MedleySolver: Online SMT Algorithm Selection

Satisfiability modulo theories (SMT) solvers implement a wide range of optimizations that are often tailored to particular class problems, and differ significantly between solvers. As result, one solver may solve query quickly while another might be flummoxed completely. Predicting the performance given is difficult for users SMT-driven applications, particularly when problems they have do not fall neatly into well-understood category. In this paper, we propose an online algorithm selection framework SMT called MedleySolver predicts relative performances set on query, distributes time amongst solvers, deploys in sequence until solution obtained. We evaluate against best available alternative, offline learning technique, terms pure practical usability typical user. find with no prior training, solves 93.9% queries solved by virtual selector achieving 59.8% par-2 score most successful individual solver, which 87.3%. For comparison, alternative takes longer train than our entire 2000 queries.