R/panel_logmeanexp.R

In panelPomp: Inference for Panel Partially Observed Markov Processes

#' @include panel_loglik.R
NULL

#' @title Log-mean-exp for panels
#' @description This function computes the \code{\link[pomp]{logmeanexp}} for each column
#'    or row of a numeric matrix and sums the result. Because the loglikelihood
#'    of a \code{panelPomp} object is the sum of the loglikelihoods of its units,
#'    this function can be used to summarize replicated estimates of the
#'    \code{panelPomp} model likelihood. If \code{se = TRUE}, the jackknife SE estimates
#'    from \code{\link[pomp]{logmeanexp}} are squared, summed and the squared root is taken.
#' @param x Matrix with the same number of replicated estimates for each panel
#' unit loglikelihood.
#' @param MARGIN The dimension of the matrix that corresponds to a panel unit
#' and over which averaging occurs (1 indicates rows, 2 indicates columns).
#' @param se logical; whether to give standard errors.
#' @return
#' A \code{numeric} value with the average panel log likelihood or, when
#' \code{se = TRUE}, a \code{numeric} vector adding the corresponding standard error.
#' @seealso panel_loglik
#' @author Carles \Breto
#' @examples
#' ulls <- matrix(c(1,1,10,10),nr=2)
#' panel_logmeanexp(ulls,MARGIN=2,se=TRUE)
#' @export
panel_logmeanexp <- function (x, MARGIN, se = FALSE)
{
  logmeanexps <-apply(x,MARGIN=MARGIN,FUN=logmeanexp,se=TRUE)
  summed <- sum(logmeanexps[1,])
  if (!se) {
    summed
  } else {
    c(summed,se=sqrt(sum(unname(logmeanexps[2,])^2)))
  }
}