Re: [eclipse-users] How to ic:search/6 with symmetry breaking during search (sbds) of lib(ic_sbds)?

Hi Kish,

First of all, thanks for your answer!

In some points I wasn't really precise, so I try to clear it up a bit...

Kish Shen wrote:
> Hi Martin,
> 
> I could not really help you with the symmetry breaking part of your 
> question,  but as no one else have replied to your message yet, I 
> thought I should reply and point out some points about the ECLiPSe code 
> you presented -- I am not sure if the predicate you gave does what you 
> intended:
> 
> Martin Berger wrote:
>> rotate90(Var,Value,SymVar,SymValue) :-
>> (
>>     (not var(Value)) -> (
>>         Var is SymVar,
>>         (Value == 'left') ->  SymValue is 'rear' ; true,
>>         (Value == 'right') ->  SymValue is 'front' ; true,
>>         (Value == 'front') -> SymValue is 'left' ; true,
>>         (Value == 'rear') -> SymValue is 'right' ; true
>>     ) ; true
>> ).
>>
>>   
> Your bracketing for the if-then-else is unusual and probably does not do 
> what you intend. Normally if-then-else would be bracketed in the 
> following way:
> 
> ( Cond -> Then ; Else)
> 
> and I assume you intend the above to be:
> 
>        (Value == 'left'  ->  SymValue is 'rear' ; true),
>        (Value == 'right' ->  SymValue is 'front' ; true),
>        (Value == 'front' -> SymValue is 'left' ; true),
> 
> which is not what your bracketing does.
> 

You are absolutely right! I forgot to put the closing brackets into the 
example code! ;-)

> Secondly, I am not sure if
> 
> SymValue is 'rear'
> 

You are right again! I really should have pointed out, that 'rear' in my 
example code represents a numeric value, i.e. the value 3. I tried to 
make the example code more intuitive and "speaking" by replacing the 
numeric value with the actual meaning of the value. Sorry for bringing 
you on the wrong track!

> is doing what you intended. is/2 is used to evaluate arithmetic 
> expressions, and  'rear'  is treated as a function call to rear/1, i.e. 
> the above is equivalent to:
> 
> rear(SymValue)
> 
> and it is probably clearer to write it that way.
> 
> If your intention is to assign the atom rear (you don't need the quotes 
> for simple atoms like these) to SymValue, you should use =/2 instead of 
> is/2:
> 
> SymValue = real
> 
> The last point is that I would be somewhat surprised if SymValue can be 
> assigned to non-numeric values, as variables for lib(ic) can only take 
> on numeric values (i.e. symbolic finite domain is not directly 
> supported). However, this may just reflect my lack of knowledge for the 
> symmetry breaking library.
> 

See the antecedent explanation for the is/2 and =/2 issue.

 >
 >
 >

Some further observations of mine on a smaller straightforward test 
problem instance: I run my code on a problem where I know that there are 
2 nonsymmetrical solutions. Without sbds my code is given me 4 solutions 
where 2 of them are symmetrically equivalent to the former ones. When I 
plug in sbds with my 8 symmetry predicates, the search comes up with 
14317 solutions (only duplicates of the 4 solutions I got without sbds)!

Somehow even a lot more solutions are instantiated. Is it because I only 
exchange (in a symmetrical sense) the values and not the variables?

Anyway, Kish thanks for pointing out the inprecise parts of my explanation!

Cheers,

Martin


> Cheers,
> 
> Kish
> 
> 
>> In my problem, I am facing a pure value symmetry, at least as far as I 
>> know and understand it! Thats why "Var is SymVar" and the value is set 
>> to a symmetrical one. The symmetrical group is the dihedral group, the 
>> symmetry group of a square. The group elements are: identity, 
>> rotate90/180/270, reflextionX, reflextionY, flipCorner1, flipCorner2.
>>
>> However now, search is given my a lot more symmetrical solutions than 
>> without sbds. I am not sure if the problem are the symmetry predicates 
>> or there is any conflict/sideeffect with the rest of my code. Am I 
>> doing or understand something completely wrong?
>>
>> I would happy to see more examples (of you) or a tutorial of how to 
>> write symmetry predicates for sbds. Or can anybody of you give me some 
>> advice or hints?
>>
>> Thank you for your help!
>>
>> Best regards,
>> Martin Berger
>>
>> _______________________________________________
>> ECLiPSe-Users mailing list
>> ECLiPSe-Users_at_crosscoreop.com
>> http://www.crosscoreop.com/mailman/options/eclipse-users
>>