R/RLearner_classif_bst.R

#' @export
makeRLearner.classif.bst = function() {
  makeRLearnerClassif(
    cl = "classif.bst",
    package = c("bst", "rpart"),
    par.set = makeParamSet(
      makeNumericLearnerParam(id = "cost", default = 0.5, lower = 0, upper = 1),
      makeDiscreteLearnerParam(id = "family", values = c("gaussian", "hinge")),
      makeIntegerLearnerParam(id = "mstop", default = 50L, lower = 1L, when = "both"),
      makeNumericLearnerParam(id = "nu", default = 0.1),
      makeLogicalLearnerParam(id = "twinboost", default = FALSE),
      makeUntypedLearnerParam(id = "f.init", default = NULL),
      makeUntypedLearnerParam(id = "xselect.init", default = NULL),
      makeLogicalLearnerParam(id = "center", default = FALSE),
      makeLogicalLearnerParam(id = "trace", default = FALSE, tunable = FALSE),
      makeIntegerLearnerParam(id = "numsample", default = 50L, lower = 1L),
      makeIntegerLearnerParam(id = "df", default = 4L, lower = 1L),
      makeIntegerLearnerParam(id = "minsplit", default = 20L, lower = 1L),
      makeIntegerLearnerParam(id = "minbucket", lower = 1L),
      makeNumericLearnerParam(id = "cp", default = 0.01, lower = 0, upper = 1),
      makeIntegerLearnerParam(id = "maxcompete", default = 4L, lower = 0L),
      makeIntegerLearnerParam(id = "maxsurrogate", default = 5L, lower = 0L),
      makeDiscreteLearnerParam(id = "usesurrogate", default = 2L, values = 0:2),
      makeIntegerLearnerParam(id = "xval", default = 10L, lower = 0L, tunable = FALSE),
      makeDiscreteLearnerParam(id = "surrogatestyle", default = 0L, values = 0:1),
      makeIntegerLearnerParam(id = "maxdepth", default = 1L, lower = 1L, upper = 30L),
      makeDiscreteLearnerParam(id = "Learner", default = "ls", values = c("ls", "sm", "tree"))
      # FIXME bst() seems to have no default for Learner. But if "ls" is correct, it is not needed in par.vals
    ),
    par.vals = list(Learner = "ls", maxdepth = 1L, xval = 0L),
    # FIXME par.vals default for maxdepth is the same as the default of LearnerParam and the function
    properties = c("numerics", "twoclass"),
    name = "Gradient Boosting",
    short.name = "bst",
    note = 'Renamed parameter `learner` to `Learner` due to nameclash with `setHyperPars`. Default changes: `Learner = "ls"`, `xval = 0`, and `maxdepth = 1`.',
    callees = c("bst", "bst_control", "rpart.control")
  )
}

#' @export
trainLearner.classif.bst = function(.learner, .task, .subset, .weights = NULL, mstop, nu, twinboost,
  f.init, xselect.init, center, trace, numsample, df, minsplit, minbucket, cp, maxsurrogate,
  usesurrogate, surrogatestyle, maxdepth, xval, Learner, ...) {
  d = getTaskData(.task, .subset, target.extra = TRUE, recode.target = "-1+1")
  ctrl = learnerArgsToControl(bst::bst_control, mstop, nu, twinboost, f.init, xselect.init, center, trace, numsample, df)
  control.tree = learnerArgsToControl(list,  minsplit, minbucket, cp, maxsurrogate, usesurrogate, surrogatestyle, maxdepth, xval)
  bst::bst(x = d$data, y = d$target, ctrl = ctrl, control.tree = control.tree, learner = Learner, ...)
}

#' @export
predictLearner.classif.bst = function(.learner, .model, .newdata, ...) {
  levs = c(.model$task.desc$negative, .model$task.desc$positive)
  p = predict(.model$learner.model, .newdata, ...)
  as.factor(ifelse(p > 0, levs[2L], levs[1L]))
}
Najah-lshanableh/R-data-mining2 documentation built on May 6, 2019, 10:11 a.m.