Function main

examples/one_class_classifiers_ex.cpp:35–243

}

int main()
{
    // We will use column vectors to store our points.  Here we make a convenient typedef
    // for the kind of vector we will use.
    typedef matrix<double,0,1> sample_type;

    // Then we select the kernel we want to use.  For our present problem the radial basis
    // kernel is quite effective.
    typedef radial_basis_kernel<sample_type> kernel_type;

    // Now make the object responsible for training one-class SVMs.
    svm_one_class_trainer<kernel_type> trainer;
    // Here we set the width of the radial basis kernel to 4.0.  Larger values make the
    // width smaller and give the radial basis kernel more resolution.  If you play with
    // the value and observe the program output you will get a more intuitive feel for what
    // that means.
    trainer.set_kernel(kernel_type(4.0));

    // Now sample some 2D points.  The points will be located on the curve defined by the
    // sinc() function.
    std::vector<sample_type> samples;
    sample_type m(2);
    for (double x = -15; x <= 8; x += 0.3)
    {
        m(0) = x;
        m(1) = sinc(x);
        samples.push_back(m);
    }

    // Now train a one-class SVM.  The result is a function, df(), that outputs large
    // values for points from the sinc() curve and smaller values for points that are
    // anomalous (i.e. not on the sinc() curve in our case).
    decision_function<kernel_type> df = trainer.train(samples);

    // So for example, let's look at the output from some points on the sinc() curve.  
    cout << "Points that are on the sinc function:\n";
    m(0) = -1.5; m(1) = sinc(m(0)); cout << "   " << df(m) << endl;  
    m(0) = -1.5; m(1) = sinc(m(0)); cout << "   " << df(m) << endl;  
    m(0) = -0;   m(1) = sinc(m(0)); cout << "   " << df(m) << endl;  
    m(0) = -0.5; m(1) = sinc(m(0)); cout << "   " << df(m) << endl;  
    m(0) = -4.1; m(1) = sinc(m(0)); cout << "   " << df(m) << endl;  
    m(0) = -1.5; m(1) = sinc(m(0)); cout << "   " << df(m) << endl;  
    m(0) = -0.5; m(1) = sinc(m(0)); cout << "   " << df(m) << endl;  

    cout << endl;
    // Now look at some outputs for points not on the sinc() curve.  You will see that
    // these values are all notably smaller. 
    cout << "Points that are NOT on the sinc function:\n";
    m(0) = -1.5; m(1) = sinc(m(0))+4;   cout << "   " << df(m) << endl;
    m(0) = -1.5; m(1) = sinc(m(0))+3;   cout << "   " << df(m) << endl;
    m(0) = -0;   m(1) = -sinc(m(0));    cout << "   " << df(m) << endl;
    m(0) = -0.5; m(1) = -sinc(m(0));    cout << "   " << df(m) << endl;
    m(0) = -4.1; m(1) = sinc(m(0))+2;   cout << "   " << df(m) << endl;
    m(0) = -1.5; m(1) = sinc(m(0))+0.9; cout << "   " << df(m) << endl;
    m(0) = -0.5; m(1) = sinc(m(0))+1;   cout << "   " << df(m) << endl;

    // The output is as follows:
    /*