[Cbc] CBC rejecting its own solution as MIP start

[Haroldo Gambini Santos]

Hi,

I confirm the bug.

It seems that when mipstart is informed (and a good bound is available 
at the start) cut generators "cut too much" in this particular problem 
and it becomes infeasible.

This fixation of the "one" variable should not interfere...

Cbc0045I Trying just fixing integer variables.
Cbc0045I MIPStart provided solution with cost 181978
Cgl0002I 301 variables fixed
Cgl0003I 63 fixed, 0 tightened bounds, 275 strengthened rows, 0 
substitutions
Cgl0003I 0 fixed, 0 tightened bounds, 124 strengthened rows, 0 substitutions
Cgl0003I 0 fixed, 0 tightened bounds, 146 strengthened rows, 0 substitutions
Cgl0000I Cut generators found to be infeasible! (or unbounded)
Pre-processing says infeasible or unbounded
Total time (CPU seconds):       0.28   (Wallclock seconds): 0.28


Em 12/01/2017 13:22, Rudi Araújo escreveu:
>
> John,
>
> Regarding the other thread (assertion failure), your fix seems to 
> work, thanks.
>
> I have unearthed this week-old thread because we came across a 
> different situation where CBC reports infeasibility, even though the 
> problem is indeed feasible. See files here:
>
> https://drive.google.com/open?id=0B-5kPpSmX5imVllVaUdxVWY3TGc
>
> Breaking it down:
>
> ·cbc -import test.lp -mipstart ms.before.txt
>
> Reports infeasibility.
>
> ·cbc -import test.lp -mipstart ms.txt
>
> Solves problem to optimality, but rejects MIP start.
>
> The files ms.txt and ms.before.txt have exactly the same solution.
>
> We traced the issue to this constraint:
>
> __one = 1
>
> If we remove it, the .before MIP start is accepted and the model is 
> solved to optimality.
>
> In this particular case, we can do without the __onevar, but in the 
> general case we add it because it is sometimes convenient for us to 
> pass a constant term in the objective function.
>
> Can you take a look at it?
>
> Thanks,
>
> Rúdi Araújo.
>
> *From:*John Forrest [mailto:john.forrest at fastercoin.com]
> *Sent:* quinta-feira, 5 de janeiro de 2017 15:29
> *To:* Rudi Araújo <rudi.araujo at siscog.pt>
> *Cc:* 'Tiago Maduro Dias' <tiago at siscog.pt>; 'Luís Borges de Oliveira' 
> <lbo at siscog.pt>
> *Subject:* Re: CBC rejecting its own solution as MIP start
>
> Rudi,
>
> On 04/01/17 17:39, Rudi Araújo wrote:
> >
>
> > John,
>
> >
>
> >
>
> >
>
> > Thank you the fast reply.
>
> >
>
> >
>
> >
>
> > The .before hint did the trick for the LP I sent you. However, we
>
> > came across a different situation where the mipstart is accepted but
>
> > CBC reports infeasibility and exits without finding an integer
>
> > solution, even though it exists:
>
> >
>
> >
> That was more interesting - a bug which was unlikely to occur in 
> normal use, but could.  As the code had a cutoff value, it added that 
> as a constraint in preprocessing and there was a bug there.
>
> > By the way, what is the reason for building the initial solution only
>
> > after pre-processing by default? Would it make sense to do it the
>
> > other way around - by default before, with an option to do it only
>
> > after?
>
> >
>
> >
>
> I did not write that bit of code and don't really want to do too much 
> to it.  Some information which could get used in heuristics can not 
> easily be passed across when creating solution before - so on the 
> whole it is more efficient to try and do work on preprocessed problem.
> >
>
> > One last thing: trunk appears to be broken. We tried compiling it and
>
> > tripped over this error:
>
> >
>
> >
> My error (you picked it up before I was told off this morning after 
> standard Coin builds were done).  I am doing some development in trunk 
> and thought I might as well commit some extra changes instead of 
> taking them out temporarily.  However I forgot that I would also need 
> to commit changes from CoinUtils.
>
> John Forrest
>
>
>
> _______________________________________________
> Cbc mailing list
> Cbc at list.coin-or.org
> https://urldefense.proofpoint.com/v2/url?u=http-3A__list.coin-2Dor.org_mailman_listinfo_cbc&d=CwICAg&c=Ngd-ta5yRYsqeUsEDgxhcqsYYY1Xs5ogLxWPA_2Wlc4&r=pLOfVNEEHf-xhIqn1-uzYcZ6Q7UefG6Bg6rXCKTMiAA&m=oSk3Nz0-nrEwX6MYB8eXDIM-Y0Jlja90xNe1YzpB5PM&s=GjKkwW0LZkNDCwXZMjGBa3SQjQxO5zSjElNzcMAI2Ik&e=

-- 
==================================================
Haroldo Gambini Santos
D.Sc, Computer Science
Universidade Federal de Ouro Preto
http://www.decom.ufop.br/haroldo/

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/cbc/attachments/20170112/5af6b7d0/attachment-0001.html>