* Fix: Made `FilterEnsemble` tests deterministic and more robust. * Fix: Made tests for `PipeOpLearnerCV` deterministic. * feat: All imputation PipeOps now support feature types `Date` and `POSIXct`. * Fix: `PipeOpTextVectorizer` now uses coercion to `TsparseMatrix` instead of deprecated `dgTMatrix` to avoid `Matrix` deprecation warnings. * New method `$predict_newdata_fast()` for `GraphLearner`. Note that currently this is only a thin wrapper around `$predict_newdata()` to maintain compatibility, but in the future it may get optimized to enable faster predictions on new data. * feat: `PipeOpRenameColumns`'s hyperparameter `renaming` can now also take a function transforming old column names to new column names. * feat: Added new hyperparameters `filter_score_transform`, `result_score_transform`, and `aggregator` to `FilterEnsemble`. BREAKING CHANGE: The default behavior for handling NA scores in the aggregation has changed. Previously, NA scores were simply ignored and weights were not changed. Now, `weighted.mean` is used, which normalizes the weights for all non-NA scores. * feat: Added new hyperparameters `weights_learner` and `weights_measure` to `PipeOpClassWeights` to allow specification which type of weight column to add to the `Task`. * New PipeOp `PipeOpClassWeightEx` extends the functionality of `PipeOpClassWeights` to also support multiclass classification tasks as well as several methods of automatically determining weights based on the target class of a sample. * New PipeOp `PipeOpSplines` that expands numeric features into spline basis columns.

* Pretty-printing some info using the `cli` package now. * New PipeOp `PipeOpInfo` prints or logs info about objects passing through. * New Pipeop `PipeOpIsomap` implements isomap embedding from `dimRed::embed` * feat: allow dates in datefeatures pipe op and use data.table for date feature generation. * feat: `PipeOpLearnerCV` can reuse the cross-validation models during prediction by averaging their outputs (`resampling.predict_method = "cv_ensemble"`). * feat: `PipeOpRegrAvg` gets new `se_aggr`, `se_aggr_rho`, `prob_aggr`, and `prob_aggr_eps` hyperparameters and now allows different forms of prob / SE aggregation. * feat: `FilterEnsemble` implements Binder et al. (2020) *Multi-Objective Hyperparameter Tuning and Feature Selection using Filter Ensembles* * Fix: `PipeOpRemoveConstants` now avoids integer overflow when evaluating relative tolerances for near-`integer.max` data. * Fix: Added support for internal validation tasks to `PipeOpFeatureUnion`. * Fix: Added internal workaround for `PipeOpNMF` attaching `Biobase`, `BiocGenerics`, and `generics` to the search path during training, prediction or when printing its `$state`. * Compatibility with new testthat version 3.3.0

* Breaking change: Removed initialization of `PipeOpImputeConstant`'s `constant` hyperparameter since it was incompatible with other defaults and would lead to not recommended usage (creating an empty level). * Removed compatibility for old `paradox` versions pre-1.0.0. * Added `empty_level_control` argument to `PipeOpImpute` allowing control over edge cases for `factor`/`ordered` columns. * Set new construction argument `empty_level_control` to `"param"` for `PipeOpImputeOOR` and to `"always"` for `PipeOpImputeConstant`. * Untrained `PipeOp`s that take `NULL` as input during training now automatically perform training during prediction. * `PipeOpImputeConstant`, `PipeOpImputeMode`, `PipeOpImputeOOR`, and `PipeOpImputeLearner` can now handle `factor` or `ordered` features with zero levels. * `PipeOpImputeConstant` now gives a more informative error message if `check_levels` is `TRUE` and a new level would be created through imputation. * Fix: `PipeOpImputeOOR` now imputes `".MISSING"` for `factor`/`ordered` features with only `NA`s instead of sampling from the feature's levels. * Fix: `PipeOpImputeLearner` no longer adds `"factor"` or `"ordered"` levels for these feature types arbitrarily and instead updates levels correctly in certain edge-cases. * Fixed the error message for unexpected Multiplicities in the input and output type checking during `PipeOp`s training and prediction. * Fixed a grammatical error in `PipeOp`'s error message wrapper: now correctly says "This happened *in* ...".

