Abstract MAPCMASampler provides an implementation of the MAP-IGO (maximum a posteriori information geometric optimization) framework, which extends the CMA-ES rank-one-update. This sampler adds momentum-based updates to the standard CMA-ES, following the MAP-IGO algorithm.
Class or Function Names MAPCMASampler(mean: dict[str, Any] | None = None, sigma0: float | None = None, seed: int | None = None, popsize: int | None = None, cov: np.ndarray | None = None, momentum_r: float | None = None, search_space: dict[str, BaseDistribution] | None = None, independent_sampler: BaseSampler | None = None) mean: Initial mean of MAPCMA.
Abstract SafeCMASampler provides an implementation of SafeCMA, a variant of CMA-ES that incorporates safety constraints. This sampler extends the standard CMA-ES algorithm to handle constrained optimization problems where certain regions of the search space should be avoided. SafeCMA uses Gaussian Process models to estimate Lipschitz constants and manage trust regions, ensuring that the optimization process respects safety constraints while exploring the search space efficiently. Please refer to the original paper for more details.