R/gs4.R

In BNPdensity: Ferguson-Klass Type Algorithm for Posterior Normalized Random Measures

#' Resampling Ystar function
#'
#' This function resamples the distinct Ystar in the semiparametric model.
#'
#' For internal use.
#'
#' @keywords internal
#' @examples
#'
#' ## The function is currently defined as
#' function(ystar, x, idx, distr.k, sigma.k, distr.p0, mu.p0, sigma.p0) {
#'   r <- length(ystar)
#'   nstar <- as.numeric(table(idx))
#'   for (j in seq(r)) {
#'     id <- which(!is.na(match(idx, j)))
#'     xj <- x[id]
#'     xbar <- sum(xj) / nstar[j]
#'     y2star <- rk(1, distr = distr.k, mu = xbar, sigma = sigma.k / sqrt(nstar[j]))
#'     f.ratio <- rfyzstar(y2star, ystar[j], xj,
#'       distr = distr.k, sigma = sigma.k,
#'       distr.p0 = distr.p0, mu.p0 = mu.p0, sigma.p0 = sigma.p0
#'     )
#'     k.ratio <- dk(ystar[j],
#'       distr = distr.k,
#'       mu = xbar, sigma = sigma.k / sqrt(nstar[j])
#'     ) / dk(y2star,
#'       distr = distr.k, mu = xbar, sigma = sigma.k / sqrt(nstar[j])
#'     )
#'     q2 <- min(1, f.ratio * k.ratio)
#'     ystar[j] <- ifelse(runif(1) <= q2, y2star, ystar[j])
#'   }
#'   return(ystar)
#' }
gs4 <-
  function(ystar, x, idx, distr.k, sigma.k, distr.p0, mu.p0, sigma.p0) {
    r <- length(ystar)
    nstar <- as.numeric(table(idx))
    for (j in seq(r)) {
      id <- which(!is.na(match(idx, j)))
      xj <- x[id]
      xbar <- sum(xj) / nstar[j]
      y2star <- rk(1, distr = distr.k, mu = xbar, sigma = sigma.k / sqrt(nstar[j]))
      f.ratio <- rfystar(y2star, ystar[j], xj,
        distr.k = distr.k,
        sigma.k = sigma.k, distr.p0 = distr.p0, mu.p0 = mu.p0,
        sigma.p0 = sigma.p0
      )
      k.ratio <- dk(ystar[j], distr = distr.k, mu = xbar, sigma = sigma.k / sqrt(nstar[j])) / dk(y2star,
        distr = distr.k, mu = xbar, sigma = sigma.k / sqrt(nstar[j])
      )
      q2 <- min(1, f.ratio * k.ratio)
      ystar[j] <- ifelse(runif(1) <= q2, y2star, ystar[j])
    }
    return(ystar)
  }