Crazy MIP search tree exploration

Answered

Comments

1 comment

  • Eli Towle

    Gurobi is only deterministic with respect to solving the same model on the same computer with the same parameters. Even changing the order of variables/constraints can lead to a different solution path. While it is unfortunate that the model takes so much longer to solve on your PC, this difference in solve time can likely be attributed to performance variability. There are much more extreme cases of performance variability, where (for example) Gurobi happens to run a key heuristic routine at the root node on one machine, resulting in a near-optimal incumbent solution very early in the solving process. However, this heuristic never runs on another machine, and the solver spends hours and hours working through a huge branch-and-bound tree. You can read more about this here.

    What happens if you solve the model with different values of the Seed parameter? Changing the Seed parameter changes Gurobi's solution path, allowing you to see how path changes can affect the solve time. Restarting your machine and re-installing Gurobi shouldn't change the solution path, since the Gurobi library and machine environment remain the same.

    I see you have a support ticket open about this issue, so we'll keep you updated there.

    0
    Comment actions Permalink

Please sign in to leave a comment.

Powered by Zendesk