[Scip] about implementing the branch and price under Windows

Marco Lübbecke marco.luebbecke at rwth-aachen.de
Fri Mar 28 15:35:32 CET 2014


Hi,

that GCG is faster than (plain) SCIP (on certain instances, under certain
assumptions) is because in GCG the branch-and-price code is already
functional, while in SCIP you would need to first code your
pricer/branching. Then, both codes would probably be at par; however, GCG
is made for *generic* problems, that is, there is an overhead probably not
needed for a particular problem -- and GCG may even lose against SCIP with
B&P implemented.

Yet, you can try GCG on small examples of your problem (providing LP files
with an "original" MIP formulation of your problem to GCG) and try what
happens. When it starts generating columns, all is well; if the bound at
the root node is better than SCIP's at the root node, things are promising.
Then go, and use SCIP to implement your B&P code (if you needed the
additional GCG features, you would already know). If GCG gives optimal
solutions to your instances, be happy, and send us a reference of your
application to be included in the website ;-)



Final advise: switch to Linux first (regardless which way you go ;-))

--Marco




2014-03-28 15:04 GMT+01:00 Xiangyong Li <lixiangyong at 163.com>:

> Dear all,
>
> I want to implement my own branch and price algorithms under Windows.
> Should I use SCIP or GCG as a framework for implementing the branch and
> price?
>
> The "INSTALL" file of GCG said GCG is currently only tested on GNU/Linux
> and may not compile under Microsoft Windows.
>
> The thesis of Gerald Gamrath demonstrated while implementing branch and
> price, GCG works much faster than the SCIP.
>
>
> Can I use GCG to implement my own branch and price algorithm under windows?
>
>
> Thanks,
>
>
> Xiangyong Li
>
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> http://listserv.zib.de/mailman/listinfo/scip
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.zib.de/pipermail/scip/attachments/20140328/0c44a217/attachment.html>


More information about the Scip mailing list