Max-value Entropy Search for Multi-Objective Bayesian Optimization

This repository contains the python implementation for MESMO from the Neurips 2019 paper "Max-value Entropy Search for Multi-ObjectiveBayesian Optimization".

---

Requirements

This code is implemented in Python and requires the following dependencies:

• sobol_seq – for generating Sobol sequences
• platypus – for multi-objective evolutionary algorithms
• scikit-learn – specifically sklearn.gaussian_process for GP modeling
• pygmo – for parallel optimization algorithms

You can install the required packages using:

pip install sobol_seq platypus-opt scikit-learn pygmo

---

Running MESMO

python main.py <function_names> <d> <seed> <initial_number> <total_iterations> <sample_number>

Here's an example command you could run from bash:

python main.py branin,Currin 2 0 5 100 10

Explanation of arguments:

1. function_names: names of the benchmark functions separated by a comma
2. d: number of input dimensions
3. seed: random seed
4. initial_number: number of initial of evaluations
5. total_iterations: number of BO iterations
6. sample_number: number of samples to use for entropy estimation

---

Citation

If you use this code please cite our papers:

@inproceedings{belakaria2019max,
  title={Max-value entropy search for multi-objective bayesian optimization},
  author={Belakaria, Syrine and Deshwal, Aryan},
  booktitle={International Conference on Neural Information Processing Systems (NeurIPS)},
  year={2019}
}

@article{belakaria2021output,
  title={Output Space Entropy Search Framework for Multi-Objective Bayesian Optimization},
  author={Belakaria, Syrine and Deshwal, Aryan and Doppa, Janardhan Rao},
  journal={Journal of Artificial Intelligence Research},
  volume={72},
  pages={667-715},
  year={2021}
}