Drools › Drools: User forum

the n best solutions on a very simple test case

Classic

List

Threaded

4 messages Options

Laurent Michenaud

the n best solutions on a very simple test case

Hi,

We have implemented a best score recaller to keep
the n best found solutions.
It seems to be ok except on very simple
test cases :

- On the first step, 4 moves are tested. They all
return a feseable score. The best one is kept and
sent to the solution recaller.
- After each of these moves, no other moves are
possible and the other solutions of the first step
are not sent to the solution recaller.

So, in our solution recaller, we have only one
solution.

Thanks for your help
Michenux.

_______________________________________________
rules-users mailing list
[hidden email]
https://lists.jboss.org/mailman/listinfo/rules-users

ge0ffrey

Re: the n best solutions on a very simple test case (drools-solver)

Administrator

Only the solution which is reached by each chosen step is send to the
BestSolutionRecaller. If you do only one step, you 'd have to do some
dirty hack to get those other 4.

Maybe you can use your current Finish combined with AND and a at least 4
steps taken finish, to force at least (4 - 1 = )3 steps. Combine it with
a completeSolutionTabu of at least 3 (usually I take a 1000) and you 'll
have 4 distinct solutions. Your starting solution will also count as one.

But I am interested in this use case :)

1) Why do you want to recall you best 4 solutions? Only the best
solution seems relevant to me (but I could be wrong). If this is a
genuine use case, I can add a MultiBestSolutionRecaller in the
drools-solver code.

2) Extending a BestSolutionRecaller is easy, but how did you hook it up
into the framework?
a) Did you build your solver and then do a setBestSolutionRecaller()
b) Or did you use xstream ability to configure in your
BestSolutionRecaller implementation directly?

BTW: other feed-back on how to improve drools-solver is welcome

With kind regards,
Geoffrey De Smet

Laurent Michenaud schreef:

> Hi,
>
> We have implemented a best score recaller to keep
> the n best found solutions.
> It seems to be ok except on very simple
> test cases :
>
> - On the first step, 4 moves are tested. They all
> return a feseable score. The best one is kept and
> sent to the solution recaller.
> - After each of these moves, no other moves are
> possible and the other solutions of the first step
> are not sent to the solution recaller.
>
> So, in our solution recaller, we have only one
> solution.
>
> Thanks for your help
> Michenux.
>
>
>
> _______________________________________________
> rules-users mailing list
> [hidden email]
> https://lists.jboss.org/mailman/listinfo/rules-users
>

_______________________________________________
rules-users mailing list
[hidden email]
https://lists.jboss.org/mailman/listinfo/rules-users

Laurent Michenaud

RE : Re: the n best solutions on a very simple test case(drools-solver)

Hi,

This is a genuine test case. It is like this in our application
because drools is not the final decision maker. Drools is here
to find and propose the n best solutions, the customer will pick
up one among these. The score of each solution is printed to the
customer, so he knows that this one is best than the other one but he
is not forced to choose it and can choose another one as well.

We build the solver and we set the solution Recaller using setBestSolutionRecaller.

I have looked at the examples and i don't see how to configure
"at least n steps taken finish". Can u help ?

Best regards
Michenux

-------- Message d'origine--------
De: [hidden email] de la part de Geoffrey De Smet
Date: jeu. 02/07/2009 20:43
À: [hidden email]
Objet : [rules-users] Re: the n best solutions on a very simple test case(drools-solver)

Only the solution which is reached by each chosen step is send to the
BestSolutionRecaller. If you do only one step, you 'd have to do some
dirty hack to get those other 4.

Maybe you can use your current Finish combined with AND and a at least 4
steps taken finish, to force at least (4 - 1 = )3 steps. Combine it with
a completeSolutionTabu of at least 3 (usually I take a 1000) and you 'll
have 4 distinct solutions. Your starting solution will also count as one.

But I am interested in this use case :)

1) Why do you want to recall you best 4 solutions? Only the best
solution seems relevant to me (but I could be wrong). If this is a
genuine use case, I can add a MultiBestSolutionRecaller in the
drools-solver code.

