BARK

This repository contains the code for the experiments for the paper BARK: A Fully Bayesian Tree Kernel for Black-box Optimization.

Installation

To set up a Python environment, we recommend the simplest approach below (for Windows, different platforms may vary). After cloning this repository:

python -m venv barkvenv
barkvenv\Scripts\activate
python -m pip install .

You can also use uv, poetry, or conda if you'd prefer.

Since we use Gurobi to optimize the acquisition function, you will need a license (you can get a free academic license, if applicable).

Running experiments

To run the experiments, you will use the configuration files in configs/, and the scripts in examples/. For example, to reproduce the Bayesian optimization for the BARK model on the TreeFunction benchmark:

python examples/bayes_opt/bark_study_strategy.py
    -s 42 # random seed for initial data
    -c configs/benchmark_configs/treefunction_config.yaml
    -m configs/model_configs/bark_config.yaml
    -o results/