R/posterior_logspline.R

In metaBMA: Bayesian Model Averaging for Random and Fixed Effects Meta-Analysis

posterior_logspline <- function(
    stanfit,
    parameter,
    prior
) {

  if (missing(stanfit) || is.null(stanfit)) {
    warning(
      "MCMC/Stan samples missing: To approximate the posterior density",
      "\n  by MCMC samples, one of the available priors must be used (see ?prior)",
      "\n and the argument 'sample' must be larger than zero!"
    )
  }

  if (inherits(stanfit, "stanfit")) {
    ss <- extract(stanfit, parameter)[[parameter]]
  } else {
    ss <- stanfit
  }

  bnd <- bounds_prior(prior)
  mini <- max(-Inf, bnd[1])
  maxi <- min(Inf, bnd[2])

  args <- list(
    "x" = ss,
    "knots" = quantile(ss, probs = c(.20, .50, .80)),
    "maxknots" = 5
  )
  if (mini != -Inf) args$lbound <- mini
  if (maxi != Inf) args$ubound <- maxi

  lspline <- do.call("logspline", args)
  dens <- function(x) dlogspline(x, lspline)
  dens
}