Changelog

Version 1.9.1 - 2026-08-24

Added

  • Implemented Gradient Descent.

Version 1.9.0 - 2026-08-03

Added

  • Implemented x_tol, f_tol, and maxiters stopping control.

Version 1.8.0 - 2026-07-29

Changed

  • Reworked MinionResult to expose an enum-based termination status with readable string output in both C++ and Python.

  • Improved convergence and termination reporting so algorithms distinguish convergence, evaluation-budget exhaustion, callback stops, stagnation, numerical errors, and runtime errors more consistently.

  • Updated callback handling so returning true / True stops optimization and reports callback_stopped.

  • Updated examples, documentation, notebooks, and tests to use result.status directly.

Fixed

  • Fixed algorithms that could previously report convergence when they actually exhausted the evaluation budget.

  • Fixed callback stop propagation in several algorithm families, including DE variants, CMA-ES variants, Dual Annealing, and L-BFGS methods.

Version 1.7.0 - 2026-07-07

Added

  • Added benchmark functionality for CEC and BBOB2009 suites.

  • Added BBOB2009 benchmark support to the Python and C++ APIs.

Fixed

  • Fixed memory leaks in the CEC benchmark implementations.

Version 1.6.1 - 2026-06-13

Changed

  • Removed tol / relTol from the Minion API.

  • Made algorithm name handling more robust across common naming variants.

  • Updated CEC2011 to more closely mirror the original MATLAB implementation.

  • Streamlined the CEC API.

Version 1.5.0 - 2026-03-07

Changed

  • Reverted back to compiled libarary style.

  • Fix compilation issue on Windows MSVC due to Minmax

Version 1.2.0 - 2026-03-05

Added

  • Added RCMAES

  • Streamlined CMake usage.

  • Header only style library

Version 1.1.0 - 2025-11-23

Added

  • Added IMODE and AGSK

  • CEC2011 is now completely rewritten in C++. No MATLAB!

Version 1.0.1 - 2025-11-07

Changed

  • Fixed memory leaks.

  • Version 1.0.1 now instead of 0.2.9

Version 0.2.8 - 2025-11-06

Changed

  • More correct implementations of LSHADE_cnepsin and ARRDE.

Version 0.2.7 - 2025-10-10

Changed

  • More correct implementations of LSHADE, jSO, JADE, DE, and ARRDE.

Version 0.2.6 - 2025-10-06

Added

  • Implemented BIPOP_aCMAES

Version 0.2.5 - 2025-10-05

Added

  • Implemented PSO, SPSO-2011, DMSPSO, LSHADE_cnepsin, and CMAES

Version 0.2.4 - 2025-10-02

Changed

  • GIL handling during python calls.

  • An improved mutation implementation.

Version 0.2.3 - 2025-08-21

Changed

  • Classical algorithms such as L-BFGS pick the best guess from the given guesses during initialization.

  • Stability improvements in L-BFGS and L-BFGS-B

Version 0.2.2 - 2025-07-31

Changed

  • Initial guess x0 is now a list of guesses. Thus, it is possible now to have more than one initial guesses.

Version 0.2.1 - 2025-03-01

Added

  • Fixed “none” unrecognized as bound_strategy

Version 0.2.0 - 2025-02-24

Added

  • Added “minimum_population_size” option for ARRDE

Version 0.1.9 - 2025-02-19

Fixed

  • Fixed L-BFGS-B violate bounds when calculating derivatives.

Version 0.1.8 - 2025-02-19

Added

  • Implemented L-BFGS

  • L-BFGS and L-BFGS-B use noise-robust Lanczos derivative

Changed

  • Step size during derivative calculation in L-BFGS-B and L-BFGS is now adaptive.

Fixed

  • Fixed some typos in the docs.

Version 0.1.7 - 2025-02-13

Added

  • Implemented Process_Parallel and Thread_Parallel for multiprocessing and multithreading support.

  • Implemented MinimizerBase::getBestFromHistory.

Changed

  • MinimizerBase::getBestFromHistory is used for most algorithms now.

Fixed

  • Resolved wrong function value when calling optimize function of LSRTDE and NLDSHADE_RSP.

Version 0.1.6 - 2025-02-11

Added

  • Implemented L_BFGS_B

  • Improved DA with L_BFGS_B as the local search.

Changed

  • Fine-tuned DA hyperparameters.

Fixed

  • Fixed problem with GitHub CI for wheel generation.