Best Heuristic Identification for Constraint Satisfaction

Authors: Frederic Koriche, Christophe Lecoutre, Anastasia Paparrizou, Hugues Wattez

IJCAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Comparative experiments are reported in Section 4. Finally, we conclude with some perspectives of research in Section 5.Our experiments have been performed with the constraint solver ACE4, by keeping the default option for most parameters.
Researcher Affiliation Academia 1CRIL, Univ. Artois & CNRS 2LIX CNRS, Ecole Polytechnique, Institut Polytechnique de Paris
Pseudocode Yes Algorithm 1: Adaptive Successive Halving (ASH) and Algorithm 2: Adaptive Single Tournament (AST)
Open Source Code No The paper mentions the constraint solver ACE with a GitHub link (https://github.com/xcsp3team/ace), but this is for a third-party tool used, not explicitly for the authors' own implementation code for the described methodology (AST). A HAL archive link is provided for proofs and additional experiments, but it does not specify code.
Open Datasets Yes For the set P, we have considered all CSP instances selected for the XCSP3 competitions from 2017 to 2019.3 and 3http://www.xcsp.org/competitions/
Dataset Splits No The paper uses CSP instances from XCSP3 competitions but does not specify any training, validation, or test dataset splits for its experiments.
Hardware Specification Yes All experiments have been conducted on a 3.3 GHz Intel XEON E5-2643 CPU, with 32 GB RAM, with a timeout set to 2, 400 seconds.
Software Dependencies No The paper mentions using the 'constraint solver ACE' but does not specify its version number or any other software dependencies with version information.
Experiment Setup Yes AST uses a parameter m for testing the behavior of the constraint solver on candidate heuristics. [...] for AST, we have evaluated the algorithm using m {1, 2, 4, 8, 16}. Our experiments have been performed with the constraint solver ACE4, by keeping the default option for most parameters. [...] each cutoff σluby(t) of Luby s sequence was rescaled to u σluby(t), where u = 150, corresponding to u wrong decisions (conflicts) per cutoff unit.