Multi-scenario returns infinity objective value for some feasible scenarios
Answered
I'm trying to debug some very strange behavior on a model with thousands of scenarios (each scenario corresponds to a subproblem in a Dantzig-Wolfe-like decomposition). All scenarios are minimizations, and should be feasible with finite optimum.
The overall model's status after optimizing is GRB.OPTIMAL, but for many (maybe all) of the scenarios, ScenNObjVal and ScenNObjBound are inf. Based on example code (e.g. https://www.gurobi.com/documentation/9.0/examples/sensitivity_py.html#subsubsection:sensitivity.py) this seems to indicate that the scenario is infeasible.
However, if I extract the scenario with singleScenarioModel and optimize that, the model is feasible and has a finite optimum just like I would expect.
-
Your expectations are correct. An overall model status of OPTIMAL means that there was at least one feasible scenario, but there can also be infeasible scenarios. Whether a particular scenario is feasible or infeasible can be seen by looking at ScenNObjVal and ScenNObjBound.
What you describe seems to indicate that there might be a bug in our code. But in order to find out, I would need to investigate what is going on. Can you share your model (preferably in *.mps file format)? Please send it to achterberg (at) gurobi (dot) com. Can you reproduce the wrong behavior using the command line tool gurobi_cl applied to your *.mps file?
Thanks,
Tobias
0 -
I finally found the cause of the issue. It actually has nothing to do with the MIP scenario analysis feature. It is a bug in handling SOS constraints with continuous variables inside the symmetry detection code.
The bug will be fixed in the upcoming Gurobi 9.0.3 fixpack release.
Thanks again for sending the model and pointing out the issue.
Best regards,Tobias
0
Please sign in to leave a comment.
Comments
2 comments