Author: JNoomen
Date: 03:20:03 01/01/06
Go up one level in this thread
On January 01, 2006 at 04:44:52, Majd Al-Ansari wrote: Hi Madj, Interesting observations, please let us know about your analysis! Concerning the book questions: A) Indeed I have seen that Fruit doesn't like the Fritz 9 book. When I let Fruit play with this book, the results went down sharply by around 15%. B) There will be a tuned book for Rybka quite soon. Maybe you can do the tournament again with all engines using their own books? Best wishes for 2006, Jeroen >I have just finished a 300 game tournament with the top programs at 5 min + 3sec >and I have to say I am extremely impressed with Rybka. I will post highlights >of the games as soon as I finish my analysis of all the games. Rybka came way >out on top with Shredder and Fritz 9 comming way behind with Fruit at the >bottom. This match might have been unfair for Fruit since book used was the >Fritz 9 book. What is very impressive is that Rybka was able to win so easily >even though it has HUGE gaps in some endgame knowledge. Mind you it can play >some fantastic endings and beat the best of the best in some endgames. But in >some endgame positions it plays less than a 1200 elo player. I am sure that >this will be easily fixed especially since a superb source such as Fruit 2.1 is >available. My first impressions about Rybka is that with a little more endgame >knowledge, a specially tuned book and a little better time management, Rybka >could be around 50 go 100 elo points even stronger. I will post why I think >that as soon as I finish my analysis (I am checking all 300 games). The most >notable thing I noticed is that Rybka plays so fast yet sooooo STRONG. The >other engines are struggling to figure out the positions yet Rybka seems to play >effortlessly and accurately. I remember Kramnik mentioning that computers are >good at searching but are terrible at searching in the critical lines. I think >Rybka is exploiting this somehow. The way it pushes pawns and plays unbalanced >positions is truly something that will impress even the most seasoned super GM. >It almost seems like a super GM is watching over and telling Rybka which >critical lines it should search instead of wasting cpu cycles on the wrong path. > If it doesn't find a quick finish through search or things are unclear, it just >adds another move that intuitevely seems strong without really going through all >the consequences of the move. Somehow it has the intuition to just load more >pressure without really calculating till a huge plus. > >I would like to say that Fritz 9 seems to best the most mature engine. It >immediately realizes a position is drawn while the other software programs think >is won (Rybka is especially notorious for that). Shredder is also a very well >rounded program and is extremely strong and I think with a better tuned >evaluation will be even stronger. Fruit 2.2.1 my previous favourite is also a >fantastic engine but seems to play really poorly in positions it does not like >(it needs a tuned book to perform well). It is just that Rybka is playing a >different type of chess that are just not understandable to the engines out >there today. > >If you are impressed with Rybka now, let me warn you .... you ain't seen nothing >yet!
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.