I-DLV+MS: Preliminary Report on an Automatic ASP Solver Selector

Current ASP solvers feature diverse optimization techniques that highly influence their performance, causing systems to outperform each other depending on the domain at hand. We present I-DLV+MS, a new ASP system that integrates an efficient grounder, namely I-DLV, with an automatic solver selector: machine-learning techniques are applied to inductively choose the best solver, depending on some inherent features of the instantiation produced by I-DLV. In particular, we define a specific set of features, and build our classification method for selecting the solver that is supposed to be the “best” for each input among the two state-of-the-art solvers clasp and wasp. Despite its prototypical stage, performance of the new system on benchmarks from the 6th ASP Competition are encouraging both against the state-of-the-art ASP systems and the best established multi-engine ASP system, ME-ASP.