[Coin-discuss] Settings for Cbc with a Osi solver (not Clp)

[Bo Jensen]

I also do not understand how Cbc uses warmstarts, I sometimes see the
basis does not have full rang i.e it is missing basic variables. Of
course a partial basis might be parsed in some cases, but just before
the call to setWarmstart there is a call to setColsolution, can anyone
explain this ? 

Here is a debugging dump (where I possibly loose 2 basic variables (from
210 to 208) ) :

OsiMskSolverInterface::loadProblem(0xbfac8280, 0x92376b0,
0x92d38c0, 0x9233760, 0x927a820, 0x927af58)  
OsiMskSolverInterface::setColSolution(0x935e260) 
OsiMskSolverInterface::setWarmStart(0x9050fd8) numcols 1600
numrows 210 numbas 208

The function setColSolution is really messy, how is it supposed to
work ? I mean what if you have a fixed variable in the basis, then it
is not clear from the numerical value if it is in or outside the
basis. On the other side it is not clear what level a super basic
variable should have in setWarmStart.  

Regards
Bo

On Thu, 29 Nov 2007 09:12:06 +0100
Bo Jensen <bo.jensen at mosek.com> wrote:

> Hi,
> I am in the process of updating OsiMsk for the new verson of mosek, I
> have build a Cbc version with OsiMsk from the CbcMain.cpp ->
> cbc-generic to stress test it. But I have difficulties in finding the
> right default settings, Cbc(Clp) runs much faster build from
> CoinSolve.cpp -> cbc than from CbcMain.cpp -> cbc-generic. The code in
> CoinSolve.cpp seams very Clp specific, can we get similar
> performance/settings with another Osi solver than Clp ?
> I have a few changes to your build system so Cbc can build with mosek
> 5, who can I talk to off list to add these changes ? 
> Regards
> Bo
> _______________________________________________
> Coin-discuss mailing list
> Coin-discuss at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/coin-discuss