Extensive documentation is available here
This repository is based on the paper EXPObench: Benchmarking Surrogate-based Optimisation Algorithms on Expensive Black-box Functions. The purpose of this repository is to benchmark different surrogate algorithms on expensive real life optimization problems. It contains different problems, where the goal is to minimize some objective function, and approaches, which solve the minimization problem through the use of surrogate models. The documentation above contains a list of the problems and approaches, as well as instructions for how to add new problems or approaches.
If you make use of EXPObench in your scientific work, please cite us:
@article{EXPObench,
title = {Benchmarking surrogate-based optimisation algorithms on expensive black-box functions},
journal = {Applied Soft Computing},
year = {2023},
issn = {1568-4946},
doi = {https://doi.org/10.1016/j.asoc.2023.110744},
author = {Laurens Bliek and Arthur Guijt and Rickard Karlsson and Sicco Verwer and Mathijs {de Weerdt}}
}This repository requires a working installation of python 3. A quick test to see if everything works is to run the following code. This runs random search (not a surrogate algorithm) on the rosenbrock problem (which is not an expensive optimization problem):
python expensiveoptimbenchmark/run_experiment.py --max-eval=100 rosenbrock --n-cont=10 randomsearch
The required packages for this minimal example are:
- numpy
- scipy
- hyperopt
All these packages can be installed using pip install <package>. Other problems and approaches in this repository may require additional packages.
The results of the minimal working example above will be put in the results folder. For more information, please refer to the documentation.
Some of the problems, namely ESP and Pitzdaily, require the use of a Docker container. Information on how to run these problems can be found in the documentation.
If the use of Docker or Singularity is not desired, we recommend the use of Poetry to manage the packages that are required for the different problems and approaches.
- A virtualenv can be created through the use of Poetry, which will automatically incorporate the neccesary dependencies:
poetry install - Once finished, you can open a shell in the virtualenv by running
or run a singular command using
poetry shellpoetry run python ...
Besides the regular ways of contact on github, if you have any questions please contact l.bliek@tue.nl.