Merge branch 'filterensemble_trafos' of https://github.com/mlr-org/mlr3pipelines into filterensemble_trafos

Author: advieser

.github/workflows/pkgdown.yml

@@ -44,7 +44,7 @@ jobs:
 
       - name: Deploy
         if: github.event_name != 'pull_request'
-        uses: JamesIves/github-pages-deploy-action@v4.7.4
+        uses: JamesIves/github-pages-deploy-action@v4.8.0
         with:
           clean: false
           branch: gh-pages

NEWS.md

@@ -5,7 +5,8 @@
 * feat: All imputation PipeOps now support feature types `Date` and `POSIXct`.
 * Fix: `PipeOpTextVectorizer` now uses coercion to `TsparseMatrix` instead of deprecated `dgTMatrix` to avoid `Matrix` deprecation warnings.
 * New method `$predict_newdata_fast()` for `GraphLearner`. Note that currently this is only a thin wrapper around `$predict_newdata()` to maintain compatibility, but in the future it may get optimized to enable faster predictions on new data.
-- feat: Added new hyperparameters `filter_score_transform`, `result_score_transform`, and `aggregator` to `FilterEnsemble`. BREAKING CHANGE: The default behavior for handling NA scores in the aggregation has changed. Previously, NA scores were simply ignored and weights were not changed. Now, `weighted.mean` is used, which normalizes the weights for all non-NA scores. 
+* feat: `PipeOpRenameColumns`'s hyperparameter `renaming` can now also take a function transforming old column names to new column names.
+* feat: Added new hyperparameters `filter_score_transform`, `result_score_transform`, and `aggregator` to `FilterEnsemble`. BREAKING CHANGE: The default behavior for handling NA scores in the aggregation has changed. Previously, NA scores were simply ignored and weights were not changed. Now, `weighted.mean` is used, which normalizes the weights for all non-NA scores. 
 
