/gp-imprecision-in-bo

This repository contains Prior-RObust Bayesian Optimization (PROBO) as introduced in our paper "Accounting for Gaussian Process Imprecision in Bayesian Optimization"

Primary LanguageRMIT LicenseMIT

Prior-RObust Bayesian Optimization (PROBO)

Introduction, TOC

This repository contains Prior-RObust Bayesian Optimization (PROBO) as introduced in our paper "Accounting for Gaussian Process Imprecision in Bayesian Optimization" (Julian Rodemann, Thomas Augustin). More precisely,

  • PROBO contains implementation of PROBO
  • benchmarking provides files for experiments (section 4), in order to reproduce results, see setup below
  • files in data allow recreating visualizations of data and functions used in the benchmark experiments, see below
  • files in univariate-benchmark-functions allow visualization of synthetic test functions mentioned in section 4

Tested with

  • R 4.1.6
  • R 4.0.3

on

  • Linux Ubuntu 20.04
  • Linux Debian 10
  • Windows 10 Build 20H2
  • MacOS (only visualizations)

Setup

First and foremost, please clone this repo (and install required packages as indicated by your IDE)

In order to reproduce figure 2 showing the papers' key results (and visualizations of further results not included but only mentioned in the paper on page 10)

Please find optional (currently commented out) visualizations in lines 118-159 of this very file. In order to rerun all simulations described in section 4 (PROBO on graphene data), please

  • source this file to kick off the simulation study (estimated time on 64-bit-core (linux gnu): 11h)
  • results are saved automatically
  • source this file to visualize the retrieved results

Data

Find files to read in data and create target functions in folder data. E.g. source data/make-kapton-rf.R to read in graphene data (source is here) and reproduce figure 1 of the paper