Choose the boundary factor and number of basis functions

Settings not given in gp_opts() are chosen from the lengthscale prior using gp_approx_constants(). The boundary factor is set by the upper (95%) prior quantile, as longer lengthscales need a wider boundary. The number of basis functions is then set by the lower (5%) prior quantile, as shorter lengthscales need more basis functions. The periodic kernel has no boundary, so only the number of basis functions is chosen.

Usage

gp_basis_settings(gp, n, gp_type)

Arguments

gp

A <gp_opts> object.

n

Length of the Gaussian process.

gp_type

Integer kernel type as used in the Stan model (0: squared exponential, 1: periodic, 2: Matern).

Value

A list with the boundary factor L and number of basis functions M.