ALEBO

This is the code associated with the paper "Re-Examining Linear Embeddings for High-Dimensional Bayesian Optimization"

If you find this code useful please cite it as

@inproceedings{Letham2020Re,
    author    = {Letham, Benjamin and Calandra, Roberto and Rai, Akshara and Bakshy, Eytan},        
    title     = {Re-Examining Linear Embeddings for High-Dimensional {B}ayesian Optimization},
    booktitle   = {Advances in Neural Information Processing Systems 33},
    year      = {2020},
    series = {NeurIPS},
}

Installation

To install the code clone the repo and install the dependencies as

git clone https://github.com/facebookresearch/alebo.git 
cd alebo
pip install -r requirements.txt

Some of the baselines require additional packages that can not be pip-installed. Detailed instructions can be found inside each file of the benchmarks/ folder.

Using ALEBO for optimizing a function

See quickstart.ipynb for a simple example of how to use ALEBO to optimize a function. ALEBO is built using the Ax platform; see instructions there on how to install via pip. You will need version 0.1.17 or later.

Reproducing the experiments

This repository contains the code required to run the benchmark experiments and generate the figures in the paper. The only exception are the DAISY figures, since the simulator is not yet open source.

Generating figures:

The figs/ directory contains a file to generate each of the figures in the paper, as indicated by the file name. Some figures show the results of simulations; in these cases the file contains code to both run the simulation and create the figure. For example, executing figs/fig_4.py will run the P_opt simulation described in the paper, will store the simulation results in figs/data/, and will then generate the figure based on those results. The pdf for Fig. 4 will be saved in figs/pdfs/.

Running benchmark experiments

The benchmarks/ directory contains code for running the benchmark BO experiments described in the paper. The benchmark problems are defined in benchmark_problems.py. Each method has its own script for evaluating that method on the appropriate set of benchmark problems: run_{method}_benchmarks.py, where {method} is:

• ax, for our implementations of ALEBO, HeSBO, and REMBO
• addgpucb for Add-GP-UCB via Dragonfly
• cmaes for CMA-ES
• ebo for Ensemble Bayesian Optimization
• linebo for LineBO
• smac for SMAC
• turbo for TuRBO

See the paper for references for each of these methods. Each file explains what needs to be done in order to run the experiments for that method. For instance, run_cmaes_benchmarks.py requires installing cma from pip; run_ebo_benchmarks.py requires cloning a repository. See each file for its instructions.

The file run_rrembo_benchmarks.R provides a similar script in R for running the benchmark experiments for k-\Psi REMBO variants. These use the R package RRembo, and results are stored in json.

All benchmark results are stored in benchmark/results/ (the json files produced by each run of each method are not shipped in this repo). Once all of the run_*_benchmarks.* files have been run, compile_benchmark_results.py is used to compile the results from all of the different methods into a single file for each experiment. These files are benchmarks/results/*_aggregated_results.json and are included in this repository as the benchmark results used in the paper.

Executing figs/fig_5.py loads these aggregated results and generates the benchmark results figure in the paper.

A separate script benchmarks/run_nasbench.py contains all of the code for running the NASBench experiment.

The ALEBO model and generation code

The actual implementation of the ALEBO method is at: https://github.com/facebook/Ax/blob/master/ax/models/torch/alebo.py

License

This code is licensed under CC-by-NC, as found in the LICENSE file.