Any configurable parameters for the nonconvex quadratic solver?

Answered

Comments

2 comments

  • Jaromił Najman

    Hi Niclas,

    Gurobi's performance seems pretty good given the number of bilinear terms and the problem size.

    Do Gurobi's MIP-related parameters apply to the nonconvex spatial branching algorithm as well? Or are there other parameters I can set, and if so where can I find information about them?

    All MIP specific parameters also directly affect non-convex problems. There are a few specific parameters, which focus on quadratic problems only, e.g., PreMIQCPForm. You can have a look at the parameter descriptions to get more information. Parameters aiming for quadratic problems often have the capital letter "Q" in their name.

    I have also noticed that you are using version 9.0.3. You should try upgrading to the latest Gurobi version 9.1.1 if possible. In the latest version, you could try the new No Relaxation Heuristic, which specifically focuses on finding feasible points before the root relaxation has been computed. The documentation of most important parameters might help here.

    Best regards,
    Jaromił

    0
    Comment actions Permalink
  • Niclas Mattsson

    Thank you, that was helpful. Now I know I can play around with all the MIP parameters, and if they don't seem to do anything then it's probably something problem specific.

    I actually had 9.1.1 installed but for some reason Julia's Gurobi interface wasn't detecting the latest version, but I resolved that issue yesterday. I'll try 9.1.1 on my problem and see if it helps.

    I have another unrelated question but I'll start another thread for that (to help maintain googlability). 

    0
    Comment actions Permalink

Please sign in to leave a comment.

Powered by Zendesk