[Cbc] Why unmaintained OsiCbc?

[Ted Ralphs]

Great, thanks! Let me know if you need anything from me.

Cheers,

Ted

On Mon, Nov 18, 2019 at 2:02 AM Jan-Willem Goossens <jhmgoossens at hotmail.com>
wrote:

> I can make a few PRs to start to use cbcmain in osicbc 'branchandbound'
> and add libosicbc project back to msvisualstudio v16 folder.
> Saves me from keeping track of these changes is my fork :-)
>
> On 18 Nov 2019, at 03:57, Ted Ralphs <ted at lehigh.edu> wrote:
>
>
> More configuration options are not a reason to abandon Osi and I find it
> very unfortunate that no one has tried to make the performance of OsiCbc
> match the performance of direct calls to Cbc. Cbc is the only solver for
> which this is the case. OsiCpx, OsiGrb, etc. all offer the same performance
> through Osi as through their direct interfaces. I don't think it would be
> difficult to get OsiCbc up to speed, especially now that there is a C
> interface! I would like to do it, but there are always just too many other
> urgent things on my COIN-OR TODO list.
>
> It is true that MIP solvers have more configuration options and more
> options that are specialized to one solver than LP solvers, but
> configuration options can be easily set for any solver, even if the solver
> is being called through Osi. For many applications, though, it is still a
> huge win to implement using Osi, even if some configuration options need to
> be set separately on a per-solver basis.
>
> I have several codes that call Cbc and where using Osi is preferable. If
> any is interested in taking this on as a small volunteer project, it would
> be super helpful.
>
> Cheers,
>
> Ted
>
> On Sun, Nov 17, 2019 at 5:50 PM Haroldo Gambini Santos <
> haroldo at ufop.edu.br> wrote:
>
>> My impression is that OSI interfaces are only well maintained for linear
>> programming solvers. Why ?  MIP solvers have much more configuration
>> options and OSI 1.0 did not evolve to cope with these things. More
>> specifically to OsiCbc,  I think that we don't have anyone working
>> actively on it. I think that OSI 2.0 developers are investing now more
>> on MIP.
>>
>> Regarding the use of CbcMain0 and CbcMain1 instead of branch and bound
>> of CbcModel:  the performance of the pure branch and bound (without cuts
>> and well tuned heuristics) is horrible. A good set of parameter settings
>> is maintained in CbcMain0 and CbcMain1  ( the recommended way of calling
>> cbc, which includes the same configuration of the standalone command
>> line solver).
>>
>> Cheers
>>
>> On 17/11/2019 19:10, Jan-Willem Goossens wrote:
>> >
>> > Hi, a few comment threads discuss how OsiCbc is unmaintained, but I
>> > couldn’t understand why?
>> >
>> > For example,
>> > https://github.com/coin-or/CyLP/issues/34#issuecomment-185988123
>> >
>> > My code (https://github.com/coin-or/SONNET) is also calling solvers
>> > via Osi, including OsiCbc. In case of OsiCbc, Sonnet calls CbcSolver
>> > to apply some of the ‘special sauce’ mentioned in the comments above.
>> >
>> > Also, why not use CbcSolver in OsiCbc instead of the plain
>> > branchAndBound, like the comment
>> >
>> > “To enjoy the full performance of Cbc, use the CbcSolver interface.”
>> > mentions?
>> >
>> > Is the best way forward to plan to stop using OsiCbc, or to try to get
>> > OsiCbc maintained?
>> >
>> > Regards,  Jan-Willem
>> >
>> >
>> > _______________________________________________
>> > Cbc mailing list
>> > Cbc at list.coin-or.org
>> > https://list.coin-or.org/mailman/listinfo/cbc
>>
>> --
>> =============================================================
>> Haroldo Gambini Santos
>> Computing Department
>> Universidade Federal de Ouro Preto - UFOP
>> email: haroldo at ufop.edu.br
>> home/research page: www.decom.ufop.br/haroldo
>>
>>
>> It has long been an axiom of mine that the little things are infinitely
>> the most important.
>> -- Sir Arthur Conan Doyle, "A Case of Identity"
>>
>> _______________________________________________
>> Cbc mailing list
>> Cbc at list.coin-or.org
>> https://list.coin-or.org/mailman/listinfo/cbc
>>
>
>
> --
> Dr. Ted Ralphs
> Professor, Industrial and Systems Engineering
> Lehigh University
> (610) 628-1280
> ted 'at' lehigh 'dot' edu
> coral.ie.lehigh.edu/~ted
>


-- 
Dr. Ted Ralphs
Professor, Industrial and Systems Engineering
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/cbc/attachments/20191118/3660392c/attachment.html>