Andrew,
I had previously thought that it was possible to do the following:
- create an LP instance
- solve with lpx_simplex()
- add rows with lpx_add_rows() and related functions
- re-solve starting from the same basis with lpx_simplex() with
LPX_K_DUAL set
However, this doesn't work for me. This issue arose in testing the
COIN-OR OSI interface, which expects that this is possible.
In fact, lpx_simplex may even abort without solving on the second
attempt (it realizes that it's not a "fresh" problem, but the basis
isn't there, or something like that).
Here are my questions:
- What behavior do you currently expect if rows are added and then
lpx_simplex is solved? (Solve again from existing basis, solve again as
if it were a new instance, something else)
- Would you be interested in a patch that allows re-solving from the
existing basis if possible? (I might be able to write such a patch.)
- Do you want me to create a small test case to demonstrate the issue?
Thanks,
Brady
--
Brady Hunsaker
Assistant Professor
Industrial Engineering
University of Pittsburgh
http://www.engr.pitt.edu/hunsaker/
_______________________________________________
Help-glpk mailing list
[email protected]
http://lists.gnu.org/mailman/listinfo/help-glpk
