In braunm/sparseMVN: Multivariate Normal Functions for Sparse Covariance and Precision Matrices

Using the sparseMVN package

The signatures of the key sparse matrix functions are

rmvn.sparse(n, mu, CH, prec=TRUE) dmvn.sparse(x, mu, CH, prec=TRUE, log=TRUE)

| Parameter | Description | |-------|-------| | x | A numeric matrix. Each row is an MVN sample.| | mu | A numeric vector. The mean of the MVN random variable.| | CH |Either a dCHMsimpl or dCHMsuper object representing the Cholesky decomposition of the covariance/precision matrix.| | prec | Logical value that identifies CH as the Cholesky decomposition of either a covariance ($\Sigma$, ) or precision($\Sigma^{-1}$, ) matrix.| | n | Number of random samples to be generated.| | log | If , the log density is returned.|

The only part of this that is substantially different from other standard MVN functions is CH, which must be a dCHMsimpl or dCHMsuper matrix computed from the Matrix package.

If your matrix is structurally sparse (lots of zeros), but stored as a dense R matrix, you first need to get it into a sparse format.

braunm/sparseMVN documentation built on Jan. 3, 2022, 4:47 p.m.