Release 1.2.0

.github/workflows/rhub.yaml

@@ -7,20 +7,24 @@
 # It is unlikely that you need to modify this file manually.
 
 name: R-hub
-run-name: "${{ github.event.inputs.id }}: ${{ github.event.inputs.name || format('Manually run by {0}', github.triggering_actor) }}"
+run-name: >-
+  ${{ github.event.inputs.id || format('Manual run by @{0}', github.triggering_actor) }}:
+  ${{ github.event.inputs.name || github.event.inputs.config }}
 
 on:
   workflow_dispatch:
     inputs:
       config:
-        description: 'A comma separated list of R-hub platforms to use.'
+        description: >-
+          Comma-separated list of R-hub platforms to use.
+          Full list: https://r-hub.github.io/containers/
         type: string
         default: 'linux,windows,macos'
       name:
-        description: 'Run name. You can leave this empty now.'
+        description: 'Run name. You can leave this empty.'
         type: string
       id:
-        description: 'Unique ID. You can leave this empty now.'
+        description: 'Unique ID. You can leave this empty.'
         type: string
 
 jobs:

DESCRIPTION

@@ -1,6 +1,6 @@
 Package: xplainfi
 Title: Feature Importance Methods for Global Explanations
-Version: 1.1.0.9000
+Version: 1.2.0
 Authors@R: 
     person("Lukas", "Burk", , "cran@lukasburk.de", role = c("aut", "cre", "cph"),
            comment = c(ORCID = "0000-0001-7528-3795"))

NEWS.md

@@ -1,4 +1,4 @@
-# xplainfi 1.1.0.9000 (development version)
+# xplainfi 1.2.0
 
 ## Behavior changes
 
@@ -24,6 +24,7 @@
 
 ## Bug fixes
 
+- `$importance()` with a test-based `ci_method` (`"lei"`, `"cpi"`) now warns and returns `NA` inference fields for a feature whose hypothesis test errors, instead of aborting the whole call; this guards against an R-devel (2026-05) change to `stats::wilcox.test` that errors on many-ties or many-zeros inputs.
 - `relation = "ratio"` importances now return `NA` (with a warning) instead of `Inf`/`NaN` for features whose baseline score is `0`.
 - `PerturbationImportance` no longer calls `$obs_loss()` without `measure`, which errored when `measures` was not the task default.
 - `ConditionalARFSampler$sample()` now errors when `parallel = TRUE` but no parallel backend is registered, e.g. after deserializing a sampler in a new session.
@@ -41,7 +42,7 @@
 
 - Use of a pre-trained `mlr3` learner is now supported in `PerturbationImportance` (`PFI`, `CFI`, `RFI`) and `SAGE` methods.
   - Requires the provided `Resampling` to be instantiated and consist of a single iteration, e.g. there must be only 1 test set.
-  - The `rsmp_all_test(task)` utility can be used to construct a single-iteration `Resampling` object from a given `Task` where all observations are alligned to the test set and the train set is empty. We will likely refine the API around this in the future.
+  - The `rsmp_all_test(task)` utility can be used to construct a single-iteration `Resampling` object from a given `Task` where all observations are aligned to the test set and the train set is empty. We will likely refine the API around this in the future.
   - Internally, a `ResampleResult` will be constructed from the given `learner`, `task`, and `resampling` arguments, which is then consistent with the previous default of performing `resample()` to get trained learners for each resampling iteration.
 
 ## Inference
@@ -95,7 +96,7 @@ The major version bump is largely to mark the occasion that the package is now c
 - Clean up various documentation issues and other metadata.
 - Adjusted the `min_permutations` default in `SAGE` methods to 10 rather than 3, since the previous value was found to lead to spurious early stopping.
 - Fix `sim_dgp_ewald` lading to erroneous variances when compared to their settings.
-- Reduce runtime of tests (mostly by using less ARF and mor Gaussian sampling)
+- Reduce runtime of tests (mostly by using less ARF and more Gaussian sampling)
 - Remove `KnockoffSequentialSampler` as the `seqknockoff` package is not available on CRAN or R-universe. `KnockoffSampler` with the corresponding `knockoff_fun = seqknockoff::knockoffs_seq` still works.
 
 # xplainfi 0.2.1

R/SAGE-marginal.R

@@ -10,10 +10,10 @@
 #' task = tgen("friedman1")$generate(n = 100)
 #' sage = MarginalSAGE$new(
 #'   task = task,
-#'   learner = lrn("regr.ranger", num.trees = 50),
+#'   learner = lrn("regr.ranger", num.trees = 20),
 #'   measure = msr("regr.mse"),
-#'   n_permutations = 3L,
-#'   n_samples = 20
+#'   n_permutations = 2L,
+#'   n_samples = 10
 #' )
 #' sage$compute()
 #' @export

tests/testthat.R

@@ -6,6 +6,10 @@
 # * https://r-pkgs.org/testing-design.html#sec-tests-files-overview
 # * https://testthat.r-lib.org/articles/special-files.html
 
+# Cap threads before any BLAS/OpenMP init: CRAN's Debian machine links a multithreaded (OpenMP) BLAS whose
+# spin-pool inflates the CPU/elapsed ratio. Must be set here, before library() and any matrix op, to take effect.
+Sys.setenv("OMP_THREAD_LIMIT" = 2)
+
 library(testthat)
 library(xplainfi)
 

tests/testthat/setup.R

@@ -3,6 +3,13 @@
 # Load mlr3 to avoid mlr3:: namespace clutter in tests
 library(mlr3)
 
+# CRAN policy: at most 2 cores in tests. data.table (OpenMP) and ranger (own thread pool via arf) need capping;
+# ranger uses std::thread, so OMP_THREAD_LIMIT in testthat.R does not cover it. Restored in teardown.
+old_dt_threads = data.table::getDTthreads()
+old_ranger_threads = getOption("ranger.num.threads")
+data.table::setDTthreads(2)
+options(ranger.num.threads = 2)
+
 # Get current log threshold
 logger = lgr::get_logger("mlr3")
 old_threshold = logger$threshold
@@ -18,6 +25,8 @@ withr::defer(
   {
     lgr::get_logger("mlr3")$set_threshold(old_threshold)
     xplain_opt(verbose = old_opts$verbose, progress = old_opts$progress)
+    data.table::setDTthreads(old_dt_threads)
+    options(ranger.num.threads = old_ranger_threads)
   },
   teardown_env()
 )

vignettes/articles/perturbation-importance.Rmd

@@ -314,7 +314,7 @@ digraph confounding_dag {
 
 The red arrows show the confounding paths: the hidden confounder creates spurious correlations between x1, proxy, and y. The **blue arrows** show true direct causal effects. Note that `independent` is truly independent (no confounding) while `proxy` provides a noisy measurement of the confounder.
 
-In the observable confounder scenaro (used later), the confounder H would be included as a feature in the dataset, allowing direct conditioning rather than relying on the noisy proxy.
+In the observable confounder scenario (used later), the confounder H would be included as a feature in the dataset, allowing direct conditioning rather than relying on the noisy proxy.
 
 ```{r viz-confounding}
 #| echo: false