Algorithm Configuration Data Mining for CMA Evolution Strategies

This repository contains the data and analysis for the paper Algorithm Configuration Data Mining for CMA Evolution Strategies.

Introduction

The Python Notebook contains all code that was used to analyse the accompanying pre-processed dataset in the /arff folder.

The original dataset was created using this repository, which builds on top of the ModEA framework, as published in 'Evolving the Structure of Evolution Strategies'

Dependencies

This project was written for Python 2.7. All required packages from PyPI are specified in the requirements.txt.

NOTE: This list of packages includes the pydot package, for which Graphviz has to be installed on your operating system. This package is optional to complete the analysis, but is included to visualize the Decision Trees that are created.

Use

The easiest way to re-run the notebook is to use the Binder link which will load an online environment with all dependencies pre-installed.

Alternatively, you can clone the repository, install the requirements.txt and run the notebook locally:

``` git clone https://github.com/sjvrijn/cma-es-configuration-data-mining.git cmaes-config-dm cd cmaes-config-dm pip install --user -r requirements.txt

pip install --user jupyter notebook # <-- only required if you don't yet have jupyter installed

jupyter notebook ```

Contact

If there are any questions or issues, please come to the Gitter chatroom or open an issue here on Github.

Citation

To cite this repository, you can use the following bibtex entry:

@misc{vanRijn2017-github, author = {van Rijn, Sander}, title = {Github repository: CMA-ES Configuration Data Mining}, year = {2017}, url = {https://github.com/sjvrijn/cma-es-configuration-data-mining}, }

To cite the corresponding paper, you can use the following bibtex entry instead:

@inproceedings{vanRijn:2017:ACD:3071178.3071205, author = {van Rijn, Sander and Wang, Hao and van Stein, Bas and B\"{a}ck, Thomas}, title = {Algorithm Configuration Data Mining for CMA Evolution Strategies}, booktitle = {Proceedings of the Genetic and Evolutionary Computation Conference}, series = {GECCO '17}, year = {2017}, isbn = {978-1-4503-4920-8}, location = {Berlin, Germany}, pages = {737--744}, numpages = {8}, url = {http://doi.acm.org/10.1145/3071178.3071205}, doi = {10.1145/3071178.3071205}, publisher = {ACM}, keywords = {empirical study, evolution strategies, metaheuristics, parameter tuning, performance measures}, }

References

• Evolving the Structure of Evolution Strategies. Sander van Rijn, Hao Wang, Matthijs van Leeuwen, Thomas Bck. IEEE SSCI December 6-9 2016, Athens, Greece.
  • Framework available as the ModEA Python package,
  • experiments and analysis code available in a separate repository
• Algorithm Configuration Data Mining for CMA Evolution Strategies. Sander van Rijn, Hao Wang, Bas van Stein, Thomas Bck. GECCO July 15-19 2017, Berlin, Germany.
• Towards an Adaptive CMA-ES Configurator. Sander van Rijn, Carola Doerr, Thomas Bck. PPSN XV September 8-12 2018, Coimbra, Portugal.