MALIBO: Meta-learning for Likelihood-free Bayesian Optimization

This repository is an original implementation for our ICML 2024 paper: MALIBO: Meta-learning for Likelihood-free Bayesian Optimization by Jiarong Pan, Stefan Falkner, Felix Berkenkamp and Joaquin Vanschoren. The code allows the users to use our implementation of MALIBO, and to reproduce the results for Forrester function and HPO-B benchmarks in the paper.

Purpose of the project

This software is a research prototype, solely developed for and published as part of the publication. It will neither be maintained nor monitored in any way.

Setup

conda create env -f environment.yml
conda activate malibo

Run

To run the optimization for Forrester function:

python run_forrester.py

To run HPO-B benchmark:

cd benchmarks
git clone https://github.com/machinelearningnuremberg/HPO-B.git

You need to download the HPO-B benchmarks data and put it into benchmarks/HPO-B/hpob-data, after that you can run:

python run_hpob.py --search_space_id 4796 --test_seed test0 --no-continuous --evaluations 100 --output results/hpob

You can visualize the generated results using the plotting functions in HPO-B.

python benchmarks/generate_json_hpob.py
python benchmarks/benchmark_plot_hpob.py

Cite

If you find this code useful in your research, please cite the paper:

@inproceedings{pan2024malibo,
  title = {{MALIBO}: Meta-learning for Likelihood-free {B}ayesian Optimization},
  author = {Pan, Jiarong and Falkner, Stefan and Berkenkamp, Felix and Vanschoren, Joaquin},
  booktitle = {Proceedings of the 41st International Conference on Machine Learning},
  pages = {39102--39134},
  year = {2024},
}

License

MALIBO is open-sourced under the AGPL-3.0 license. See the LICENSE file for details.

For a list of other open source components included in MALIBO, see the file 3rd-party-licenses.txt.