mlr3 package updates - q3/2022

This posts gives an overview by listing the recent release notes of mlr3 packages from the last quarter. Cover photo by Etienne Girardet.


Sebastian Fischer


October 18, 2022

Due to the high amount of packages in the mlr3 ecosystem, it is hard to keep up with the latest changes across all packages. This posts gives an overview by listing the recent release notes of mlr3 packages from the last quarter. Note that only CRAN packages are listed here and the sort order is alphabetically.

bbotk 0.5.4

Description Black-Box Optimization Toolkit

  • feat: Add OptimizerFocusSearch that performs a focusing random search.

mlr3 0.14.0

Description Machine Learning in R - Next Generation

  • Added multiclass measures: mauc_aunu, mauc_aunp, mauc_au1u, mauc_au1p.
  • Measure classif.costs does not require a Task anymore.
  • New converter: as_task_unsupervised()
  • Refactored the task types in mlr_reflections.

mlr3 0.13.4

  • Added new options for parallelization ("mlr3.exec_random" and "mlr3.exec_chunk_size"). These options are passed down to the respective map functions in package future.apply.
  • Fixed runtime measures depending on specific predict types (#832).
  • Added head() and tail() methods for Task.
  • Improved printing of multiple objects.

mlr3benchmark 0.1.4

Description Analysis and Visualisation of Benchmark Experiments

  • Add friedman_global argument to posthoc tests and to autoplots to allow methods and plots to run even if the global Friedman test fails (i.e. don’t reject null)
  • New maintainer: Sebastian Fischer
  • Fix documentation

mlr3cluster 0.1.4

Description Cluster Extension for ‘mlr3’

  • code refactoring

mlr3data 0.6.1

Description Collection of Machine Learning Data Sets for ‘mlr3’

  • Fixed documentation and CRAN notes.

  • Added simplified version of the penguins data set as penguins_simple.

  • Added labels to data sets.

mlr3db 0.5.0

Description Data Base Backend for ‘mlr3’

  • Support for parquet files as Backend via DuckDB.
  • New converter as_duckdb_backend().

mlr3fairness 0.3.1

Description Fairness Auditing and Debiasing for ‘mlr3’

  • Minor update to improve stability of unit tests and vignette building on CRAN.

mlr3filters 0.6.0

Description Filter Based Feature Selection for ‘mlr3’

  • Add FilterCarSurvScore (#120, (mllg?))
  • Use featureless learner instead of rpart as default learner for FilterImportance and FilterPerformance (#124)
  • Add documentation for PipeOpFilter
  • Add mlr3pipelines examples to help pages (#135, (sebffischer?))
  • Add label arg to Filter class (#121, (mllg?))

mlr3fselect 0.7.2

Description Feature Selection for ‘mlr3’

  • docs: Re-generate rd files with valid html.

mlr3hyperband 0.4.2

Description Hyperband for ‘mlr3’

  • docs: Re-generate rd files with valid html.

mlr3learners 0.5.4

Description Recommended Learners for ‘mlr3’

  • Added regr.nnet learner.
  • Removed the option to use weights in classif.log_reg.
  • Added default_values() function for ranger and svm learners.
  • Improved documentation.

mlr3measures 0.5.0

Description Performance Measures for ‘mlr3’

  • Added some observation-wise loss functions: ae, ape, se, sle, and zero_one,

mlr3oml 0.6.0

Description Connector Between ‘mlr3’ and ‘OpenML’


  • Add R6 classes for OMLCollection, OMLRun, OMLFlow.
  • Added function benchmark_grid_oml that allows for easier creation of benchmark designs from OpenML task-resampling pairs.
  • Added sugar functions oml_flow, oml_data, oml_task, oml_run, oml_collection for all OpenML objects.
  • Conversion from OpenML to mlr3 objects is now only possible with the usual s3-converters as_<object>. This improves consistency by ensuring that the subcomponents of OpenML objects are always OpenML objects and not suddenly mlr3 objects.
  • Added more converter functions: as_learner, as_resample_result, as_data_backend, as_benchmark_result.
  • Added support for parquet files that were recently introduced on OpenML. The global option mlr3oml.parquet can be used to enable or disable this. By default it is FALSE. This is implemented via the duckdb backend from mlr3db.
  • Support to use the OpenML test server. This can be globally enabled using the option mlr3oml.test_server or individually for objects. Options to globally define an API-key for the test server are through the environment variable TESTOPENMLAPIKEY or the option mlr3oml.test_api_key


  • Removed support for survival tasks as mlr3proba is no longer on CRAN
  • OpenML tasks can now also be filtered according to the task type


  • Implement an arff writer and remove the arff dependency, therefore also removing the option "farff" as the mlr3oml.arff_parser
  • Increment the cache version number due to changes in the cache structure: This will flush the previous cache folder.
  • Simplified the code structure by adding OMLObject class from which all other OpenML objects like OMLData, OMLTask inherit.

mlr3pipelines 0.4.2

Description Preprocessing Operators and Pipelines for ‘mlr3’

  • Documentation: Clarified PipeOpHistBin operation.
  • Documentation: Fixed PipeOpPCA documentation of center default.
  • Added $label active binding, setting it to the help()-page title by default.
  • Made tests compatible with upcoming mlr3misc update.

mlr3spatial 0.2.1

Description Support for Spatial Objects Within the ‘mlr3’ Ecosystem

  • fix: add "space" and "time" column role from mlr3spatiotempcv

mlr3spatial 0.2.0

  • BREAKING CHANGE: TaskClassifST and TaskRegrST are used to train a learner with spatial data. The new tasks unify the work with mlr3spatiotempcv.
  • BREAKING CHANGE: Raster objects cannot be used to create tasks for training anymore.
  • BREAKING CHANGE: TaskUnsupervised is used to predict on rasters objects now. The new task type is more convenient for data without a response.
  • feat: Add as_task_regr_st() and as_task_classif_st() from spatial objects.
  • feat: Add as_task_unsupervised() from raster objects.
  • feat: Task leipzig with land cover target.
  • feat: data("leipzig") loads an sf object with land cover in Leipzig.
  • feat: GeoTIFF and GeoPackage of Leipzig in extdata folder.
  • refactor: Vector data is handled with DataBackendDataTable now and DataBackendVector is removed.
  • BREAKING CHANGE: DataBackendRaster cannot be created from RasterLayer objects anymore.
  • fix: spatial_predict() returned an unnamed response.
  • fix: spatial_predict() wrote predictions to the wrong cell.
  • BREAKING CHANGE: Remove demo_raster(), demo_stack_spatraster(), demo_stack_rasterbrick() and demo_rasterbrick() functions.
  • feat: Prediction layer contains NA at raster cells with NA values in one or more feature layers.

mlr3tuning 0.14.0

Description Tuning for ‘mlr3’

  • feat: Add option evaluate_default to evaluate learners with hyperparameters set to their default values.
  • refactor: From now on, the default of smooth is FALSE for TunerGenSA.

mlr3viz 0.5.10

Description Visualizations for ‘mlr3’

  • Improved documentation.
  • Make checks run without suggested packages.

paradox 0.10.0

Description Define and Work with Parameter Spaces for Complex Algorithms

  • Reset .has_extra_trafo to FALSE when trafo is set to NULL.
  • rd_info.ParamSet collapses vector with "\n" due changes in roxygen 7.2.0
  • Add method set_values() to conveniently add parameter values.