2) Extending a BestSolutionRecaller is easy, but how did you hook it up
into the framework?
a) Did you build your solver and then do a setBestSolutionRecaller()
b) Or did you use xstream ability to configure in your
BestSolutionRecaller implementation directly?

BTW: other feed-back on how to improve drools-solver is welcome

With kind regards,
Geoffrey De Smet

Laurent Michenaud schreef:

_______________________________________________
rules-users mailing list
[hidden email]
https://lists.jboss.org/mailman/listinfo/rules-users

_______________________________________________
rules-users mailing list
[hidden email]
https://lists.jboss.org/mailman/listinfo/rules-users

winmail.dat (5K) Download Attachment

ge0ffrey

Re: RE : Re: the n best solutions on a very simple test case(drools-solver)

Administrator

Try something like this:

<finish>
<finishCompositionStyle>AND</finishCompositionStyle>
<maximumStepCount>4</maximumStepCount>
... your current finish configuration ...
</finish>

With kind regards,
Geoffrey De Smet

Laurent Michenaud schreef:

> Hi,
>
> This is a genuine test case. It is like this in our application
> because drools is not the final decision maker. Drools is here
> to find and propose the n best solutions, the customer will pick
> up one among these. The score of each solution is printed to the
> customer, so he knows that this one is best than the other one but he
> is not forced to choose it and can choose another one as well.
>
> We build the solver and we set the solution Recaller using setBestSolutionRecaller.
>
> I have looked at the examples and i don't see how to configure
> "at least n steps taken finish". Can u help ?
>
> Best regards
> Michenux
>
> -------- Message d'origine--------
> De: [hidden email] de la part de Geoffrey De Smet
> Date: jeu. 02/07/2009 20:43
> À: [hidden email]
> Objet : [rules-users] Re: the n best solutions on a very simple test case(drools-solver)
>
> Only the solution which is reached by each chosen step is send to the
> BestSolutionRecaller. If you do only one step, you 'd have to do some
> dirty hack to get those other 4.
>
> Maybe you can use your current Finish combined with AND and a at least 4
> steps taken finish, to force at least (4 - 1 = )3 steps. Combine it with
> a completeSolutionTabu of at least 3 (usually I take a 1000) and you 'll
> have 4 distinct solutions. Your starting solution will also count as one.
>
> But I am interested in this use case :)
>
> 1) Why do you want to recall you best 4 solutions? Only the best
> solution seems relevant to me (but I could be wrong). If this is a
> genuine use case, I can add a MultiBestSolutionRecaller in the
> drools-solver code.
>
> 2) Extending a BestSolutionRecaller is easy, but how did you hook it up
> into the framework?
> a) Did you build your solver and then do a setBestSolutionRecaller()
> b) Or did you use xstream ability to configure in your
> BestSolutionRecaller implementation directly?
>
> BTW: other feed-back on how to improve drools-solver is welcome
>
> With kind regards,
> Geoffrey De Smet
>
>
> Laurent Michenaud schreef:
>> Hi,
>>
>> We have implemented a best score recaller to keep
>> the n best found solutions.
>> It seems to be ok except on very simple
>> test cases :
>>
>> - On the first step, 4 moves are tested. They all
>> return a feseable score. The best one is kept and
>> sent to the solution recaller.
>> - After each of these moves, no other moves are
>> possible and the other solutions of the first step
>> are not sent to the solution recaller.
>>
>> So, in our solution recaller, we have only one
>> solution.
>>
>> Thanks for your help
>> Michenux.
>>
>>
>>
>> _______________________________________________
>> rules-users mailing list
>> [hidden email]
>> https://lists.jboss.org/mailman/listinfo/rules-users
>>
>
> _______________________________________________
> rules-users mailing list
> [hidden email]
> https://lists.jboss.org/mailman/listinfo/rules-users
>
>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> rules-users mailing list
> [hidden email]
> https://lists.jboss.org/mailman/listinfo/rules-users

_______________________________________________
rules-users mailing list
[hidden email]
https://lists.jboss.org/mailman/listinfo/rules-users