> I have added lpx_scale_prob(lp) and lpx_set_int_parm(lp, LPX_K_PRESOL, 1); > to my code and things seem to have stabilised somewhat and some large data > sets that were originally problematic can be solved now. There are, however, > some huge data sets that remain problematic. A file with problem data in > fixed MPS format will be on its way soon.
Unfortunately the rar archive you posted to me is damaged (it has invalid crc), so I cannot roll out it. Please check the archive and repeat posting. > > For now, if I do not get a solution after 15 mins or if the "numerical > instability" error message appears, I restart the process. Is this > reasonbale? No, because lpx_simplex itself does that. I need to look at the problem data to say something more intelligent. > > And if after a long time the "objval" is still at an extremely small > number, say 1.5 when I expect the solution to be around 15, should I restart > the process? No. This may happen due to high degeneracy of your problem. > > Is the final results scaled in some way? That is, should I multiply the > "OPTIMAL SOLUTION FOUND" say "Z=1.000" by a constant factor? No. All resulting data provided by glpk rouines are not scaled, i.e. scaling is transparent for the user. P.S. Please check settings in your mail agent. It sends multiple copies of the same message. _______________________________________________ Help-glpk mailing list Help-glpk@gnu.org http://lists.gnu.org/mailman/listinfo/help-glpk
