Hi Andrew, > I thought about that. Converting glp_prob to internal primal/dual > simplex structures takes a tiny percentage of the overall solution time, > so I decided not to complicate the interface.
I am a bit more optimistic: Using the draft API, I saw some cases of ~20% improvement in pseudocost initialisation time. Of course this was for large problems and for just 30 dual simplex iterations, but I expect some gain. I'll try it when I have some time, to get actual numbers, and get back to you. >> Therefore I still think than a different internal API is needed, with >> this procedure (and others) implemented on top of it. I'll think about >> it some more and get back to you with a more detailed proposal for >> this API. I think that offering something like the eval_degrad() function of glpios09.c from the simplex routines is the best approach to implement the procedure you outlined. If you want, I can make a mock-up using the bfd_copy() operation which can be later simplified to work with the Schur complement basis factorisation. Best Regards, Chris Matrakidis _______________________________________________ Help-glpk mailing list [email protected] https://lists.gnu.org/mailman/listinfo/help-glpk
