[gecode-users] staged search

[Christian Schulte]

Ah, okay. Yes, "solution" I used in the sense that the branching is done not
that the variables are all assigned. Remember, what a search engine returns
is just a space where all branching have been tried.

What should work in principle (that might be an idea to have in Gecode 3.0):
One has a branching that has a single choice point with a single
alternative. The single alternative executes a method of the space
(actually, as branchings can be programmed one can already do this and
that's not even difficult...), so the branching just stores which message is
to be executed (hmmm, maybe not so easy have to think some more. Gecode used
to have first class closures combining a space and a method but as I did not
understand what it could be useful for, I kicked it out even before 1.0.0).
Just executing a statically defined method is okay). Then in that method you
post your additional constraints and new branching (good knows whether you
can create new branching in a branching but I think yes as we had this
before on this list). This approach utilizes the fact that you can have
multiple branching that are tried in order of creation.

Some examples have custom-programmed branching: so that could be a starting
point.

Okay, sorry for thinking in public.

Cheers
Christian

-----Original Message-----
From: users-bounces at gecode.org [mailto:users-bounces at gecode.org] On Behalf
Of Denys Duchier
Sent: Wednesday, October 08, 2008 10:20 PM
To: users at gecode.org
Subject: Re: [gecode-users] staged search

"Christian Schulte" <cschulte at kth.se> writes:

> Yes,  but what do you want to do exactly? Take a solution of the first
stage
> and then continue with the second stage?

more or less, except of course it might not be a "complete" solution
(only the vars distributed over would be determined, but there might be
other vars that would be taken care of in the second stage).

I was more precisely wondering whether such a staged search could
integrate transparently into gist: the first stage would produce a
prefix of the search tree, while the second stage would produce the
extension of that search tree to "full" solutions.

Probably the answer is no, but it would be so amazingly nice if it were
yes ;-)

Cheers,

--Denys

_______________________________________________
Gecode users mailing list
users at gecode.org
https://www.gecode.org/mailman/listinfo/gecode-users