llikGP: Calculate a GP log likelihood

In laGP: Local Approximate Gaussian Process Regression

Calculate a GP log likelihood

Description

Calculate a Gaussian process (GP) log likelihood or posterior probability with reference to a C-side GP object

Usage

llikGP(gpi, dab = c(0, 0), gab = c(0, 0))
llikGPsep(gpsepi, dab = c(0, 0), gab = c(0, 0))

Arguments

gpi	a C-side GP object identifier (positive integer); e.g., as returned by newGP
gpsepi	similar to gpi but indicating a separable GP object
dab	ab for the lengthscale parameter, see Details
gab	ab for the nugget parameter, see Details

Details

An “ab” parameter is a non-negative 2-vector describing shape and rate parameters to a Gamma prior; a zero-setting for either value results in no-prior being used in which case a log likelihood is returned. If both ab parameters are specified, then the value returned can be interpreted as a log posterior density. See darg for more information about ab

Value

A real-valued scalar is returned.

Author(s)

Robert B. Gramacy [e-mail]

See Also

mleGP, darg

Examples

## partly following the example in mleGP
if(require("MASS")) {

  ## motorcycle data and predictive locations
  X <- matrix(mcycle[,1], ncol=1)
  Z <- mcycle[,2]

  ## get sensible ranges
  d <- darg(NULL, X)
  g <- garg(list(mle=TRUE), Z)
  
  ## initialize the model
  gpi <- newGP(X, Z, d=d$start, g=g$start)

  ## calculate log likelihood
  llikGP(gpi)
  ## calculate posterior probability
  llikGP(gpi, d$ab, g$ab)

  ## clean up
  deleteGP(gpi)
}

laGP documentation built on March 31, 2023, 9:46 p.m.