A problem in research in mathematical optimization is the lack of publicly available real-life cases for benchmarking. Available benchmarks are often generated according to specific distributions that are not representative of real cases, and focus on these test cases has caused a divide to occur between established research and practical requirements. The difference in qualitative properties of generated and real cases can give a skewed view of performance of certain algorithms. At ORTEC we have an arsenal of data at our disposal, and by sharing this data we aim to bridge the divide, and contribute to the scientific community. This platform was launched on the 24th of May, 2018 at the 15th ESICUP meeting.
The benchmarks that will be published on this platform can be divided into the following categories:
- Load Building:
- The problem of Load Building involves the placement of items in three dimensional space according to certain constraints. Typical problems involve minimizing the number of required containers, or maximizing the number of placed items subject to constraints with respect to, for example, the weight distribution or support. A well-known mathematical model of a simple load building problem is the bin packing problem.
- In Routing problems we seek to find a route (or multiple routes) that in some sense has a low cost, and also meeting certain constraints. Typically we seek to minimize the number of deployed vehicles, travel time, or travelled distance subject to, for example, precedence or time window constraints. A few well-known formulations of routing problems are the travelling salesman problem and the vehicle routing problem.
- Workforce Planning:
- In Workforce Planning the goal is to efficiently assign members of a workforce to tasks such that certain tasks are performed. Typical problems involve minimizing the cost, or maximizing some measure of coverage subject to, for example, employee preferences and holidays or legal constraints. Two well-known formulations of simple workforce planning problems are the assignment problem and the nurse scheduling problem.
The main purpose of this website is to make real-life cases available for usage by the scientific community. However, in line with the GDPR, measures have been taken to protect privacy sensitive data: customer data has been anonymized. The core principle of our anonymization process is to remain true to the spirit of the case with respect to complexity, structure, and size, while altering the data to be untraceable to any individual or company. From this core principle different measures have been derived for, and applied to, the different categories of problems.
If you have used one or more of our benchmark cases in your research, we would like you to refer to the cases you have used. Each set of instances has its own unique identifier, following the format: "osb-[category id]-[set id]". Where the category identifier of Load Building, Routing, and Workforce Planning instances are respectively lb, r, and wp. When referring to sets of instances, add the postfix "_[instance ids]", where [instance ids] is a comma separated list of ranges of instance identifiers. A few examples of citations are given below:
"ORTEC Scientific Benchmarks: osb-lb-efr_1-101". benchmarks.ortec.com. Retrieved on 2018-05-24. "ORTEC Scientific Benchmarks: osb-lb-hpi_1,3-5". benchmarks.ortec.com. Retrieved on 2018-05-24.
where efr is the set identifier of the Efficient Fill Rate set, and hpi is the set identifier of the Heavy Palletized Items set.