Athanor: High-Level Local Search Over Abstract Constraint Specifications in Essence

Authors: Saad Attieh, Nguyen Dang, Christopher Jefferson, Ian Miguel, Peter Nightingale

IJCAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental A series of case studies show the performance of ATHANOR, benchmarked against several local search solvers on a range of problem classes.
Researcher Affiliation Academia 1School of Computer Science, University of St Andrews, UK 2Department of Computer Science, University of York, UK
Pseudocode Yes Algorithm 1 presents the entire search procedure used in the current version of ATHANOR.
Open Source Code Yes Source code and all data used in this work are publicly available as a github repository 1. 1https://github.com/athanor
Open Datasets Yes The instances used in this work come from popular benchmarking datasets whenever possible, and are randomly generated otherwise. The instances for TSP, CVRP, Knapsack and Social Golfers are taken from TSPLIB [Reinelt, 1995], VRP-REP [Mendoza et al., 2014], Pisinger s hard knapsack [Pisinger, 2005], and CSPLib [Miguel et al., 2000], respectively. All instances or reference links to them are available in the github repository of ATHANOR6.
Dataset Splits No The paper evaluates solvers on problem instances but does not describe training, validation, or test splits for datasets, as it focuses on solving constraint programming problems rather than machine learning model training.
Hardware Specification No The paper mentions using a 'Cirrus UK National Tier-2 HPC Service at EPCC', but this is a general service name and does not provide specific details such as GPU/CPU models or memory.
Software Dependencies Yes The Mini Zinc backend of SAVILE ROW along with Mini Zinc 2.1.7 [Nethercote et al., 2007] were used to produce Flat Zinc inputs for Chuffed, Oscar-CBLS and Yuck.
Experiment Setup Yes For each problem class, all solvers were run ten times (except Chuffed, as it is deterministic). For optimisation problems... the best objective found by each solver is logged every second. ...I 10, Z 1.3, L 500 algorithm constants.