TR-BEACON: Shedding Light on Efficient Behavior Discovery in High-Dimensional Spaces with Bayesian Novelty Search over Trust Regions

This is the repo containing codes and results for TR-BEACON: Shedding Light on Efficient Behavior Discovery in High-Dimensional Spaces with Bayesian Novelty Search over Trust Regions paper published at the NeurIPS 2024 Workshop on Bayesian Decision-making and Uncertainty.

TR-BEACON is the extension of the Bayesian optimization inspired novelty search algorithm BEACON. Please refer to the original BEACON paper and github repo for more detail.

Installation

pip install -r requirements.txt

Usage

We provide the code scripts for executing TR-BEACON, BEACON, and other SOTA novelty search algorithm on a 20-d Ackley function. Noted that TR-BEACON and BEACON requires the usage of ThompsonSampling.py file to perform efficient Thompson sampling strategy proposed in this work.

Running Experiments

Run the following commands to execute TR-BEACON:

python Continuous_TR_BEACON.py

Run the following commands to execute BEACON:

python Continuous_BEACON.py

Run the following commands to execute MaxVar:

python Continuous_MaxVar_RS.py

Run the following commands to execute sobol:

python Continuous_Sobol.py

Run the following commands to execute logEI:

python Continuous_LogEI.py

Run the following commands to execute NS-FS:

python Continuous_NS-FS.py