make.gaussian: Gaussian distribution objects

In SamplerCompare: A Framework for Comparing the Performance of MCMC Samplers

Gaussian distribution objects

Description

Gaussian distribution objects

Usage

make.gaussian(mean, sigma=NULL, rho=NULL)
N2weakcor.dist
N4poscor.dist
N4negcor.dist

Arguments

mean	The mean of the distribution as a numeric vector; implicitly specifies the dimension.
sigma	The covariance of the distribution.
rho	The marginal correlations between parameters.

Details

make.gaussian returns a distribution object representing a multivariate normal distribution. If sigma is specified, that is taken to be its covariance. Otherwise, if rho is specified, the covariance is taken to be a matrix with ones on the diagonal and rho on the off-diagonal elements. To preserve positive definiteness, rho must be between -1/(length(mean)-1) and 1.

N2weakcor.dist, N4poscor.dist, and N4negcor.dist are predefined distributions generated with make.gaussian. They are intended to be used as test cases with compare.samplers. The examples below show how they are defined. N2weakcor.dist is a weakly positively correlated two-dimensional Gaussian. N4poscor.dist is a highly positively correlated four-dimensional Gaussian. N4negcor.dist is a highly negatively correlated four-dimensional Gaussian. N4poscor.dist and N4negcor.dist are similarly conditioned, but N4poscor.dist has one large eigenvalue and three small ones, while N4negcor.dist has one small eigenvalue and three large ones.

See Also

compare.samplers, make.dist

Examples

  N2weakcor.dist <- make.gaussian(c(0,0), rho=0.8)
  N4poscor.dist <- make.gaussian(c(1,2,3,4), rho=0.999)
  N4negcor.dist <- make.gaussian(c(1,2,3,4), rho=-0.3329)

SamplerCompare documentation built on April 24, 2023, 9:09 a.m.