[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Bug-glpk] FPUMP heuristic pass bug GLPK4.60
From: |
Andrew Makhorin |
Subject: |
Re: [Bug-glpk] FPUMP heuristic pass bug GLPK4.60 |
Date: |
Wed, 06 Apr 2016 14:19:58 +0300 |
> > OK. It is feature but it seems not a good feature.
>
> > I tried to modify the npass (default 5) to npass <=10 and fpump finds
> > a solution in npass 9 and it starts again from pass 1 to pass 10. It
> > takes a long time to finish the job.
> >
>
> Well, all primal heuristic work in this way--usually it is reasonable to
> spent some time to find a good initial solution rather than to spent a
> lot of time to close a large integrality gap during the b&b search. The
> only question is when to stop. May be to add a time limit like for PROXY
> heuristic. On the other hand, if your instance is not hard, i.e. if the
> b&b solver finds an initial solution quickly, using primal heuristic(s)
> is not needed.
>
Also you can run glpsol with the option "--save filename" to save the
best integer feasible solutions found to a file and cancel the solver at
any time you want. Then you may use that saved solution as an initial
one with the option "--use filename",