calculate_posterior_VL: Vecchia Laplace extension of GPVecchia for non-Gaussian data

In GPvecchia: Scalable Gaussian-Process Approximations

Vecchia Laplace extension of GPVecchia for non-Gaussian data

Description

Vecchia Laplace extension of GPVecchia for non-Gaussian data

Usage

calculate_posterior_VL(
  z,
  vecchia.approx,
  likelihood_model = c("gaussian", "logistic", "poisson", "gamma", "beta", "gamma_alt"),
  covparms,
  covmodel = "matern",
  likparms = list(alpha = 2, sigma = sqrt(0.1)),
  max.iter = 50,
  convg = 1e-06,
  return_all = FALSE,
  y_init = NA,
  prior_mean = rep(0, length(z)),
  verbose = FALSE
)

Arguments

z	an array of real numbers representing observations
vecchia.approx	a vecchia object as generated by vecchia_specify()
likelihood_model	text describing likelihood model to be used for observations. Can be "gaussian","logistic", "poisson", "gamma", or "beta"
covparms	covariance parameters as a vector
covmodel	type of the model covariance or selected elements of the covariance matrix
likparms	likelihood parameters for the likelihood_model, as a list. Default values are sqrt(.1) for Gaussian noise and 2 for the alpha parameter for Gamma data.
max.iter	maximum iterations to perform
convg	convergence criteria. End iterations if the Newton step is this small
return_all	Return additional posterior covariance terms, TRUE or FALSE
y_init	Specify initial guess for posterior mode
prior_mean	specify the prior latent mean
verbose	if TRUE messages about the posterior estimation will be displayed

Value

multivariate normal posterior parameters calculated by the Vecchia-Laplace approximation

Examples

z=rnorm(10); locs=matrix(1:10,ncol=1); vecchia.approx=vecchia_specify(locs,m=5)
calculate_posterior_VL(z,vecchia.approx,"gaussian",covparms=c(1,2,.5))

GPvecchia documentation built on Oct. 25, 2022, 1:06 a.m.