Nigel;
I think the best approach for the solving successively with the simplex method is to use glp_get_row_stat()/glp_get_col_stat() and glp_set_row_stat()/glp_set_col_stat(). Does this make sense? Jeff From: Nigel Galloway [mailto:[email protected]] Sent: Thursday, April 28, 2011 8:50 AM To: Kelly, Jeff (ON0F); [email protected] Subject: Re: [Help-glpk] Calling GLPK successively/iteratively/recurisvely for nonlinear problems. Try glp_save_mps to save the basis and and glp_read_mps to load the warm-start.. Depending on what you are trying to do, you may wish to use a solver which can understand XML. This can be edited outside of the solver, and hence the problem can be modified and solved without programming. Of course MPS is text based and you could try to edit it (I wish you good luck). I am currently considering an example of combining five constraint matricies to produce a single solution. As pointed out in earlier communications you need rules for the combination. When the CMs are in XML then XLST provides a way to provide the rules. -- Nigel Galloway [email protected] On Tue, 26 Apr 2011 09:42 -0400, "Kelly, Jeff (ON0F)" <[email protected]> wrote: Nigel – thanks. Is there anyway to save the basis so that from major iteration to major iteration it can use a “warm-start” when solving with the simplex method? Thanks - Jeff From: Nigel Galloway [mailto:[email protected]] Sent: Tuesday, April 26, 2011 6:36 AM To: Kelly, Jeff (ON0F); [email protected] Subject: Re: [Help-glpk] Calling GLPK successively/iteratively/recurisvely for nonlinear problems. See t1.cs in the examples directory -- Nigel Galloway [email protected] On Mon, 25 Apr 2011 07:33 -0400, "Kelly, Jeff (ON0F)" <[email protected]> wrote: All; I am wondering if anybody calls GLPK iteratively/successively/recursively to solve nonlinear problems similar to a sequential linear programming algorithm? My concern is calling GLPK multiple times where each LP subproblem is technically unrelated to a previous major iteration but successive calls to glp_set_row_bnds(), glp_set_col_bnds(),glp_set_obj_coef() and glp_load_matrix() are required. Although the structure of the LP subproblems do not change from major iteration to major iteration i.e., only the row/column bounds and objective/matrix coefficients change, I am wondering if there would be any problems using GLPK for this? Presumably if I call glp_delete_prob() after each major iteration has completed, then theoretically there should be no problem other than the overhead of setting up the LP subproblem each time. Any comments on this would be appreciated - Jeff _______________________________________________ Help-glpk mailing list [email protected] https://lists.gnu.org/mailman/listinfo/help-glpk -- http://www.fastmail.fm - A no graphics, no pop-ups email service _______________________________________________ Help-glpk mailing list [email protected] https://lists.gnu.org/mailman/listinfo/help-glpk -- http://www.fastmail.fm - The professional email service
_______________________________________________ Help-glpk mailing list [email protected] https://lists.gnu.org/mailman/listinfo/help-glpk
