Debugging JuMP optimization models using graph theory

07/28/2023, 7:10 PM — 7:20 PM UTC
32-155

Abstract:

Writing a large optimization model is a time-consuming and error-prone task. If and when a modeling error is suspected, the developer often must re-consider every constraint in the model for faulty assumptions causing singularities in the optimization model, which impede convergence of solvers. We introduce a package that computes the Dulmage-Mendelsohn and block triangular partitions of incidence graphs of JuMP models, which can be used to detect sources of structural and numeric singularities.

Description:

The package we implement is named "JuMP-Incidence" or "JuMPIn.jl". It depends on JuMP and Graphs.jl, and may be of interest to the JuMP and Julia-Graphs communities. This package is of particular interest as model debugging is a notorious pain point in mathematical optimization. In the presentation, I intend to to describe the algorithms implemented, describe their use cases, and demonstrate their use to detect a modeling error in an optimization problem from chemical engineering.

Platinum sponsors

JuliaHub

Gold sponsors

ASML

Silver sponsors

Pumas AIQuEra Computing Inc.Relational AIJeffrey Sarnoff

Bronze sponsors

Jolin.ioBeacon Biosignals

Academic partners

NAWA

Local partners

Postmates

Fiscal Sponsor

NumFOCUS