Machine Learning Demo

Example using different classifiers.

K-class classification data of 2D points (training data)

K = 3;  % number of classes (pick a number between 2 and 6)
[X,Y] = cv.createConcentricSpheresTestSet(200*K, 2, K);
Y = Y(:);

create grid of points to cover domain of data (testing data)

mn = min(X);
mx = max(X);
N = 100;
[X1,X2] = meshgrid(linspace(mn(1),mx(1),N), linspace(mn(2),mx(2),N));
X1X2 = [X1(:) X2(:)];

some colors for plots

if true
    clrBG = [1.0 0.7 0.7; 0.7 1.0 0.7; 0.7 0.7 1.0; ...
            0.7 1.0 1.0; 1.0 0.7 1.0; 1.0 1.0 0.7];
    clrFG = [1 0 0; 0 1 0; 0 0 1; 0 1 1; 1 0 1; 1 1 0];  % 'rgbcmy'
else
    clrBG = brighten(lines(6), 0.7);
    clrFG = lines(6);
end

different machine learning models

models = {'ANN_MLP', 'DTrees', 'KNearest', ...
    'LogisticRegression', 'NormalBayesClassifier', 'SVM'};
figure('Position',get(0, 'DefaultFigurePosition').*[0.5 0.5 1.8 1.8])
for i=1:numel(models)
    % create ML model
    switch models{i}
        case 'ANN_MLP'
            obj = cv.ANN_MLP();
            obj.LayerSizes = [2 4 4 K];
            obj.setActivationFunction('Sigmoid', 'Param1',1.0, 'Param2',1.0);
            obj.TermCriteria.maxCount = 2000;
            obj.TermCriteria.epsilon = 1e-4;
        case 'DTrees'
            obj = cv.DTrees();
        case 'KNearest'
            obj = cv.KNearest();
        case 'LogisticRegression'
            obj = cv.LogisticRegression();
        case 'NormalBayesClassifier'
            obj = cv.NormalBayesClassifier();
        case 'SVM'
            obj = cv.SVM();
    end

    % train
    if strcmp(models{i}, 'ANN_MLP')
        % ANN_MLP doesn't handle categorical data, we use 1-of-N encoding
        YY = bsxfun(@eq, Y, 0:(K-1));  % [Y==0 Y==1 .. Y==(K-1)]
    else
        YY = int32(Y);
    end
    tic
    obj.train(X, YY);
    toc

    % predict
    YYhat = obj.predict(X1X2);
    if strcmp(models{i}, 'ANN_MLP')
        [~,Yhat] = max(YYhat, [], 2);
    else
        Yhat = int32(YYhat);
    end

    % plot predictions and actual classes
    subplot(2,3,i)
    imagesc(X1(1,:), X2(:,1), reshape(double(Yhat)+1, size(X1)))
    hold on
    if mexopencv.isOctave()
        %HACK: GSCATTER not implemented in Octave
        scatter(X(:,1), X(:,2), 6, clrFG(double(Y)+1,:), 'filled')
    else
        gscatter(X(:,1), X(:,2), Y, clrFG, '.', 6, 'off')
    end
    hold off
    colormap(clrBG(1:K,:))
    axis equal, axis([mn(1) mx(1) mn(2) mx(2)])
    title(models{i}, 'Interpreter','none')
end

Elapsed time is 0.053490 seconds.
Elapsed time is 0.002736 seconds.
Elapsed time is 0.001419 seconds.
Elapsed time is 1.274223 seconds.
Elapsed time is 0.003751 seconds.
Elapsed time is 0.007012 seconds.