Globalizing constraint models

We present a method to detect implicit model patterns (such as global constraints) that might be able to replace parts of a combinatorial problem model that are expressed at a low-level. This can help non-expert users write higher-level models that are easier to reason about and often yield better performance. Our method generates candidate model patterns by analyzing both the structure of the model – its constraints, variables, parameters and loops – and the input data from one or more data files. Each candidate is scored by comparing a sample of its solution space with that of the part of the model it is intended to replace. The top-scoring candidates are presented to the user through an interactive display, which shows how they could be incorporated into the model. The method is implemented for the MiniZinc modeling language and available as part of the MiniZinc distribution.

Authors

• Kevin Leo Department of Data Science & AI (DSAI), Monash University, Australia
• Christopher Mears
• Guido Tack Monash University
• Maria Garcia de la Banda Department of Data Science & AI (DSAI), Monash University, Australia ARC Training Centre in Optimisation Technologies, Integrated Methodologies, and Applications (OPTIMA), Australia

Keywords

• Constraint programming
• Constraint acquisition
• Automated modeling
• Constraint modeling
• Global constraints