R/computeMode.R

In BBmisc: Miscellaneous Helper Functions for B. Bischl

#' Compute statistical mode of a vector (value that occurs most frequently).
#'
#' Works for integer, numeric, factor and character vectors.
#' The implementation is currently not extremely efficient.
#'
#' @param x [\code{vector}]\cr
#'   Factor, character, integer, numeric or logical vector.
#' @param na.rm [\code{logical(1)}]\cr
#'   If \code{TRUE}, missing values in the data removed.
#'   if \code{FALSE}, they are used as a separate level and this level could therefore
#'   be returned as the most frequent one.
#'   Default is \code{TRUE}.
#' @param ties.method [\code{character(1)}]\cr
#'   \dQuote{first}, \dQuote{random}, \dQuote{last}: Decide which value to take in case of ties.
#'   Default is \dQuote{random}.
#' @return Modal value of length 1, data type depends on data type of \code{x}.
#' @export
#' @examples
#' computeMode(c(1,2,3,3))
computeMode = function(x, ties.method = "random", na.rm = TRUE) {
  assertAtomicVector(x)
  assertChoice(ties.method, c("first", "random", "last"))
  assertFlag(na.rm)
  tab = as.data.table(x)[,  .N, by = x]
  if (na.rm) tab = na.omit(tab)
  ind = (tab$N == max(tab$N))
  mod = tab$x[ind]
  if (is.factor(mod))
    mod = as.character(mod)
  if (length(mod) > 1L)
    ind = switch(ties.method, first = mod[1L], random = sample(mod, 1L), last = mod[length(mod)])
  else mod
}