[Cbc] confusing solution

[John Forrest]

Mike,

Fix is in svn 2.9 - won't be in a release just yet.

John Forrest
On 10/02/15 13:41, Mike Steglich wrote:
> Hi John,
>
> I tested Cbc  2.9.2 on OS X with the attached free mps file and found 
> the same issue for this maximisation problem:
>
> /Coin0505I Presolved problem not optimal, resolve after postsolve
> Cgl0013I Postprocessed model is infeasible - possible tolerance issue 
> - try without preprocessing
> 7 relaxed row infeasibilities - summing to 1604
> 7 relaxed row infeasibilities - summing to 1604
> 7 relaxed row infeasibilities - summing to 1604
>
> Result - Optimal solution found
>
> Objective value:                56640.00000000
> Enumerated nodes:               0/
>
> It is shown that an optimal solution was found ... but as for the 
> example I sent Yesterday .... one of the non-negative variables 
> (x[1,1]) is shown in the solution file with a negative value.
> Without preprocessing the problem is stated as infeasible.
>
> Thanks,
>
> Mike
>
>
>
>> Am 10.02.2015 um 11:33 schrieb John Forrest 
>> <john.forrest at fastercoin.com <mailto:john.forrest at fastercoin.com>>:
>>
>> Mike,
>>
>> The problem is due to the elements of 1.0e9 in model. In 
>> postprocessing this led to values of 1.0e-9 which is just on a 
>> particular tolerance.
>>
>> Without preprocessing my tests gave correct solution.
>>
>> It was a bug - I have made postprocessing more robust in 2.9.
>>
>> John Forrest
>>
>> On 09/02/15 11:39, Mike Steglich wrote:
>>> Hello,
>>>
>>> I've got an interesting behaviour with the attached mps file. When I 
>>> run this problem with the CBC binary (2.8.9 on OS X and 2.9.0 on 
>>> Win8.1) then I got the following information:
>>>
>>> ...
>>> /Coin0505I Presolved problem not optimal, resolve after postsolve
>>> Cgl0013I Postprocessed model is infeasible - possible tolerance 
>>> issue - try without preprocessing
>>> 16 relaxed row infeasibilities - summing to 193
>>> 16 relaxed row infeasibilities - summing to 193
>>> 16 relaxed row infeasibilities - summing to 193
>>>
>>> Result - Optimal solution found
>>>
>>> Objective value:                813.00000000/
>>> ...
>>>
>>> The solution file (It doesn't matter whether I use the solu or gsolu 
>>> option) shows an different objective function value of 1925 and 
>>> values for the variables that are obviously infeasible.  When I set 
>>> the preprocessing off then the problem is stated as infeasible.
>>>
>>> It could be possible that the problem formulation and therefore the 
>>> mps file is stupid. But when I use the same mps file with glpk, scip 
>>> and Cplex then I'm able to find in any case an optimal solution with 
>>> the objective function value 813 (that is also shown by Cbc).
>>>
>>>
>>>
>>>
>>>
>>> Thanks,
>>>
>>> Mike
>>>
>>>
>>> _______________________________________________
>>> Cbc mailing list
>>> Cbc at list.coin-or.org
>>> http://list.coin-or.org/mailman/listinfo/cbc
>>
>> _______________________________________________
>> Cbc mailing list
>> Cbc at list.coin-or.org <mailto: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/20150210/752e6bf0/attachment-0001.html>