Modern heuristic solvers can tackle difficult computational problems, but each solver performs well only on certain tasks. An algorithm portfolio uses empirical knowledge---past experience of each solver's behavior---to run the best solvers on each task.
The borg project includes a practical algorithm portfolio for decision problems, a set of tools for algorithm portfolio development, and a research platform for the application of statistical models and decision-theoretic reasoning to the algorithm portfolio setting.
The project web site is:
http://nn.cs.utexas.edu/pages/research/borg/
The best description of the research surrounding the borg project can be found in the PhD dissertation of the primary author.
Borg is provided under the non-copyleft open-source "MIT" license. The complete legal notice can be found in the included LICENSE file.