3 présentations

• 
  08h30 - 08h52

  The journey of creating a Modelling Language

  • Bardhyl Miftari, prés., University of Liege
  • Guillaume Derval , University of Liege
  • Quentin Louveaux , University of Liege
  • Damien Ernst , University of Liege

  In this presentation, we walk through the journey of creating a modeling language, the Graph-Based Optimization Modelling Language (GBOML), from its design ideas to the final version of the tool and future work. We start the presentation by highlighting the issues encountered with previous modelling tools and how we addressed them in GBOML. The development of the tool was articulated around encapsulation, structure exploitation and reuse, functionalities that were missing in other tools.

  GBOML encodes problems in the form of a hierarchical hypergraph made of nodes and hyperedges. We explain we leverage this structure for efficient problem encoding and generation. We also provide an overview of the future work as we take a step forward and think of functionalities that modelling languages should have. We talk about using the predefined structure to implement automatic presolving by using the syntax tree, node templating and automation of sensitivity analysis methods.

• 
  08h52 - 09h14

  Automated Conversion of Optimization Problems in the MP Solver Interface Library

  • Robert Fourer, prés., AMPL Optimization Inc.
  • Gleb Belov, AMPL Optimization Inc.

  The range of expressions offered by modeling languages and recognized by solvers has been steadily extended, in ways that make optimization models easier to describe, validate, and maintain — but that make conversion possibilities ever more numerous and complex. Following an introductory survey, this presentation describes a variety of challenges that have been faced in analyzing objective and constraint expressions and in converting them to forms that solvers require. Topics include correct and efficient implementation, side-effects of numerical tolerances, and especially the difficulties presented by the evolution of solver features. Examples are taken from the new MP solver interface library designed for the AMPL modeling language.

• 
  09h14 - 09h36

  Duality and decomposition in Google’s OR-Tools

  • Juan Pablo Vielma , prés., Google

  OR-Tools MathOpt is a software library for algebraic modeling of mathematical optimization problems. MathOpt supports solver-independent modeling of a wide range of optimization problems including continuous and mixed-integer problems with linear or quadratic objectives and various classes of constraints (including linear, quadratic, second order cone and some specialized constraints). In this talk, we describe various aspects of MathOpt’s design and implementation with a particular focus on duality and decomposition methods.