NEWS

mlr3mbo 1.1.1 (2026-04-24)

fix: acqo() now correctly returns the mlr_acqoptimizers dictionary when called with no arguments (#211).
fix: AcqFunctionEILog now provides a more informative error message when the surrogate is not configured with the correct output transformation.
fix: AcqOptimizerDirect and AcqOptimizerLbfgsb now correctly enforce the max_restarts limit in all cases.
fix: SurrogateLearner and SurrogateLearnerCollection now correctly apply their output transformation after imputing running evaluations.

feat: Added mlr_acqoptimizers dictionary with pre-defined acquisition function optimizers (AcqOptimizerDirect, AcqOptimizerLbfgsb, AcqOptimizerLocalSearch, AcqOptimizerRandomSearch).
perf: Default surrogate model, acquisition function, optimizer, and further settings of OptimizerMbo are now empirically derived from a large-scale benchmark study, significantly improving out-of-the-box optimization performance.
feat: Added Mlr3ErrorMbo* condition classes.

fix: logger changes related to bbotk.
fix: assure that candidates after acquisition function optimization are always within bounds.
perf: minor changes to speed up predictions with SurrogateLearner and SurrogateLearnerCollection.
feat: added supported for input and output transformations (see InputTrafo, OutputTrafo and the related classes).
refactor: dropped functionality to assert insample performance of the surrogate model completely.

chore: silence rush logger and fixed some partial matches, depend on mlr3 >= 0.22.1.
test: fix test_AcqFunctionMulti, robustify helper and loading.
test: fix test_ResultAssignerArchive and test_ResultAssignerSurrogate due to upcoming changes of internal tuned values in mlr3tuning 1.3.1.

docs: gracefully exit examples of OptimizerAsyncMbo, OptimizerADBO, TunerAsyncMbo, and TunerADBO if Redis is not available.
test: skip tests involving asynchronous logic if Redis is not available.

refactor: refactored SurrogateLearner and SurrogateLearnerCollection to allow updating on an asynchronous Archive.
feat: added experimental OptimizerAsyncMbo, OptimizerADBO, TunerAsyncMbo, and TunerADBO that allow for asynchronous optimization.
feat: added AcqFunctionStochasticCB and AcqFunctionStochasticEI that are useful for asynchronous optimization.
docs: minor changes to highlight differences between batch and asynchronous objects related to asynchronous support.
refactor: AcqFunctions and AcqOptimizer gained a reset() method.

docs: move vignette to mlr3book.
feat: add AcqFunctionMulti that can wrap multiple acquisition functions resulting in a multi-objective acquisition function problem.
feat: support callbacks in AcqOptimizer.
feat: allow AcqFunctionEI to be adjusted by epsilon to strengthen exploration.

refactor: compatibility with upcoming paradox upgrade.
feat: OptimizerMbo and TunerMbo now update the Surrogate a final time after the optimization process finished to ensure that the Surrogate correctly reflects the state of being trained on all data seen during optimization.
fix: AcqFunction domain construction now respects Surrogate cols_x field.
feat: support more than one candidate point as a result of acquisition function optimization even for non-batch acquisition functions.
feat: added default_gp and default_rf helpers that allow for construction of a default Gaussian Process and random forest as for example used within default_surrogate.
refactor: changed Gaussian Process and random forest defaults (in default_gp and default_rf and therefore also in default_surrogate). Gaussian Process now uses a "matern5_2" kernel. Random forest now uses 100 trees. The number of trees used in the fallback random forest was reduced to 10.

docs: updated some references in vignette.
refactor: minor clean up of the internal structure of all loop functions.
perf: default initial design constructed based on a Sobol sequence in all loop functions.
refactor: no longer depend on mlr3tuning but import instead.
refactor: srlrn sugar function now can construct both a SurrogateLearner and SurrogateLearnerCollection; dropped srlrnc.
feat: added AcqFunctionSD, AcqFunctionEHVI and AcqFunctionEHVIGH, introduced bayesopt_emo loop function.
feat: AcqFunctions now include a $packages field stating required packages which are checked for whether their namespace can be loaded prior to optimization.
fix: fixed bug in fix_xdt_missing() helper function.
BREAKING CHANGE: renaming default_loopfun -> default_loop_function, default_acqfun -> default_acqfunction, default_acqopt -> default_acqoptimizer.
BREAKING CHANGE: result_functions now replaced by ResultAssigners.
BREAKING CHANGE: renamed $model field of all Surrogate classes to $learner.
BREAKING CHANGE: For all Surrogate and AcquisitionFunction classes fields *_cols renamed to cols_* (e.g., x_cols to cols_x).

refactor: adapt to mlr3tuning 0.18.0.
feat: acquisition functions now assert whether surrogates match their required predict type.
fix: unloading mlr3mbo removes optimizers and tuners from the dictionaries.
docs: faster examples.
feat: characters in surrogate regression tasks are no longer automatically converted to factors. default_surrogate now respects this and gained an appropriate pipeline step.
feat: AcqFunctionAEI added.
docs: fix of docs, README and bibentries.