load('P300data.mat')
epochs = extract_epochs(P300Data(:,4),flash_times,flash_ids,0.1*Fs,0.6*Fs);

p2p_vals = peak2peak(epochs.data, 2);
epochs.data = epochs.data(p2p_vals < 100, :);
epochs.ids = epochs.ids(p2p_vals < 100);

features_target = epochs.data(epochs.ids == 2, :);
features_non_target = epochs.data(epochs.ids == 1, :);

target_erp = mean(features_target, 1);
nontarget_erp = mean(features_non_target, 1);

hold on
times = ((1:size(epochs.data,2)) - 0.1*Fs)/Fs;
plot(times, target_erp)
plot(times, nontarget_erp)
xlabel('Time (ms)')
ylabel('Potential (uV)')
legend('Target', 'Non-target')

var = epochs.data;
features = p300_features(var');
features2 = p300_features(features_target');
features3 = p300_features(features_non_target');

features2_labels = ones( length(features2), 1 );
features3_labels = 2 * ones( length(features2), 1 );

both = [features2; features3];
both_labels = [features2_labels; features3_labels];

figure
scatter3( features2(:, 1), features2(:, 2), features2(:, 3), 'g' ) ;
hold on
scatter3( features3(:, 1), features3(:, 2), features3(:, 3), 'r' ) ;
hold off
legend('target', 'non-target')

model = fitcdiscr( both,  both_labels, 'DiscrimType', 'linear') ;
CVmodel = crossval(model, 'KFold', 5);
CVerror = kfoldLoss(CVmodel);

model2 = fitcsvm( both,  both_labels, 'KernelFunction', 'RBF');
CVmodel1 = crossval(model2, 'KFold', 5);
CVerror1 = kfoldLoss(CVmodel1);