Hi Zichen,
Yes, you are right the minimodeling layer (that is expr()) should do a
better job at avoiding posting reified constraints if it is not necessary
(as you did in your own improved variants).
I'll write that down for a future release.
Let me try to explain what happens: with your first expression you create
three Boolean variables b1, b2, b3 and then three reified propagators:
!_A[i] <=> b1
(x[index] != val) <=> b2
b1 || b2 = b3
and expr() returns b3. Unfortunately the constraints might stay on even
though the disjunction is already satisified (which it is in you example).
Embarrasingly enough, I am the co-author of a paper (done using Gecode) that
explains the problem and shows how to eliminate these constraints but it has
not yet made it into Gecode:
http://web.it.kth.se/~cschulte/paper.php?id=SchulteStuckey:PPDP:2008
I hope that helps.
Best
Christian
--
Christian Schulte, Professor of Computer Science, KTH,
www.ict.kth.se/~cschulte/
From: users-boun...@gecode.org [mailto:users-boun...@gecode.org] On Behalf
Of Zhu Zichen's cse
Sent: Tuesday, September 03, 2013 4:38 AM
To: users@gecode.org
Subject: [gecode-users] Expr() in GECODE
Hi,
I have a question.
For GECODE,
When I use expressions:
BoolVar sofar = expr( home,( !_A[i] || x[index] != val ));// where
_A[i] is a BoolVar, and x is IntVarArray, index and val are integer.
If I rewrite it as:
if(_A[i].one())
sofar = expr( home, x[index] != val);
else
sofar = expr( home,( !_A[i] || x[index] != val));
There will be a great improvement on runtime, say 40%.
If I rewrite it as:
if(_A[i].one())
x[index].nq(val);
else
sofar = expr( home,( !_A[i] || x[index] != val));
There will be another improvement on the latest one for runtime, say 30%.
For ILOG,
IlcConstraint sofar = (! _A[i]) || x[index] != val ) ;
Rewrite as:
if(_A[i].isTrue()==IlcTrue)
sofar = x[index] != val;
else
sofar = (! _A[i]) || x[index] != val ) ;
Running times of the above two methods are the same.
Is this because the "inefficient" of expr() to handle Boolean expressions
and relations in GECODE?
Thanks!
Zichen
_______________________________________________
Gecode users mailing list
users@gecode.org
https://www.gecode.org/mailman/listinfo/gecode-users
