Official implementation of NeurIPS'24 paper "Monte Carlo Tree Search based Space Transfer for Black Box Optimization".
This repository contains the Python code for MCTS-Transfer , an search space transfer algorithm for expensive Black-Box Optimization. The code is implemented based on LA-MCTS. Data generation code is based on RIBBO.
Ubuntu == 20.04
Python == 3.8.0
pip install -r requirements.txt
Download data from the link to the directory /data.
Pretrained MCTS models can be downloaded from link or generated automatically.
Download HPO-B surrogates from the link to directory /functions/hpob/saved-surrogates.
# test on Sphere2D
bash experiments/run_sphere.sh
# test on BBOB
bash experiments/run_bbob.sh
# test on real-world problem
bash experiments/run_real.sh
# test on design-bench
bash experiments/run_design_bench.sh
# test on hpob
bash experiments/run_hpob.sh
@inproceedings{mcts-transfer,
author = {Shu-kuan Wang , Ke Xue, Song Lei, Xiao-bin Huang, Chao Qian},
title = {Monte Carlo Tree Search based Space Transfer for Black Box Optimization},
booktitle = {Advances in Neural Information Processing Systems 38 (NeurIPS’24)},
year = {2024},
address={Vancouver, Canada}
}