[gecode-users] Any Advise on Delta

[Christian Schulte]

Hi David,

I just checked the code for minus_r and I was a little surprised: I just
forgot to add the detection of this special case for minus_r, the other
iterator-based variable operations check for special cases (precisely to get
a more accurate delta). I'll jot it down to my list and it should be in
3.0.0 as it is not hard (Mikael, sorry for contradicting you here ;-) ).

Cheers
Christian

--
Christian Schulte, www.ict.kth.se/~cschulte/


-----Original Message-----
From: users-bounces at gecode.org [mailto:users-bounces at gecode.org] On Behalf
Of Mikael Zayenz Lagerkvist
Sent: Wednesday, January 21, 2009 3:48 PM
To: David Rijsman
Cc: users at gecode.org
Subject: Re: [gecode-users] Any Advise on Delta

Hi,

In the case of cumulatives, the domain information could be reported
in a better way since singleton ranges are removed, mapping naturally
onto the integer delta abstraction. Unfortunately, this would
complicate the code for the minus_r operation, and thus it is not
clear that we will do it.

If your interest is to check if the upper bound of a variable has
changed, then caching the maximum value in the advisor is simple and
rather efficient. This is probably what I would do, and it is
independent of the delta-information that is computed.

Cheers,
Mikael

On Wed, Jan 21, 2009 at 2:38 PM, David Rijsman
<David.Rijsman at quintiq.com> wrote:
> Christian,
>
> in a previous conversation on 'Any Advise on assignment delta' you
mentioned "the other operations that lead to 'any' are iterator-based
variable domain updates which in most cases just remove a single value or a
range (and hence the delta information is accurate).". Concerning this I
have the following question:
>
> I have been implementing a propagator which implements its own advise
method. The intention of the advise method is to only schedule the invoking
propagator if the maximum has changed. This is relatively easy to implement
but I noticed that the 'delta' reporting is rather 'rough' implemented and
my propagator gets scheduled quite often to conclude nothing. That is I am
getting a lot of 'any' hits when there could have been more information.
Specifically in my use case I have a cumulatives constraint whose propagator
uses View::minus_r a lot and this propagator only reports 'any' deltas (as
far as I can conclude). In the example in this e-mail clearly shows we could
be more informative in the delta, it streams 1-0 (any) although I would
expect 0-50.
>
> My question is if this is done for a reason and if so if this is done for
many operators on the views?
>
> //----------------------------------
>
> #include "gecode/kernel.hh"
> #include "gecode/int.hh"
> #include "gecode/iter.hh"
>
> namespace Gecode
> {
>  class MyPropagator : public Propagator
>  {
>  public:
>    MyPropagator(Space *home, Int::IntView v):
>        Propagator(home),
>        MView(v)
>    {
>      MView.subscribe(home, new (home)
ViewAdvisor<Int::IntView>(home,this,MCouncil,MView));
>    }
>
>    MyPropagator* copy (Space* home, bool share)
>    {
>      return 0;
>    }
>
>    ExecStatus propagate(Space *home, ModEventDelta med)
>    {
>      return ES_OK;
>    }
>
>    PropCost  cost (ModEventDelta med) const
>    {
>      return PC_LINEAR_LO;
>    }
>
>    ExecStatus  advise (Space *home, Advisor *a, const Delta *d)
>    {
>      ViewAdvisor<Int::IntView>& advisor =
static_cast<ViewAdvisor<Int::IntView>&>(*a);
>
>      std::cout << advisor.view().min(d) << "-" << advisor.view().max(d) <<
std::endl;
>
>      return ES_OK;
>    }
>  private:
>    Int::IntView MView;
>    Council<ViewAdvisor<Int::IntView>> MCouncil;
>  };
>
>  class MySpace : public Space
>  {
>  public:
>    MySpace()
>    {
>      MX.init(this,0,100);
>
>    }
>
>    MySpace(bool share, MySpace &s):
>      Space(share,s)
>    {
>       MX.update(this,share,s.MX);
>    }
>
>    MySpace* copy(bool share)
>    {
>      return new MySpace(share, *this);
>    }
>
>    IntVar MX;
>  };
> }
>
> int main()
> {
>  Gecode::MySpace space;
>
>  new (&space) Gecode::MyPropagator(&space,space.MX);
>
>  Gecode::Int::IntView view(space.MX);
>
>  Gecode::Iter::Ranges::Singleton d(0,50);
>
>  view.minus_r(&space,d,false);
> }
>
> //----------------------------------
>
> David
>
> Quintiq Conference "Quintessence 09" Tuesday May 12th, 2009, Country
Estate Duin & Kruidberg, near Amsterdam Schiphol, The Netherlands - for more
information visit www.quintiq.com
>
> This message contains information that may be privileged or confidential
and is the property of Quintiq. It is only intended for the person to whom
it is addressed. If you are not the intended recipient, you are not
authorized to read, print, retain, copy, disseminate, distribute or use this
message or any part thereof. If you have received this message in error,
please notify the sender immediately and delete all copies of this message.
Please note that e-mails are susceptible to change, therefore they are not
binding.
>
> _______________________________________________
> Gecode users mailing list
> users at gecode.org
> https://www.gecode.org/mailman/listinfo/gecode-users
>



-- 
Mikael Zayenz Lagerkvist, http://www.ict.kth.se/~zayenz/

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