MPMMine: A Benchmark Suite for Constraint Acquisition and Mathematical Programming Model Validation
A new benchmark suite named MPMMine has been launched to fill the gap in effective benchmarks for Constraint Acquisition (CA) and the related field of Mathematical Programming (MP) model validation and improvement using domain knowledge artifacts. Current benchmarks, primarily created for evaluating solvers, are poorly organized, inconsistently address individual problems, and lack the necessary domain knowledge artifacts for CA methods, which hinders reproducibility and comparability across studies. MPMMine evaluates algorithms that discover, validate, and enhance MP models with various domain knowledge artifacts, adhering to principles of consistency, standardization, completeness, extensibility, openness, and version control. It utilizes a consistent structure and open formats like MiniZinc, CommonMark, and JSON, offering multiple models for each problem. The suite intends to promote the development of CA methods through thorough evaluation and comparison across research.
Key facts
- MPMMine is a benchmark suite for Constraint Acquisition and Mathematical Programming model validation.
- Existing benchmarks were designed for solver evaluation, not for CA algorithms.
- Current benchmarks are loosely organized and omit domain knowledge artifacts.
- MPMMine uses open formats: MiniZinc, CommonMark, and JSON.
- The suite provides multiple models per problem.
- It is guided by consistency, standardization, completeness, extensibility, openness, and version control.
- The deficiency in benchmarks impedes reproducibility and cross-study comparability.
- MPMMine aims to assess algorithms that discover, validate, and enhance MP models.
Entities
—