fix: hash for filters

DESCRIPTION

@@ -78,7 +78,7 @@ Suggests:
     lme4,
     mlbench,
     bbotk (>= 0.3.0),
-    mlr3filters (>= 0.1.1),
+    mlr3filters (>= 0.8.1),
     mlr3learners,
     mlr3measures,
     nloptr,

NEWS.md

@@ -9,6 +9,7 @@
 * New PipeOp `PipeOpDecode` / `po("decode")` to reverse one-hot or treatment encoding.
 * Fix: Columns that are `feature` and something else no longer lose the other column role during training or predicting of `PipeOp`s inheriting from `PipeOpTaskPreproc`.
 * Fix: Made tests for `PipeOpBLSmote` deterministic.
+* Fix: Corrected hash calculation for `PipeOpFilter`.
 
 # mlr3pipelines 0.7.1
 

R/PipeOpFilter.R

@@ -191,7 +191,7 @@ PipeOpFilter = R6Class("PipeOpFilter",
     .transform = function(task) {
       task$select(self$state$features)
     },
-    .additional_phash_input = function() class(self$filter)
+    .additional_phash_input = function() self$filter$hash
   )
 )
 

tests/testthat/test_dictionary.R

@@ -84,8 +84,6 @@ test_that("Dictionary contains all PipeOps", {
 
   # the loop now checks whether we can construct each pipeop from the dictionary *and* by itself
   for (idx in seq_along(dictnames)) {
-    if (dictnames[[idx]] == "filter") next  # TODO: remove this when https://github.com/mlr-org/mlr3filters/issues/162 is solved
-
     pogen = get(pipeops[idx], pkgenv)  # the constructor, as found in the package namespace
     dictname = dictnames[idx]  # the "key" in the mlr_pipeops dictionary
     args = initargs[[pipeops[idx]]] %??% list()  # the required arguments, if any. e.g. 'outnum' for PipeOpCopy.