Guillaume Marques | Design and features of Coluna.jl


Guillaume Marques, a Ph.D. student at the Université de Bordeaux, gave his talk that was accepted for JuMP-dev 2020 on the design and features of the Coluna.jl package during June 2020’s JuMP monthly developer call.

Abstract

Coluna.jl is a branch-and-cut-and-price framework written in Julia. The user introduces a formulation of his problem using JuMP and BlockDecomposition which is then reformulated by Coluna based on the annotations provided through BlockDecomposition. Then, the framework optimises the reformulation using the algorithm chosen by the user. We aim Coluna to be very flexible. Coluna thus provides an interface and atomic algorithmic routines to let the user develop his own optimisation procedures. Moreover, we built Coluna on top of MathOptInterface as we may optimise parts of reformulations with MILP solvers. We will present the design of the package and some features through few classic problems of operations research.