CPO: Composable Preprocessing Operators

.gitignore

@@ -16,4 +16,10 @@ ehthumbs.db
 Icon?
 Thumbs.db
 *xgboost.model
+.#*
+*~
+\#*#
+.project
+.settings
+.ipynb_checkpoints
 man/*.Rd

DESCRIPTION

@@ -165,4 +165,4 @@ LazyData: yes
 ByteCompile: yes
 Version: 2.12
 VignetteBuilder: knitr
-RoxygenNote: 6.0.1.9000
+RoxygenNote: 6.0.1

NAMESPACE

@@ -786,6 +786,9 @@ export(brier.scaled)
 export(calculateConfusionMatrix)
 export(calculateROCMeasures)
 export(capLargeValues)
+export(changeData)
+export(checkLearner)
+export(checkPredictLearnerOutput)
 export(cindex)
 export(cindex.uno)
 export(configureMlr)
@@ -939,14 +942,19 @@ export(lsr)
 export(mae)
 export(makeAggregation)
 export(makeBaggingWrapper)
+export(makeBaseWrapper)
+export(makeChainModel)
 export(makeClassifTask)
+export(makeClassifTaskDesc)
 export(makeClassificationViaRegressionWrapper)
 export(makeClusterTask)
+export(makeClusterTaskDesc)
 export(makeConstantClassWrapper)
 export(makeCostMeasure)
 export(makeCostSensClassifWrapper)
 export(makeCostSensRegrWrapper)
 export(makeCostSensTask)
+export(makeCostSensTaskDesc)
 export(makeCostSensWeightedPairsWrapper)
 export(makeCustomResampledMeasure)
 export(makeDownsampleWrapper)
@@ -976,6 +984,7 @@ export(makeMultilabelDBRWrapper)
 export(makeMultilabelNestedStackingWrapper)
 export(makeMultilabelStackingWrapper)
 export(makeMultilabelTask)
+export(makeMultilabelTaskDesc)
 export(makeOverBaggingWrapper)
 export(makeOversampleWrapper)
 export(makePrediction)
@@ -989,12 +998,15 @@ export(makeRLearnerMultilabel)
 export(makeRLearnerRegr)
 export(makeRLearnerSurv)
 export(makeRegrTask)
+export(makeRegrTaskDesc)
 export(makeRemoveConstantFeaturesWrapper)
 export(makeResampleDesc)
 export(makeResampleInstance)
 export(makeSMOTEWrapper)
 export(makeStackedLearner)
 export(makeSurvTask)
+export(makeSurvTaskDesc)
+export(makeTaskDescInternal)
 export(makeTuneControlCMAES)
 export(makeTuneControlDesign)
 export(makeTuneControlGenSA)

R/BaseWrapper.R

@@ -1,3 +1,22 @@
+#' Exported for internal use only.
+#' @param id [\code{character(1)}]\cr
+#'   Id string for object. Used to display object.
+#' @param type [\code{character(1)}]\cr
+#'   Learner type.
+#' @param next.learner [\code{Learner}]\cr
+#'   Learner to wrap.
+#' @param package [\code{character}]\cr
+#'   Packages to load when loading learner.
+#' @param par.set [\code{ParamSet}]\cr
+#'   Parameter set.
+#' @param par.vals [\code{list}]\cr
+#'   Optional list of named (hyper)parameter values.
+#' @param learner.subclass [\code{character}]\cr
+#'   Class to assign the new object.
+#' @param model.subclass [\code{character}]\cr
+#'   Class to assign learner models.
+#' @keywords internal
+#' @export
 makeBaseWrapper = function(id, type, next.learner, package = character(0L), par.set = makeParamSet(),
   par.vals = list(), learner.subclass, model.subclass) {
   if (inherits(next.learner, "OptWrapper") && is.element("TuneWrapper", learner.subclass))

R/ChainModel.R

@@ -1,3 +1,11 @@
+#' Only exported for internal use.
+#' @param next.model [\code{WrappedModel}]\cr
+#'   The next model.
+#' @param cl [\code{character}]\cr
+#'   Subclass to assign to the resulting model.
+#'
+#' @keywords internal
+#' @export
 makeChainModel = function(next.model, cl) {
   setClasses(list(next.model = next.model), c(cl, "ChainModel", "WrappedModel"))
 }

R/ClassifTask.R

@@ -31,6 +31,8 @@ makeClassifTask = function(id = deparse(substitute(data)), data, target, weights
   addClasses(task, "ClassifTask")
 }
 
+#' @export
+#' @rdname makeTaskDesc
 makeClassifTaskDesc = function(id, data, target, weights, blocking, positive, spatial) {
   levs = levels(data[[target]])
   m = length(levs)

