[Couenne] Printing best solution found when time limit exceeded

[Ouanes, Nesrine]

Giacomo, Pietro,


"is the solution that you obtain feasible?"


Yes, it is. Although I was hoping couenne would give a better solution than the local solver does.


"Can you see any non-Cbc log line that reports the solution with 0.25 objective, or the 0.28 objective?"


There are no solver log lines, other than cbc and only the 0.25 Objective is reported.


I will try to improve my model and hope that couenne would find a better solution more quickly. Thank you very much for you fast and helpful reply!


Best,

Nesrine

________________________________
Von: Pietro Belotti <petr.7b6 at gmail.com>
Gesendet: Donnerstag, 26. April 2018 19:43
An: Giacomo Nannicini
Cc: Ouanes, Nesrine; couenne at list.coin-or.org
Betreff: Re: [Couenne] Printing best solution found when time limit exceeded

Just to add to what Giacomo correctly said: Couenne and Cbc have, for several reason, a different idea of what makes a solution feasible, and sometimes it happens in the opposite sense of what you observed (i.e. Cbc logging a worse solution than what Couenne holds). Can you see any non-Cbc log line that reports the solution with 0.25 objective, or the 0.28 objective?

What would be needed is a single methods for flexibility checking, which Cbc allows for but hasn't been implemented in Couenne yet.

Regards,
Pietro

On Thu, Apr 26, 2018 at 12:24 PM, Giacomo Nannicini <giacomo.n at gmail.com<mailto:giacomo.n at gmail.com>> wrote:
Nesrine,
is the solution that you obtain feasible?
In my experience, the log is wrong: Cbc (the branch-and-bound part of
Couenne) thinks that it found a feasible solution of value 0.25, but
Couenne internally rejects it according to its own tolerances.
At the end of the search, Couenne prints the best solution it found -
which is worse than the "fictitious" 0.25 solution.

Giacomo

On Wed, Apr 25, 2018 at 2:26 PM, Ouanes, Nesrine
<nesrine.ouanes at campus.tu-berlin.de<mailto:nesrine.ouanes at campus.tu-berlin.de>> wrote:
> Hello,
>
>
> I'm solving a NLP Problem with couenne. My model is written in Python and
> Pyomo.
>
>
> In my case, Couenne takes very long without exiting with an optimal
> solution. The best solution is given after a few iterations and doesn't
> change after hours of execution:
>
>
> Cbc0010I After 7100 nodes, 3364 on tree, 0.25239475 best solution, best
> possible 0.15 (14.72 seconds)
>
>
> Therefore, I've set the option time_limit to a smaller value (100 seconds),
> to be able to retrieve the best solution found so far and print it to a
> file. However couenne prints another solution than the one stated in the
> solver log:
>
>
> instead of 0.25239475, I get 0.281323177 as a value for the objective
> function.
>
>
> The value printed by Couenne is the same one I get if I solve my model using
> Ipopt.
>
>
> 1) Do you have any idea, why this occurs?  Is it possible that Couenne is
> printing the best solution found by ipopt, since it's used in Couenne?
>
>
> 2) Is there a way to obtain the better solution given in the solver log?
>
>
> I look forward to your reply.
>
>
> Best,
>
> Nesrine
>
>
>
>
>
>
>
>
>
>
>
>
> _______________________________________________
> Couenne mailing list
> Couenne at list.coin-or.org<mailto:Couenne at list.coin-or.org>
> https://urldefense.proofpoint.com/v2/url?u=https-3A__list.coin-2Dor.org_mailman_listinfo_couenne&d=DwICAg&c=Ngd-ta5yRYsqeUsEDgxhcqsYYY1Xs5ogLxWPA_2Wlc4&r=8FaVxIx99H879Ebkij38NpcRZoYLK-7QMl1LqyvBccA&m=TKDQWNLB4YWq5VR5iJ4Bpb5iCCr7hDVDR-UF_vMD9ao&s=s7IjlPTu9bAJDemY_xIfISefaKeijm2HjN-3_ozHi1c&e=

_______________________________________________
Couenne mailing list
Couenne at list.coin-or.org<mailto:Couenne at list.coin-or.org>
https://list.coin-or.org/mailman/listinfo/couenne

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/couenne/attachments/20180427/a8814f15/attachment.html>