[Cbc] Antw: Re: Cut efficiency in cbcSolve for versions 2.0.0, 2.1.0, 2.2.2 and 2.3.0

[Torsten Fahle]

John,

Thanks for the fast reply. I've tried -probing forceOnStrong with all versions.
Basically, the newer versions CBC 2.2.2 and 2.3.0 show no effect, whereas
the older versions CBC 2.0.0 and 2.1.0 find even better bounds than before
(see details below)

Maybe there is an issue in the newer version?
I'll send you the MPS file in a separate mail directly to you.
You'll find all details in the trac system at 
<https://projects.coin-or.org/Cbc/ticket/78>
Thanks for looking at this! Let me know if I can be 

Best regards

  Torsten




Testresults in details 
call: cbcSolve testfile.mps -cuts off -probing on|forceOnStrong -branch

results in
Initial Lower Bound 160.236, optimal solution 246.296 (all versions)
Bound improvement in Root node:
CBC 2.0.0  probing on                       LB: 195.305  (14 row cuts, 6 active, 28 col cuts)
                  probing forceOnStrong    LB: 245.235  (720 row cuts, 33 active, 44 col cuts)

CBC 2.1.0  probing on                       LB: 215.421  (104 row cuts, 12 active, 41 col cuts)
                  probing forceOnStrong    LB: 245.235  (823row cuts, 43 active, 45 col cuts)

CBC 2.2.2  probing on                       LB: 160.236  (0 row cuts, 0 active, 0 col cuts)
                  probing forceOnStrong   LB: 160.236  (0 row cuts, 0 active, 0 col cuts)

CBC 2.3.0  probing on                       LB: 160.236  (0 row cuts, 0 active, 0 col cuts)
                  probing forceOnStrong   LB: 160.236  (0 row cuts, 0 active, 0 col cuts)

all tests under WinXP using Visual Studio 2005 in debug mode





-- 

Dr. Torsten Fahle
Airport Systems Division
INFORM GmbH, Pascalstr.23, 52076 Aachen, Germany
Tel. (+49) 24 08 - 94 56 24  FAX: -94 56 25
e-mail: Torsten.Fahle at inform-ac.com   http://www.groundstar.de 
INFORM Institut fuer Operations Research und Management GmbH
Registered AmtsG Aachen HRB1144 Gfhr.Adrian Weiler


>>> John J Forrest <jjforre at us.ibm.com> schrieb am Donnerstag, 27. August 2009 um
09:29 in Nachricht
<OF30599A72.CF0984BF-ON8525761F.00287904-8525761F.00293226 at us.ibm.com>:

> Torsten,
> 
> Techniques such as probing can be very expensive and default settings have
> been changed between versions.  What happens if you try and force more
> probing e.g. -probing forceonstrong?  Also it could be a bug of sorts.
> 
> If you want, you can send me the problem for me to see if there is a bug
> (as long as it is not too large).
> 
> John Forrest
> 
> 
>                                                                        
>   From:       "Torsten Fahle" <Torsten.Fahle at inform-ac.com>            
>                                                                        
>   To:         <cbc at list.coin-or.org>                                   
>                                                                        
>   Date:       08/26/2009 11:32 AM                                      
>                                                                        
>   Subject:    [Cbc] Cut efficiency in cbcSolve for versions 2.0.0, 
> 2.1.0,2.2.2and 2.3.0
>                                                                        
>   Sent by:    cbc-bounces at list.coin-or.org                             
>                                                                        
> 
> 
> 
> 
> 
> Hi,
> 
> When testing with older cbcSolver versions we were surprised  to see that
> the lower bound in the root node after adding cuts
> differs significantly between different cbc releases:
> 
> On a test case the root node after cuts had a lower bound of
> 195.305  (CBC 2.0.0)
> 221.561  (CBC 2.1.0)
> 160.296  (CBC 2.2.2 and 2.3.0)
> 
> The initial LP bound is 160.236, the optimal solution 246.296.
> This means almost no improvement by root cuts in CBC 2.2.2 and 2.3.0. The
> initial gap is 53%
> In CBC 2.0.0 and 2.1.0 there is a significant improvement, the initial gap
> is 26%, and 11%, resp.
> 
> It seems that probing is not as efficient in cbcSolve 2.2.2 and 2.3.0 as it
> was in cbcSolve 2.0.0 and 2.1.0( after some tuning). Maybe there is a small
> bug in probing or some
> other setting stops probing from being as effective as in older versions?
> 
> All tests were performed using MS Visual Studio 2005 on an Intel QuadCore
> CPU using Debug mode.
> There are numerical differences between release and debug version in this
> setting. I hope the effect
> is nevertheless still reproducible.
> 
> I've filed a ticket (#78) with more details on the case. Maybe someone can
> investigate this case and find
> 
> Thanks !
> 
>   Torsten
> 
> --
> 
> Dr. Torsten Fahle
> Airport Systems Division
> INFORM GmbH, Pascalstr.23, 52076 Aachen, Germany
> Tel. (+49) 24 08 - 94 56 24  FAX: -94 56 25
> e-mail: Torsten.Fahle at inform-ac.com   http://www.groundstar.de
> INFORM Institut fuer Operations Research und Management GmbH
> Registered AmtsG Aachen HRB1144 Gfhr.Adrian Weiler
> 
> 
> 
> _______________________________________________
> Cbc mailing list
> Cbc at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/cbc