R/relative_entropy.R

In hlennon/LCTMtools: Latent Class Trajectory Models: Tools for checking adequacy

#' The Relative Entropy
#'
#' @param p  is the posterior probabilities of assignment of dimensions, K columns and N rows
#' @return  Relative Entropy - where values close to 1 indicate lowest classification uncertainty. In the special case when there is most uncertainty and each individual has equal probability of belonging to each class, E_K=0. Jedidi et al., describes relative entropy as a relative measure of ‘fuzziness’, and suggested cause concern when close to zero, as this implies that the latent class centroids are not sufficiently separated.
#' @examples
#' \dontrun{relative_entropy(p)}
#' @export


relative_entropy <- function(p) {
    K <- ncol(p)
    n <- nrow(p)
    
    relEntropy <- 1 + (sum(p * log(p), na.rm = TRUE)/(n * log(K)))
    
    return(relEntropy)
}