[Cbc] CBC misses optimal solution depending on activation of cuts

[Haroldo Santos]

Thanks for the super fast fix.

On Thu, Oct 20, 2011 at 8:23 AM, John Forrest
<john.forrest at fastercoin.com>wrote:

>  Haroldo,
>
> Thanks for finding this.  Because only one variable had a cost, Cbc decides
> in its wisdom that it is worth trying strong branching on that variable even
> if it is satisfied.  This got into a part of the code where it expects
> ceil(value)-value to be nonzero and bad things happened.
>
> Should be fixed in trunk and 2.7
>
> John Forrest
>
>
> On 19/10/11 21:31, Haroldo Santos wrote:
>
> Dear Forrest/Ted,
>
> I have this surprisingly small MIP (few hundred of variables and
> constraints)  where CBC misses the optimal solution if I deactivate cuts.
>
> http://www.decom.ufop.br/haroldo/files/C_I_6_4_S_1-124_1.lp
>
> CPLEX and GLPK instantly found  the optimal solution (77) no matter if
> cuts/heuristics are activated or not.
>
> Cbc, however, found a sub-optimal solution (136 or 124, depending on the
> machine) when I deactivate cuts and declares is as optimal. There are no
> very large/small coefficients so this is not due to numerical instability.
>
> cbc C_I_6_4_S_1-124_1.lp cuts off solve      (wrong result)
>
> If I call cbc with cuts
>
> cbc C_I_6_4_S_1-124_1.lp cuts on solve     (correct result)
>
> Cheers,
>
> Haroldo
>
> --
> =============================================================De
> Haroldo Gambini Santos
> Computing Department - Universidade Federal de Ouro Preto - UFOP
> email: haroldo [at ] iceb.ufop.br
> home/research page: www.decom.ufop.br/haroldo/
>
> "Computer science is no more about computers than astronomy
> is about telescopes." Edsger Dijkstra
>
>
>
> _______________________________________________
> Cbc mailing listCbc at list.coin-or.orghttp://list.coin-or.org/mailman/listinfo/cbc
>
>
>
> _______________________________________________
> Cbc mailing list
> Cbc at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/cbc
>
>


-- 
=============================================================
Haroldo Gambini Santos
Computing Department - Universidade Federal de Ouro Preto - UFOP
email: haroldo [at ] iceb.ufop.br
home/research page: www.decom.ufop.br/haroldo/

"Computer science is no more about computers than astronomy
is about telescopes." Edsger Dijkstra
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/cbc/attachments/20111020/95752ee5/attachment.html>