Constrained Two-step Look-Ahead Bayesian Optimization

Authors: Yunxiang Zhang, Xiangyu Zhang, Peter Frazier

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In numerical experiments, 2-OPT-C typically improves query efficiency by 2x or more over previous methods, and in some cases by 10x or more.
Researcher Affiliation Academia Yunxiang Zhang Cornell University yz2547@cornell.edu Xiangyu Zhang Cornell University xz556@cornell.edu Peter I. Frazier Cornell University pf98@cornell.edu
Pseudocode Yes Pseudocode for using 2-OPT-C is provided in the supplement.
Open Source Code No The paper does not provide any explicit statements about releasing source code or links to a code repository for the methodology described.
Open Datasets Yes The benchmark problems include three synthetic problems from [1], named P1, P2, and P3, and two real-world problems, portfolio optimization and robot pushing. Detailed descriptions are in the supplement.
Dataset Splits No The paper describes sampling initial points and the number of function evaluations, but does not provide specific train/validation/test dataset splits for model training.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts) used for running the experiments.
Software Dependencies No The paper mentions using 'GPy [38]' but does not provide a specific version number for this or any other software dependency.
Experiment Setup Yes The 2-OPT-C implementation uses GPs with a constant zero-mean prior and ARD square-exponential kernels for both objectives and constraints. GP hyperparameters are obtained by maximizing the marginal likelihood using GPy [38]. For the initialization of each experiment, we randomly sample three points with at least one feasible point from a Latin hypercube design. We run N = 40 function evaluations for P1 and P2, N = 60 for P3, N = 30 for portfolio optimization problem, and N = 50 for robot pushing problem. We use batch size of 1 for all five experiments.