[gecode-users] Reified Constraints do not propagate
Guido Tack
tack at gecode.org
Tue Apr 17 01:20:58 CEST 2012
Hi,
I just added a fix to the svn trunk for the case of a reified equality between an IntVar and an int. The default ICL for that case is now ICL_DOM (consistent with just using rel directly).
Other cases (e.g. equality between two variables) may still ignore the ICL, but we'll have a look at that, too.
Cheers,
Guido
--
Guido Tack
http://www.csse.monash.edu/~guidot/
On 13/04/2012, at 9:17 PM, Max Ostrowski wrote:
> Thanks and sorry for bugging you again.
>
> I tried to analyze the minimodel code.
> I think that posting LinExpr instead of LinRel won't change the problem, right?
> So is there any interface so that one can still use the very handy LinExpr class?
> I do parse some quite complicated formulas and want to post a linear() constraint.
> LinExpr as far as I understood, optimizes this formula and reformulates it to fit into the form that is needed for the linear constraint.
> Is there a way to get this linear constraint out of it again to post it manually and preserve the ICL?
>
> Or do i have to generate the linear constraint (array of variables and numbers) myself to get the full propagation power. Is there a logical reason why the ICL is lost during this "translation"?
>
> Best,
> Max
>
> On 04/10/2012 01:19 PM, Christian Schulte wrote:
>>
>> Maybe, not so clear yet. We will change reification for the next major release and might at that time also think how to fix passing the consistency level.
>>
>> For the time being you should stick to what Guido has suggested.
>>
>> Christian
>>
>> --
>> Christian Schulte, www.ict.kth.se/~cschulte/
>>
>> From: users-bounces at gecode.org [mailto:users-bounces at gecode.org] On Behalf Of Max Ostrowski
>> Sent: Tuesday, April 10, 2012 12:42 PM
>> To: Guido Tack
>> Cc: users at gecode.org
>> Subject: Re: [gecode-users] Reified Constraints do not propagate
>>
>> Thank you very much for the insight.
>> Unfortunately this could be a problem,
>> as i have to change my data structures.
>>
>> Will this be fixed in future versions?
>>
>> Best,
>> Max
>>
>> On 04/10/2012 12:32 PM, Guido Tack wrote:
>> This is a current shortcoming of the minimodel library. For some linear constraints (such as x==5 in this case) the IntConLevel is ignored. To get the full propagation, you have to post the constraint directly, like this:
>>
>> Gecode::rel(*test, test->x_[1], IRT_EQ, 5, test->b_[1], ICL_DOM);
>>
>> Cheers,
>> Guido
>>
>> On 10 Apr 2012, at 19:20, Max Ostrowski wrote:
>>
>>
>> I have a small example where i do not understand why no propagation takes place.
>>
>> test is a pointer to the space instance
>> x_ is an array of 10 IntVars(0,10)
>> b_ is an array of 2 BoolVars(0,1)
>>
>>
>> LinRel rel1(test->x_[0],IRT_EQ,5);
>> rel1.post(*test,true,ICL_DEF);
>>
>> Gecode::distinct(*test,test->x_,ICL_DEF);
>>
>> LinRel rel2(test->x_[1],IRT_EQ,5);
>> Gecode::rel(*test, rel2 == test->b_[1],ICL_DEF);
>>
>> test->status();
>> if (test->failed())
>> std::cout << "Failed" << std::endl;
>> std::cout << test->x_ << std::endl;
>> std::cout << test->b_ << std::endl;
>>
>>
>> So i post das one of my elements x_[0] is 5, and all elements are distinct.
>> Now i have a reified constraint that x_[1] == 5.
>> The distinct constraint removes 5 from he domain of x_[1],
>> but the propagator does not fire.
>> The result is:
>> {5, {0..4,6..10}, {0..4,6..10}, {0..4,6..10}, {0..4,6..10}, {0..4,6..10}, {0..4,6..10}, {0..4,6..10}, {0..4,6..10}, {0..4,6..10}}
>> {[0..1], [0..1]}
>>
>> Even with ICL_DOM the boolean variable is not touched if values are removed from the domain.
>> Also i think this could be easily done with "Advisors" ?
>> I have no experience with this, so my question:
>> Is this behavior the desired one?
>> Can this be changed if i have reifications like mine where i compare to constants?
>>
>> Thanks a lot,
>>
>> Max
>> _______________________________________________
>> Gecode users mailing list
>> users at gecode.org
>> https://www.gecode.org/mailman/listinfo/gecode-users
>>
>> --
>> Guido Tack,
>> http://www.csse.monash.edu/~guidot/
>>
>>
>>
>>
>
> _______________________________________________
> Gecode users mailing list
> users at gecode.org
> https://www.gecode.org/mailman/listinfo/gecode-users
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.gecode.org/pipermail/users/attachments/20120417/588a3ac6/attachment.htm>
More information about the users
mailing list