GitHub - mlysy/mniw: Simulation and Inference with the Matrix-Normal-Inverse-Wishart Distribution. (original) (raw)

mniw: The Matrix-Normal Inverse-Wishart Distribution

Martin Lysy, Bryan Yates

Description

Density evaluation and random number generation for the Matrix-Normal Inverse-Wishart (MNIW) distribution, as well as the the Matrix-Normal, Matrix-T, Wishart, and Inverse-Wishart distributions. Core calculations are implemented in a portable (header-only) C++ library, with matrix manipulations using the Eigen library for linear algebra. Also provided is a Gibbs sampler for Bayesian inference on a random-effects model with Matrix-Normal observations.

Installation

To install the CRAN version (1.0.1):

install.packages("mniw", INSTALL_opts = "--install-tests")

To install the latest development version: first install the devtools, then:

devtools::install_github("mlysy/mniw", INSTALL_opts = "--install-tests")

Usage

The primary advantage of the mniw package is that it "vectorizes" over its input arguments. Take for example the simulation of a Wishart distribution, which can be done with the built-in R function stats::rWishart():

n <- 10 p <- 3 nu <- 6

produces an array of size p x p x n

Psi <- stats::rWishart(n = n, df = nu, Sigma = diag(p))

Now suppose we want to generate Wishart random variables each with a different Sigma:

Vectorizing over the 'Sigma' argument

X <- apply(Psi, 3, stats::rWishart, n = 1, df = nu) X <- array(X, dim = c(p, p, n))

However, the code above is both slow for large n, and inconvenient due to the reshaping of the apply() output. The equivalent code using mniw is:

X <- rwish(n, df = nu, Psi = Psi) # produces an array of size p x p x n

It is both simpler, and much faster for large n and p.

The other functions in mniw behave much the same way. A complete description of the distributions provided by the package is available here.