[Cbc] CBC and Gurobi give different solutions

[John Forrest]

San, Noli,

Bug not where I thought.  Preprocessing was (correctly) combining a L 
row and a G row into a ranged row.  CglFlowCover then gave incorrect 
results.

If you add -flow off then gets correct answer.

I have put quick fix in CglFlowCover (ignore ranged rows) - if anyone 
wants to, it would be fairly easy to do better - but case is not common.

John Forrest

On 11/07/14 13:05, Noli Sicad wrote:
> Hi John,
>
> Here's zip file. The zip contents are:
>
> 1. GLPK/MathProg - runs and solve in GLPK/MathProg
> 2. CPLEX lp file, converted from GLPK/MathProg
> 3. FreeMps file which also converted using GLPK/MathProg.
>
> The solution is 24.
>
> Regards, Noli
>
>
>
>
> On 7/11/14, John Forrest <john.forrest at fastercoin.com> wrote:
>> I do not have ampl, but if someone can convert the .mod file to a .nl
>> file, then I can look at it.  It might be an error in reading .nl files.
>>
>> John Forrest
>>
>> On 07/07/14 13:54, Tú San Pha.m wrote:
>>> Dear all,
>>>
>>> I submitted the same model to NEOS using Gurobi and CBC. Gurobi gave
>>> me the optimal solution with objective value equals to 24 (this is the
>>> optimal solution, I have checked with CPLEX), while CBC gave me the
>>> one with objective value equals to 4 (for the same instance: 6*6 matrix)
>>>
>>> Attached is the model that I used. Could you please give me any idea
>>> why it goes wrong?
>>>
>>> Thanks a lot
>>>
>>> San
>>>
>>>
>>> _______________________________________________
>>> Cbc mailing list
>>> Cbc at list.coin-or.org
>>> http://list.coin-or.org/mailman/listinfo/cbc
>>