Skip to main content

too high gap

Answered
Tanzin Ahmed
  • Gurobi-versary
  • First Comment
  • First Question

I received too high gap even after running for 1 hour. Could you please help, how can I reduce the gap?

0

Comments

7 comments

Date Votes
  • Riley Clement
    • Gurobi Staff

    Hi Tanzin,

    Non-convex MINLP can be pretty tough to solve to global optimality.

    As suggested by the log, setting tighter bounds on variables would be a good idea if you can.

    Our beta release for v13 is out, and there are significant improvements for non-linear models.  Can you give it a try and let us know how it goes?

    https://support.gurobi.com/hc/en-us/articles/39159696273809-How-do-I-access-the-Gurobi-13-beta

    - Riley

    0
  • Tanzin Ahmed
    • Gurobi-versary
    • First Comment
    • First Question

    Hi,

    I am now using temporary WLS license but solver is still taking v12 not v13. 

    0
  • Riley Clement
    • Gurobi Staff

    Hi Tanzin,

    How do you call Gurobi?  Are you using gurobipy or something else?

    - Riley

    0
  • Tanzin Ahmed
    • Gurobi-versary
    • First Comment
    • First Question

    I am using python [conda env:base]

    0
  • Riley Clement
    • Gurobi Staff

    Ah ok, the beta has not been published to Anaconda.  You can still install it in a conda environment with pip, i.e. with your conda environment activated (maybe create a new environment for v13), run

    python -m pip install gurobipy==13.0.0b1
    0
  • Tanzin Ahmed
    • Gurobi-versary
    • First Comment
    • First Question

    Hi,

    still too high gap in V13, it is reduced but not significant %. 

    Time limit reached Best objective 2.269228814313e+08, best bound 2.722777909074e+08, gap 19.9869%

    0
  • Riley Clement
    • Gurobi Staff

    Next recommendation I have is to consult the following resources, and see what can be done with parameters:

    Introduction to Performance Tuning
    Tech Talk - A Practical Tour Through Non-convex Optimization

    I would recommend experimenting with Presolve, MIPFocus, VarBranch, BranchDir and NoRelHeurTime to begin with.  These are documented in our Parameter Reference.

    Warning: Model contains variables with very large bounds participating
            in product terms.

    While this warning persists, I expect you will not be able to make good progress on the MIP gap.  To understand why, see the 33 minute mark of the Tech Talk linked above where Jaromił discusses McCormick Envelopes.

    - Riley

     

    0

Please sign in to leave a comment.

Didn't find what you were looking for?

New post