Documentation/Modules/OptAlgCMA

Summary

This optimization module is an implementation of the Evolution Strategy with Covariance Matrix Adaptation (CMA-ES).

Among the tree evolutionary operators (Recombination, Mutation, Selection), the mutation is considered the most important. Mutation is performed by sampling a Covariance Matrix. The parameters of the matrix (variances and correlation coefficients) are adapted by tracking the path of successful mutations.

Properties

General

Algorithm	stochastic - generates new solutions by sampling a probability function, however deterministic adaptation of the covariance matrix.
Design Variables	Written for continuous variables. No discrete or mixed variables are possible.
Objectives	single-objective for minimization.
Constraint handling	no
Boundary handling	yes
Initialization	-

Connections

Starting at this module	Module requires exactly one connection of type optimization.
Ending at this module	-

Actions

Name	Description
Run	starts the optimization.

Options

The options are currently only described in the references (see below).

Module Description

Initialization

The algorithm distinguishes several cases, depending on the problem properties (initial solution, initial search region, bounds).

Optimization

The algorithm contains stochastic processes and operates with a population. Parallelization on the basis of the population size is implemented.

Usage

... todo

Source Code

ToDo:Link to SVN

References

Detailed information is given at http://www.lri.fr/~hansen/ .