[SCIP] resolve without pricing first in price&cut loop

[Jakob Witzig]

Hi Jens,

your idea is not so easy to realize with SCIP.

A possible way could be to do not price variables and to return 
SCIP_DIDNOTRUN. The problem is, that every time cuts are added to the LP 
your prices will be called. That means you have to decide within your 
pricer whether you want to price or not. Moreover, you have to ensure 
that your pricer runs if the LP is infeasible.

Another possible solution could be to extend your separation method with 
an external method that returns whether cut were added to the LP or not. 
Afterwards, based of the return value of the method your pricer can 
decide if new variables are needed.

Best,
Jakob


On 20.10.2015 09:48, Jens Leoff wrote:
> Hi,
>
> I have a Branch, Price & Cut algorithm and in some cases I observe the 
> following behaviour:
> When solving the LP for some node I end up in a loop where
>
> 1 violated constraint is found and added to the lp.
> The LP is resolved with column generation, but no new variables are 
> found.
>
> As pricing takes a lot longer than separation in my case, solving the 
> lp takes very long.
> Often no new variables are added in pricing, so I expect considerable 
> speed up when solving the Price & Cut loop the following way:
>
> After separation the LP is solved without pricing and a new separation 
> round is started.
> Only if the lower bound increased or no further violated constraints 
> are detected the lp is resolved with pricing.
>
> Is this currently possible with SCIP?
>
> Kind regards
>
> Jens Leoff
>

-- 
Jakob Witzig

Konrad-Zuse-Zentrum für
Informationstechnik Berlin (ZIB)

Division Mathematical Optimization and Scientific Information
Research Group Mathematical Optimization Methods

Takustrasse 7
14195 Berlin

Tel. : +49 (0)30 84185-416
Fax  : +49 (0)30 84185-269
email: witzig at zib.de