Increasing NumericFocus makes problem infeasible
Awaiting user inputHello,
I have a simulation model in Python where I iteratively solve an optimization model using gurobipy. In one of the iterations, I encountered the problem discussed here, where I had a nonzero gap for the optimal solution. As the workaround proposed, I set the value of NumericFocus to 2 (NumericFocus = 1 did not help in my case). That solves the issue of the nonzero gap.
However, in another iteration (i.e., same model, different set of values), my model becomes infeasible. I set the NumericFocus back to the default value and the problem becomes feasible. For this particular iteration, the coefficient statistics are as follows:
Matrix range [1e-02, 9e+01]
Objective range [3e+01, 1e+02]
Bounds range [1e+00, 9e+01]
RHS range [1e+00, 2e+02]
Given the magnitude of the coefficients, I was not expecting any numerical issues. I know how to go around this issue in this specific case but given that increasing NumericFocus is supposed to "shift the focus towards being more careful in numerical computations," now I am concerned if my problem is infeasible to begin with. So now, my question is, is it possible increasing NumericFocus makes a feasible problem infeasible? or the problem is in fact infeasible?
I am using gurobipy 9.1.1. I have the mps of the problem here.
Any help is much appreciated.
-
This issue has been handled in an internal ticket and will be fixed in an upcoming release of Gurobi.
0 -
Hi Jaromil, I am using Gurobi 10.0.1 and having a similar issue; with smaller NumericFocus my model is being solved, but with higher NumericFocus my model is found to be infeasible. Was the issue fixed in 10.0.1 ?
0 -
Hi Kazi,
This particular issue has been fixed already for v9.1.2.
Could you please share the model where you experience the problematic behavior? In order to generate a model file, please use the write method. Note that uploading files in the Community Forum is not possible but we discuss an alternative in Posting to the Community Forum.
Best regards,
Jaromił0
Please sign in to leave a comment.
Comments
3 comments