Communication Dans Un Congrès Année : 2025

MH-Builder: a C++ framework for designing adaptive metaheuristics for single and multi-objective optimization

Résumé

Heuristics and metaheuristics are essential for solving optimization problems, especially when exact methods are impractical due to resource limitations [8]. These methods prioritize finding satisfactory solutions within a reasonable timeframe rather than optimal ones. However, the lack of standardization across programming languages, data structures, and optimizations limits reproducibility and fair comparison of metaheuristics. Several frameworks exist (such as JMetalPy [2] or ParadisEO [4]) that address this issue by integrating multiple metaheuristics, ensuring fair and consistent evaluations through reproducibility. Despite their strengths, both have static designs and lack a dynamic environment for composing or modifying metaheuristics at runtime. Some work on adaptive metaheuristics issues is emerging [1], but no corresponding framework for reproducibility is currently available.

To overcome these limitations, MH-Builder [6] is an open-source, adaptive C++ framework developed by the ORKAD team at the CRIStAL laboratory of the University of Lille. Designed for single-and multi-objective optimization problems, it was initially created in 2019 to support ORKAD members develop metaheuristics within a unified environment that promotes reproducibility in an adaptive and dynamic setting [3, 5, 7]. In early 2024, it became opensource, broadening its support for researchers in fields like operations research and machine learning. As of November 2024, it had grown to include 248 classes, 525 commits, 144 unit tests, and a team of 10 contributors. Its GitLab repository [6] and documentation are publicly available. With ongoing development, MH-Builder aims to serve as a valuable framework for integrating heuristics and metaheuristics in an adaptive environment.

Like other available frameworks [2, 4], MH-Builder embraces two core principles of software development: modularity and flexibility. These principles are key to designing robust, scalable systems ensuring reproducibility, enhancing user customization and problem-solving capabilities. In MH-Builder, modularity is achieved through C++ templates and operator overloads, enabling easy customization of components. Heuristics and metaheuristics are organized into modular components that can be easily assembled to solve optimization problems. This modular foundation provides the flexibility to tackle both single-and multi-objective optimization problems with minimal code changes.

Building on these principles, MH-Builder stands out with a dynamic architecture that integrates adaptability. All components are configurable and can be exchanged dynamically, enabling users to modify or replace elements at runtime. For instance, users can switch the neighbor search operation (e.g., from the swap operator to 2-opt) or adjust the solution acceptance criteria (e.g., always accepting an improving solution to applying the Metropolis criterion [8]). This adaptability enables MH-Builder to deal effectively with a wide range of optimisation problems. As an extensible framework, it also enables users to define problem constraints and use various ready-to-use metaheuristics, ensuring easy reproducibility of previous studies.

MH-Builder provides a range of heuristics, local searches, and metaheuristics for single-and multi-objective optimization problems. It includes local search operators like swap, insertion,

Fichier principal
Vignette du fichier
MH-Builder a C++ framework for designing adaptive.pdf (186.01 Ko) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Licence

Dates et versions

hal-05234174 , version 1 (01-09-2025)

Licence

Identifiants

  • HAL Id : hal-05234174 , version 1

Citer

Amadeu Almeida Coco, Olivier Caron, Laetitia Jourdan, Marie-Eléonore Kessaci, Nadarajen Veerapen. MH-Builder: a C++ framework for designing adaptive metaheuristics for single and multi-objective optimization. 26ème édition du congrès annuel de la Société Française de Recherche Opérationnelle et d'Aide à la Décision, ROADEF, Feb 2025, Champs-sur-Marne, France. ⟨hal-05234174⟩
434 Consultations
163 Téléchargements

Partager

  • More