
wolffd@0:     out(ci).ok_config = ok_config;
wolffd@0:     out(ci).ok_test = ok_test;
wolffd@0:     out(ci).ok_train = ok_train;
wolffd@0: end
wolffd@0: 
wolffd@0: % ---
wolffd@0: % show results for different C
wolffd@0: % ---
wolffd@0: if numel([out.mean_ok_test]) > 1 && show
wolffd@0:     
wolffd@0:     % plot means  % plot std = sqrt(var) % plot training results
wolffd@0:     figure;
wolffd@0:     boxplot([out.mean_ok_test], sqrt([out.var_ok_test]), [out.mean_ok_train]);
wolffd@0:     title (sprintf('Performance for all configs'));
wolffd@0: end
wolffd@0: 
wolffd@0: % --- 
wolffd@0: % write max. test success
wolffd@0: % ---
wolffd@0: mean_ok_test = [out.mean_ok_test];
wolffd@0: [val, idx] = max(mean_ok_test(1,:));
wolffd@0: if show
wolffd@0:     fprintf(' --- Maximal test set success: nr. %d, %3.2f percent. --- \n', idx, val * 100)
wolffd@0: end
wolffd@0: 
wolffd@0: % save
wolffd@0: save([hash(strcat(dirin{:}),'md5') '_summary'], 'out');
wolffd@0: 
wolffd@0: end
wolffd@0: 
wolffd@0: 
wolffd@0: 
wolffd@0: function boxplot(mean, std, train);
wolffd@0: 
wolffd@0:     bar([train; mean]', 1.5);
wolffd@0:     hold on;
wolffd@0:     errorbar(1:size(mean,2), mean(1,:), std(1,:),'.');
wolffd@0: %     plot(train,'rO');
wolffd@0:     colormap(spring);
wolffd@0:     axis([0 size(mean,2)+1 max(0, min(min([train mean] - 0.1))) max(max([train mean] + 0.1))]);
wolffd@0: end