updt.range: Update range parameter via Metropolis-Hastings

In lbelzile/mgp: Multivariate generalized Pareto models

Update range parameter via Metropolis-Hastings

Description

This function, adapted from gev.update.lambda, uses a Laplace approximation to build a proposal for the range parameter of an exponential covariance model with precision \alpha and range \lambda.

Usage

updt.range(tau, alpha, lambda, di, a, b, lb = 0.01, discount = 0.2,
  maxstep = Inf)

Arguments

tau	current value of centered random effects
alpha	precision of the random effect
lambda	range parameter to be updated
di	matrix of pairwise distances between points
a	shape parameter of the Gamma prior for \lambda
b	rate parameter of the Gamma prior for \lambda
lb	lower bound for admissible value of \lambda. Default to 0.01.
discount	size of discount in Newton update; default to 0.2.
maxstep	maximum discount size for the truncated Normal proposal

Details

The difference with the spatial.gev.bma package is that proposals are truncated Gaussian, so may be accepted even if the mean of the Newton step proposal is negative. If the current \lambda value is in a region where the Hessian matrix is not negative, the proposal is automatically rejected. The argument discount controls the size of the Newton steps.

Value

new value of lambda

Author(s)

code from spatial.gev.bma by Alex Lenkoski

See Also

gev.update.lambda

lbelzile/mgp documentation built on Aug. 5, 2023, 2:34 a.m.