R/sarsop_plus.R

In boettiger-lab/pomdpplus: Planning and Learning in Uncertain Systems

#' sarsop_plus
#'
#' @inheritParams sim_plus
#' @param mc.cores number of parallel cores to use
#' @param log_data a data frame with information to be logged, one row for each model. Leave NULL if not logging
#' @importFrom sarsop sarsop
#' @return A list of alpha vector matrices
#' @export
#'
#' @examples \dontrun{
#' #' source(system.file("examples/K_models.R", package="pomdpplus"))
#' alphas <- sarsop_plus(models, discount, precision = 1)
#' unif <- compute_plus_policy(alphas, models, c(0.5, 0.5))
#' }
sarsop_plus <- function(models, discount, state_prior = NULL,
                        mc.cores = 1L, log_data = NULL, ...){

  n_states <- dim(models[[1]]$transition)[[1]]
  if(is.null(state_prior)) state_prior <- rep(1, n_states) / n_states

  parallel::mclapply(1:length(models), function(i){

    m <- models[[i]]
    if(!is.null(log_data) && dim(log_data)[1] > 1)
      log_dat <- log_data[i,]
    else
      log_dat <- log_data

    sarsop::sarsop(m$transition, m$observation, m$reward, discount, state_prior, log_data = log_dat, ...)
  }, mc.cores = mc.cores)

}