[SCIP] Ubiquity Generator for branch-and-cut-and-price

Yuji Shinano shinano at zib.de
Mon May 20 22:33:38 CEST 2019


Hi list,

Eddie and I discussed the subject in person some time ago, and a SCIP team member
asked me to share them in this ML. So, I forward them in below:

-----
Hi Eddie,

I think it is possible by using ug[SCIP,*]. It is also possible to branching on constraints.
SCIP can do it and ug[SCIP-*,*]-libraries can transfer it automatically, if you implemented it in SCIP.
The primal bound also distributed automatically. Please check STP example code in SCIP application
and UG application.

Cheers,
Yuji


On 2019/05/16 0:27, Edward Lam wrote:
Hi Yuji,

Actually I don’t need to transfer the variables to the other CPUs because they can be generated again easily. I just need to communicate some data on the open nodes and the current primal bound. The node data will contain information on the branching decisions of its ancestors. I’m branching on constraints but all I need is some node data to create the same constraints on the other CPUs.

So my question is: is there a way to just communicate the data of the open nodes and the current primal bound? If not, any tips on coding this myself? Thanks in advance.

Cheers
Eddie

On 15 May 2019, at 6:31 pm, Yuji Shinano <shinano at zib.de> wrote:

Hi Eddie,

Unfortunately, Gregor is right. UG does not support to adding variables during solving process.

Best,
Yuji


On 2019/05/08 3:03, Edward Lam wrote:
> Hi all,
>
> I’m trying to solve a problem using distributed parallel branch-and-cut-and-price on a cluster. Before I get started, I wanted to know whether it is possible using UG and SCIP. I was just chatting to Gregor who suggested that UG requires all variables to be declared up front. Can someone confirm?
>
> Also some documentation and/or examples would be great. :) Thanks in advance.
>
> Cheers
> Eddie
>
>
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> https://listserv.zib.de/mailman/listinfo/scip



More information about the Scip mailing list