Compute the symmetrized kernel matrix Ωs = ½(Ω + a·Ω*)

Used by the symmetric LS-SVR model (Model 4). Entry [k, l] of Ω* is K(xk, -xl), so negation is applied to the columns of X (i.e., to X2).

Usage

sym_kernel_matrix(K, X, a)

Arguments

K

A kernel function from make_kernel().

X

Numeric training matrix (N × p).

a

Symmetry parameter: 1 (even) or -1 (odd).

Value

Numeric N × N matrix Ωs = ½(Ω + a·Ω*).