[Osi] Identifying offending constraints
mjs at clemson.edu
Wed Feb 10 17:36:03 EST 2010
On Wed, 2010-02-10 at 14:14 -0800, Paul De La Musica wrote:
> I am new to LP and OSI/CLP.
> Our flow calls for solving the LP problem first, if it fails, we
> modify the offending constraints and solve again. But how can we know
> which constraints are the "offending constraints"?
It rather depends on what you mean by "offending".
I could be wrong (I just started on the maintenance team for CLP and I
haven't studied the code that much yet), but I don't think CLP has an
infeasibility analyzer along the lines of CPLEX's IIS.
So if phase I terminates with infeasibility, you've got a solution that
happens to violate some constraints and/or bounds, and minimizes some
function of distances to variable and constraint bounds--exactly what
function depends on exactly how phase I is implemented. About the best
you can do with that is look at the variables in the phase I solution
(including the logical variables) and check which ones violate their
bounds. I'd have to spend some time to figure out what the exact
incantation would be.
What IIS does is try to identify a set of constraints that is minimally
infeasible. That might be more useful for trying to diagnose where the
infeasibility comes from, but again, I don't believe that CLP can do
that at this point.
> Osi mailing list
> Osi at list.coin-or.org
Clemson University Math Sciences
mjs AT clemson DOT edu
More information about the Osi