This code repository accompanies the paper "Bayesian Optimization with Cost-varying Variable Subsets".

The easiest way to install the required dependencies is to use Anaconda on Linux (code was tested on Ubuntu 20.04.4 LTS). In this directory, run

conda env create -f environment.yml

The environment can then be used with

conda activate bocvs

Alternatively, the dependencies can be installed manually using environment.yml as reference.

To run the experiments, first generate the desired list of experiments to run with

python job_creator.py

This will create a jobs directory with jobs.txt containing a list of experimental settings. After this, run

python job_runner.py

job_runner.py will run indefinitely until all jobs in jobs.txt are exhausted. Multiple job_runner.py can be run at the same time to run experiments in parallel.

Once all experiments have completed, plot the results with

python results.py with OBJ_NAME

where OBJ_NAME is one of {'gpsample', 'hartmann', 'plant', 'airfoil'}. The results will then be plotted in summary_results.