* Added missing error for predicting with untrained `PipeOp`s / `Graph`s. * Fix: Corrected typo in the hyperparameter name `use_parallel` of `PipeOpVtreat`. * Fix: Do not overwrite initial hyperparameter settings of `bbotk::OptimizerBatchNLoptr` in `LearnerClassifAvg` / `LearnerRegrAvg`'s internal `optimize_weights_learneravg` function. * Added new convenience function `preproc()` for easier training of or prediction with `PipeOp`s or `Graph`s. * Fix: `PipeOpVtreat`, `PipeOpEncodeImpact`, and `PipeOpEncodeLmer` now accept the more precise `TaskSupervised` instead of `Task` as input for training and prediction. * Docs: Added missing documentation for the `task_type` of the input and output channels of `PipeOp`s that inherit from `PipeOpTaskPreproc` and set a non-default `task_type`. * Fix: `PipeOpEncodeLmer`, `PipeOpADAS`, `PipeOpBLSmote`, `PipeOpSmote`, and `PipeOpSmoteNC` no longer throw an error in case of empty target levels during training. * Fix: `PipeOpClassBalancing` now handles unseen target levels by ignoring them during upsampling instead of producing `NA`s.

* New parameter `no_collapse_above_absolute` for `PipeOpCollapseFactors` / `po("collapse_factors")`. * Fix: `PipeOpCollapseFactors` now correctly collapses levels of ordered factors. * Fix: `LearnerClassifAvg` and `LearnerRegrAvg` hyperparameters get the `"required"` tag. * New parameter `use_groups` (default `TRUE`) for `PipeOpSubsampling` to respect grouping (changed default behaviour for grouped data) * New parameter `new_role_direct` for `PipeOpColRoles` / `po("colroles")` to change column roles by role instead of by column. * Dictionary sugar functions `po()` / `pos()` / `ppl()` / `ppls()` now make suggestions for entries in both `mlr_pipeops` as well as `mlr_graphs` when an object by the given name could not be found in the respective dictionary. * 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`. * New PipeOps `PipeOpEncodePLQuantiles` and `PipeOpEncodePLTree` that implement piecewise linear encoding with two different binning methods. * Compatibility with new `R6` release. * Docs: Performed cleanup and standardization. * Docs: Performed cleanup of reference index page on website. * Docs: Fixed parsing of examples on website for `PipeOpNMF` and `PipeOpLearnerPICVPlus`. * Fix: `PipeOpTargetMutate` and `PipeOpTargetTrafoScaleRange` no longer drop unseen factor levels of features or targets during train and predict. * Simplified parameter checks and added internal type checking for `PipeOpTargetMutate`.

* Compatibility fix for upcoming `mlr3` * New down-sampling PipeOps for inbalanced data: `PipeOpTomek` / `po("tomek")` and `PipeOpNearmiss` / `po("nearmiss")` * New PipeOp `PipeOpLearnerPICVPlus / po("learner_pi_cvplus")` * New PipeOp for Quantile Regression `PipeOpLearnerQuantiles` / `po(learner_quantiles)` * `GraphLearner` has new active bindings/methods as shortcuts for active bindings/methods of the underlying `Graph`: `$pipeops`, `$edges`, `$pipeops_param_set`, and `$pipeops_param_set_values` as well as `$ids()` and `$plot()`.

* New PipeOp `PipeOpRowApply` / `po("rowapply")` * Empty `PipeOp` IDs now explicitly forbidden. * Bugfix: `Graph$tran()` / `Graph$predict()` with `single_input = FALSE` now correctly handles `PipeOp`s with multiple inputs. * `GraphLearner$base_learner()` now works with `PipeOpBranch`, and is generally more robust. * `GraphLearner` now supports `$importance`, `$selected_features()`, `$oob_error()`, and `$loglik()`. These are computed from the underlying `Learner`. * `GraphLearner$impute_selected_features` option added: `$selected_features()` is reported even if the underlying base learner does not report it; in this case, the full feature set as seen by that learner is returned. * `GraphLearner$predict_type` handling more robust now. * `PipeOpThreshold` and `PipeOpTuneThreshold` now have the `$predict_type` `"prob"`. They can be set to `"response"`, in which case the probability predictions are discarded, potentially saving memory. * Bugfix for handling multiplicities in PipeOps with vararg channels. * Bugfix: `PipeOpImputeOOR` now retains the `.MISSING` level in factors during prediction that were imputed during training, but had no missing values during prediction. * `as_data_table(po())` now works even when some `PipeOp`s can not be constructed. For these `PipeOp`s, `NA` is reported in most columns. * Compatibility with upcoming `mlr3` release. * New PipeOps for handling inbalanced data: `PipeOpADAS` / `po("adas")`, `PipeOpBLSmote` / `po("blsmote")` and `PipeOpSmoteNC` / `po("smotenc")`

* Compatibility with new `bbotk` release. * Added marshaling support to `GraphLearner` * Support internal tuning and validation