Cods for the manuscript "Computationally Efficient High-Dimensional Bayesian Optimization via Variable Selection", the arXiv version is https://arxiv.org/abs/2109.09264. This work has been accepted in AutoML 2023, the camera ready version is https://openreview.net/forum?id=QXKWSM0rFCK1.

Our method can be run on python==3.9

To install the dependencies as:

pip install -r requirements.txt

We also need to use an R package tmvtnorm (https://cran.r-project.org/web/packages/tmvtnorm/index.html), please refer to https://www.dataquest.io/blog/install-package-r/ for instructions on how to install a package in R.

The detailed description of ALEBO can be found in https://github.com/facebookresearch/alebo. The dependencies required for ALEBO are already in the file requirements.txt.

The detailed description of HeSBO can be found in https://github.com/aminnayebi/HesBO. The dependencies required for HeSBO are already in the file requirements.txt.

To run MOPTA08 function, first decompress the file mopta_libs.zip. There are two libraries, libmopta.so and moptafunc.so, in this .zip file, please put them into the same folder of the running script. For detailed instructions on compiling these two .so files, please see https://gist.github.com/denis-bz/c951e3e59fb4d70fd1a52c41c3675187.

To run VS-BO, use the following command: python VSBO_run.py --obj_func Branin --method VSBO (obj_func can also be Hartmann6/StyblinskiTang4/rover/mopta). it will automatically create an output path ./Branin/VSBO/. VS-BO has seven output files, and all these output files will be in the output path:

• X_*.npy: The query obtained for each iteration
• Y_*.npy: The output value obtained for each iteration
• Time_*.npy: The accumulated wall clock time for each iteration
• Time_process_*.npy: The accumulated CPU time for each iteration
• F_importance_val_*.npy: The importance scores of variables for each variable selection step
• F_rank_*.npy: Sorted variables for each variable selection step based on their importance scores
• F_chosen_*.npy: the chosen variables for each variable selection step.

To run SAASBO, please see the instructions in the folder ./saasbo_scripts

To run other baselines, use the command such as: python baseline_run.py --obj_func Branin --method vanillaBO (methods can also be rembo/rembo_interleave/alebo/hesbo). It will automatically create an output path ./Branin/vanillaBO/, which contains the following four output files:

• X_*.npy
• Y_*.npy
• Time_*.npy
• Time_process_*.npy

To plot the regret curve versus iterations, use the following command: python plot_iterations.py --input_path ./Branin/VSBO/ --output_path ./ --method VSBO --runs 20 --iterations 200 --best_value -0.44165457, it will generate a figure called curve_iterations.png.

To plot the regret curve versus wall clock times, use the following command: python plot_times.py --input_path ./Branin/VSBO/ --output_path ./ --method VSBO --runs 20 --T_max 600 --best_value -0.44165457 --wct, it will generate a figure called curve_wct.png.

To plot the regret curve versus CPU times, use the following command: python plot_times.py --input_path ./Branin/VSBO/ --output_path ./ --method VSBO --runs 20 --T_max 8000 --best_value -0.44165457, it will generate a figure called curve_cput.png.

Please choose T_max to be an integer value such that it can be divided by 100.

To plot the bar figure of the total frequency of being chosen as important for each variable, use the following command: python plot_important_freq.py --input_path ./Branin/VSBO/ --output_path ./ --iterations 200 --runs 20 --vs_freq 20