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 4 of
wolffd@0: %
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: % 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: % some iterations of this script, and selection of RBMs according to
wolffd@0: % their training set performance.
wolffd@0: % This takes considerable time, and for the sake of usability of this script
wolffd@0: % is left our. % It is exemplarily done for hidnum = 1000  in the script "table 1" in the
wolffd@0: % correpsonding folder.
wolffd@0: %
wolffd@0: % % The train and test performances are plotted in a figure
wolffd@0: %
wolffd@0: % For convenicence, The precomputed RBM features are stored in the files
wolffd@0: % accompaining this script.
wolffd@0: % In order to compute new SVM features, delete these files.
wolffd@0: % ---
wolffd@0: 
wolffd@0: global globalvars;
wolffd@0: globalvars.debug = 3;
wolffd@0: 
wolffd@0: % ---
wolffd@0: % vary feature parameters of mixed features
wolffd@0: % ---
wolffd@0:         
wolffd@0: ftype = 'MTTMixedFeatureSonRBM'; %'MTTMixedFeatureStober11Genre';
wolffd@0: 
wolffd@0: fparams_all = struct(...
wolffd@0:             ... % ---
wolffd@0:             ... % these are SONfeatRaw parameters
wolffd@0:             ... % ---
wolffd@0:             'son_filename',{{'rel_music_raw_features+simdata_ISMIR12.mat'}}, ...
wolffd@0:             'son_conf', 2, ...
wolffd@0:              ... % --- 
wolffd@0:             ... % Following: RBM params
wolffd@0:             ... % ---
wolffd@0:             'norm_pre_rbm', 0, ... % norm before RBM?
wolffd@0:             'norm_post_rbm',0, ... % norm before RBM?
wolffd@0:             'rbm_hidNum',[30 50 100 500 1000], ...   % number of hidden units % 500
wolffd@0:             'rbm_eNum', 100, ...
wolffd@0:             'rbm_bNum', 1, ...
wolffd@0:             'rbm_gNum', 1, ...
wolffd@0:             'rbm_lrate1' , [0.05], ... % initial learning rate % 0.01
wolffd@0:             'rbm_lrate2', [0.17], ... %  learning rate, %0.05  
wolffd@0:             'rbm_momentum', [0.1], ... % 0.5
wolffd@0:             'rbm_cost', [0.00002], ... % cost function
wolffd@0:             'rbm_N', 50, ...
wolffd@0:             'rbm_MAX_INC', 10 ...
wolffd@0:         );
wolffd@0: 
wolffd@0: % ---
wolffd@0: % vary parameters for svmlight
wolffd@0: % ---    
wolffd@0: 
wolffd@0: trainparams_all = struct(...
wolffd@0:             'C', [1], ...%    
wolffd@0:             'weighted', [0], ...
wolffd@0:             'dataset', {{'comp_partBinData_ISMIR12_01.mat'}}, ...
wolffd@0:             'inctrain', 0 ...
wolffd@0:             ... % this optional
wolffd@0:             ... %'deltafun', {{'conv_subspace_delta'}}, ...
wolffd@0:             ... %'deltafun_params', {{{[1],[0]},{[5],[1]},{[10],[1]},{[20],[1]},{[30],[1]},{[50],[1]},{[70],[1]}}} ... % normalisation improves results
wolffd@0:             );
wolffd@0: 
wolffd@0: % set training function
wolffd@0: trainfun = @svmlight_wrapper;
wolffd@0: 
wolffd@0: 
wolffd@0: % create test directory
wolffd@0: akt_dir = migrate_to_test_dir();
wolffd@0: 
wolffd@0: 
wolffd@0: % call eval
wolffd@0: out = test_generic_features_parameters_crossval...
wolffd@0:     (fparams_all, trainparams_all, trainfun, ftype);
wolffd@0: 
wolffd@0: % get number of Hidden Units
wolffd@0: fparams = [out(:).fparams];
wolffd@0: hidNum = [fparams(:).rbm_hidNum];
wolffd@0: [hidNum,idx] =sort(hidNum);
wolffd@0: 
wolffd@0: % get sorted test data results
wolffd@0: svm_test_results = [out(idx).mean_ok_test];
wolffd@0: svm_test_results = svm_test_results(1,:);
wolffd@0: 
wolffd@0: % get sorted train data results
wolffd@0: svm_train_results = [out(idx).mean_ok_train];
wolffd@0: svm_train_results = svm_train_results(1,:);
wolffd@0: 
wolffd@0: % ---
wolffd@0: % plot results
wolffd@0: % ---
wolffd@0: figure;
wolffd@0: plot(hidNum,svm_train_results,'r-');
wolffd@0: hold on
wolffd@0: plot(hidNum,svm_test_results,'b');
wolffd@0: legend('Training','Test');
wolffd@0: title ('Figure 4: SVM results for different hidNum');