R/rk.R

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

#' Kernel density sampling function
#'
#' This function simulates from a density.  There are 4 density options (1 =
#' Gaussian, 2 = Gamma, 3 = Beta, 4 = double exponential, 5 = lognormal). All
#' densities are parameterized in terms of mean and standard deviation.
#'
#' For internal use.
#'
#' @keywords internal
#' @examples
#'
#' ## The function is currently defined as
#' function(n, distr = NULL, mu = NULL, sigma = NULL) {
#'   if (is.null(distr)) {
#'     stop("Argument \"distr\" should be defined numeric with possible values 1,2,3,4 or 5")
#'   }
#'   else if (distr == 1) {
#'     a <- ifelse(is.null(mu), 0, mu)
#'     b <- ifelse(is.null(sigma), 1, sigma)
#'     rk <- rnorm(n, mean = a, sd = b)
#'   }
#'   else if (distr == 2) {
#'     a <- ifelse(is.null(mu), 0, mu)
#'     b <- ifelse(is.null(sigma), 1 / sqrt(2), sigma / sqrt(2))
#'     rk <- a + b * sample(c(-1, +1), size = n, replace = TRUE) *
#'       rexp(n)
#'   }
#'   else if (distr == 3) {
#'     a <- ifelse(is.null(mu), exp(1 / 2), log(mu / sqrt(1 + (sigma / mu)^2)))
#'     b <- ifelse(is.null(sigma), exp(1) * (exp(1) - 1), sqrt(log(1 +
#'       (sigma / y)^2)))
#'     rk <- rlnorm(n, meanlog = a, sdlog = b)
#'   }
#'   else if (distr == 4) {
#'     a <- ifelse(is.null(mu), 1, mu^2 / sigma^2)
#'     b <- ifelse(is.null(sigma), 1, mu / sigma^2)
#'     rk <- rgamma(n, shape = a, rate = b)
#'   }
#'   else if (distr == 5) {
#'     a <- ifelse(is.null(mu), 0.5, (1 - mu) * (mu / sigma)^2 -
#'       mu)
#'     b <- ifelse(is.null(sigma), 1 / sqrt(12), (mu * (1 - mu) / sigma^2 -
#'       1) * (1 - mu))
#'     if (any(c(a, b) <= 0)) {
#'       stop(paste(
#'         "\nNegative Beta parameters:\n a =", a,
#'         ";\t b =", b
#'       ))
#'     }
#'     rk <- rbeta(n, shape1 = a, shape2 = b)
#'   }
#'   else {
#'     stop("Argument \"distr\" should be defined numeric with possible values 1,2,3,4 or 5")
#'   }
#'   return(rk)
#' }
rk <-
  function(n, distr = NULL, mu = NULL, sigma = NULL) {
    msg <- "Argument \"distr\" should be defined numeric with possible values 1 (normal), 2 (gamma), 3 (beta), 4 (exponential), 5 (lognormal), 6 (half-Cauchy), 7 (half-normal), 8 (half-student), 9 (uniform) and 10 (truncated normal)"
    if (is.null(distr)) {
      stop(msg)
    }
    else if (distr == 1) {
      rk <- rnorm(n, mean = mu, sd = sigma)
    }
    else if (distr == 2) {
      a <- ifelse(is.null(mu), 1, mu^2 / sigma^2)
      b <- ifelse(is.null(sigma), 1, mu / sigma^2)
      rk <- rgamma(n, shape = a, rate = b)
    }
    else if (distr == 3) {
      a <- ifelse(is.null(mu), 0.5, (1 - mu) * (mu / sigma)^2 -
        mu)
      b <- ifelse(is.null(sigma), 1 / sqrt(12), (mu * (1 - mu) / sigma^2 -
        1) * (1 - mu))
      if (any(c(a, b) <= 0)) {
        stop(paste(
          "\nNegative Beta parameters:\n a =", a,
          ";\t b =", b
        ))
      }
      rk <- rbeta(n, shape1 = a, shape2 = b)
    }
    else if (distr == 4) {
      a <- ifelse(is.null(mu), 0, mu)
      b <- ifelse(is.null(sigma), 1 / sqrt(2), sigma / sqrt(2))
      rk <- a + b * sample(c(-1, +1), size = n, replace = TRUE) *
        rexp(n)
    }
    else if (distr == 5) {
      a <- ifelse(is.null(mu), exp(1 / 2), log(mu / sqrt(1 + (sigma / mu)^2)))
      b <- ifelse(is.null(sigma), exp(1) * (exp(1) - 1), sqrt(log(1 +
        (sigma / mu)^2)))
      rk <- rlnorm(n, meanlog = a, sdlog = b)
    }
    else if (distr == 6) {
      rk <- rhalfcauchy(n, location = ifelse(is.null(mu), 0,
        mu
      ), scale = ifelse(is.null(sigma), 1, sigma))
    }
    else if (distr == 7) {
      rk <- rhalfnorm(n,
        mean = ifelse(is.null(mu), 0, mu),
        sd = ifelse(is.null(sigma), 1, sigma)
      )
    }
    else if (distr == 8) {
      rk <- rhalft(n, df = 10, mean = ifelse(is.null(mu), 0,
        mu
      ), sd = ifelse(is.null(sigma), 1, sigma))
    }
    else if (distr == 9) {
      rk <- runif(n, min = ifelse(is.null(mu), 0, mu), max = ifelse(is.null(sigma),
        1, sigma
      ))
    }
    else if (distr == 10) {
      rk <- rtnorm(n, mean = ifelse(is.null(mu), 0, mu), sd = ifelse(is.null(sigma),
        1, sigma
      ), lower = 0.1)
    }
    else {
      stop(msg)
    }
    return(rk)
  }