Rybka Chess Community Forum
What do you guys think?
(In a middle game position with queens on),which engine has the best evaluation for Rook+Bishop vs Rook+pawn.
Is it :
b) Rybka3 Dynamic?
c) Rybka Larry Kaufman (Rybka 2)
d) or are there any other engines, that can evaluate unusual material balances better than Rybka?
I'd vote for Dynamic, though much more goes on in a position than just "unusual material balances."
> c) Rybka Larry Kaufman (Rybka 2)
You'd like to ask for the other one, basically the engines we're using now are "Rybka 3 LK", because his terms became default.
> or are there any other engines, that can evaluate unusual material balances better than Rybka?
Perhaps, but even then, they'll miss tactical/positional shots that Rybka wouldn't, so I wouldn't advice for Rybkaless analysis.
Not all of his terms, Rybka 3 Human is basically a Rybka also with terms that should work but didn't add playing strength, so it's more a LK version, since it includes more algorithms by him.
Wasn't it that LK became the default for the next version (2.3.1 I believe), and then he proposed new terms that didn't work but should (at least at long time controls), so they only put them on Rybka Human?
If I understand correctly, LK version is more prominent in Rybka Human, (with no increase in playing strength at normal tournament time controls). is this correct?
LK wrote a very interesting article about dynamic material evaluations in Chess Life Magazine 3-4 years ago. If his evaluations are in Rybka Human, that may be the strongest program with unusual material balances.
If my understanding is right, Larry's evaluation functions for Rybka 2 LK became default so that they've been in the next Rybka's and they're in Rybka 3. For Rybka Human he came up with new terms or terms that never worked before without an explanation, and their theory was that such terms would hurt performance at the hyperbullet time controls they were using but that at long time controls Rybka Human would use the additional knowledge to choose better moves.
> If his evaluations are in Rybka Human, that may be the strongest program with unusual material balances.
Someone has to test this, I haven't seen signs of it and it remains only theoretical.
The different Rybka personalities use different eval weights but the eval terms are the same. Rybka 3 Human is more materialistic and also values minor pieces a bit more compared to major pieces. There is no real theory behind it, it's just a different personality.
Well, I still hope Rybka 4 brings the return of Human and Dynamic while providing new personalities for our pleasure (I'm looking forward to that one that has a crazy playing style while keeping Rybka 3 strength.)
Somehow, the R3 version seems to be stronger than HUMAN and DYNAMIC personalities in my machine, keeping hash size and other things constant.
> Somehow, the R3 version seems to be stronger than HUMAN and DYNAMIC personalities in my machine, keeping hash size and other things constant.
That's why is the default. But there are not big strength differences among these versions. In fact, strength differences are almost imperceptible.
Yes, for short time controls *40 moves/2 hours) Rybka # is strongest. But what if time was more avaliable (couple hours/move).
Wouldn't inclusion of more algorythms make Human stronger?
(But if I understand correctly, Vas says they are exactly the same, with different evaluations for material)
> But what if time was more avaliable (couple hours/move).
The strength is diminished relative to other chess entities that are able to draw more games because they find better moves and make less mistakes, but Rybka Default still remains on top on the tests I've seen.
And, yup, I was misunderstanding, Human and Dynamic only use different weights (not only of material.)
I had for some time the theory that less materialistic play would work better at shorter time controls, but it wasn't supported by the data.
Though the data showed that higher contempt does work better at shorter time controls (against non Rybka engines, which maybe is uninteresting because basically Rybka beats them no matter what.)
Did anybody thought before Rybka times that discussions on the future about what was the best chess engine would be about different personalities of the same engine?
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill