[gecode-users] mzn-gecode search strategy

[Guido Tack]

On 24/04/2013, at 20:32, Tias Guns <tias.guns at cs.kuleuven.be> wrote:

> On Thu, 18 Apr 2013 09:36:13 +0200, Guido Tack <tack at gecode.org> wrote:
> 
>> Hi Tias,
>> 
>> it's slightly complicated (and it has changed in Gecode 4).  The default search is implemented in gecode/flatzinc/flatzinc.cpp in function createBranchers, and the defaults in 4.0.0 are
>> 
>>    branch(*this, iv_sol, INT_VAR_AFC_SIZE_MAX(0.99), INT_VAL_MIN());
>>    branch(*this, bv_sol, INT_VAR_AFC_MAX(0.99), INT_VAL_MIN());
>>    branch(*this, fv_sol, FLOAT_VAR_SIZE_MIN(), FLOAT_VAL_SPLIT_MIN());
>>    branch(*this, sv_sol, SET_VAR_AFC_SIZE_MAX(0.99), SET_VAL_MIN_INC());
>> 
>> where iv_sol are the variables mentioned in the output statement and not mentioned in an explicit search annotation, and
>> 
>>      branch(fzs,fzs.iv_aux,INT_VAR_AFC_SIZE_MAX(),INT_VAL_MIN());
>>      branch(fzs,fzs.bv_aux,INT_VAR_AFC_MAX(),INT_VAL_MIN());
>>      branch(fzs,fzs.sv_aux,SET_VAR_AFC_SIZE_MAX(),SET_VAL_MIN_INC());
>>      branch(fzs,fzs.fv_aux,FLOAT_VAR_AFC_SIZE_MAX(),FLOAT_VAL_SPLIT_MIN());
>> 
>> where iv_aux are variables introduced by mzn2fzn that are not functionally defined.
>> 
>> I'm planning to introduce annotations default_{int,bool,set,float}_search that you can stick on the solve item to change the default for the different variable types - would that be useful?
> 
> Yes.
> 
> But since your asking, would there be a way to express the exact search strategy (on the *_sols) in minizinc itself, instead of introducing a new default_* annotation?
> 
> For example:
> int_search(IntVars, size_afc_max(0.99), indomain_min, complete);
> 
> (I guess the only difference with what is available now is adding the 0.99?)
> 

Yes, that's possible. I'll add support for this next week.

Cheers,
Guido


> 
> Kind regards,
> Tias
> 
> Disclaimer: http://www.kuleuven.be/cwis/email_disclaimer.htm
> 
> _______________________________________________
> Gecode users mailing list
> users at gecode.org
> https://www.gecode.org/mailman/listinfo/gecode-users