CMA-ES

Evolutionary.CMAESType

Covariance Matrix Adaptation Evolution Strategy Implementation: (μ/μ_{I,W},λ)-CMA-ES

The constructor takes following keyword arguments:

  • μ/mu is the number of parents
  • λ/lambda is the number of offspring
  • c_1 is a learning rate for the rank-one update of the covariance matrix update
  • c_c is a learning rate for cumulation for the rank-one update of the covariance matrix
  • c_mu is a learning rate for the rank-$\mu$ update of the covariance matrix update
  • c_sigma is a learning rate for the cumulation for the step-size control
  • c_m is the learning rate for the mean update, $c_m \leq 1$
  • σ0/sigma0 is the initial step size σ
  • weights are recombination weights, if the weights are set to $1/\mu$ then the intermediate recombination is activated.
  • metrics is a collection of convergence metrics.
source

Description

The Covariance Matrix Adaptation Evolution Strategy (CMA-ES) is a stochastic derivative-free numerical optimization algorithm for difficult (non-convex, ill-conditioned, multi-modal, rugged, noisy) optimization problems in continuous search spaces [1].

The current CMA-ES algorithm implementation based on a simplified outline[2].

References