Cornflow: a flexible open source deployment server for optimization engines

Speaker: Franco Peschiera, PhD, Baobab soluciones

In the last decade, the cloud has motivated many business in OR to offer their optimization services through REST APIs. From modellers to solver suppliers, each platform provides access to the owner’s product. Most of the times this access is tied to the use of the product.

In combinatorial optimization there is no silver-bullet (or free lunch), each solution method (MIP model, CP model, meta-heuristic, add-hoc heuristic, etc.) is suited to solve different problems. Only offering a limited range of solution methods risks losing the power to solve some problem efficiently.

Finally, except a few exceptions (VRP, TSP), most problems do not have a clear data interface that is standardized. This makes it hard to benchmark different approaches to a given problem.

To address these problems (vendor lock, no-free-lunch, lack of standardization), Cornflow was created as a free (as in freedom) problem-agnostic and solver-agnostic deployment server for optimization problems. It hosts problems as flexible data-structures (json schemas) and solution methods as code (usually python code but not exclusively). This allows full customization of problems and their solution methods. For example a VRP-like problem can be solved with Concorde, a CP-SAT model, column-generation with gurobi or a 2-opt heuristic. While a Nurse-Rostering-like problem can be solved with a traditional MIP model, Timefold or some clever matheuristic.

We believe that by keeping the deployment server open, we will make it easier for problem providers (i.e., industry) and solution providers (i.e., academia) to build on top of each other’s work and collaborate more easily.

About the speaker:

Franco Peschiera is an engineer, a programmer and a scientist. He loves solving puzzles, organizing stuff and customer service. During his free (and not-so-free) time he’s a father, a cyclist and a professional skeptic.

Franco has a double degree in industrial engineering from the PUCP (Lima) and UPM (Madrid) universities and a PhD in Industrial Engineering from ISAE-SUPAERO (Toulouse). He founded a successful OR software (baobab soluciones) company based in Madrid where he worked 10 years as various roles, including project leader, chief technical officer, and consultant in several optimization projects in the car, aerospace, oil and pharmaceutic industries, among many others.

He also founded a yet-to-be-succesful clothing-rental company based in Lima/Rome (HackPacking), is the proud maintainer of the PuLP mathematical modelling library and maintains several other open source tools for Operations Research. You can find him in: LinkedIn and github.