[Coin-discuss] Controlling printing for CbcStrategyDefault

[John J Forrest]

Kish,

I can not reproduce error messages.




                                                                           
             Kish Shen                                                     
             <kish.shen at crossc                                             
             oreop.com>                                                 To 
             Sent by:                  Discussions about open source       
             coin-discuss-boun         software for Operations Research    
             ces at list.coin-or.         <coin-discuss at list.coin-or.org>     
             org                                                        cc 
                                                                           
                                                                   Subject 
             07/09/06 02:17 AM         Re: [Coin-discuss] Controlling      
                                       printing for CbcStrategyDefault     
                                                                           
             Please respond to                                             
             Discussions about                                             
                open source                                                
               software for                                                
                Operations                                                 
                 Research                                                  
             <coin-discuss at lis                                             
              t.coin-or.org>                                               
                                                                           
                                                                           




Hi,

On Saturday 08 July 2006 07:31, Kish Shen wrote:

> One question about this preprocessing: do you expect it to improve the
> performance? In my case, I have already called CglPreprocess on the
problem
> when it was first copied from the OsiClpSolver to the CbcModel, and in
the
> rather limited test examples I have been running (mainly used to test if
my
> changes are working, rather than for measuring performances), where some
> MIPLIB problems are solved (but they are modelled in our ECLiPSe
> programming language, rather than read in from an MPS file), doing this
> second preprocessing in CbcStrategy does not improve the solving times,
and
> in fact for one problem (noswot), it slows the solving time time from 33s
> to 54s.
>

I decided to try out some more tests with my code. With the
CbcStrategyDefault
set for the CbcModel  (model is a CbcModel*):

   (code taken and modified from sample2.cpp in Cbc Samples directory)
    // Set up pre-processing to find sos if wanted
    if (lpd->presolve) strategy.setupPreProcessing(2);
    //strategy.setupPrinting(*model, 0);
    model->setStrategy(strategy);
    model->branchAndBound();

this generally solved the problems faster than if the strategy
preprocessing
and setStrategy code is commented out. However, for a couple of the
problems
I tried (versions of MIPLIB's cap6000 and aflow30a), I got what looks like
error messages from Clp:

Clp6002E 1 bad bound pairs or bad objectives were found - first at C3013
Clp6002E 1 bad bound pairs or bad objectives were found - first at C1987
Clp6002E 1 bad bound pairs or bad objectives were found - first at C3863

the actual objective value for the MIP problem after the search appears to
be
correct though. Is the above a potential problem, or are such messages
expected?

aflow30a, in addition to generating Clp error messages like the above, also

ran much slower with the CbcStrategyDefault than without -- slowing down
from
2000 seconds to 6500 seconds, the only problem to do so. Is this related to

the above messages?

Cheers,

Kish Shen
_______________________________________________
Coin-discuss mailing list
Coin-discuss at list.coin-or.org
http://list.coin-or.org/mailman/listinfo/coin-discuss