[Scip] Pricing generates an already existing column

[Cristina Núñez del Toro]

Hello Martin,

I think I understand the problem. I just have a couple of questions:


   1. I can not see clearly how this situation can be fixed by implementing
   my own branching rule
   2. I have checked that there some branching rules that are currently
   available on SCIP.  How can I know which branching rule does SCIP uses
   by default? Because these branching rules seem to be quite general, I would
   like to try first on swapping among them. Is there a way to make a callback
   to force the use of one of them instead of the default one? If not, is the
   only way to implement my personalized branching rule by adding explicitly a
   constraint handler?

Thanks in advance.



---
Cristina Nuñez


2014-05-28 20:46 GMT+02:00 Martin Bergner <bergner at or.rwth-aachen.de>:

> Hi Cristina,
>
> somewhere in your detailed explanation, you write:
> > Even
> > that your comment about branching was focus on variables with upper
> > bound = 0, now I am thinking that maybe the branching rule is somehow
> > forbidding some of the variables. How can I check this behaviour?  I do
> > not use any special rule for the branching. All the branching procedure
> > is set as default.
> This is probably the issue:
> When you don't specify a branching rule, SCIP branches on the priced
> variables. If they are binary, they are fixed to one in one branch and
> zero in the other. If the variable is fixed to zero, then nothing in the
> pricing problem forbids it from being generated again.
>
> Suppose now you have a non-degenerate variable in the basis that is
> removed by fixing it to zero. Since nothing in the pricing forbids it
> from doing so, it regenerates that variable that has been fixed. After
> all, this variable is highly attractive (negative reduced cost) and
> should be put into the basis, it was there before it was forcibly
> removed by branching.
>
> Do you see the problem? Whenever SCIP fixes a variable to zero, your
> pricing problem generates it again (and you thus have the problem you see).
>
> The remedy is: You have to (and in fact everyone  when using
> branch-and-price with SCIP with integral priced variables should)
> implement your own branching rule (maybe Ryan-Foster or something more
> general, depending on your problem at hand).
>
> I hope I could clarify the issue.
>
> Regards,
> Martin
>
>
>


-- 
---
Cristina Nuñez
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.zib.de/pipermail/scip/attachments/20140603/990d8630/attachment.html>