Toward Efficient Mixed-Integer Black-Box Optimization via Evolution Strategies with Plateau Handling Techniques

This repository contains the official code for our paper, "Toward Efficient Mixed-Integer Black-Box Optimization via Evolution Strategies with Plateau Handling Techniques" accepted at GECCO 2025.

Setup

1. Clone the repository and navigate to the root directory.

   git clone https://github.com/nAuTahn/eMI-BBO
   cd eMI-BBO

2. Make sure to install the required libraries. Besides NumPy, you also need SciPy to perform some calculations.

Performance evaluation on standard MIP benchmark functions.

You can reproduce the results in our paper by running the following scripts. Hyperparameters can be adjusted based on the papers or customized by the user.

python test.py --func ellipsoid_int --dim 120  --dim_co 80 --max_evals 100000 --target 1e-10 --sigma_VD 0.5 --sigma_NES 0.5 --step_size_control "TPA"

Another argument --list_funcs is added to support displaying the names of some available benchmark functions. Additionally, users can add other objective functions in test.py for evaluation.

Citation

If you use our source code, please cite our work as:

@inproceedings{AnhGecco2025,
  author       = {Tuan Anh Nguyen and Ngoc Hoang Luong},
  title        = {{Toward Efficient Mixed-Integer Black-Box Optimization via Evolution Strategies with Plateau Handling Techniques}},
  booktitle    = {GECCO '25: Proceedings of the Genetic and Evolutionary Computation Conference},
  address      = {Málaga, Spain},
  publisher    = {{ACM}},
  year         = {2025}
}

Acknowledgement

Our source code is inspired by:

• Comparison-Based Natural Gradient Optimization in High Dimension
• Fast Moving Natural Evolution Strategy for High-Dimensional Problems
• Natural Evolution Strategy for Mixed-Integer Black-Box Optimization