[Coin-discuss] Time limit on presolve?

[Stefan Vigerske]

Hi,

as far as I know there are not artifical limits on model complexity 
(whatever that is). If Glpk hits the time limit, then it would say so, 
and not primal infeasible.

Can you send the complicating mps file directly to me (not the mailing 
list) so I can try it?

Running directly Glpk can behave different than running it through Osi 
since parameter values are changed in OsiGlpk. Also the column or row 
order might be different due to the different mps readers. Or even the 
called B&B solver is different: There are two in Glpk, and I forgot 
which are the defaults in glpsol and in OsiGlpk, respectively.

Best,
Stefan

El Dono schrieb:
> Hello everybody,
>     I'm trying to use the Osi interface but I have problems with some models. I 
> load the model and the data from a mps file and I call the 
> branchAndBound() method of my GlpkSolverInterface, but I get as output a message 
> like "primal infisible". So I've tried loading the same mps file directly from 
> glpsolve and there I get the correct solution. I've tried to recreate the mps 
> file using the same model but different data (I have a mod file and various dat 
> files and convert them into mps files when I need it). This time both glpsolve 
> and the Osi Interface gave me the same result. So I thought the problem could be 
> in the number of variables / constraints and did some experiments but I 
> discovered that this is not the problem: sometimes a dat file with over 700 
> variables has been solved in milliseconds while smaller problems produced the 
> "primal infeasible" messagge (obviously I'm sure they are feasible...). So I 
> think the problem is that somehow there is a limit on the complexity of the 
> model or on the computational time to get the solution but I can't figure out 
> what's this limit and how to change it.
> Could somebody explain me what's the matter? what am I doing wrong?
> Thanks in advance,
> Domenico S.
>  
> P.S.: sorry for my english, if I made some mispellings...
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Coin-discuss mailing list
> Coin-discuss at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/coin-discuss


-- 
Stefan Vigerske
Humboldt University Berlin, Numerical Mathematics
http://www.math.hu-berlin.de/~stefan