Skip to main content

Advice on performance tuning of MILP

Answered

Comments

3 comments

  • Marika Karbstein
    • Gurobi Staff Gurobi Staff

    Hi Jesse,

    In the log file, you can see that the best bound does not improve after the root node is solved; it stays at the value 6812335.05. Only incumbents improve the gap from 73% to 3%.
    You could try parameters that focus on finding good solutions, for example

    • set MIPFocus=1
    • experiment with Heuristics, for example, Heuristics=0.5
    • test the no-relaxation heuristic by setting the parameter NoRelHeurTime to the number of seconds the heuristic should run, for example, NoRelHeurTime=600

    Additionally, you could experiment with the parameter PrePasses to try to reduce the time spent in Presolve (without losing too much performance) or try to reduce Presolve with Presolve=1 (but this might already be the default here).

    Best regards,
    Marika

    1
  • Jesse Mommersteeg
    • Gurobi-versary
    • First Question
    • First Comment

    Hi Marika,

    Thank you so much for your quick response, I will try playing around with these parameters immediately ;)

    Regards,

    Jesse

    0
  • Jesse Mommersteeg
    • Gurobi-versary
    • First Question
    • First Comment

    In case anyone is dealing with a comparable situation and manages to find this post: the suggestion above worked and the solution is now found within 1000 seconds generally ;) A huge improvement!

    1

Please sign in to leave a comment.