R/getEDAAlgoPerf.R

In aslib: Interface to the Algorithm Selection Benchmark Library

# helper to retrieve algo data for plotting
#  - imputes it (in standard way)
#  - aggregates replications
#  - returns in wide or long format
#  - returns also range of orig data and how often runs where successful

getEDAAlgoPerf = function(asscenario, measure, impute.failed.runs, jitter,
  impute.zero.vals, check.log, format, with.instance.id) {

  assertClass(asscenario, "ASScenario")
  desc = asscenario$desc
  measure = checkMeasure(measure, desc)
  assertFlag(impute.failed.runs)
  assertFlag(jitter)
  assertFlag(impute.zero.vals)
  assertChoice(format, choices = c("wide", "long"))
  assertFlag(with.instance.id)

  # compute range and success.rate now
  origdata = asscenario$algo.runs
  range = range(origdata[, measure], na.rm = TRUE)
  success.rate = ddply(origdata, "algorithm", function(d) {
    mean(as.character(d$runstatus) == "ok")
  })
  success.rate = setNames(success.rate$V1, success.rate$algorithm)

  if (impute.failed.runs) {
    jitter2 = ifelse(jitter, 0.00, 0)
    # for runtime scenarios set to cutoff, otherwise use default, which is min or max value of perfs
    base = if (desc$performance_type[[measure]] == "runtime" && !is.na(desc$algorithm_cutoff_time))
      desc$algorithm_cutoff_time
    else
      NULL
    # potentially set zero values to something small here
    algo.runs = imputeAlgoPerf(asscenario, measure = measure,
      base = base, range.scalar = 0.3, jitter = jitter2, impute.zero.vals = impute.zero.vals)
  } else {
    # only impute zeros as extra operation
    algo.runs = imputeZeroVals(asscenario$algo.runs, measure, impute.zero.vals)
  }
  # include fake repetition col, so we can convert to wide
  algo.runs = aggregateStochasticAlgoRuns(algo.runs, measure = measure, with.repetition = TRUE)

  checkLogarithm(check.log, algo.runs, measure)
  data = if (format == "wide")
    convertAlgoPerfToWideFormat(algo.runs, desc = desc)
  else
    algo.runs
  # remove fake repetition col
  data$repetition = NULL
  if (!with.instance.id)
    data = dropNamed(data, "instance_id")
  list(data = data, measure = measure, range = range,
    success.rate = success.rate)
}