|

R202
Benchmark on DAOOPT and GUROBI with the PASCAL2 Inference Challenge Problems
Junkyu Lee, William Lam, and Rina Dechter

Abstract
We report the PASCAL2 benchmark for DAOOPT and GUROBI on MPE task with 330 optimally solved instances from 8 benchmark domains. DAOOPT outperformed GUROBI in 3 domains, while GUROBI was faster than DAOOPT in the rest of the 5 domains. We show that DAOOPT performed well in domains where it could have high quality initial solutions for pruning the AND/OR search space, or skip search when the heuristic upper bounds were converged to the optimal due to MPLP/JGLP algorithms. GUROBI presented excellent performance if cutting planes were applied progressively and its heuristic algorithms could find the optimal solution at the root of branch-and-cut tree.

[pdf]