[Cbc] Cutlength value

[John Forrest]

Yves,

The intended use for cutlength was just for efficiency not for 
accuracy.  However your problems seem to pose more problems on accuracy 
because you have general integer variables, continuous variables and 
many <= or >= rows.  The computations in CglGomory allow for the slight 
relaxation of a cut rhs when there are more variables in cut, BUT it did 
not look at the extra work done for >= rows at bound - this can lead to 
cancellation due to very small values, so the number of variables in cut 
does not totally reflect the possible inaccuracy.  Well .... anyway 
something like that.

I have updated CglGomory to try and reflect this.  It does not seem to 
do any harm in my tests, so I am checking the change into trunk.  If 
anyone notices significant degradation in their work, then tell me and 
go back to previous way by defining CGL_GOMORY_OLD_RELAX in your configure.

John Forrest



On 03/12/14 17:59, Yves Touchard wrote:
> Hello,
>
> There is a constraint file 
> (https://www.dropbox.com/s/e1x1xilj4gr200u/timing1.lp?dl=0) which is 
> not solved when cutlength is 40:
>
> /bin/cbc timing1.lp threads 104 ratio 0.05 cutlength 40 multiple 4 
> dives 301 branch printingOptions rows solution timing1.sol/
>
> but has an optimal solution if cutlength is 30:
>
> /bin/cbc timing1.lp threads 104 ratio 0.05 cutlength 30 multiple 4 
> dives 301 branch printingOptions rows solution timing1.sol/
>
> Not sure to know which value is fine for my set of problems.
>
> Regards
>
> Yves
>
> [Tests done with Rev #2100 and #2099 of Cbc trunk]
>
> ------------------------------------------------------------------------------
> This message and any attachments (the "message") are confidential and 
> intended solely for the addressee(s). Any unauthorized use or 
> dissemination is prohibited. E-mails are susceptible to alteration. 
> Neither DxO Labs nor any of its subsidiaries or affiliates shall be 
> liable for the message if altered, changed or falsified.
> Ce message et toutes les pièces jointes (ci-après le "message") sont 
> confidentiels et établis à l'intention exclusive de ses destinataires. 
> Toute utilisation ou diffusion non autorisé est interdite. Tout 
> message électronique est susceptible d'altération. DxO Labs et ses 
> filiales déclinent toute responsabilité au titre de ce message s'il a 
> été altéré, modifié ou falsifié.
>
>
> _______________________________________________
> Cbc mailing list
> Cbc at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/cbc

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/cbc/attachments/20141205/2e4be0b1/attachment.html>