Comparison of derivative-free optimization algorithms

This page accompanies the paper by Luis Miguel Rios and Nikolaos V. Sahinidis Derivative-free optimization: A review of algorithms and comparison of software implementations, Journal of Global Optimization, Volume 56, Issue 3, pp 1247-1293, 2013. The paper presents results from the solution of 502 test problems with 22 solvers. Here, we provide all test problems and detailed results that can be used to (a) reproduce the results of the paper and (b) facilitate comparisons with other derivative-free optimization algorithms.

Models in GAMS format:
convex nonsmooth || convex smooth || nonconvex nonsmooth || nonconvex smooth
one or two variables || three to nine variables || ten to thirty variables || over thirty one variables

Models in C/Fortran format:
The globallib and princetonlib models are available in C source code. Random instances of Richtarik's, Conn et al.'s and Nesterov's models are available in C source code. The Luksan and Vlcek models and are available in FORTRAN source code.

problemdata files for models: problemdata.zip
Results for all solvers are included in the following file: dfo_results.zip
Information about the format of the result files is given in the following README

Results for individual solvers:
asa || bobyqa || cma-es || dakota-direct || dakota-ea || dakota-pattern || dakota-soliswets || dfo || fminsearch || global || hopspack || imfil || mcs || newuoa || nomad || pswarm || sid-psm || snobfit || tomlab-glccluster || tomlab-lgo || tomlab-multimin || tomlab-oqnlp

Refinement ability results for individual solvers:
asa || bobyqa || cma-es || dakota-direct || dakota-ea || dakota-pattern || dakota-soliswets || dfo || fminsearch || global || hopspack || imfil || mcs || newuoa || nomad || pswarm || sid-psm || snobfit || tomlab-glccluster || tomlab-lgo || tomlab-multimin || tomlab-oqnlp

For the convenience of further testing with the above problems, we provide a complete listing of test problems and model statistics.