R/kl.R

In brandmaier/semtree: Recursive Partitioning for Structural Equation Models

# symmetric version of kl divergence
# @export
klsym <- function(mu1, cov1, mu2, cov2) {
  return( kl(mu1, cov1, mu2, cov2)+kl(mu2, cov2,mu1, cov1))
}

#' Distances
#' 
#' Divergence measures for multivariate normal distributions as used in the
#' diversityMatrix function.
#' 
#' 
#' @aliases hellinger klsym
#' @param mu1 Mean vector
#' @param mu2 Mean vector
#' @param cov1 Covariance matrix
#' @param cov2 Covariance matrix
#' @export
kl <- function(mu1, cov1, mu2, cov2) {
  
  d <- nrow(cov1)
  
  mudiff <- mu2 - mu1
  
  icov2 <- cov2^-1
  
  dist <- log(det(cov2))-log(det(cov1))+tr(icov2 %*% cov1)-d+
    t(mudiff) %*% icov2 %*% mudiff
  
  return(dist)
}