 # mlr3pipelines 0.10.0
 

R/PipeOpRenameColumns.R

@@ -28,17 +28,21 @@
 #'
 #' @section Parameters:
 #' The parameters are the parameters inherited from [`PipeOpTaskPreproc`], as well as:
-#' * `renaming` :: named `character`\cr
-#'   Named `character` vector. The names of the vector specify the old column names that should be
-#'   changed to the new column names as given by the elements of the vector. Initialized to the empty
-#'   character vector.
+#' * `renaming` :: named `character` | `function`\cr
+#'   Takes the form of either a named `character` or a `function`.
+#'   For a named `character` vector, the names of the vector elements specify the
+#'   old column names and the corresponding element values give the new column names.
+#'   A `function` specifies how the old column names should be changed to the new column names.
+#'   The function must return a `character` vector with one entry per input column name so that each selected column receives a new name.
+#'   To choose columns use the `affect_columns` parameter.
+#'   Initialized to `character(0)`.
 #' * `ignore_missing` :: `logical(1)`\cr
 #'   Ignore if columns named in `renaming` are not found in the input [`Task`][mlr3::Task]. If this is
 #'   `FALSE`, then names found in `renaming` not found in the [`Task`][mlr3::Task] cause an error.
 #'   Initialized to `FALSE`.
 #'
 #' @section Internals:
-#' Uses the `$rename()` mutator of the [`Task`][mlr3::Task] to set the new column names.
+#' Uses the `$rename()` mutator of the [`Task`][mlr3::Task] to set new column names.
 #'
 #' @section Fields:
 #' Only fields inherited from [`PipeOp`].
@@ -56,36 +60,53 @@
 #' task = tsk("iris")
 #' pop = po("renamecolumns", param_vals = list(renaming = c("Petal.Length" = "PL")))
 #' pop$train(list(task))
+#'
+#' pof = po("renamecolumns", param_vals = list(renaming = function(colnames) {
+#'   sub("Petal", "P", colnames)
+#' }))
+#' pof$train(list(task))
+#' 
 PipeOpRenameColumns = R6Class("PipeOpRenameColumns",
   inherit = PipeOpTaskPreprocSimple,
   public = list(
     initialize = function(id = "renamecolumns", param_vals = list()) {
       ps = ps(
         renaming = p_uty(
-          custom_check = crate(function(x) check_character(x, any.missing = FALSE, names = "strict") %check&&% check_names(x, type = "strict"),
-                               .parent = topenv()),
+          custom_check = crate(function(x) (check_character(x, any.missing = FALSE, names = "strict") %check&&% check_names(x, type = "strict")) %check||% check_function(x)),
           tags = c("train", "predict", "required")
         ),
         ignore_missing = p_lgl(tags = c("train", "predict", "required"))
       )
       ps$values = list(renaming = character(0), ignore_missing = FALSE)
-      super$initialize(id, ps, param_vals = param_vals, can_subset_cols = FALSE)
+      super$initialize(id, ps, param_vals = param_vals, can_subset_cols = TRUE)
     }
   ),
   private = list(
+    .get_state = function(task) {
+      if (is.function(self$param_set$values$renaming)) {
+        new_names = self$param_set$values$renaming(task$feature_names)
+        assert_character(new_names, any.missing = FALSE, len = length(task$feature_names), .var.name = "the value returned by `renaming` function")
+        names(new_names) = task$feature_names
+        list(old_names = task$feature_names, new_names = new_names)
+      } else {
+        pv = self$param_set$get_values(tags = "train")
+        new_names = pv$renaming
+        innames = names(new_names)
+        nontargets = task$col_roles
+        nontargets$target = NULL
+        takenames = innames %in% unlist(nontargets)
+        if (!pv$ignore_missing && !all(takenames)) {
+          # we can't rely on task$rename because it could also change the target name, which we don't want.
+          stopf("The names %s from `renaming` parameter were not found in the Task.", str_collapse(innames[!takenames]))
+        }
+        list(old_names = innames[takenames], new_names = new_names[takenames])
+      }
+    },
     .transform = function(task) {
-      if (!length(self$param_set$values$renaming)) {
+      if (!length(self$state$new_names)) {
         return(task)  # early exit
       }
-      innames = names(self$param_set$values$renaming)
-      nontargets = task$col_roles
-      nontargets$target = NULL
-      takenames = innames %in% unlist(nontargets)
-      if (!self$param_set$values$ignore_missing && !all(takenames)) {
-        # we can't rely on task$rename because it could also change the target name, which we don't want.
-        stopf("The names %s from `renaming` parameter were not found in the Task.", str_collapse(innames[!takenames]))
-      }
-      task$rename(old = innames[takenames], new = self$param_set$values$renaming[takenames])
+      task$rename(old = self$state$old_names, new = self$state$new_names)
     }
   )
 )

man/mlr_pipeops_renamecolumns.Rd

@@ -39,3 +39,23 @@ test_that("error handling", {
   op$param_set$values$ignore_missing = TRUE
   expect_equal(task$data(), op$train(list(task))[[1]]$data())
 })
+
+test_that("PipeOpRenameColumns - errors for renaming function", {
+  task = mlr_tasks$get("iris")
+  expect_error(po("renamecolumns", param_vals = list(renaming = 1 + 1)))
+
+  op = po("renamecolumns", param_vals = list(renaming = function(x) "a"))
+  expect_error(op$train(list(task)), "value returned by `renaming` function.*length")
+})
+
+test_that("PipeOpRenameColumns - renamimg by function", {
+  task = mlr_tasks$get("iris")
+  
+  op = po("renamecolumns", param_vals = list(renaming = function(colnames) sub("Petal", "P", colnames)))
+  result = op$train(list(task))
+  expect_equal(result[[1]]$feature_names, c("P.Length", "P.Width", "Sepal.Length", "Sepal.Width"))
+
+  op$param_set$set_values(affect_columns = selector_name("Petal.Length"))
+  result = op$train(list(task))
+  expect_equal(result[[1]]$feature_names, c("P.Length", "Petal.Width", "Sepal.Length", "Sepal.Width"))                                          
+})