Generation of Rule-Based Constraint Solvers: Combined Approach

Inductive Constraint Solving is a subfield of inductive machine learning concerned with the automatic generation of rule-based constraint solvers. In this paper, we propose an approach to generate constraint solvers given the definition of the constraints that combines the advantages of generation by construction with generation by testing. In our proposed approach, semantically valid rules are constructed symbolically, then the constructed rules are used to prune the search tree of a generate and test method. The combined approach leads in general to more expressive and efficient constraint solvers. The generated rules are implemented in the language Constraint Handling Rules.