Friday, January 20, 2017

Opossum wins the COIN-OR Cup 2016

Opossum (OPtimizatiOn System with SUrrogate Models)—a new optimization plug-in for Grasshopper available on food4rhino —and RBFOpt—a model-based, black-box optimization library—have jointly won the COIN-OR Cup 2016!

Computational Infrastructure for Operations Research (COIN) is an international initiative to spur the development of open-source software for the operations research community. The award committee recognized the use of state-of-the-art mathematical optimization tools in a context that is historically skeptical about them, the effort to increase awareness of the potential of mathematical optimization within the architecture community and its easy-to-use interface for non-experts.

Opossum is the first model-based optimization tool for Grasshopper. It is based on the RBFOpt library and has a GUI similar to Galapagos. Opossum uses advanced machine learning techniques to find good solutions with a small number of evaluations. Its specialty is finding close to optimal solutions in 100-300 evaluations, while other types of algorithms typically need many more steps.

This paper presents benchmark results comparing an early version of Opossum, Galapagos and Goat.

No comments: