RBFOpt: an open-source library for black-box optimization with costly function evaluations

作     者：Costa, Alberto Nannicini, Giacomo 

作者机构：Swiss Fed Inst Technol Future Cities Lab Program Singapore Singapore IBM TJ Watson Res Ctr Yorktown Hts NY 10598 USA 

出 版 物：《MATHEMATICAL PROGRAMMING COMPUTATION》 (数学规划计算)

年 卷 期：2018年第10卷第4期

页      面：597-629页

核心收录：

学科分类：1201[管理学-管理科学与工程(可授管理学、工学学位)] 08[工学] 0835[工学-软件工程] 0701[理学-数学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：SUTD-MIT International Design Center [IDG21300102] NRF's Campus for Research Excellence and Technological Enterprise (CREATE) program [FI 370074011, FI 370074016] 

主　　题：Black-box optimization Derivative-free optimization Global optimization Radial basis function Open-source software Mixed-integer nonlinear programming 

摘      要：We consider the problem of optimizing an unknown function given as an oracle over a mixed-integer box-constrained set. We assume that the oracle is expensive to evaluate, so that estimating partial derivatives by finite differences is impractical. In the literature, this is typically called a black-box optimization problem with costly evaluation. This paper describes the solution methodology implemented in the open-source library RBFOpt, available on COIN-OR. The algorithm is based on the Radial Basis Function method originally proposed by Gutmann (J Glob Optim 19:201-227, 2001. https://***/10.1023/A:1011255519438), which builds and iteratively refines a surrogate model of the unknown objective function. The two main methodological contributions of this paper are an approach to exploit a noisy but less expensive oracle to accelerate convergence to the optimum of the exact oracle, and the introduction of an automatic model selection phase during the optimization process. Numerical experiments show that RBFOpt is highly competitive on a test set of continuous and mixed-integer nonlinear unconstrained problems taken from the literature: it outperforms the open-source solvers included in our comparison by a large amount, and performs slightly better than a commercial solver. Our empirical evaluation provides insight on which parameterizations of the algorithm are the most effective in practice. The software reviewed as part of this submission was given the Digital Object Identifier (DOI) https://***/10.5281/zenodo.597767.