[gecode-users] different nr of solutions with and without GIST

Gustavo Gutierrez ggutierrez at atlas.puj.edu.co
Fri Jan 9 14:02:04 CET 2009

Gustavo Gutierrez

On Fri, Jan 9, 2009 at 10:42 AM, Guido Tack <tack at ps.uni-sb.de> wrote:

> Gustavo Gutierrez wrote:
>> I am experiencing almost the same problem. There is a difference in the
>> number of solutions when i run my problem instance with or without gist. The
>> difference with the case reported is that for me the number of solutions
>> reported by gist is really large compared to the number of solutions of the
>> command line search. In my case the search without gist reports the right
>> number of solutions.
>> As i am working with a new constraint system i think this is caused by
>> some bug in my implementation, but the question is: if it is a bug then both
>> search strategies should exhibit it, am i right?
> Not necessarily.  If it's due to non-monotonicity of a propagator, anything
> can happen - the shape and size of the tree might even depend on memory
> layout and other influences that are impossible to control.  If the
> propagator is in addition not correct, also the number of solutions can
> differ substantially.
> In your case, does Gist return solutions that shouldn't be accepted as
> solutions, or does it return solutions more than once?

I removed all the propagators from my problem. Currently i am only branching
on one variable and i am getting different result from gist and the other
search. As there are no propagators, the shape of the search tree is easy
to guess and should have only 80 solutions. When i run the gist search i get
more than that number of solutions but an interesting thing is that clicking
or passing the mouse over a solution that should not be there leads to a
segmentation fault. Actual solutions to the problem are shown when double
clicking without any problem.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.ps.uni-sb.de/pipermail/users/attachments/20090109/64ff3b2c/attachment.htm>

More information about the gecode-users mailing list