A High Level Language for Solver Independent Model Manipulation and Generation of Hybrid Solvers

摘要

This paper introduces a high level language that allows for the specification and manipulation of solver independent models and allows for easily generating complex solvers in the Comet language. As Constraint Programming (CP) techniques have increased in complexity, it has become more difficult and time consuming to implement models that take advantage of state-of-the-art modeling techniques and search heuristics. This is particularly problematic for problems that have not been well studied as it is often unclear a priori which modeling technologies and search strategies will be effective.This work builds on previous solver independent languages by introducing a more general framework based on abstract models and model operators. Model operators represent complex model transformations that can be applied in various combinations to yield a wide array of concrete solvers, including hybrid solvers. Furthermore, Local Search (LS) is fully supported allowing for sequential and parallel bounds-passing hybrids that have not been possible in previous solver independent languages. Large Neighborhood Search (LNS) and column generation based models are also demonstrated.