Hi Christian,
To me, legal but not sensible values should be allowed -- some of the
variable and value selection strategies implemented by Gecode probably
are not very sensible also. In addition, the AFC decay factor can be set
to 0 already using afc_decay(0.0), so making the initialision of an
IntAFC consistent with that seems reasonable to me.
Also, the decay factor for IntActivity can be set to 0, although I guess
it might make more sense here, as AFC is only changed when a failure occurs.
Cheers,
Kish
On 23/05/2014 08:49, Christian Schulte wrote:
Hi Kish,
Yes, a decay factor of zero makes normally no sense. If you want me to, I
can change that.
Yes, you are right, there is only one global AFC decay factor.
Cheers
Christian
Christian Schulte, KTH, web.it.kth.se/~cschulte/
-----Original Message-----
From: users-boun...@gecode.org [mailto:users-boun...@gecode.org] On Behalf
Of Kish Shen
Sent: Friday, May 23, 2014 04:13 AM
To: users@gecode.org
Subject: Re: [gecode-users] AFC decay with global AFC in search engines
Hi Christian,
Thanks for this. I decided to change my code to ude INT_VAR_AFC_* with a
decay argument, and then I tried to rerun the test. However, I now get a
Gecode exception "AFC: Illegal decay factor" for a factor of 0.0 with
IntAFC afc(*solver, vars, 0.0);
while previously, I was able to set the decay factor to 0 with
afc_decay(0.0).
I assume this is because of a > vs >= test for the valid value. Should this
be changed? It seems 0 should be a legal value (even if it is not too
sensible).
On a related note, from the description in the MPG, I assume there is only a
single AFC decay factor, so setting the decay factor with
IntAFC afc(*solver, vars, decay)
will also change the "global" decay factor, i.e. the one obtained by
afc_decay(). Is this correct?
Cheers,
Kish
_______________________________________________
Gecode users mailing list
users@gecode.org
https://www.gecode.org/mailman/listinfo/gecode-users
_______________________________________________
Gecode users mailing list
users@gecode.org
https://www.gecode.org/mailman/listinfo/gecode-users
