Skip to main content

OneClassSVM

#include <Skigen/SVM>

template <typename Scalar = double>
class Skigen::OneClassSVM(kernel=Kernel::RBF, degree=3, gamma=0, coef0=0, nu=0.5, tol=1e-3, max_passes=50, random_state=std::nullopt)

Unsupervised outlier detection with a one-class SVM.

Solves the one-class nu-SVM dual with a dedicated SMO variant and exposes sklearn-style scoring: decision_function, score_samples, predict (+1 inlier, -1 outlier), and offset_.

Mirrors the dense core of sklearn.svm.OneClassSVM.

Attributes:

  • kernel : Kernel

  • gamma : Scalar

  • nu : Scalar

  • support : const std::vector< Eigen::Index > &

  • n_support : int

  • dual_coef : VectorType

  • offset : Scalar

Methods​

SKIGEN_PARAMS()​

decision_function(X)​

Signed distance to the separating hyperplane (rho-shifted).

score_samples(X)​

Unshifted kernel score for each sample.

predict(X)​

Predict +1 for inliers and -1 for outliers.

Example​

Eigen::MatrixXd X(12, 2);
for (int i = 0; i < 10; ++i) {
    X(i, 0) = 0.2 * std::sin(static_cast<double>(i));
    X(i, 1) = 0.2 * std::cos(static_cast<double>(i));
}
X(10, 0) = 6.0;  X(10, 1) = 6.0;
X(11, 0) = -6.0; X(11, 1) = 5.0;

using K = Skigen::OneClassSVM<double>::Kernel;
Skigen::OneClassSVM<double> det(K::RBF, 3, 0.0, 0.0, /*nu=*/0.2);
det.fit(X);
const Eigen::VectorXi labels = det.predict(X);