Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: New Rybka book in test: STUNNING

Author: Stephen Ham

Date: 10:35:28 02/28/06

Go up one level in this thread


On February 28, 2006 at 13:14:31, Salvador H Cresce wrote:

>On February 28, 2006 at 12:54:20, Uri Blass wrote:
>
>>On February 28, 2006 at 11:09:32, Majd Al-Ansari wrote:
>>
>>>Letting Rybka use an optimized book is almost not fair for other programs.
>>
>>Rybka cannot use that book not under chessbase so I consider it as not part of
>>rybka'a package.
>>
>>I think that the ssdf should not use it in their testing of rybka when they test
>>rybka because they should test commercial products and the combination between
>>.ctg book and rybka is combination of 2 commercial products.
>>
>>I am lucky to have Fritz8 and Junior9 as chessbase engines but I simply do not
>>care about book today and I am not going to use it.
>>
>><snipped>
>>>   Rybka was already beating any engine with its own native book very
>>>handily, so now I guess things will really get ugly.  They might get even uglier
>>>once the last few weaknesses of Rybka are covered (endgame knowledge).
>>
>>You are wrong if you think that lack of some endgame knowledge is the only
>>weakness of Rybka.
>>
>>based on my experience in correspondence games it has more weaknesses.
>>
>>Uri
>
>What are the weaknesses you found on correspondence games? Where do you play
>correspondence games? I play on ICCF.
>
>Salvador

Dear readers,

I agree with Uri.

Before I go further, I should mention that that my analyses of Rybka's
performances have generally been at very long time controls, in tournaments and
matches against other engines. Yes, I'm a correspondence chess player too, but
don't use Rybka to generate moves. I have been using it though to analyze
completed game and positions that have already occurred. I do the latter partly
for my own benefit, but primarily to test the engines against positions that I
think I comprehend very well, having studied it/them for hours.

Long story short, Rybka did complete a tournament run at relatively fast
controls (e.g. 45/40, 25/40, 10/game), partly to compare its results versus
those at very long time controls (e.g. 240/40, 120/40, 60/game, etc). In this
specific tournament versus Junior 9, Shredder 9, Fruit 2.2.1, and the top Toga,
Rybka scored about 50%. Yes, there was too little data. But, I saw specific
evidence of need for improvement in certain areas, which I also saw at long
time-controls. So, I forwarded the games to Vaz with some general comments, to
which he responded with agreement.

One problem that I've seen repeatedly is Rybka doesn't know how to deal with IQP
positions. It's unable to determine when IQPs are an asset or a liability, and
it apparently has no "knowledge" how to use an IQP for advantage, nor how to
attack it. It's my opinion that engines in general struggle here, but Rybka
seems to perform worse than others on this specific issue.

And, there are a few other issues too, which I also showed to Vaz. Still, count
me as a huge Rybka supporter and fan. While the engine still has a lot of room
for improvement, it's the strongest on the market.

All the best,

Steve



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.