Originally posted by: Eumpfenbach

I have a difficult to solve MIP, but I have a heuristic that provides a very good initial starting solution. I want to find the optimal solution down to a fraction of a percent, though. The starting solution is within a couple percent of the optimal, but maybe 100% away from the root relaxation.

Any suggestions on parameter settings let me take advantage of this near optimal feasible starting solution? I am thinking that setting the search emphasis to best bound is obvious. Is there anything else?
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

You're adding the heuristic solution as a MIP start, right? You could try turning on solution polishing as early as possible (multiple parameters control solution polishing; if you look in the docs under "Topical list of parameters", they have their own subsection). You could also set the RINS heuristic frequency (RINSheur parameter) so that it runs at most/all nodes.

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: Eumpfenbach

I tried the RINS frequency. Found a lot of feasible solutions but they were all lower quality than my initial one. Even though my initial is reasonably near-optimal, the true optimum might be far away in the tree. So those solutions don't really do a ton for me. Just really need to get that bound down quicker (but I guess that is 90% of the battle).

I'll read up on solution polishing. Thanks.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

Did you try changing MIPEmphasis to best bound?

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: Eumpfenbach

Yes, but weirdly, that increases solution time. I guess it 100% forces cplex to choose the node with the best bound, when it is better to just let it do it's own thing and get out of the way.

Thanks for the suggestions.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

Did you try aggressive cut settings? If I understand correctly then your LP relaxation is poor and you want to improve that? Aggressive cuts may help doing that.
#CPLEXOptimizers
#DecisionOptimization