Solving the dual as Marc suggested works much better. There were no failures even w/ 10,000 points whereas the original formulation consistently fails w/ 1000 points.
This may be hopelessly naive of me, but would it make sense to generate the dual and restart if an IPM problem fails? Or, alternatively, is there a criterion that could be applied at startup that would allow determining when solving the dual would be more appropriate? BTW I'm using the scripts for this as an example of Unix batch execution for the wikibook. Unfortunately, a system hang at my end resulted in losing an hour's work just before I was going to save it. I've started redoing it but have to rewrite the explanation and add the model that Marc contributed. Have Fun! Reg _______________________________________________ Help-glpk mailing list [email protected] https://lists.gnu.org/mailman/listinfo/help-glpk
