[gecode-users] Using IntActivity with space

[Christian Schulte]

Hi Kish,

As said, don't wait up for Activity with dynamically adding variables...

Yes, activity information must be initialized (either by restarts or even
better by randomized probing) to be useful.

Variables do not have a member function for activity, that's the whole
point. Check 
	gecode/kernel/brancher-merit.hpp
how to use activity. The access is by the index of the variable in the
variable array used for initializing the activity object.

Yes, activity must be updated during cloning via an update() function.

Yes, recomputation has an effect on activity. But that's okay, activity is
just a measure and does not need to be 100% accurate.

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 Kish Shen
> Sent: Thursday, July 11, 2013 12:46 AM
> Cc: users at gecode.org
> Subject: Re: [gecode-users] Using IntActivity with space
> 
> Hi Christian,
> 
> Thanks. It would be great if activity can be supported by dynamically
added
> variables.
> 
> However, if the user chose to do the search in Gecode, then I guess I can
create
> a IntActivity for those variables just before doing the search.
> 
> I assume the activity information would be very poor at the start, since
it would
> not include any propagation done during the modelling phase.
> 
> It may also be possible to use activity for my (that is the interface's)
variable
> selection, which does the selection at the C++ level, so the activity
information
> need not be associated with all variables, as the ECLiPSe user would not
be able
> to access the activity information themselves.
> 
> I will still need to access the activity for a variable in an IntActivity
though, and
> looking at the class reference for IntActivity, I can't see a method to
get this.
> Also, as cloning and recomputation is done by the interface code, do I
need to
> do anything with this IntActivity object for cloning?
> 
> About recomputation, there might be an issue -- both activity and AFC are
based
> on the history of the computation, e.g. AFC is the accumulated failure so
far in
> the search space, and not just the current branch.
> However, recomputation will recompute previous computation, and probably
> should not again for AFC and activity; I assume it is OK for AFC as
recomputation
> does not fail, but I guess activity might be affected?
> 
> Thanks and cheers,
> 
> Kish
> 
> On 10/07/2013 09:45, Christian Schulte wrote:
> > Hi Kish,
> >
> > No, there is no easy way to get activity for dynamically added
variables.
> > Doable but this would be a very different implementation. In contrast
> > to AFC, activity lives completely outside the kernel.
> >
> > Cheers
> > Christian
> >
> > --
> > Christian Schulte, Professor of Computer Science, KTH,
> > www.ict.kth.se/~cschulte/
> >
> >
> > -----Original Message-----
> > From: users-bounces at gecode.org [mailto:users-bounces at gecode.org] On
> > Behalf Of Kish Shen
> > Sent: Tuesday, July 09, 2013 8:49 PM
> > To: users at gecode.org
> > Subject: [gecode-users] Using IntActivity with space
> >
> > Hi,
> >
> > I am now looking at adding the variable activity selection to the
> > interface I am writing.
> >
> >   From reading the MPG, it looks like IntActivity has a similar API to
> > IntAFC, and so cannot be easily used in my code. So is there anyway to
> > setup and use it in a similar way to what Christian suggested for AFC
> previously?
> > That is,
> >
> > 1) to setup the recording of activity for all variables in the space,
> >      including variables that are dynamically added
> > 2) to obtain the activity of a variable outside of the search engine
> > 3) change the decay for all variables
> >
> > For 2), I could not find an equivalent function to afc() in the
> > Gecode::IntVar Member List page that Christian pointed to previously
> > (which does list the afc method).
> >
> > Thanks in advance for any information!
> >
> > Cheers,
> >
> > Kish
> >
> > _______________________________________________
> > Gecode users mailing list
> > users at gecode.org
> > https://www.gecode.org/mailman/listinfo/gecode-users
> >
> 
> 
> _______________________________________________
> Gecode users mailing list
> users at gecode.org
> https://www.gecode.org/mailman/listinfo/gecode-users