Converters for Resampling <-> data.table

NAMESPACE

@@ -39,6 +39,7 @@ S3method(as_resample_result,ResampleResult)
 S3method(as_resample_result,ResultData)
 S3method(as_resample_result,list)
 S3method(as_resampling,Resampling)
+S3method(as_resampling,data.table)
 S3method(as_resamplings,default)
 S3method(as_resamplings,list)
 S3method(as_task,Task)

R/Resampling.R

@@ -46,6 +46,17 @@
 #' Next, the grouping information is replaced with the respective row ids to generate training and test sets.
 #' The sets can be accessed via `$train_set(i)` and `$test_set(i)`, respectively.
 #'
+#' @section Conversion to and from Tabular Data:
+#' Resamplings can be converted to a plain `data.table` with `as.data.table()`, resulting in a table with the
+#' following columns:
+#'
+#' * `set`: factor with levels `"train"` and `"test"`
+#' * `iteration`: non-negative integer
+#' * `row_id`: integer row ids
+#'
+#' This table can serve as exchange format with other toolkits. A table with this exact structure can also be
+#' converted to a [ResamplingCustom] again via [as_resampling()]. Note that while the training and test sets are
+#' preserved for all iterations, the elements inside the set may be reordered.
 #'
 #' @template seealso_resampling
 #' @export

R/as_resampling.R

@@ -3,6 +3,12 @@
 #' @description
 #' Convert object to a [Resampling] or a list of [Resampling].
 #'
+#' For the conversion of a `data.table`, the following columns (as returned by
+#' [as.data.table.Resampling()]) are mandatory to create a [ResamplingCustom]:
+#' * `set`: factor with levels `"train"` and `"test"`
+#' * `iteration`: non-negative integer
+#' * `row_id`: integer row ids
+#'
 #' @inheritParams as_task
 #' @export
 as_resampling = function(x, ...) { # nolint
@@ -15,6 +21,26 @@ as_resampling.Resampling = function(x, clone = FALSE, ...) { # nolint
   if (isTRUE(clone)) x$clone() else x
 }
 
+
+#' @export
+#' @rdname as_resampling
+as_resampling.data.table = function(x, ...) { # nolint
+  assert_data_table(x, min.rows = 1L, col.names = "unique")
+  assert_names(names(x), permutation.of = c("set", "iteration", "row_id"))
+  assert_factor(x$set, any.missing = FALSE)
+  assert_set_equal(levels(x$set), c("train", "test"))
+  assert_integerish(x$iteration, lower = 1L, any.missing = FALSE)
+  assert_integerish(x$row_id, any.missing = FALSE)
+
+  row_id = NULL
+  resampling = ResamplingCustom$new()
+  resampling$instance = list(
+    train_sets = x[list("train"), list(ids = list(row_id)), by = "iteration"]$ids,
+    test_sets = x[list("test"), list(ids = list(row_id)), by = "iteration"]$ids
+  )
+  resampling
+}
+
 #' @export
 #' @rdname as_resampling
 as_resamplings = function(x, ...) { # nolint

tests/testthat/test_Resampling.R

@@ -104,6 +104,13 @@ test_that("as.data.table.Resampling", {
   expect_integer(tab$iteration, any.missing = FALSE)
   expect_factor(tab$set, levels = c("train", "test"), any.missing = FALSE)
   expect_integer(tab$row_id, any.missing = FALSE)
+
+  r2 = as_resampling(tab)
+  expect_r6(r2, "ResamplingCustom")
+  for (i in seq_len(r$iters)) {
+    expect_set_equal(r$train_set(i), r2$train_set(i))
+    expect_set_equal(r$test_set(i), r2$test_set(i))
+  }
 })
 
 test_that("custom_cv", {