[Cbc] Crash when disabling preprocess

[John Forrest]

Nicolas,

I am unable to reproduce the errors (on linux).

Crash.lp says it is infeasible - even just using clp to solve 
continuous.  With scaling it says there is an infeasibility of 5.0e-5 
while with scaling off there are two summing to 0.002 which might be 
easier to track down.  I suspect tiny infeasibilities are the reason you 
got the messages with preprocessing.

If you can track down infeasibility, it may be worth adding extra 
variables or relaxing rhs to keep feasible - and then you might be able 
to go back to using preprocessing.

Erreur.lp looks alright (although coefficients of -1.0e15 are not 
healthy - but I think they transfer to objective).

John Forrest
On 06/07/16 16:01, nicolas.derhy at engie.com wrote:
>
> Hi everyone,
>
> I want to report problems with CBC when I disable preprocessing.
>
> The version I use is CBC 2.9.8 (x86) on Win7.
>
> In the command line, I add “-ratio 0.05 -seconds 30” (for example : 
> cbc "Crash.lp" -ratio 0.05 -preprocess off -seconds 30 -solve 
> -printingOptions all -solution "solution_crash.txt")
>
> I have attached two files :
>
> -Crash.lp : CBC stops suddenly
>
> -Error.lp : first, it seems optimization was completed successfully. 
> But the objective function is clearly wrong and when I export 
> solution, optimal value of some binary variables are non integer 
> (vDebitNul(Interco_0,Noeud_9,Noeud_2) = 0.168948)
>
> I tried to modify some parameters (preprocess, presolve) but nothing 
> works.
>
> If I activate preprocessing, everything works.
>
> I disabled preprocessing previously because I obtained weird results 
> and could not use optimal solution provided by CBC  in some cases. I 
> noticed CBC generated the following message :
>
> “Cgl0013I Postprocessed model is infeasible - possible tolerance issue 
> - try without preprocessing”
>
> So, it was the reason for disabling preprocessing. After that (and 
> until these new problems), everything worked better.
>
> If anyone has an idea to improve stability of CBC when preprocess is 
> off, it would be great.
>
> Best regards,
>
> Nicolas
>
> ENGIE Mail Disclaimer: http://www.engie.com/disclaimer/disclaimer-fr.html
>
>
>
> _______________________________________________
> 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/20160707/8ce52622/attachment.html>