Confidence Region Bayesian Optimization

This is the companion code for the benchmarking study reported in the paper Cautious Bayesian Optimization for Scalable and Efficient Policy Search by Lukas Fröhlich et al., L4DC 2021. The paper can be found here. The code allows the users to reproduce and extend the results reported in the study. Please cite the above paper when reporting, reproducing or extending the results.

Purpose of the project

This software is a research prototype, solely developed for and published as part of the publication cited above. It will neither be maintained nor monitored in any way.

Installation guide

In the root directory of the repository execute the following commands:

conda env create -f environment.yml
conda activate crbo
pip install -e .

Basic functionality

Note: This implementation assumes that you want to minimize a function.

We have prepared a jupyter notebook to run CRBO on the CartPole task. Before we start, make sure to download the pre-trained RL agents. For your convenience, just execute the following commands:

# makes the script executable
chmod +x download_agents.sh
# downloads the files (only a few kB) and stores to disk
./download_agents.sh

Now, to open the notebook execute the following commands:

jupyter notebook run_experiment.ipynb

If the browser window does not open automatically, copy the generated link and paste it manually into your browser.

License

Confidence Region Bayesian Optimization is open-sourced under the AGPL-3.0 license. See the LICENSE file for details.