[gecode-users] Accumulated failure count and variables part of a failure

[Christian Schulte]

Hi David,

Let me try to give some reasonable answers:

 - I am not sure whether I really get the first question. But yes, in this
situation the AFC of a _propagator_ can only increase.

 - You are right, the AFC of a _variable_ can in fact decrease in the cases
you describe. It even makes sense: the variable has become less difficult as
less propagators that have failed in the past depend on it.

Actually, your idea of deducing something from the AFC before and after
sounds quite interesting. I would be very interested to know whether this
works for you.

Best
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 David Rijsman
Sent: Wednesday, June 23, 2010 10:57 AM
To: users at gecode.org
Subject: [gecode-users] Accumulated failure count and variables part of a
failure

Hi All,

I am trying to expose as many as possible variables which are part of a
propagator which triggered a failure. Before I execute a branch I store the
accumulated failure count of all variables. Next I execute the branch and
assume this one fails. Right after failing I query the accumulated failure
count of the variables in the failed space, if it has increased I know the
variable is part of the propagator that triggered the failure (if no
constraints have been added). 

Is it true that the accumulated failure count of the variable can only
increase because it has been part of the propagator which was part of the
propagator which triggered the failure? 

I can see some problems which is if a propagator cancels an advisor on a
variable or if the propagator is entailed, this decreases the accumulated
failure count. This can happen before the failure was triggered and therefor
the afc of a variable can decrease although it is part of a propagator which
failed. Is this correct? 

If I would query the space which has been 'recomputed' to execute the
alternative of the branch which triggered the failure, the afc compared to
before the failure happened would be a good indicator if the variable was
part of a failure, do you agree?

David












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