R/wtd.pctiles.fast.R

In ejanalysis/analyze.stuff: Miscellaneous Tools for Analyzing Data in Rows and Columns

#' @title Show the values at 100 weighted percentiles
#'
#' @description
#' Get a quick look at a weighted distribution by seeing the 100 values that are the weighted percentiles 1-100
#' @details
#' Provides weighted percentiles without using wtd.quantile, see [Hmisc::wtd.Ecdf()] \cr \cr
#' #  NOTE: THIS ONLY SHOWS PERCENTILES AND MEAN FOR THE VALID (NOT NA) VALUES !
#' Defining these types as type=1 and type="i/n" will create simple discontinuous quantiles, without interpolation where there are jumps in the values analyzed.
#' *** WARNING: Unless set type=1, the default type=7 in which case stats::quantile() FUNCTION INTERPOLATES, WHICH ISN'T OBVIOUS IN EVERY DATASET!
#' use type=1 to avoid interpolation.
#' and pctiles() rounded results so interpolation would be even less apparent.\cr
#' The quantile function will NOT interpolate between values if type=1:\cr
#' stats::quantile(1:12, probs=(1:10)/10, type=1)\cr
#'  10%  20%  30%  40%  50%  60%  70%  80%  90% 100% \cr
#'  2    3    4    5    6    8    9   10   11   12 \cr
#'###########################\cr
#' **** IMPORTANT ***\cr
#'###########################\cr
#' *** WARNING: The Hmisc::wtd.quantile function DOES interpolate between values, even if type='i/n'\cr
#' There does not seem to be a way to fix that for the Hmisc::wtd.quantile() function.
#' For example, \cr
#' Hmisc::wtd.quantile(1:12, probs=(1:10)/10, type='i/n', weights=rep(1,12))\cr
#'  10%  20%  30%  40%  50%  60%  70%  80%  90% 100% \cr
#'  1.2  2.4  3.6  4.8  6.0  7.2  8.4  9.6 10.8 12.0 \cr
#' @param x Required numeric vector of values whose distribution you want to look at.
#' @param wts NULL by default, or vector of numbers to use as weights in Hmisc::wtd.quantile
#' @param na.rm Logical optional TRUE by default, in which case NA values are removed first.
#' @return Returns a data.frame
#' @seealso  [pctiles()] [pctiles.exact()] [pctiles.a.over.b()] [wtd.pctiles.exact()] [wtd.pctiles()] [wtd.pctiles.fast()]
#' @export
wtd.pctiles.fast <- function(x, wts=NULL, na.rm=TRUE) {

  wts.as.pcts <- wts / sum(wts, na.rm=TRUE) # done just once if same wts used for many fields

  # for each column (field), sort (rank) the raw data alongside wts.as.pcts
  i <- order(x) # i is a vector saying which of the x is lowest, which is next, etc. in rank small to large
  pctiles <- cumsum(wts.as.pcts[i]) # ranked low to high, and with ties given differing pctiles

  # NOW NEED TO FIX THE TIED CASES!
  #mytabs <- table(x)
  #tied <- mytabs > 1
  #x[tied] <-  ??? max of that group
  #
  # make a value's pctile change to be the next higher pctile if corresponding raw equals the next higher raw (tied)
  pctiles[ x[1:(length(pctiles) - 1)]==x[2:length(pctiles)]] <- pctiles[2:length(pctiles)]

  return( pctiles[order(i)] )
}