gp_fit: Find Maximum Posterior solution for a Gaussian Process model.

In svdataman/gin: Gaussian Processes for time series inference

Description

gp_fit returns the posterior mode for a Gaussian Process model.

Usage

gp_fit(theta.0, acv.model = NULL, dat = NULL, logPrior = NULL,
  method = "Nelder-Mead", trace = 0, theta.scale = NULL,
  maxit = 5000, chatter = 0, PDcheck = FALSE)

Arguments

acv.model	(name) name of the function to compute ACV(tau | theta)
dat	(matrix) an N * 3 matrix of data: 3 columns
logPrior	(name) Name of the function returning the log Prior density.
method	- choice of method for optim
theta.scale	- passed as parscale to optim.
maxit	- passed to optim.
chatter	(integer) higher values give more run-time feedback
PDcheck	(logical) use Matrix::nearPD to coerse the matrix

Value

A list with components (similar to optim):

par	parameter values (maximum likelihood estimates)
err	std. dev. of MLEs (based on Hessian matrix)
acv.model	name of function used to compute ACV
value	value of gp_logLikelihood at maximum
covergence	convergence output from optim
nfunction.calls	counts output from optim

Notes

Find the posterior mode for a GP model, given data dat. See gp_logLikelihood for details of the input data form. The user must supply the name of a suitable ACV function and some intial values for the ACV's parameters (the hyper-parameters of the GP). The parameters are then optimised using optim. (This is essentially a wrapper function applying optim on gp_logPosterior.) If no logPrior function is supplied, the result is equivalent to Maximum Likelihood Estimation.

See Also

gp_logLikelihood, gp_logPosterior

svdataman/gin documentation built on March 12, 2021, 7:37 a.m.