Hi Mark,
I am not sure to understand fully the problem to solve, but my impression is
that your constraint model makes the evaluation of the preferences quite
difficult to compute. Let me propose a different model, and see whether it
might fit in your solution.
The problem consists in determining how to order a set of scenes. Your
solution is a list of scene indices; it defines a mapping from position in
the list to scene index. It might be more efficient to consider the inverse
mapping, from scene index to position. Consider something like this:
% a list of scene identifiers (indexes or atoms)
Scenes = ...
% maps each scene to its position in the sequence
Position = {FD.record position Scenes 1#{Length Scenes}}
{FD.distinct Position}
% the following tells that scene X must precede scene Y
Position.X <: Position.Y
The constraint above is a hard constraint: it must be satisfied. You can
make it soft by reifying '<:', which introduces an FD boolean variable
(which is 1 iff the constraint is satisfied). Summing those boolean
variables tells how many preferences are satisfied.
To determine the preference constraints, simply iterate over all pairs of
scenes, and post the required constraint. If constraints are reified, use
the 'collect' feature in the for loop to collect all boolean preference
variables.
for X in Scenes do
for Y in Scenes do
if X\=Y andthen ... % a condition on the scene
descriptions
then
% post constraint
end
end
end
Hope this helps...
Cheers,
Raphael
On Fri, Feb 19, 2010 at 3:35 PM, mark richardson <[email protected]>wrote:
> Hi,
>
> I have a program which is very similar to the 'aligning for a photo'
> example. My 'people' are scenes in a narrative, each scene is a tuple which
> describes a scene by type, start time and participants. A scene might
> require precedent scenes, which have the participants in various different
> configurations, to have occurred before. In my script, Alignment is a list
> of integers which represent indexes to the actual scnenes. The order in
> which the scenes should occur is the order of the list. I am trying to order
> the scenes so that as many as possible of the precedences are satisfied.
>
> I have functions which create a tuple with 'wild-cards' and can then match
> this tuple against actual scenes. So my current approach is to loop through
> each item in 'Alignment', locating the corresponding scene and building
> 'wild-card' tuples for any precedence required. I then check each preceding
> item in Alignment to see if the corresponding scene matches any of these
> tuples. Finally a successful match is converted to a 1, unsuccessful to a 0.
>
> In effect it's similar to aligning people for a photo so that each person
> has preferences on who should be to the left of them in the line, which may
> be none or more people.
>
> I'm having enormous problems with my program, but if someone could suggest
> how the photo example could be modified to add the 'to the right of'
> preference, I might be able to convert that to suit my purposes.
>
> Thank you in advance for any help
>
> Regards
>
> Mark
>
> _________________________________________________________________________________
> mozart-users mailing list
> [email protected]
> http://www.mozart-oz.org/mailman/listinfo/mozart-users
>
_________________________________________________________________________________
mozart-users mailing list
[email protected]
http://www.mozart-oz.org/mailman/listinfo/mozart-users
