Originally posted by VarenkaI'm certainly not an expert regarding chess engines, but the latest scuttlebutt I've read regarding Rybka's strength is interesting. When Rybka first came out, people noticed Rybka's extremely low node count. The author of Rybka, Vasik Rajlich, claimed that Rybka incorporated lots of evaluation code, which slowed down its speed. However, the latest rumors seem to imply that Vasik's explanation was a smokescreen to divert attention from the real reason, which was that Vasik somehow found a way to greatly optimize the engine's code. So the rumor is that Rybka is good because its code executes extremely fast. (The rumor is that Vasik must have divided the true node count by some arbitrary number.)
Test groups such as CCRL and CEGT test the latest engines and the games can be downloaded.
The latest Rybka is much stronger than Fritz 5.32. However, if this was based on tactical ability alone, the difference is much smaller. In fact, I wouldn’t be surprised if Fritz 5.32 is still slightly better in tactical tests. So, how do you categorise the fac ...[text shortened]... Would you term it “positional ability”, “strategic ability” or what? It’s not tactical ability.
Originally posted by Varenkayeah, the rybka-hiarcs games on CCRL seem a lot more like 'normal' chess... there were some funny endings, but otherwise the games I looked at seemed pretty okay...
Test groups such as CCRL and CEGT test the latest engines and the games can be downloaded.
The latest Rybka is much stronger than Fritz 5.32. However, if this was based on tactical ability alone, the difference is much smaller. In fact, I wouldn’t be surprised if Fritz 5.32 is still slightly better in tactical tests. So, how do you categorise the fac ...[text shortened]... Would you term it “positional ability”, “strategic ability” or what? It’s not tactical ability.
2 edits
Originally posted by droflaceOK, I just started a 300-game tournament between two identical Toga II 1.4 Beta 5c engines on my PC. Both engines are using Scorpio 5-piece bitbases. Time control is 40 moves in 4 minutes, repeating. The opening book is the rather smallish mainbook that comes with Arena 1.1. No idea if a larger book would make much of a difference, though.
Inspired by the white vs. black thread that was recently revived, I began to wonder what the results look like when the same engine on a computer with the same specs are matched against each other?
I know that engines vary their openings, but once they get out of book do they play exactly the same way each time? What is the W/L/D record over a large number of games? What does this tell us about the nature of chess--is it a forced draw?
It may take a day or two to get a feel for the draw percentage. I expect over 50 percent draws, but I have no idea how much over 50 percent.
Stay tuned, same time, same Bat channel...
Originally posted by Mad RookAfter 10 games, Engine 2 is off to a good start with 4 wins, 1 loss, and 5 draws (+108 elo). Of course, that advantage won't hold. At least the draw percentage is in the ballpark of expectations.
OK, I just started a 300-game tournament between two identical Toga II 1.4 Beta 5c engines on my PC. Both engines are using Scorpio 5-piece bitbases. Time control is 40 moves in 4 minutes, repeating. The opening book is the rather smallish mainbook that comes with Arena 1.1. No idea if a larger book would make much of a difference, though.
It may take a ...[text shortened]... raws, but I have no idea how much over 50 percent.
Stay tuned, same time, same Bat channel...
I won't bother with many (if any) more updates until the final result, as this stuff is really kind of boring.
Originally posted by Mad RookI have another engine tournament to run, so I'm pulling the plug on this one at 100 games. Final score for Engine 2 was 28 wins, 18 losses, and 54 draws. That's a score of 55 percent (+35 elo) with a draw rate of 54 percent.
After 10 games, Engine 2 is off to a good start with 4 wins, 1 loss, and 5 draws (+108 elo). Of course, that advantage won't hold. At least the draw percentage is in the ballpark of expectations.
I won't bother with many (if any) more updates until the final result, as this stuff is really kind of boring.
Since the two engines are identical, I have to assume that the score would have eventually approached 50 percent. CCRL states that if 200 games are played, the error margin is about plus or minus 40 elo. (And still plus or minus 25 elo at 500 games.) So at least I was within the error margin.
The draw rate varied between the mid 40 percent to mid 50 percent range. It would be interesting to run a similar tournament with the opening books disabled to see how many identical or nearly identical games would be played.