[Coin-discuss] accessing the Clp solver from Cbc/Clp

Kish Shen kish.shen at crosscoreop.com
Fri Jun 30 00:05:53 EDT 2006


Lou,
On Wednesday 28 June 2006 18:25, Lou Hafer wrote:
> Kish,
>
> > For the interface to OSI/Coin solvers that I am developing, I am hoping
> > to do most things via the OSI interface, so that we can access as wide a
> > range of solvers as possible, and then provide `specialised' code to
> > improve performance/functionality for specific solvers, such as Cbc/Clp.
>
> 	You're entering relatively uncharted territory here.  (See the earlier
> thread in coin-discuss about cbc vs.  cbc-generic, from June 22.)  Given
> demand from Coin users, support for other OSIs can move up the priority
> scale, but right now it's low and occasionally broken.  Getting this right
> will require effort at both the Cbc and OsiXXX levels.  Your feedback will
> be valuable here.
>
> 							Lou
>
> _______________________________________________
> Coin-discuss mailing list
> Coin-discuss at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/coin-discuss

I guess for my specific case, I don't really need OSI CPLEX or XPRESS, as our 
own interface already directly use the CPLEX and XPRESS C API already. 
I have actually been trying out the OSI on and off for the last two years, but 
the first time I tried it, I mainly tried using GLPK, and I was somewhat 
disappointed with the performance. Last year I tried again, and I was quite 
impressed with the improvements and the performance of CLP/CBC, and so now I 
am doing the proper conversion of our interface. 

In terms of functionality, it would be quite nice if the difference between 
the linear solver and CBC be more hidden, as I need to solve both linear and 
MIP (and actually also QP and MIQP) problems using one solver. As things 
stands now, I need to extract both the CBC and CLP model pointers to do 
different things. More support for MIP in the OsiSolverInterface would be 
useful.

Cheers,

Kish



More information about the Coin-discuss mailing list