Instance-Based Selection of Policies for SAT Solvers

Abstract

Abstract. Execution of most of the modern DPLL-based SAT solvers is guided by a number of heuristics. Decisions made during the search process are usually driven by some fixed heuristic policies. Despite the outstanding progress in SAT solving in recent years, there is still an appealing lack of techniques for selecting policies appropriate for solving specific input formulae. In this paper we present a methodology for instancebased selection of solver's policies that uses a data-mining classification technique. The methodology also relies on analysis of relationships between formulae, their families, and their suitable solving strategies. The evaluation results are very good, demonstrate practical usability of the methodology, and encourage further efforts in this direction.

Simple algorithm portfolio for SAT

“…We presented a strikingly simple algorithm portfolio for SAT stemming from our work on ArgoSmArT (Nikolić et al, 2009). The presented system, ArgoSmArT k-NN, benefits from the SATzilla system in several ways: it uses a subset of SATzilla features for representation of instances, a selection of SAT solvers, SATzilla solving data for the training corpus, and fragments of SATzilla implementation.…”

“…This virtual SATzilla system will be just referred to as SATzilla. In our experimental comparison, we included MXC08 (the best single solver on the training set), SATzilla, the ArgoSmArT system based on (Nikolić et al, 2009) with 13 SATzillasolvers instead of ArgoSAT configurations, ArgoSmArT 1-NN, and ArgoSmArT 9-NN. Also, we compare to the virtual best solver -a virtual solver that solves each instance by the fastest available solver for that instance (showing the upper achievable limit).…”

“…In this paper we present an algorithm portfolio system, based on the k-nearest neighbors method, that is conceptually significantly simpler and more efficient than SATzilla. It derives from our earlier research on solver policy selection (Nikolić, Marić, & Janičić, 2009).…”

Portfolio approaches in constraint programming

“…Furthermore, even assuming to have fewer solvers than cores, it is likely that -due to synchronization and memory consumption issues-running in parallel all the solvers on the same multicore machine is actually far from running the same solvers on different machines [162]. A naive multiprocessing approach could even result in wasting a huge amount of resources for little gains (e.g., see the examples of portfolios consisting of nearly 60 solvers in [147]). If we look at the results of the MiniZinc Challenge 2014 we notice that -despite the availability of eight logical cores-the performance of parallel solvers were not much better than those of sequential ones.…”

“…Some lazy approaches have been studied and evaluated, see for instance [154,151,147,72,164]. Avoiding (or at least lightening) the training phase can be advantageous in terms of simplicity and flexibility: it facilitates the treatment of new incoming problems, solvers and features.…”