R/ClusterTask.R

@@ -13,6 +13,8 @@ makeClusterTask = function(id = deparse(substitute(data)), data, weights = NULL,
   addClasses(task, "ClusterTask")
 }
 
+#' @export
+#' @rdname makeTaskDesc
 makeClusterTaskDesc = function(id, data, weights, blocking, spatial) {
   target = character(0L)
   td = makeTaskDescInternal("cluster", id, data, target, weights, blocking, spatial)

R/CostSensTask.R

@@ -37,6 +37,8 @@ makeCostSensTask = function(id = deparse(substitute(data)), data, costs, blockin
   addClasses(task, "CostSensTask")
 }
 
+#' @export
+#' @rdname makeTaskDesc
 makeCostSensTaskDesc = function(id, data, target, blocking, costs, spatial) {
   td = makeTaskDescInternal("costsens", id, data, target, weights = NULL, blocking = blocking, spatial)
   td$class.levels = colnames(costs)

R/ImputeMethods.R

@@ -218,7 +218,8 @@ imputeNormal = function(mu = NA_real_, sd = NA_real_) {
 imputeHist = function(breaks, use.mids = TRUE) {
   if (missing(breaks)) {
     breaks = "Sturges"
-  } else {
+  }
+  if (!identical(breaks, "Sturges")) {
     breaks = asCount(breaks)
   }
   assertFlag(use.mids)

R/MultilabelTask.R

@@ -30,6 +30,8 @@ print.MultilabelTask = function(x, ...) {
   print(sums)
 }
 
+#' @export
+#' @rdname makeTaskDesc
 makeMultilabelTaskDesc = function(id, data, target, weights, blocking, spatial) {
   levs = target
   td = makeTaskDescInternal("multilabel", id, data, target, weights, blocking, spatial)

R/RegrTask.R

@@ -22,6 +22,8 @@ makeRegrTask = function(id = deparse(substitute(data)), data, target, weights =
   addClasses(task, "RegrTask")
 }
 
+#' @export
+#' @rdname makeTaskDesc
 makeRegrTaskDesc = function(id, data, target, weights, blocking, spatial) {
   addClasses(makeTaskDescInternal("regr", id, data, target, weights, blocking, spatial), c("RegrTaskDesc", "SupervisedTaskDesc"))
 }

R/SurvTask.R

@@ -42,6 +42,8 @@ makeSurvTask = function(id = deparse(substitute(data)), data, target, weights =
   addClasses(task, "SurvTask")
 }
 
+#' @export
+#' @rdname makeTaskDesc
 makeSurvTaskDesc = function(id, data, target, weights, blocking, spatial) {
   td = makeTaskDescInternal("surv", id, data, target, weights, blocking, spatial)
   addClasses(td, c("SurvTaskDesc", "SupervisedTaskDesc"))

R/Task.R

@@ -102,6 +102,23 @@
 #' }
 NULL
 
+#' Exported for internal use.
+#' @param id [\code{character(1)}]\cr
+#'   task id
+#' @param data [\code{data.frame}]\cr
+#'   data
+#' @param target [\code{character}]\cr
+#'   target columns
+#' @param weights [\code{numeric}]\cr
+#'   weights
+#' @param blocking [\code{numeric}\cr
+#'   task data blocking
+#' @param spatial [\code{logical(1)}]\cr
+#'   whether data is spatial
+#' @keywords internal
+#' @name makeTaskDesc
+NULL
+
 makeTask = function(type, data, weights = NULL, blocking = NULL, fixup.data = "warn", check.data = TRUE, spatial = FALSE) {
   if (fixup.data != "no") {
     if (fixup.data == "quiet") {

R/TaskDesc.R

@@ -34,6 +34,23 @@
 #' @rdname TaskDesc
 NULL
 
+#' Exported for internal use.
+#' @param type [\code{character(1)}]\cr
+#'   Task type.
+#' @param id [\code{character(1)}]\cr
+#'   task id
+#' @param data [\code{data.frame}]\cr
+#'   data
+#' @param target [\code{character}]\cr
+#'   target columns
+#' @param weights [\code{numeric}]\cr
+#'   weights
+#' @param blocking [\code{numeric}\cr
+#'   task data blocking
+#' @param spatial [\code{logical(1)}]\cr
+#'   whether data is spatial
+#' @keywords internal
+#' @export
 makeTaskDescInternal = function(type, id, data, target, weights, blocking, spatial) {
   # get classes of feature cols
   cl = vcapply(data, function(x) class(x)[1L])

R/Task_operators.R

@@ -421,6 +421,19 @@ subsetTask = function(task, subset = NULL, features) {
 
 
 # we create a new env, so the reference is not changed
+#' Change Task Data
+#'
+#' Mainly for internal use. Changes the data associated with a task, without modifying other task properties.
+#'
+#' @template arg_task
+#' @param data [\code{data.frame}]\cr
+#'   The new data to associate with the task. The names and types of the feature columns must match with the old data.
+#' @param costs [\code{data.frame}\cr
+#'   Optional: cost matrix.
+#' @param weights [\code{numeric}]\cr
+#'   Optional: weight vector.
+#' @keywords internal
+#' @export
 changeData = function(task, data, costs, weights) {
   if (missing(data))
     data = getTaskData(task)

R/checkLearner.R

@@ -1,6 +1,15 @@
 # props checks if Learner has one or more properties specified in props as a
 # character vector.
 
