R/createPrior.default.R

In jmcurran/Bolstad: Functions for Elementary Bayesian Inference

#' Create prior default method
#' 
#' @param x a vector of x values at which the prior is to be specified (the support of the prior). This should contain
#' unique values in ascending order. The function will sort values if x is unsorted with a warning, and will halt if x contains any
#' duplicates or negative lag 1 differences.
#' @param wt a vector of weights corresponding to the weight of the prior at the
#'   given x values.
#' @param \dots optional exta arguments. Not currently used.
#' @return a linear interpolation function where the weights have been scaled so
#'   the function (numerically) integrates to 1.
#' @export
#' 
createPrior.default = function(x, wt, ...){
  if(length(x) != length(wt))
    stop("x and wt must be of equal length")
  if(any(wt < 0)){
    stop("All weights must be >= 0")
  }
  
  if(is.unsorted(x)){
    warning("x is not in ascending order. We have sorted it for you, but this might have unintended consequences")
    o = order(x)
    x = x[o]
    wt = wt[o]
  }
  
  if(any(diff(x) <= 0)){
    stop("x has duplicated values or negative differences. x should contain unique values in ascending order.")
  }
  
  fx = approxfun(x, wt, yleft = 0, yright = 0, rule = 2)
  A = integrate(fx, x[1], x[length(x)])$value
  if(A != 1){
    if(!Bolstad.control(...)$quiet){
      cat("Normalizing prior. Normalizing constant: ")
      cat(A, "\n")
    }
    fx = approxfun(x, wt/A, yleft = 0, yright = 0, rule = 2)
  }
  
  return(fx)
}