bri.gpr: Gaussian Process Regression in 1D

In julianfaraway/brinla: Bayesian Regression with INLA

Gaussian Process Regression in 1D

Description

Gaussian Process Regression in 1D

Usage

bri.gpr(x, y, pcprior, nbasis = 25, degree = 2, alpha = 2,
  xout = x, sigma0 = sd(y), rho0 = 0.25 * (max(x) - min(x)))

Arguments

x	the predictor vector
y	the response vector
pcprior	limites for the penalised complexity prior (optional). If specified should be a vector of the form c(r,s) where P(range < r = 0.05) and P(SD(y) > s = 0.05)
nbasis	- number of basis functions for the spline (default is 25)
degree	- degree for splines (default is 2) - allowable possibilities are 0, 1 or 2.
alpha	- controls shape of the GP kernel (default is 2) - 0 < alpha <=2 is possible
xout	- grid on which posterior will be calculated (default is x)
sigma0	- prior mean for the signal SD (default is SD(y))
rho0	- prior mean for the range

Value

list consisting of xout, the posterior mean, the lower 95% credibility band, the upper 95% credibility band and the INLA object containing the fit

Author(s)

Julian Faraway, [e-mail]

julianfaraway/brinla documentation built on April 6, 2023, 2:33 p.m.