MPILS: An automatic tuner for MILP solvers

The parameter configuration problem consists of finding a that gives particular algorithm the best performance. This paper introduces new multi-phase tuner based on iterated local search meta-heuristic. addresses for deterministic MILP solvers are used to solve challenging industrial optimization problems. Further, proposed offers strategy three ideas. First, instead tuning in entire space induced by set, focuses small pool is dynamically enriched with promising parameters. Second, it leverages gathered knowledge during using statistical learning forbid less combinations. Third, tunes single instance provided earlier clustering instances. A computational study widely-used commercial solver CPLEX instances from MIPLIB library and real large-scale highlights potential tuner.