[Cbc] Very different results with CBC/CLP 2.7.7

[John Forrest]

Herve,

preOption 12 should switch off two parts of presolve CoinDoubleton and 
CoinTripleton.  You can check this by running with -pOption 3 which 
should give printout on what actions are done in presolve. Without 
preOption you should see printout like

CoinPresolveDoubleton(4) - 0 rows, 0 columns dropped in time 0.004, 
total 0.720045
CoinPresolveTripleton(8) - 0 rows, 0 columns dropped in time 0, total 
0.720045

while with it set you should not see those lines.

CoinPresolve failing/giving wrong answer is a bug.

Can you put model (compressed) somewhere so I can download it. Does Clp 
give correct result without any presolve?  If not please give me correct 
value of objective.

John Forrest

On 11/03/13 13:24, Hervé BOUVART wrote:
> Dear John,
>
> I solve a large problem (470000 rows, 655000 cols), in continuous 
> variables (now, after numerous attempts with integer variables).
> It appears that the solution, also given by other solvers is ok using 
> version 2.7.7 of CLP, if I provide -preO 12 as an option (infeasible 
> or unbounded if presolver is used without option).
> With version 2.8, I get a solution which is 25% different! Moreover 
> the -preO options does not seem to help anymore.
> My .MPS file is over 134 MB, so I don't attach it here.
>
> What is the meaning of the figure after -preO (presolve options) command?
>
> I use CLP and CBC compiled via MS Visual C++ 2010 Express.
>
> Thanks in advance,
>
> Herve
>
>
>
>
> _______________________________________________
> Cbc mailing list
> Cbc at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/cbc

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/cbc/attachments/20130311/36bce221/attachment-0001.html>