rtmvt: Random number generation for truncated multivariate Student's...

In tmvmixnorm: Sampling from Truncated Multivariate Normal and t Distributions

Description

rtmvt simulates truncated multivariate (p-dimensional) Student's t distribution subject to linear inequality constraints. The constraints should be written as a matrix (D) with lower and upper as the lower and upper bounds for those constraints respectively. Note that D can be non-full rank, which generalizes many traditional methods.

Usage

rtmvt(n, Mean, Sigma, nu, D, lower, upper, int = NULL, burn = 10, thin = 1)

Arguments

n	number of random samples desired (sample size).
Mean	location vector of the multivariate Student's t distribution.
Sigma	positive definite dispersion matrix of the multivariate t distribution.
nu	degrees of freedom for Student-t distribution.
D	matrix or vector of coefficients of linear inequality constraints.
lower	lower bound vector for truncation.
upper	upper bound vector for truncation.
int	initial value vector for Gibbs sampler (satisfying truncation), if NULL then determine automatically.
burn	burn-in iterations discarded (default as 10).
thin	thinning lag (default as 1).

Value

rtmvt returns a (n*p) matrix (or vector when n=1) containing random numbers which follows truncated multivariate Student-t distribution.

Examples

# Example for full rank
d <- 3
rho <- 0.5
nu <- 10
Sigma <- matrix(0, nrow=d, ncol=d)
Sigma <- rho^abs(row(Sigma) - col(Sigma))

D1 <- diag(1,d) # Full rank

set.seed(1203)
ans.t <- rtmvt(n=1000, Mean=1:d, Sigma, nu=nu, D=D1, lower=rep(-1,d), upper=rep(1,d),
burn=50, thin=0)

apply(ans.t, 2, summary)

tmvmixnorm documentation built on Sept. 19, 2020, 1:07 a.m.