started to add support for test rows

DESCRIPTION

@@ -87,20 +87,22 @@ Suggests:
  methods,
  vtreat,
  future
+Remotes:
+ mlr-org/mlr3@feat/train-predict
 ByteCompile: true
 Encoding: UTF-8
 Config/testthat/edition: 3
 Config/testthat/parallel: true
 NeedsCompilation: no
 Roxygen: list(markdown = TRUE, r6 = FALSE)
-RoxygenNote: 7.2.3
+RoxygenNote: 7.3.1
 VignetteBuilder: knitr
 Collate:
  'Graph.R'
- 'GraphLearner.R'
- 'mlr_pipeops.R'
  'multiplicity.R'
  'utils.R'
+ 'GraphLearner.R'
+ 'mlr_pipeops.R'
  'PipeOp.R'
  'PipeOpEnsemble.R'
  'LearnerAvg.R'

NAMESPACE

@@ -148,6 +148,7 @@ import(mlr3)
 import(mlr3misc)
 import(paradox)
 importFrom(R6,R6Class)
+importFrom(data.table,as.data.table)
 importFrom(digest,digest)
 importFrom(stats,setNames)
 importFrom(utils,bibentry)

R/GraphLearner.R

@@ -40,6 +40,7 @@
 #' so `TRUE` (default) is recommended. In particular, note that the `$state` of `$graph` is set to `NULL` by reference on
 #' construction of `GraphLearner`, during `$train()`, and during `$predict()` when `clone_graph` is `FALSE`.
 #'
+#' @include utils.R
 #' @section Fields:
 #' Fields inherited from [`PipeOp`], as well as:
 #' * `graph` :: [`Graph`]\cr
@@ -102,7 +103,7 @@ GraphLearner = R6Class("GraphLearner", inherit = Learner,
  feature_types = mlr_reflections$task_feature_types,
  predict_types = names(mlr_reflections$learner_predict_types[[task_type]]),
  packages = graph$packages,
- properties = mlr_reflections$learner_properties[[task_type]],
+ properties = setdiff(mlr_reflections$learner_properties[[task_type]], "uses_test_rows"),
  man = "mlr3pipelines::GraphLearner"
  )
 
@@ -173,6 +174,9 @@ GraphLearner = R6Class("GraphLearner", inherit = Learner,
  }
  ),
  private = list(
+ .contingent_properties = function() {
+ if (some(self$graph$pipeops, uses_test_rows)) "uses_test_rows" else character(0)
+ },
  .graph = NULL,
  deep_clone = function(name, value) {
  # FIXME this repairs the mlr3::Learner deep_clone() method which is broken.
@@ -187,6 +191,15 @@ GraphLearner = R6Class("GraphLearner", inherit = Learner,
 
  .train = function(task) {
  on.exit({self$graph$state = NULL})
+
+ if (!("uses_test_rows" %in% self$properties)) {
+ # PipeOps like PipeOpTaskPreproc will always make predictions for the test rows during train
+ # To avoid doing this unless we need it, we remove those row roles temporarily if no pipeop needs them
+ on.exit({task$row_roles$test = prev_test}, add = TRUE) # nolint
+ prev_test = task$row_roles$test
+ task$row_roles$test = integer(0)
+ }
+
  self$graph$train(task)
  state = self$graph$state
  state

R/PipeOp.R

@@ -335,6 +335,10 @@ PipeOp = R6Class("PipeOp",
  ),
 
  active = list(
+ properties = function(rhs) {
+ assert_ro_binding(rhs)
+ character(0)
+ },
  id = function(val) {
  if (!missing(val)) {
  private$.id = val

R/PipeOpImpute.R

@@ -148,6 +148,7 @@ PipeOpImpute = R6Class("PipeOpImpute",
 
  .train = function(inputs) {
  intask = inputs[[1]]$clone(deep = TRUE)
+ task = intask$clone(deep = TRUE)
 
  affected_cols = (self$param_set$values$affect_columns %??% selector_all())(intask)
  affected_cols = intersect(affected_cols, private$.select_cols(intask))
@@ -189,8 +190,29 @@ PipeOpImpute = R6Class("PipeOpImpute",
 
  intask$select(setdiff(intask$feature_names, colnames(imputanda)))$cbind(imputanda)
 
+ test_rows_exist = length(task$row_roles$test) > 0
+
  self$state$outtasklayout = copy(intask$feature_types)
 
