R/convertToLlamaCVFolds.R

In aslib: Interface to the Algorithm Selection Benchmark Library

#' Convert an ASScenario scenario object to a llama data object with cross-validation folds.
#'
#' For features, mean values are computed across repetitions.
#' For algorithms, repetitions are not supported at the moment and will result in an error.
#'
#' @param asscenario [\code{\link{ASScenario}}]\cr
#'   Algorithm selection scenario.
#' @param measure [\code{character(1)}]\cr
#'   Measure to use for modelling.
#'   Default is first measure in scenario.
#' @param feature.steps [\code{character}]\cr
#'   Which instance feature steps are allowed?
#'   Default are the default instance feature steps or all steps
#'   in case no defaults were defined.
#' @param algorithm.feature.steps [\code{character}]\cr
#'   Which algorithm feature steps are allowed?
#'   Default are the default algorithm feature steps or all steps
#'   in case no defaults were defined.
#' @param cv.splits [\code{data.frame}]\cr
#'   Data frame defining the split of the data into cross-validation folds,
#'   as returned by \code{\link{createCVSplits}}.
#'   Default are the splits \code{asscenario$cv.splits}
#' @return Result of calling \code{\link[llama]{input}} with data partitioned into folds.
#' @export
convertToLlamaCVFolds = function(asscenario, measure, feature.steps, algorithm.feature.steps, cv.splits) {
  assertClass(asscenario, "ASScenario")
  if (missing(measure))
    measure = asscenario$desc$performance_measures[1]
  else
    assertString(measure)
  if (missing(cv.splits))
    cv.splits = asscenario$cv.splits
  else
    assertClass(cv.splits, "data.frame")
  allsteps = names(asscenario$desc$feature_steps)
  algo.allsteps = names(asscenario$desc$algorithm_feature_steps)
  if (missing(feature.steps)) {
    feature.steps = getDefaultFeatureStepNames(asscenario)
  }
  else
    assertSubset(feature.steps, c(allsteps, algo.allsteps))

  reps = max(cv.splits$repetition)
  if (reps > 1L)
    stopf("LLAMA does not handle repeated cross validations!")

  folds = cv.splits

  llamaFrame = convertToLlama(asscenario, measure = measure,
      feature.steps = feature.steps)

  nfolds = length(unique(folds$fold))
  rownames(folds) = folds$instance_id
  splitFactors = folds[match(llamaFrame$data$instance_id, folds$instance_id), "fold"]
  parts = split(1:nrow(llamaFrame$data), splitFactors)

  retval = c(llamaFrame,
            list(train = lapply(1:nfolds, function(x) {
                    return(unlist(parts[-x]))
                }),
                 test = lapply(1:nfolds, function(x) {
                    return(parts[[x]])
                 })))
  class(retval) = "llama.data"
  attr(retval, "hasSplits") = TRUE
  return(retval)
}