[gecode-users] propagation consistency of count (single value) constraint

[Christian Schulte]

Hmm, I had a quick look at the code and I think it should rewrite to the
int/int case. I am travelling right now, so I will look more carefully when
I am back in a week or so.

Cheers
Christian

--
Christian Schulte, www.gecode.org/~schulte 
Professor of Computer Science, KTH, cschulte at kth.se
Expert Researcher, SICS, cschulte at sics.se

-----Original Message-----
From: users-bounces at gecode.org [mailto:users-bounces at gecode.org] On Behalf
Of Kish Shen
Sent: Thursday, August 27, 2015 05:48 PM
To: users at gecode.org
Subject: Re: [gecode-users] propagation consistency of count (single value)
constraint

As a follow up to my previous post on this, then from the reference
documentation, for

count(home, x, y, IRT_EQ, z)

where y and z are int, then there is only one consistency level supported,
and it is not domain consistent.

We have a rostering program (originally written for our ic solvder), which I
used to test the Gecode interface, and it uses count constraints, and
changing the consistency level affected the execution times significantly
for certasin problem instances. I took a look again at the code, and the
count constraints are with y and z being integers.
(the execution times are faster with ICL_BND than ICL_DOM)

The count single vsalue constraint is posted to Gecode using only one
propagator, the version with y and z being IntVar (and if they are integers,
they are posted as IntVar with a singleton domain). This is done for all
constraints that have variants that accept both IntVar and int, as this
avoid me having to write a lot of code to distinguish the argument types,
and I remember asking about this here a while ago and I think Chirstian said
that Gecode will figure out the best propagator to use for the argument.

So my questions are: is the propagator posted the one with y and z being
int? If so, is there more than consistency level for the propagator, as my
result seem to suggest? The difference is very significant, e.g. for one
instance:

ICL_BND: 0.73s  ICL_DOM: 198s

the ICL_BND times are more than 100 times faster, and the only difference
between the two runs are the count constraint posted with ICL_BND rather
than ICL_DOM (running on the same machine, of course).


Cheers,

Kish





_______________________________________________
Gecode users mailing list
users at gecode.org
https://www.gecode.org/mailman/listinfo/gecode-users