+ if (test_rows_exist) {
+ # FIXME: This is copy pasta from PipeOpTaskPreproc
+ predict_task = task$clone(deep = TRUE)
+ predict_task$row_roles$use = task$row_roles$test
+ predict_task = private$.predict(list(predict_task))
+
+ test_cols = unique(unlist(intask$col_roles))
+
+ test_data = predict_task[[1]]$data(cols = test_cols)
+ # this creates new row_ids for the test data
+ prev_use = intask$row_roles$use
+ intask$rbind(test_data)
+ intask$row_roles$test = setdiff(intask$row_roles$use, prev_use)
+ intask$row_roles$use = prev_use
+ }
+
+
+
+
  list(intask)
  },
 

R/PipeOpLearner.R

@@ -97,6 +97,10 @@ PipeOpLearner = R6Class("PipeOpLearner", inherit = PipeOp,
  }
  ),
  active = list(
+ properties = function(rhs) {
+ assert_ro_binding(rhs)
+ if ("uses_test_rows" %in% private$.learner$properties) "uses_test_rows" else character(0)
+ },
  id = function(val) {
  if (!missing(val)) {
  private$.id = val

R/PipeOpLearnerCV.R

@@ -142,6 +142,10 @@ PipeOpLearnerCV = R6Class("PipeOpLearnerCV",
 
  ),
  active = list(
+ properties = function(rhs) {
+ assert_ro_binding(rhs)
+ if ("uses_test_rows" %in% private$.learner$properties) "uses_test_rows" else character(0)
+ },
  learner = function(val) {
  if (!missing(val)) {
  if (!identical(val, private$.learner)) {

R/PipeOpTaskPreproc.R

@@ -199,8 +199,10 @@ PipeOpTaskPreproc = R6Class("PipeOpTaskPreproc",
 
  .train = function(inputs) {
  intask = inputs[[1]]$clone(deep = TRUE)
+
  do_subset = !is.null(self$param_set$values$affect_columns)
  affected_cols = intask$feature_names
+ remove_cols = NULL
  if (do_subset) {
  affected_cols = self$param_set$values$affect_columns(intask)
  assert_subset(affected_cols, intask$feature_names, empty.ok = TRUE)
@@ -210,16 +212,52 @@ PipeOpTaskPreproc = R6Class("PipeOpTaskPreproc",
  }
  intasklayout = copy(intask$feature_types)
 
+ test_rows_exist = length(intask$row_roles$test) > 0
+ if (test_rows_exist) {
+ predict_task = intask$clone(deep = TRUE)
+ }
+
  intask = private$.train_task(intask)
 
  self$state$affected_cols = affected_cols
  self$state$intasklayout = intasklayout
  self$state$outtasklayout = copy(intask$feature_types)
  self$state$outtaskshell = intask$data(rows = intask$row_ids[0])
 
+
+ if (test_rows_exist) {
+ predict_task$row_roles$use = predict_task$row_roles$test
+ predict_task = private$.predict_task(predict_task)
+
+ # FIXME: These are all the columns that a learner might use.
+ # To be on the save side, we could also add all available columns
+ test_cols = unique(c(remove_cols, unlist(predict_task$col_roles)))
+
+ test_data = predict_task$data(cols = test_cols)
+
+ # in some cases (such as class weights, different columns are added during train and predict),
+ # we fill those values with NAs
+ missing_cols = setdiff(unlist(intask$col_roles), colnames(test_data))
+ if (length(missing_cols)) {
+ missing_data = intask$data(intask$row_roles$use[1L], missing_cols)
+ missing_data[1, ] = NA
+ missing_data = missing_data[1, lapply(get(".SD"), function(col) rep(col, nrow(test_data)))]
+ test_data = cbind(missing_data, test_data)
+ }
+
+ # this creates new row_ids for the test data
+ prev_use = intask$row_roles$use
+ intask$rbind(test_data)
+ intask$row_roles$test = setdiff(intask$row_roles$use, prev_use)
+ intask$row_roles$use = prev_use
+ }
  if (do_subset) {
  # FIXME: this fails if .train_task added a column with the same name
- intask$col_roles$feature = union(intask$col_roles$feature, y = remove_cols)
+ new_features = union(intask$col_roles$feature, y = remove_cols)
+ intask$col_roles$feature = new_features
+ if (test_rows_exist) {
+ intask$col_roles$feature = new_features
+ }
  }
  list(intask)
  },

R/utils.R

@@ -134,3 +134,7 @@ dictionary_sugar_inc_mget = function(dict, .keys, ...) {
  names(objs) = map_chr(objs, "id")
  objs
 }
+
+uses_test_rows = function(pipeop) {
+ "uses_test_rows" %in% pipeop$properties
+}