[Coin-discuss] OSI with Clp, how to use interior-point solver

[John J Forrest]

Sebastian,

>
> Three questions,
>
> 1. Does the barrier/interior-point implementation in Clp behave as the
> standard LP interior point methods in the literature and will return a
> central point (as measured by some centrality measure) on the solution
face?

Yes - unless crossover to a vertex is enabled.

>
> 2. As the problem is grown using a constraint-generation strategy in
> such a way that it can become quite degenerate, is the Clp interior
> point solver as stable/robust as the simplex solver?  (And does the
> usual advice of "for degenerate problems, interior point solvers are
> quite robust" hold for the Clp implementation?)
>

Reasonably robust - Interior point is not my strong point.  If it works for
one of a class of problems it should for all.  If anything happens I can
look at it.  The main problem is that you need a cholesky code - e.g. one
from University of Florida to make it run fast.  If you pester me I can try
and get a code into ThirdParty (with a bit of help).

> 3. How to use/enable the interior point implementation when using
> OsiClpSolver?

One can use getModelPtr() but if you prefer to stay in Osi land then create

a ClpSolve options;

options.setSolveType(ClpSolve::useBarrier);
options.setPresolveType(ClpSolve::presolveOn);
// Tell OsiClpSolverInterface to pay attention
options.setSpecialOption(6,1);

model->setSolveOptions(options);
model->initialSolve();

I may not have got this quite right - but anyway it is possible.
>
>
> Thanks a lot and keep up the excellent COIN-OR projects,

Thanks

John Forrest
> Sebastian
> _______________________________________________
> Coin-discuss mailing list
> Coin-discuss at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/coin-discuss