R/simHelper.R

In KFAS: Kalman Filter and Smoother for Exponential Family State Space Models

# simulate error terms from standard normal distribution
# transformation to correct distribution is done in Fortran
#' @importFrom stats pchisq qchisq
simHelper <- function(model, nsim, antithetics) {

  epsplus <- array(0, c(attr(model, "p"), attr(model, "n"), nsim))
  etaplus <- array(0, c(attr(model, "k"), attr(model, "n"), nsim))
  aplus1 <- array(0, dim = c(attr(model, "m"), nsim))
  dtol <- 100 * .Machine$double.eps
  if (any(model$distribution != "gaussian")) {
    x <- array(TRUE, c(attr(model, "p"), attr(model, "n"), nsim))
  } else {
    x <- array(apply(model$H, 3, diag) > dtol, c(attr(model, "p"), attr(model, "n")))
    x <- array(x, c(attr(model, "p"), attr(model, "n"), nsim))
  }

  dfeps <- sum(x)/nsim
  x2 <- array(apply(model$Q, 3, diag) > dtol,
    c(attr(model, "k"), (attr(model, "n") - 1) * attr(model, "tv")[5] + 1))
  x2 <- array(x2, c(attr(model, "k"), attr(model, "n"), nsim))
  dfeta <- sum(x2)/nsim
  nonzeroP1 <- which(diag(model$P1) > dtol)
  nNonzeroP1 <- length(nonzeroP1)
  dfu <- dfeps + dfeta + nNonzeroP1
  u <- rnorm(dfu * nsim, mean = 0, sd = 1)
  if (dfeps > 0)
    epsplus[x] <- u[1:(dfeps * nsim)]
  if (dfeta > 0)
    etaplus[x2] <- u[(dfeps * nsim + 1):(dfeps * nsim + dfeta * nsim)]
  if (nNonzeroP1 > 0)
    aplus1[nonzeroP1, ] <- u[(dfeps * nsim + dfeta * nsim + 1):(dfu * nsim)]
  c2 <- numeric(nsim)
  # for second antithetic
  if (antithetics) {
    for (i in 1:nsim) {
      u1 <- c(etaplus[, , i], epsplus[, , i], aplus1[, i])
      c2[i] <- t(u1) %*% c(u1)
    }
    q <- pchisq(c2, df = dfu)
    c2 <- sqrt(qchisq(1 - q, dfu)/c2)
  }
  list(epsplus = epsplus, etaplus = etaplus, aplus1 = aplus1, c2 = c2,
    nNonzeroP1 = nNonzeroP1,
    nNonzeroP1inf = as.integer(sum(model$P1inf)))
}