R/OptPath_plotYTraces.R

In mlr-org/ParamHelpers: Helpers for Parameters in Black-Box Optimization, Tuning and Machine Learning

#' @title Plots Y traces of multiple optimization paths
#'
#' @description Can be used for only single-objective optimization paths. Useful
#' to compare runs of different algorithms on the same optimization problem. You
#' can add your own ggplot layers to the resulting plot object.
#'
#' @param opt.paths [[OptPath()] | list of [OptPath()]]\cr
#'   Object(s) to plot.
#' @param over.time [`character`]\cr
#'   Should the traces be plotted versus the iteration number or the cumulated
#'   execution time? For the later, the opt.path has to contain a extra column
#'   names exec.time. Possible values are dob and exec.time, default is `dob`.
#' @return ggplot2 plot object
renderYTraces = function(opt.paths, over.time = "dob") {

  opt.paths = ensureVector(opt.paths, cl = "OptPath", n = 1L, names = "opt.path")
  assertList(opt.paths, types = "OptPath", min.len = 1L, names = "unique")
  assertChoice(over.time, choices = c("dob", "exec.time"))
  run.name = names(opt.paths)
  y.name = NULL
  minimize = NULL
  data = data.frame()
  # combine y traces for this algo + add algo / repl + do some sanity checks
  fronts = lapply(seq_along(opt.paths), function(j) {

    run = opt.paths[[j]]
    name = names(as.data.frame(run, include.x = FALSE, include.rest = FALSE, stringsAsFactors = TRUE))
    if (j == 1L) {
      y.name <<- name
      minimize <<- run$minimize
    }
    if (length(name) != 1L) {
      stopf("Must always have 1 objective in opt path. But found: %i", length(name))
    }
    if (!y.name == name) {
      stopf("Must always have the same objective in opt path: %s (first ones taken). But found here: %s",
        y.name, name)
    }
    if (!minimize == run$minimize) {
      stopf("Must always have the same 'minimize' settings for objective in opt path: %s (first one taken).
          But found here: %s", minimize, run$minimize)
    }
    if (over.time == "dob") {
      df = data.frame(
        y = getOptPathY(op = run),
        time = getOptPathDOB(op = run),
        stringsAsFactors = TRUE
      )
    }
    if (over.time == "exec.time") {
      times = seq(0, sum(getOptPathExecTimes(run)), length.out = 128)
      df = getOptPathColAtTimes(run, times)
      df = df[, c(y.name, "time")]
    }

    # add column for algorithm and replication
    cbind(df, .algo = run.name[j], stringsAsFactors = TRUE)
  })

  data = do.call(rbind, fronts)
  mean.data = aggregate(data[[y.name]], by = list(data$time, data$.algo), FUN = mean)
  names(mean.data) = c("time", ".algo", y.name)

  pl = ggplot2::ggplot(data, ggplot2::aes(x = .data$time, y = .data[[y.name]], group = .data$.algo,
    linetype = .data$.algo, col = .data$.algo))
  if (over.time == "dob") {
    pl = pl + ggplot2::geom_point(size = 3)
  }
  pl = pl + ggplot2::geom_line(data = mean.data, linewidth = 1)

  return(pl)
}

#' @title Plots Y traces of multiple optimization paths
#'
#' @description Plot function for [renderYTraces()]
#'
#' @param opt.paths [`list`]\cr
#'   List of `OptPath` objects
#' @param over.time [`character`]\cr
#'   Should the traces be plotted versus the iteration number or the cumulated
#'   execution time? For the later, the opt.path has to contain a extra column
#'   names exec.time. Possible values are dob and exec.time, default is `dob`.
#'
#' @return [`NULL`]
#'
#' @export
plotYTraces = function(opt.paths, over.time = "dob") {
  print(renderYTraces(opt.paths, over.time))
  return(invisible(NULL))
}