Possible CBC bug (was "Re: [Coin-discuss] CBC feasibility pump heuristic")

John J Forrest jjforre at us.ibm.com
Mon Aug 6 09:39:19 EDT 2007


Sebastian,

In trunk I had made some changes to ampl interface for MINLPs and messed up
integers for linear case so solver thought there were no integer variables.

Should be OK now.

I have added Thirdparty/Glpk to get gmpl support.

John


                                                                           
             Sebastian Nowozin                                             
             <nowozin at gmail.co                                             
             m>                                                         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 
             08/06/2007 07:46          Possible CBC bug (was "Re:          
             AM                        [Coin-discuss] CBC feasibility pump 
                                       heuristic")                         
                                                                           
             Please respond to                                             
             Discussions about                                             
                open source                                                
               software for                                                
                Operations                                                 
                 Research                                                  
             <coin-discuss at lis                                             
              t.coin-or.org>                                               
                                                                           
                                                                           





Hello John,

thanks for the reply.  Today I had time to try the latest Cbc trunk
(20070806) but I think I hit on a bug in Cbc.  So far I only used Cbc
1.1.x, not trunk.

Using the trunk version, the attached AMPL model returns an infeasible
solution as optimal, after just about a second of computation.

Using the 1.1.1 version properly goes on to try to solve this hard model
(takes forever).

I also converted the model using GLPK to the MPS format (glpsol --check
--math cbc-bug-mathprog.mod --wfreemps cbc-bug-mathprog.mps).  The
resulting MPS file can be read in and solved by both 1.1.1 and today's
trunk.  (I.e. I do not obtain optimal solutions in any case but it
computes for more than one second and the output indicates reasonable
feasible objective function values).

I hope this helps in identifying whether this is a bug in cbc trunk or
not.  I attached the original model (cbc-bug.mod), the MathProg version
(cbc-bug-mathprog.mod) and the MPS file (cbc-bug-mathprog.mps.gz)

Sebastian

John J Forrest wrote:
> The answer of course is neither :-)
>
> If you use the standalone cbc then it will go round again if it gets a
> solution and try and get a better one.  With some options it will also
try
> a mini branch and bound fixing all variables which did not move on any
> pass.  I can't remember what I put in code - if you start cbc and then
type
> -
>
> feas??
> passf??
> pumpt??
>
> you will get some incomprehensible help - try pumptune 1008003 for
> example!!

(See attached file: cbc-bug-mathprog.mps.gz)(See attached file:
cbc-bug.mod)(See attached file: cbc-bug-mathprog.mod)
_______________________________________________
Coin-discuss mailing list
Coin-discuss at list.coin-or.org
http://list.coin-or.org/mailman/listinfo/coin-discuss
-------------- next part --------------
A non-text attachment was scrubbed...
Name: cbc-bug-mathprog.mps.gz
Type: application/octet-stream
Size: 18246 bytes
Desc: not available
URL: <http://list.coin-or.org/pipermail/coin-discuss/attachments/20070806/56fdbd5b/attachment.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: cbc-bug.mod
Type: application/octet-stream
Size: 3695 bytes
Desc: not available
URL: <http://list.coin-or.org/pipermail/coin-discuss/attachments/20070806/56fdbd5b/attachment-0001.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: cbc-bug-mathprog.mod
Type: application/octet-stream
Size: 3587 bytes
Desc: not available
URL: <http://list.coin-or.org/pipermail/coin-discuss/attachments/20070806/56fdbd5b/attachment-0002.obj>


More information about the Coin-discuss mailing list