IIS COMPUTATION AND CONSISTENCY
Answered
I’m working on a MIP optimization problem where the constraints have a fixed structure, but some parameter values (initial values), which are used in the constraints, are provided by an end user through an input file.
Since the user is not familiar with optimization, they may sometimes specify combinations of numerical values that make the model infeasible. Because these possible mistakes fall into a finite and known set of cases, I would like to give the user helpful feedback explaining why the model is infeasible, ideally pointing to the specific mistake so they can correct it without diving into technical details.
My idea is to use the IIS computed by Gurobi, process it to identify the violated constraint logic, and then train a classifier to map IIS patterns to user-friendly error messages.
To do this, I need to understand the behavior of Gurobi's IIS extraction. I have two questions:
1. Is the IIS computation deterministic?
If I provide the same input file and therefore generate the same model, will Gurobi always return the same IIS, or could it return different IISs (still valid, but not identical) in different runs?
2. What happens with different input files that violate the same logical rule?
If two different input files lead to two different models, but both violate the same logical constraint pattern (with different numerical parameter values), should I expect Gurobi to return the same IIS structure, or is it likely that the IIS will differ run-to-run or instance-to-instance?
Thanks in advance for any clarification.
-
-
Gurobi Staff
Hi Lorenzo,
Is the IIS computation deterministic?
Yes.
What happens with different input files that violate the same logical rule? …should I expect Gurobi to return the same IIS structure?
No, you can not expect this. Even changing the Seed parameter and using the same input file might lead to a different IIS being calculated.
- Riley
0 -
Please sign in to leave a comment.
Comments
1 comment