GRB.MAXIMISE identifies lower objective function values than GRB.MINIMIZE. How can this be avoided?
Ongoing
Hello,
I have a sufficiently big optimization problem with many bilinear terms in the objective function (MIQP). The issue that I face is that when I use GRB.MINIMIZE, Gurobi only identifies a single solution say -1000 but when I use GRB.MAXIMISE, Gurobi identifies 3 solutions say -1000, -1050 and -1084. Since the latter is a maximization, it converges to -1000. However, my question is that when, clearly, solutions with -1050 and -1084 exist, then why does GRB.MINIMIZE fail to reach them?
I believed that Gurobi can handle non-convexities due to bilinear terms. I tried varying the seed but it doesn't work. What can be done to build trust in such an optimization?
-
Hi Gaurav,
One way to check what is happening here is to create and solve a fixed model where the upper bound and lower bound of each variable is fixed to the value of the variable in the solution with the objective value -1084. If the fixed model is infeasible, you can inspect the infeasibility, for example, by running the computeIIS method.
Please take a look at this article for more details: How do I diagnose a suboptimal objective value returned as optimal by Gurobi?
Best regards,
Simran0 -
Hi Simranjit
Thank you very much for your reply. However, if GRB.MAXIMISE is able to identify the -1050 and -1084 as sub-optimal solutions then I do not understand why would the fixed model (fixed for the variable values corresponding to -1084 case) return an infeasible model.
I have another issue. I tried retrieving the variable values for suboptimal solutions of the maximisation problem using the SolNumber parameter. During the solution process Gurobi prints that it was able to find 4 solutions: -416K, -433K, -433K and -433K. However, when I set the Solution Number to 3 and retrieve the Obj Value, by using PoolObjVal, it still returns -416K instead of -433K. What could be wrong?
RegardsGaurav
0
Please sign in to leave a comment.
Comments
2 comments