[rules-users] [Planner] Planning problem - looking for tips, and advices for complicated case

[Geoffrey De Smet]

On 09-05-13 16:00, Mz wrote:
> Hello,
> I'm having a problem to plan my solver, I'm wondering if it's possible to
> configure the solver, so that it picks the best solution from multiple
> collections of planning variables. Here is an example, to be more precise:
Interesting use case. Are you free to explain the practical appliance a 
bit more so we can understand it better?
>
> I have 3 planning variables: X, Y, Z. and many collections of values of
> these variables. For example, collection A contains three X values, five Y
> values, and two Z values; other collection would contain different values
> etc. And the problem is that the values from different collections can't be
> mixed with each other, and I don't know how to prevent that.
A) One approach would be to use a custom move factory/iterator that 
generates moves
which move all 3 variables to any value for which the 3 values stem from 
the same collection.
If you only use that move selector, you wouldn't even need to check that 
the variables use the same collection with a constraint (as it would be 
build-in).

>
> I have tried different approaches but with no luck... Tried to mix all of
> the values together identifying them with a collection ID, and setting in
> rules a negative score for the ones that are not from the same collection,
> but its very inefficient,
B) Normally, I 'd recommend this approach, but indeed it will be very 
inefficient.
The metaheuristics would be continually score trapped (see latest docs 
on "score trap"), making it unlikely that they switch collection.
Mixing in custom moves (like the one from approach A) with normal moves 
would make this approach viable too.

C) A drastic, but simpler approach, if those collections are always 
small (otherwise it will run into scaling issues), is the following:
Use a single variable. Create a cross product of all 3 values in each of 
the collections and use such a combination as the value.
For example, collection A results into 3 * 5 * 2 = 15 combinations. Add 
the results of combinations of the other collections and that's the 
range for that single variable.
> and the solver never stops (despite the fact that
> the timeout was set for 120sec).
That sounds like a bug. Older versions had this problem, but recent ones 
don't (I don't recall for which version it was fixed).
Enable trace logging and see what the logging says.
>
> I was thinking about using multiple @PlanningEntities, but as far as I know
> it is not supported, so I have resigned from this idea.
I don't see how multiple @PlanningEntities would fix it.
Support for multiple @PlanningEntities is getting better since 
6.0.0.Beta1, but not perfect yet.
>
> I have also tried using @ValueRange type FROM_PLANNING_ENTITY_PROPERTY, and
> in MoveFactory I have changed planningEntityProperty according to the
> collection I was using, but it led to exception "Corrupted Undo Move", and
> all in all I think it was a bad idea.
FROM_PLANNING_ENTITY_PROPERTY had a regression in 5.5, but works well 
again since 6.0.0.Beta1
>
> My last idea, which I have not tested yet, is to make two different solvers,
> one would find best solution for the given collection, and from all of the
> best solutions from these collections, the next solver would pick the best
> one, but I think this might be a very inefficient way of doing it.
This is basically partitioning, which is a suboptimal approach. I 
wouldn't recommend it.
>
> I'm really sorry if I made this post very hard to understand, but my problem
> is a bit more complicated then it sound, and its hard to describe it :P I
> would appreciate any tips/ideas how I could configure/plan my solver, and
> rules so it would work better.
>
> Thanks in advance,
> Mateusz
>
>
>
> --
> View this message in context: http://drools.46999.n3.nabble.com/Planner-Planning-problem-looking-for-tips-and-advices-for-complicated-case-tp4023753.html
> Sent from the Drools: User forum mailing list archive at Nabble.com.
> _______________________________________________
> rules-users mailing list
> rules-users at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/rules-users
>