view core/magnatagatune/tests_evals/test_generic_display_results_absviolated.m @ 0:e9a9cd732c1e tip

first hg version after svn

function out = test_generic_display_results_absviolated(file)
% out = test_generic_display_results_absviolated(file) 
%
% displays the finalresults mat file and enables 
% further analysis and duiagnostics of the individual runs

global db_MTTAudioFeatureSlaney08;
global db_magnamixedfeat_genrebasicsm;
global db_MTTMixedFeatureGenreBasicSmPCA;


global comparison;
global comparison_ids;
        
if nargin < 1 || isempty(file) || isnumeric(file)
    u = dir();
    u = {u.name};
    [idx, strpos] = substrcellfind(u, '_finalresults.mat', 1);
    
    if exist('file','var') && isnumeric(file)
        file = u{idx(file)};
    else
        file = u{idx(1)};
    end
end

load(file);

% ---
% % get statistics for feature parameters
% Visualise the accuracy and variance
% ---
if isfield(out, 'inctrain')
    for i = 1:numel(out)
        
        % ---
        % get training and test sample sizes
        % ---
        nData = [];
        n_train_data = zeros(1,numel(out(i).inctrain.dataPartition));
        n_test_data = zeros(1,numel(out(i).inctrain.dataPartition));
        for j = 1:numel(out(i).inctrain.dataPartition)
            n_train_data(j) = mean(out(i).inctrain.dataPartition(j).TrainSize);
            n_test_data(j) = mean(out(i).inctrain.dataPartition(j).TestSize);
        end
        
        % ---
        % get lost percentages
        % ---
        mean_lost_test = 1 - [out(i).inctrain.mean_ok_test];
        mean_lost_test = mean_lost_test(1,:).* n_test_data;
        
        var_ok_test = sqrt([out(i).inctrain.var_ok_test]);
        
        mean_lost_train = 1 - [out(i).inctrain.mean_ok_train];
        mean_lost_train = mean_lost_train(1,:).* n_train_data;
        
        % plot test results
        figure;
        subplot(2,1,1)
        plot(n_train_data, mean_lost_test,'r');
        hold;
        % plot training results
        plot(n_train_data, mean_lost_train,'m');
        
        xlabel ('# training constraints');
        ylabel ('# constraints violated');
        legend ('test','training');
        
        plot(n_train_data, mean_lost_test + var_ok_test(1,:).* n_test_data,'r:');
        plot(n_train_data, mean_lost_test - (var_ok_test(1,:).* n_test_data),'r:');
        
        % ---
        % get percentage of unknown data examples learnt
        % ---
        lost_test_not_in_train = mean_lost_test - mean_lost_train;
        ntest_not_in_train = n_test_data - n_train_data;
        
        lost_test_not_in_train = lost_test_not_in_train ./ ntest_not_in_train;
    
        lost_test_not_in_train(isnan(lost_test_not_in_train)) = 0;
        
        subplot(2,1,2)
        plot(n_train_data, lost_test_not_in_train);
        
        xlabel ('# training constraints');
        ylabel ('% unknown constraints violated');
    end
end

   
% --- 
% write max. training success
% ---
mean_ok_test = [out.mean_ok_test];
[val, idx] = max(mean_ok_test(1,:));
fprintf(' --- Maximal training success: nr. %d, %3.2f percent. --- \n', idx, val * 100)

end