Originally posted by: SystemAdmin

[dfearing said:]

Hello,

Does anyone know a good way to debug infeasibilities in a solution produced by an HeuristicCallback?  I have a large scale MIP where in most cases, my HeuristicCallback produces solutions that are accepted as the new incumbent.  But, in a few of my largest and most complex scenarios, the solution produced by the HeuristicCallback is not accepted as the incumbent (even though there is no incumbent at the time).  I'm hypothesizing this is due to infeasibilities in my heuristic solution, but I'm having a tough time tracking those down.  Ideally I would like a way to get CPLEX to provide information regarding which contraints are violated when attempting to set a solution using an Heuristic Callback.  If instead there is a way to pass IloVars and corresponding values to a separate validation method that would also work, but so far I haven't been able to find any API of the kind.  Any help (or suggestions) would be greatly appreciated.

Thanks,
Doug
#DecisionOptimization
#OPLusingCPLEXOptimizer

Originally posted by: SystemAdmin

[jfk said:]

hello there,
I guess you could:
1. add constraints to the original prb: some of the variables set to the solution you got - many enough so you don't have to wait long - and run CPLEX without your heuristics and with the conflict detection facility
2. you could give your solution as a hot-start to CPLEX. If I remember well it checks first and gives you the conflicts

anyway, if you are using your code in an application it may not work so nicely or take time...
are you working on an application for an end user or is it a prototype of some sort?

I hope it helps...

cheers
#DecisionOptimization
#OPLusingCPLEXOptimizer

Originally posted by: SystemAdmin

[dfearing said:]

Hi,

Thanks for the suggestions.  I ended up writing my own validation routine which was able to track down the problem, though your second suggestion would have been interesting to try as well.  The code I'm writing is for an academic research project, so I have a lot of flexibility.

Thanks again,
Doug
#DecisionOptimization
#OPLusingCPLEXOptimizer