Skip to content

Latest commit

 

History

History
44 lines (32 loc) · 3.8 KB

README.md

File metadata and controls

44 lines (32 loc) · 3.8 KB
Raw

Gitter

Summary

This repository contains the configurable CMA-ES experiments code that uses the Modular EA framework. The framework is available separately as ModEA.

Most experiments are programmed in the main.py file, while the configurable CMA-ES setup is defined in EvolvingES.py.

Example Usage

main.py

This file is setup mostly to run either an optimizer (currently a MIES, but still labeled as 'GA') or a brute-force enumeration run over all available structure configurations.

The current behavior of executing the file depends on the number of arguments given:

  • 0: executes the 'default' option, currently set to a 5D brute-force enumeration over the Sphere function F1.
  • 2 (ndim fid): executes the 'optimization' option (currently MIES) on the given BBOB function fid in ndim dimensions
  • 3 (ndim fid run): executes the 'optimization' option (currently MIES) on the given BBOB function fid in ndim dimensions, but with an added run parameter that is used to log multiple runs without overwriting.
  • 4 (ndim fid parallel part): executes the 'brute-force enumeration' option on the given BBOB function fid in ndim dimensions, while explicitly running parallel configurations in parallel. Also, to prevent some memory issues, part specifies which part out of 4 of the entire list of possible configurations should executed. An external script (e.g. the included BF_experiment.sh file) should be used to iterate over parts 1...4.

EvolvingES.py

The most important function in EvolvingES.py isrunCustomizedES(representation, iid, rep, ndim, fid, budget). This function takes a representation from {0,1}^9 X {0,1,2}^2 (optionally extended with a set of parameter values), combined with some BBOB parameters: instance ID iid, repetition number rep, dimensionality ndim, function ID fid, and the total available budget in evaluations budget. This function takes the representation, splits it into , lambda, mu, and passes them on accordingly to the customizedES in ModEA.

To simplify large runs of experiments, the evaluateCustomizedESs(...): function also exists, which tries to automatically parallelize the running of multiple representations over multiple instance ID's of a fixed fid and ndim.

Citation

To cite this work, please use the following reference:

@INPROCEEDINGS{vanrijn2016,
    author={S. {van Rijn} and H. {Wang} and M. {van Leeuwen} and T. {Bäck}},
    booktitle={2016 IEEE Symposium Series on Computational Intelligence (SSCI)}, 
    title={Evolving the structure of Evolution Strategies}, 
    year={2016},
    doi={10.1109/SSCI.2016.7850138},
}

References