Originally posted by: lyxthe

Hello, 

as I've allready explained in several previous posts (as previous post) I'm trying to solve a difficult MILP problem and for the moment I'm not trying to find the optimum solution, but only the best one I may obtain after a given time of process through the parameter TimLimit. For the moment I set this limit to 660 or 720 s. 

In an answer of my previous question, one told me about the PolishAfterTime parameter. So I tried it and was surprised that this parameter has an extreme efficiency on my MILP problem. In fact I have the best solutions when dividing the time by 2, one half (400s) to solve then the other half to polish. 

I would like to know if that meens some things about my problem, and if it may lead to other good parameters to make even more efficient the computing. (allready try some of them, MIPEmphasis to 1, and Probe to 3 espetially)

Thanks for your attention.

Lyxthe

Originally posted by: TobiasAchterberg

I am happy to learn that this works for you. Polishing is mostly about local search. If this works for you, this means that in your problem class it is relatively easy to find feasible solutions. So, the set of feasible solutions within the full set of vectors in the domain of the variables is relatively dense (compared to other MIP problems), which enables the local search heuristics to go from one to the next, each time improving the objective value a little bit.

One part of polishing is the RINS heuristic. As I said in the other thread you should also try to play with the RINSHEUR parameter, independently of polishing. Maybe this gets you already to the same solution quality as polishing.

Other than this I don't think that there is any parameter that could have a systematic positive effect on all problem instances that feature a large set of feasible solutions.

Tobias