Contributions to automatic configuration and selection for satisfiability

Laburpena

Within the computer science community, it is well known that algorithms may exhibit completely different behaviours depending on how their parameters are configured. This is even more obvious when the problems being tackled are NP-Complete. For example it has been proved experimentally that configuring solvers for the SATisfiabilty (SAT) problem, the Maximum SATisfiability (MaxSAT) problem or for Mixed Integer Programming (MIP), can result in improvements of orders of magnitude. In this thesis, we show how automatically configured metahueristic algorithms can efficiently solve families of industrial and crafted instances of the MaxSAT problem. Then, we focus on improving the automatic algorithm configurator GGA providing a new distributed configurator which outperforms GGA on families of industrial and crafted SAT and MIP instances. Finally, in our aim of making this technology more accessible for all research communities and industry we present the tool PyDGGA that implements our new configurator. We additionally introduce the OptiLog framework for rapid prototyping of SAT-based systems that also incorporates a module for automatic configuration for the easy application of several configuration tools.