toulbar2, an exact cost function network solver - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2023

toulbar2, an exact cost function network solver

Résumé

toulbar2 is an open-source C++ solver for cost function networks, with an MIT license and a documentation describing its interfaces with C++ and python. The constraints and objective function are factorized in local functions on discrete variables. Each function returns a cost for any assignment of its variables. Constraints are represented as functions with costs in {0, T} where T is an upper bound cost associated with forbidden assignments. toulbar2 looks for a non-forbidden assignment of all variables that minimizes the sum of all functions. These functions can be nonlinear (quadratic terms or higher arity) and all variables must have a finite domain of reasonable size (less than a few hundred). Using on the fly translation, toulbar2 can also directly solve optimization problems on other graphical models such as Maximum Probability Explanation on Bayesian networks, and Maximum A Posteriori on Markov random fields. It can also read partial weighted MaxSAT problems (wcnf format), pseudo-Boolean optimization problems (opb and qpbo formats), as well as constrained satisfaction and optimization problems (CSP and COP in XCSP3 format). toulbar2 exploits an anytime hybrid best-first branch-and-bound algorithm (HBFS) that tries to quickly provide good solutions together with an upper bound on the gap between the cost of each solution and the (unknown) optimal cost. Thus, even when it is unable to prove optimality, it will bound the quality of the solution provided. It can also apply a variable neighborhood search algorithm exploiting a problem decomposition (UDGVNS). Both algorithms are complete and can be run in parallel using OpenMPI. In addition, toulbar2 can also find a sequence of diverse solutions or exhaustively enumerate solutions below a cost threshold and perform guaranteed approximate weighted counting of solutions. For stochastic graphical models, it computes the partition function, a #P-complete problem. Current work in toulbar2 is on multicriteria and bilevel optimization. The presentation will give an overview of the solver and its results on two recent competitions.
Fichier principal
Vignette du fichier
Givry23c.pdf (322.15 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Licence : CC BY - Paternité

Dates et versions

hal-04021879 , version 1 (09-03-2023)

Licence

Paternité

Identifiants

  • HAL Id : hal-04021879 , version 1

Citer

Simon De Givry. toulbar2, an exact cost function network solver. 24ème édition du congrès annuel de la Société Française de Recherche Opérationnelle et d'Aide à la Décision ROADEF 2023, Feb 2023, Rennes, France. ⟨hal-04021879⟩
10 Consultations
21 Téléchargements

Partager

Gmail Facebook X LinkedIn More