+#' Exported for internal use only.
+#' @param learner [\code{\link{Learner}} | \code{character(1)}]\cr
+#'   The learner to check, or the name of the learner to create
+#' @param type [\code{character(1)}]\cr
+#'   What type of learner to require.
+#' @param props [\code{character(1)}]\cr
+#'   What properties to require.
+#' @keywords internal
+#' @export
 checkLearner = function(learner, type = NULL, props = NULL) {
   if (is.character(learner))
     learner = makeLearner(learner)

R/checkLearnerBeforeTrain.R

@@ -27,7 +27,7 @@ checkLearnerBeforeTrain = function(task, learner, weights) {
   }
 
   if (td$n.feat["ordered"] > 0L && !hasLearnerProperties(learner, "ordered")) {
-    wrong.cols = getColNames(task, is.factor)
+    wrong.cols = getColNames(task, function(x) class(x)[1] == "ordered")
     stopf("Task '%s' has ordered factor inputs in '%s', but learner '%s' does not support that!", td$id, wrong.cols, learner$id)
   }
 

R/checkMeasures.R

@@ -1,5 +1,5 @@
 checkMeasures = function(measures, obj, aggr = NULL) {
-  if (missing(measures)) {
+  if (missing(measures) || is.null(measures)) {
     measures = list(getDefaultMeasure(obj))
   } else {
     measures = ensureVector(measures, n = 1L, cl = "Measure")

R/getHyperPars.R

@@ -14,7 +14,6 @@
 #' @family learner
 #' @export
 getHyperPars = function(learner,  for.fun = c("train", "predict", "both")) {
-  assertClass(learner, classes = "Learner")
   assertSubset(for.fun, choices = c("train", "predict", "both"))
   UseMethod("getHyperPars")
 }

R/predictLearner.R

@@ -91,6 +91,23 @@ insertLines = function(prediction, inserts) {
   return(ret)
 }
 
+#' @title Check output returned by predictLearner.
+#'
+#' @description
+#' Check the output coming from a Learner's internal
+#' \code{predictLearner} function.
+#'
+#' This function is for internal use.
+#'
+#' @param learner [\code{\link{Learner}}]\cr
+#'   The learner.
+#' @param model [\code{\link{WrappedModel}}]]\cr
+#'   Model produced by training.
+#' @param p [any]\cr
+#'   The prediction made by \code{learner}.
+#' @return [any]. A sanitized version of \code{p}.
+#' @keywords internal
+#' @export
 checkPredictLearnerOutput = function(learner, model, p) {
   cl = class(p)[1L]
   if (learner$type == "classif") {

R/setHyperPars.R

@@ -21,9 +21,8 @@
 #' print(cl2)
 setHyperPars = function(learner, ..., par.vals = list()) {
   args = list(...)
-  assertClass(learner, classes = "Learner")
-  assertList(args, names = "named", .var.name = "parameter settings")
-  assertList(par.vals, names = "named", .var.name = "parameter settings")
+  assertList(args, names = "unique", .var.name = "parameter settings")
+  assertList(par.vals, names = "unique", .var.name = "parameter settings")
   setHyperPars2(learner, insert(par.vals, args))
 }
 

tests/testthat/test_base_impute.R

@@ -109,9 +109,9 @@ test_that("Impute data frame", {
   x = impute(data, target = target, dummy.classes = "numeric", dummy.cols = "z")$data
   expect_true(setequal(names(x), c(names(data), "x.dummy", "y.dummy", "z.dummy")))
 
-  x = impute(data, target = target, classes = list(factor = imputeMode(), numeric = imputeMedian(),
+  x = impute(data, target = character(0), classes = list(factor = imputeMode(), numeric = imputeMedian(),
       integer = imputeMedian(), logical = imputeConstant(1)))
-  expect_true(all(!is.na(x)))
+  expect_true(all(!is.na(x$data)))
 
   data2 = data[1:5, ]
   x = impute(data2, target = target, dummy.classes = c("numeric", "logical", "factor"), force.dummies = TRUE)