TMVN-methods: Functions for specifying the method and corresponding options...

In mcmcsae: Markov Chain Monte Carlo Small Area Estimation

Functions for specifying the method and corresponding options for sampling from a possibly truncated and degenerate multivariate normal distribution

Description

These functions are intended for use in the method argument of create_TMVN_sampler.

Usage

m_direct(use.cholV = NULL)

m_Gibbs(slice = FALSE, eps = sqrt(.Machine$double.eps), diagnostic = FALSE)

m_HMC(Tsim = pi/2, max.events = .Machine$integer.max, diagnostic = FALSE)

m_HMCZigZag(
  Tsim = 1,
  rate = 1,
  prec.eq = NULL,
  diagnostic = FALSE,
  max.events = .Machine$integer.max,
  adapt = FALSE
)

m_softTMVN(
  sharpness = 100,
  useV = FALSE,
  CG = NULL,
  PG.approx = TRUE,
  PG.approx.m = -2L
)

Arguments

use.cholV	whether to use the Cholesky factor of the variance instead of precision matrix for sampling. If NULL the choice is made based on a simple heuristic.
slice	if TRUE, a Gibbs within slice sampler is used.
eps	small positive value to control numerical robustness of the algorithm.
diagnostic	whether information about violations of inequalities, bounces off inequality walls (for 'HMC' and 'HMCZigZag' methods) or gradient events (for 'HMCZigZag') is printed to the screen.
Tsim	the duration of a Hamiltonian Monte Carlo simulated particle trajectory. This can be specified as either a single positive numeric value for a fixed simulation time, or as a function that is applied in each MCMC iteration to generates a simulation time.
max.events	maximum number of events (reflections off inequality walls and for method 'HMCZigZag' also gradient events). Default is unlimited. Specifying a finite number may speed up the sampling but may also result in a biased sampling algorithm.
rate	vector of Laplace rate parameters for method 'HMCZigZag'. It must be a positive numeric vector of length one or the number of variables.
prec.eq	positive numeric vector of length 1 or the number of equality restrictions, to control the precision with which the equality restrictions are imposed; the larger prec.eq the more precisely they will be imposed.
adapt	experimental feature: if TRUE the rate parameter will be adapted in an attempt to make the sampling algorithm more efficient.
sharpness	for method 'softTMVN', the sharpness of the soft inequalities; the larger the better the approximation of exact inequalities. It must be a positive numeric vector of length one or the number of inequality restrictions.
useV	for method 'softTMVN' whether to base computations on variance instead of precision matrices.
CG	use a conjugate gradient iterative algorithm instead of Cholesky updates for sampling the model's coefficients. This must be a list with possible components max.it, stop.criterion, verbose. See the help for function CG_control, which can be used to specify these options. Currently the preconditioner and scale options cannot be set for this use case.
PG.approx	see sampler_control.
PG.approx.m	see sampler_control.

Value

A method object, for internal use only.

mcmcsae documentation built on April 12, 2025, 2:25 a.m.