Re: [eclipse-clp-users] How can bb_min fail after finding a solution?

From: <mskala_at_cs.toronto.edu>
Date: Sat, 16 Feb 2013 21:58:33 -0600 (CST)
On Fri, 15 Feb 2013, Joachim Schimpf wrote:
> During search, bb_min always saves the "best solution so far".  Once search
> is finished, the last saved solution is the optimum.  bb_min then instantiates
> the problem variables to this solution, and this will wake up the constraints
> one more time - which should of course succeed.  If it doesn't succeed, then

Thanks.  After further experimentation, I think what was going on was that
my problem contained some "hidden" variables I wasn't labelling directly -
intermediate values in more complicated constraints which ought to have
had their values completely determined by the variables which I was
labelling.  Propagation through those wasn't always complete, depending on
the order of assignment, so it could happen that labelling all the visible
variables would create a contradiction among the hidden ones, but it
wouldn't be detected because not all consequences had propagated; and then
the final assignment, in a different order, would detect the
contradiction.  Changes to my model made for other reasons have made the
issue disappear for the moment, but now if it happens again, I'll know
where to look.
-- 
Matthew Skala
Postdoctoral Fellow, University of Manitoba
mskala_at_cs.umanitoba.ca
Received on Sun Feb 17 2013 - 04:21:04 CET

This archive was generated by hypermail 2.2.0 : Tue Feb 19 2013 - 06:14:32 CET