Originally posted by: PaoloL

I am solving a problem with numerical stability issues and I am trying to get the condition number of the problem.
I use the methods of the class IloCplex:
cplex.getQuality(IloCplex::Kappa)
cplex.getQuality(IloCplex::ExactKappa)
But the value returned is always 0 or -1..

It is a Binary Program: I have two sets of binary variables and some numerically difficult constraints.. I add them to an IloModel that is extracted to cplex.
The value of Kappa before the call to cplex.solve() is always 0 and after the solving of the model it is always -1.
What could be the cause of this behavior?
(I am using cplex 12.1 with C++)

Thanks in advance,
Paolo
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

These measures are documented in the optim.cplex.solutionquality section of the callable library reference manual (i.e., the C API), and they apply only to simplex solutions, not to MIP solutions. If the model is a MILP, you can try solving it as a MILP, then using IloCplex::solveFixed to solve the linear portion with the integer variables fixed at their optimal values, then get the kappa value for the final simplex basis. If all the variables are integer, I suspect the best you can do is to relax the integrality restrictions, solve it as an LP, and then check kappa.

/Paul

Mathematicians are like Frenchmen: whenever you say something to them, they translate it into their own language, and at once it is something entirely different. (Goethe)
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

Please note that we added a "MIP kappa" feature to CPLEX 12.2. Since the academic initiative version was just released, you should be able to download it and try it. The "MIP kappa" feature has to be activated before solving the MIP (it makes the solving process a little bit slower because the quality data needs to be calculated and collected). Then, displaying the solution quality will give you some information regarding the LP bases that have been encountered during the MIP solve. For example, you will get the number of LPs with good/medium/poor numerical properties.

Tobias
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: PaoloL

I'll try both by relaxing integrality in cplex 12.1 and by using the MIP kappa feature as soon as I get cplex 12.2

Thanks a lot,
Paolo
#CPLEXOptimizers
#DecisionOptimization