Kernel-PCA linearity test

Generalizes a linear subspace-extraction method to a kernelized nonlinear version, then compares downstream task performance across a range of kernels (linear vs. RBF/sigmoid/polynomial/Laplace) to empirically test whether a claimed linear subspace hypothesis holds, rather than assuming linearity by construction.