Skip to main content

Gurobi not making progress on the gap

Answered

Comments

1 comment

  • Elisabeth Rodríguez Heck
    Gurobi Staff Gurobi Staff

    Hi Sophie,

    This is a huge model! 

    Some comments on your current parameter settings:

    • The combination of Presolve=2 PreSparsify=1 looks like a very good choice because presolve is able to greatly reduce the size of your model (especially the nonzeros) in little more than 1000s, so I would definitely keep these parameters.
    • Method=1: this forces Gurobi to use dual simplex to solve the root node. I would rather keep this one as default or use Method=3 in order to run primal simplex, dual simplex and barrier at the same time. 
    • MIPFocus=3: this makes Gurobi focus on improving the bound. I can see that the bound improves very slowly, but it is also possible that the bound is very good and doesn't have to be improved that much (depending on the formulation). Do you have an idea of what needs to be improved, whether the incumbent or the bound? If the bound is already very good, then I would rather use MIPFocus=1. Maybe you can guess which needs to be improved (incumbent or bound) from running your monthly model and applying the results to your yearly model. 

    The NoRel heuristic seems to be very successful, but you may improve the incumbent even more by increasing NoRelHeurTime.

    Some other parameters I would test:

    • DegenMoves=0. Because of the "Total elapsed time" messages after solving the root relaxation.
    • NodeMethod=2. The number of iterations per node (column It/Node) is huge. Trying barrier for the nodes might help decrease this.

    You can try our automatic tuner on one of your smaller models to find parameters and apply them to your larger model.

    Best regards,

    Elisabeth

    0

Please sign in to leave a comment.