Changelog

Version 2.0.1

Bugfix

• Fix ACO save_mode = "best" not reliably saving the best model. In parallel mode, the "improved" flag from workers could become stale, causing a worse model to overwrite the saved best. The flag is now recomputed on the main process. Additionally, ACO now re-runs and saves the best model at the end of the search, matching the behavior of exhaustive search.

Version 2.0.0

New Features

Custom Model Libraries

• Custom model support (library = "custom"): use your own model library instead of the built-in PK library. Provide a model_creation_func(library, filters) that returns a model file path or a Monolix lib: reference. Compatible with "exhaustive_search" and "ACO" algorithms.

• Monolix library integration: model_creation_func can return lib: references from getLibraryModelName(), enabling search over any Monolix library (PKPD, PD, TMDD, TGI, etc.) without writing model files.

• param_mapping argument: defines which parameters are available for IIV in each model variant via common_params and filter_params.

IIV and Parameter Control

• fixed_iiv argument: lock IIV on (TRUE) or off (FALSE) for specific parameters, reducing the search space. Works with all algorithms and with both iiv = TRUE and iiv = FALSE.

• param_distributions argument: override the default log-normal distribution for specific parameters (supports "normal", "logNormal", "logitNormal" with optional limits).

Parallel Execution

• Parallel model evaluation for ACO and exhaustive search using future / future.apply. Workers run models concurrently across local cores or HPC clusters.

Other

• Specify table_func to output additional information about the model.

Search and Algorithm Improvements

• settings$tasks: configure which Monolix scenario tasks to run (e.g. disable standardErrorEstimation for faster exploration).

• settings$table_func: custom function to output additional information about each run.