Abstract: Mixed Integer Linear Programming (MILP) solvers, such as GUROBI, besides problem input may also receive a set of tunable parameters that affect the performance of a solver. Although iterated racing methods (namely Irace) show significant success with such tasks, we still consider local-search methods competitive. Therefore, we investigate different sides of methods’ capabilities, such as real-time efficiency, number of solver runs, etc. Further research will define necessity for a wide-scale automatization, namely usage of machine learning and neural network applications for feature extraction based on task files and prediction of parameter configuration based on obtained features using general linear model. The research was conducted for a task-scheduling problem.

Cite: Ustyugov V.
On different methods for automated MILP solver configuration
In compilation MOTOR 2024: сборник тезисов XXIII Международной конференции «Теория математической оптимизации и исследование операций», (Омск, 30 июня – 06 июля 2024 г.). – Издательство ОмГУ., 2024. – C.97. – ISBN 978-5-7779-2691-3.

Dates:

Published print:	Jul 17, 2024
Published online:	Jul 17, 2024

Identifiers: No identifiers

Citing: Пока нет цитирований