[Coin-discuss] Another Clp QP question
Sebastian Nowozin
nowozin at gmail.com
Wed Sep 17 08:11:20 EDT 2008
Thanks John,
John J Forrest wrote:
> SLP is a fairly rugged trust region approach. If I have to solve a large
> QP I do something like -
> switch off presolve as there may be bugs
> nonlinearSLP(20,1.0e8) i.e. just a few iterations - although that can vary
> primal() which works a lot lot better given a near optimal solution
Are you sure you mean 1.0e8, or 1.0e-8?
My best working configuration right now is disabling presolve and
crossover and using the barrier method. It is quite slow but never
crashed or had some problems on all problems I tested. That does the
job for me, but in general the absence of a decent QP solver in the
open-source world is saddening.
Thanks for all the work,
Sebastian
More information about the Coin-discuss
mailing list