Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Slate's Challenge of the Day

Author: Jeremiah Penery

Date: 19:48:31 10/07/03

Go up one level in this thread


On October 07, 2003 at 20:56:00, Jaime Benito de Valle Ruiz wrote:

>Surely if you tune an engine properly it'll find "amazing" moves now and then
>(esp. when the King's safety is at stake) by over-rating the chances of an
>attack. But, what about the rest of the games? I've managed to configure lots of
>engines to find extremely hard sacrifices... but after several games, their
>overall performance become clearly much worse than normal in normal games!
>I just find it estrange that most commercial engines such as Fritz, Shredder or
>Junior (to name three) can’t see anything within a minute, but your version has
>no problems at all finding it in almost no time!!!
>If it shows a more consistent behavior, I want a copy!! Otherwise, it's not an
>useful result
>Regards,

I completely agree!  For a *long* time, I've been trying to make Crafty
understand more about king-safety and attacking, but I've had only moderate
success.  The stuff I have in there now was written quite a long time ago and I
never finished adjusting or tuning it.  You can see that my NPS is only about
1/3 of what a normal Crafty would get on my machine (AthlonXP 1.6GHz), because
of these intensive attack calculations.  Most of the time, when I let it search
for a long time it comes up with good analysis, but it is very bad at playing
blitz games, precisely for the reason that it overevaluates its attacking
chances way too often.

What I have now isn't supposed to be tuned to find material sacrifices like
this, but rather I want it to correctly evaluate attacking potential.
Obviously, there are some positions where this works wonderfully (like the
posted position here), but many positions where it fails badly.  Strangely
enough, in many of the king-attacking positions that have been posted here, this
version completely fails to see the attacking chances.  So, I have a lot to work
on. :)

If I ever manage to get something that I think works the way I want it to work,
I will announce it here and put it up for download.  If not, then I will
occasionally post search results here from it, even if they might be
meaningless.  I think it's still interesting to give the analysis - it's easier
to dismiss a posted analysis than to imagine one that's not posted at all. :)



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.