[Clp] Incorrect solution for LP problem

[Ted Ralphs]

Hi Kish,

I fixed things up so that if you have just one objective, you do not need
to specify the name. With multiple objectives, you need the names in order
to separate the expressions. There's no easy way around this, but it only
affects users of SYMPHONY's bicriteria solve anyway. With rows, the
right-hand side serves as a separator so it is possible to leave the row
names out even when there are multiple rows.

I committed the fix to trunk. Please let me know if everything seems OK.

Cheers,

Ted

On Wed, Oct 22, 2014 at 9:44 PM, Kish Shen <kisshen at cisco.com> wrote:

> Hi John and Ted,
>
> Thanks for your replies. I had wondered if the problem is with the file,
> but I forgot to mention that in my posting, as I was in a rush.
>
> Some context: I did some checking on this failing test, and it seems
> the test has been in our system for 10 years, pre-dating our interfacing
> to CBC. It is so long ago that it is difficult to be absolutely certain,
> but I think the LP file was generated by XPRESS,
> which suggest that XPRESS, at least at the time (2oo4) was generating
> objectives without a name for LP format files.
>
> Cheers,
>
> Kish
>
>
>
> On 22/10/2014 23:49, Ted Ralphs wrote:
>
>> I am the guilty party who made the changes to CoinLpIO. I was extending
>> the
>> reader so that it recognizes multiple objectives. I am also not an expert
>> on the LP format, but what is wrong here is that the objective does not
>> have a name. If you change it to:
>>
>> Minimize
>>    obj: C1
>>
>> then it should work as expected. If there is only one objective, It is
>> supposed to assign the default name of "obj" and looking at the code, it
>> seems it should still be doing that. However, it is throwing an error. I
>> will debug it and let you know.
>>
>> Cheers,
>>
>> Ted
>>
>> On Wed, Oct 22, 2014 at 2:46 PM, John Forrest <
>> john.forrest at fastercoin.com>
>> wrote:
>>
>>  Kish,
>>>
>>> Seems to be a change to CoinLpIO - objective is being ignored - something
>>> to do with no objective name possibly.  Coding has changed for CoinlpIO
>>> but
>>> I am not an expert.  Will look at it tomorrow morning, unless whoever
>>> made
>>> changes to CoinLpIO fixes problem before then.
>>>
>>> John
>>>
>>>
>>> On 22/10/14 18:28, Kish Shen wrote:
>>>
>>>  Hi,
>>>>
>>>> I updated CBC yesterday (for dealing with handling of messages via
>>>> message handlers -- thanks John!), but this CBC is failing one of our
>>>> existing tests. It looks like the problem is in CLP, as CLP seems to be
>>>> returning an incorrect objective value for the following LP problem:
>>>>
>>>> \Problem name: eclipse3
>>>>
>>>> Minimize
>>>>    C1
>>>>
>>>> Subject To
>>>> R1: C1 + C2 >= 7
>>>> R2: - C2 + C1 = 0
>>>>
>>>> Bounds
>>>> C1 free
>>>> C2 free
>>>>
>>>> End
>>>>
>>>> The objective should be 3.5, but 0.0 is returned.
>>>>
>>>> I have been able to reproduce this problem using the clp executable --
>>>> by import(ing) the LP file and then solve.
>>>>
>>>> This is from CBC 2.8 that I downloaded yesterday. The problem was not in
>>>> the earlier CBC 2.8 that I was using, which was probably downloaded
>>>> about a
>>>> year ago.
>>>>
>>>> Is this a problem in CLP, or am I doing something wrong?
>>>>
>>>> Thanks in advance for any help/information!
>>>>
>>>> Cheers,
>>>>
>>>> Kish
>>>> _______________________________________________
>>>> Clp mailing list
>>>> Clp at list.coin-or.org
>>>> http://list.coin-or.org/mailman/listinfo/clp
>>>>
>>>>
>>>>
>>>>  _______________________________________________
>>> Clp mailing list
>>> Clp at list.coin-or.org
>>> http://list.coin-or.org/mailman/listinfo/clp
>>>
>>>
>>
>>
>>
>>
>> _______________________________________________
>> Clp mailing list
>> Clp at list.coin-or.org
>> http://list.coin-or.org/mailman/listinfo/clp
>>
>>
> _______________________________________________
> Clp mailing list
> Clp at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/clp
>



-- 
Dr. Ted Ralphs
Professor, Lehigh University
(610) 628-1280
ted 'at' lehigh 'dot' edu
coral.ie.lehigh.edu/~ted
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/clp/attachments/20141023/ef682730/attachment-0001.html>