Skip to main content

shorter solving time for a harder problem instance modeled as ILP




  • Riley Clement
    Gurobi Staff Gurobi Staff

    Hi Selçuk,

    The behavior you describe is perhaps a little unusual - but this depends on what "harder" means.

    I think the first thing I would want to do is build some statistical confidence around the comparison and that means running both models with many values of Seed and then comparing distributions.  If it takes an hour for your problem to solve, then try 10 seeds, if it takes a minute, try 50.  We are quite used to seeing users make qualitative statements based on a single run of a model, which then don't hold true when running the model many different times.

    Often users also associate "hardness" with the size of the problem, and this is a weak relationship.  The "hardness" of the problem is often related to the data - for example a small vehicle routing problem with tight time windows can be much much harder to solve than a much larger problem with big time windows.

    - Riley

  • Selçuk Cevher
    First Comment
    First Question

    I got you very well, Riley. I probably should think more about what "harder" means.

    Thank you very much.


Please sign in to leave a comment.