AdaScale-TuRBO

This is the repo containing codes for the Rethinking Trust Region Bayesian Optimization in High Dimensions paper. This work has been accepted to the 2026 AISTATS OPTIMAL Workshop.

Installation

pip install -r requirements.txt

Running Experiments

Experiments can be run using the main.py script. You must specify a benchmark to run the algorithms.

Basic Command

python main.py benchmark=<benchmark_name>

• To see a list of available benchmarks, run python main_NeSTBO.py.
• Adding seed=<number> is recommended for reproducibility.

Configuration Overrides

All default settings are stored in configs/default.yaml. Since this project uses Hydra, you have the flexibility to modify these values on the fly via the command line without editing the file.

# Example: override the evaluation budget for the rastrigin benchmark
python main.py benchmark=rastrigin seed=0 benchmark.n_tot=1000

Citation

If you use this code in your research, please cite the following paper:

@article{tang2026rethinking,
  title={Rethinking Trust Region Bayesian Optimization in High Dimensions},
  author={Tang, Wei-Ting and Paulson, Joel A},
  journal={arXiv preprint arXiv:2604.22967},
  year={2026}
}