Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / What does each of the four different Rybka engines do?
- - By RookieMeister (*) Date 2009-10-26 16:26
What does each of the following four different Rybka engines do?

Rybka 3 1-cpu w32
Rybka 3 960 1-cpu w32
Rybka 3 Dynamic 1-cpu w32
Rybka 3 Human 1-cpu w32
Parent - - By Uly (Gold) Date 2009-10-26 17:13
They all do the same, that all engines do, analyze chess positions. If you refer to the differences, 960 manages castling rules differently and is capable of playing Chess960, Human is more materialistic, and Dynamic likes sacrifices.
Parent - By morphyzen (**) Date 2009-10-28 13:26
I had the same exact question! I pitted the 3 of them against each other and stockfish. Rybka, Human tied stockfish just behind and dynamic last.
Parent - - By Highendman (****) Date 2009-10-26 17:27
Last I've checked they all played chess quite well. ;)

Rybka 3 is the default engine, supposedly with the best all around coefficients / weights for its analysis.
960 is the same as R3, with the added knowledge to allow it to play chess960 (aka Fischer Random).
Rybka Dynamic tends to evaluate material / space different from R3 - seems it emphasize space / initiative more, and as a by product seems to like giving up the exchange.
Rybka Human is possibly the opposite - different weights to the material.

It's claimed that R3 has the better ELO performance vs. her siblings.
However GM L.Kaufman recently posted he believes R3 Human, in long t/c, has a more accurate evaluation function, though he couldn't explain why it doesn't perform accordingly in matches vs. R3.
Parent - - By SpiderG (***) Date 2009-10-26 23:35
Where did he post it? Here? It seems I'm missing everything... :(
Parent - - By Highendman (****) Date 2009-10-26 23:59
Yes, here. He said he's switched to using R-Human for his long t/c analysis as he now believes her evaluation is more correct.
Right after that post you could see a lot of players on playchess when playing long time control games have started using R-Human, which was quite funny.
I believe the fact most switched back to R3 'the original series' demonstrated that while the eval may be more accurate, the results did not reflect that.
Maybe next time GM Kaufman posts around here someone could jump him and ask whether he still holds that view, why, and if he has an explanation as to the results discrepancy.
Pretty interesting topic. He has a unique insight into how the engine works probably, having helped tune these values / rules or however else they are defined in Rybka.
Parent - - By SpiderG (***) Date 2009-10-27 00:05
Thanks. I have another question for you. Do you know when the next cluster game is?
Parent - By Highendman (****) Date 2009-10-27 00:22
I believe there's one scheduled vs. Buddha towards the end of this week.
As to when I get a chance to drop won positions into draws or worse simply pathetically lose - as soon as Lukas has time he'll find me there :)
I'm a very bad loser.
Parent - By Banned for Life (Gold) Date 2009-10-27 00:06
Vas has described the feature weights as "nearly linear" by which I believe he meant that the evaluation was arrived at by a weighted sum  of (possibly nonlinear) feature values.

I also recall hearing that the human variant has more features and therefor a longer eval time. This may be responsible for some of the head to head advantage enjoyed by the default variant.
Parent - By NATIONAL12 (Gold) Date 2009-10-27 00:44
it's quite a while ago since Larry gave his opinion on R3 human.i can confirm that in a match in which Bobby C and i were partners at 48hr/move deep into game R3 default and R3 human were giving diff evals,but only by 0.13.
Parent - By irulats (****) Date 2009-10-27 00:16

>Where did he post it? Here? It seems I'm missing everything... 


http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?pid=183466;hl=larry   14th post I think in this thread

lkaufman  Date 2009-09-10 21:43 Your second question is easy, "yes". I believe that Rybka 3 human is more useful than Rybka 3 for analyzing openings for human use or for analyzing human games. Whether it is better for computer-assisted play I don't know. The main point is that the material values in the default R3 are just obviously "wrong" from a human perspective, minor pieces are seriously undervalued in general. For some reason this tests well in pure engine play.
     I am no expert on mp in Aquarium. I assume that if I ask Aquarium to analyze 10 positions for five minutes each with my octal, all eight cores are devoted to each problem sequentially. This seems to be right as the time taken is indeed 50 minutes and task manager shows all cores are working. I only ask for analysis in best mode, I don't ask for "best-3" or the like. Is there anything wrong with what I am saying here?

     Best regards,
     Larry
Parent - By Quicksort (**) Date 2009-10-27 00:52
According to Larry (Kaufmann) whose algorithms play a prominent role in Rybka' s position evaluation, default Rybka, unlike Rybka Human, undervalues minor pieces.
Derived heuristics seeming to suit computer chess well.

Quicksort
Up Topic Rybka Support & Discussion / Rybka Discussion / What does each of the four different Rybka engines do?

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill