MIPLIB, short for Mixed-Integer Programming Library, is the standard test set used to compare mixed-integer solvers. It was first released in 1992 by Robert E. Bixby, E.A. Boyd, and R.R. Indovina. Naturally, as mixed-integer solvers become more accessible, more industries integrated mixed-integer solvers, and the size of instances increased, new iterations of MIPLIB were required. Releases followed in 1996, 2003, 2010, and 2017, with these releases following an unofficial seven year schedule. The customary seven years has now passed since the last release, and a new iteration of MIPLIB, MIPLIB 2024, is being put together.
In its current state, MIPLIB 2017 serves as the standard benchmark set for comparing the performance of different mixed-integer solvers. It was compiled by a group of 21 interested researchers, including representatives of the development teams of nine mixed-integer solvers, with submissions derived from 66 unique submitters (including bulk submissions from online optimization servers and optimization competitions). Nevertheless, the mixed-integer landscape has changed in seven years. New applications of mixed-integer solvers have arisen, and an increasing amount of industries are now using mixed-integer solvers. Therefore, we are advertising submissions for MIPLIB 2024.
To emphasize the importance of MIPLIB: It is the standard benchmark library for mixed-integer solvers. Having one of your instances included in this benchmark guarantees that mixed-integer solvers as a whole try and improve solution time on the included instance. If the instance is representative, then performance will at least partly generalize to other similar instances that you might have. It also guarantees that academics will regularly consider new solving techniques for your instance. Having as many industries and applications represented as possible is critical for MIPLIB 2024 to be an appropriate benchmark for mixed-integer solvers. Please submit any and all instances you are comfortable sharing to.
If you have instances in LP or MPS format, then submit these directly at https://miplibsubmissions.zib.de/
To generate an LP or MPS file from a GAMS MIP model, we suggest to use GAMS/CONVERT. For example, adding the following lines in front of a solve statement for a model named m will generate a file cplex.mps that captures the model instance for that solve statement:
$echo CplexMPS > convert.opt
option solver = convert;
m.optfile = 1;
Alternatively, you are also welcome to upload your GAMS MIP models at https://cloud.gams.com/s/p5AYL5336Kyep8H. We will then extract one or several model instances from the submission and pass them on to https://miplibsubmissions.zib.de/. Please include information on submitter, application area, references, etc., either into the uploaded file or by e-mail to svigerske@gams.com. By submitting your model, you consent that the generated instances will be published under CC BY-SA 4.0 license. The model source will not be published. On request, we can convert variable and equation names to generic names (x1, x2, …).
Deadline: 30.11.2024
Instance specifications: Anything that is classified as a mixed-integer linear program (this includes binary and integer programs). Instances containing indicator and SOS constraints are also welcome.