Author: Richard G. Fadden
Date: 09:57:31 02/20/06
Go up one level in this thread
On February 20, 2006 at 11:28:32, Richard G. Fadden wrote:
Note the following is just a warm-up example from the Rybka related
comments that are in the archives - I include this just to start the thread...
>>>>>>>>>> Cut/Paste of a 2004 posting from Vasik Rajlich >>>>>>>>>>>
From Subject: Re: Symbolic: Search, planning, and a prospective
Posted by: Vasik Rajlich on March 15, 2004
< Skipping the text from Steven Edwards>
Note also that within the alpha-beta more-or-less-brute-force approach there
seems to be quite a bit of room for improvement at guiding the search and
improving the evaluation.
For example, some attacks in chess are just crushing. You look at the position,
and you know it's over. In principle, a good evaluation would give such a
position +10 or whatever, however no current programs will do this. In fact,
none of the commercial programs which I own have very good evaluation functions.
There is a lot of room for improvement here, and if you look at
the commercial programs you will see that they do keep making progress. They
have to balance work on their evaluation function with its speed, and when they
consider high weights they have to think about the exceptions, etc, so progress
comes one drop at a time, but it's there.
I am also quite sure that the commercial programs haven't exhausted the topic of
selective search.
It seems to me that the existing framework is quite rich in possibilities, and
there are quite a few clever people who think the same.
Anyway, good luck with your program. Any additions to the computer chess
programmer's arsenal would of course be great ...
Cheers,
Vas
<<<<<<<<<<<<<<<<<<<<<<<<<<End of Cut/Paste <<<<<<<<<<<<<<
> Let's have some fun! Let's try to figure out what is in Rybka. That's
>what most intrigues me right now.
>
> I went through the log of Vasik's postings to CCC, and in 2003 and 2004 he
>very clearly spells out a set of new ideas that he indicates he will pursue in
>Rybka. At the time he reported on the performance of Rybka, and at the time it
>was a relatively fast searcher with BitBoards and most of the agreed good search
>technique. And now the numbers coming out show a slow searcher that still
>manages to go deep, so it looks like it is highly selective (yet it plays great
>chess aparently).
>
> What I find fascinating is that Vasik, in the past, spends quite a bit of
>time "talking" about his thoughts on what must be in Shredder, Fritz, and
>Junior. So he supplies a really clear statement as to the inventions and
>techniques that the authors apparently came up with to make these programs so
>great. In other words, if we spend time speculating on what is in Rybka then
>our postings look just like Vasik's from that time.
>
> Then I notice from the limited informaiton of my search that Vasik is not
>posting in the year 2005. At the end of this time period we see Rybka seemingly
>on top of Shredder, Fritz, Junior, etc., and so it looks like he has been doing
>exactly what he said he would do.
>
> I'd like to cut and paste some of these historic CCC postings in my own
>postings of this thread - as long as my fellow members don't mind.
>
>
>
> I noticed three really interesting "threads" of ideas from Vasik's
>postings, and I'll mention one now. He refers to a possibly "Non-Symmetric"
>evaluation approach. I am not the best expert on CCC concerning this ongoing
>topic, but I think I could summarize why I am intrigued with this notion.
>
>1. Imagine Rybka has a great evaluation of the prospects of passed pawns...
>Then imagine us noting that other programs clearly do not include this logic
>(just assume this is true so I can make my point)... During eval, why give
>other programs "credit" for seeing the true nature of a passed pawn... why not
>assume that the other program is dumb on this subject. So, can we possibly look
>at things in a "non-symmetric" fashion assuming for example that the opponent
>will evaluate a static position with less knowledge?
>
> I suspect that someone here is going to tell me that my description is not
>an actual example of a non-symmetric eval. That's OK, please tell me what you
>think, because I'm just starting the topic so others can pipe up and tell us
>what they know. Also, someone who has been here through '03, '04, and '05 (not
>me) could comment about the past Search and Eval threads where Vasik also
>contributed his thoughts.
>
> What do you think? Is this not fun? If not, then what is fun... (possibly
>a day on a sunny beach)?
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.