Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka 3 -- a hundred elo!
1 2 3 4 5 Previous Next  
Parent - - By Fulcrum2000 (****) Date 2008-07-09 18:06
Makes for a catchy title :  Rybka 3 -- a one hundred and one elo!  :-)
Parent - By Kapaun (****) Date 2008-07-09 23:08
111 elomatians...
Parent - - By Mark (****) Date 2008-07-09 18:18
One elo isn't that insignificant considering that the development cost to increase the strength of Rybka is probably several thousand dollars/elo!
Parent - - By lkaufman (*****) Date 2008-07-09 19:54
Well, it's not that high; I can't speak for Vas, but I don't choose to do this work because of the high monetary rewards. Also, I said "at least" one Elo, I didn't state a maximum!
Parent - - By Mark (****) Date 2008-07-09 20:47
Well, somehow you have to be able to support the Potomac lifestyle! :)
Parent - By lkaufman (*****) Date 2008-07-09 21:07
Fortunately, my house is already paid off.
Parent - - By Vempele (Silver) Date 2008-07-09 19:57
What's the weather like on your planet?

Wait, dollars?? Now that makes sense...
Parent - By Mark (****) Date 2008-07-09 20:50
I can remember a time when no one made fun of dollars. Those were the good old days... :)
Parent - - By Nelson Hernandez (Gold) Date 2008-07-09 18:26
I have to admit I envy you, Larry.  Here you are over sixty and apparently far from your mind are thoughts of retirement, where you could sit on a rocking chair on your front porch yelling at passing children to get off your lawn in vividly colorful language, while menacingly shaking a cane in their direction and threatening to call the police.  You do not seem inclined to pursue a normal, healthy, curmudgeonly trajectory.  Instead, you're intellectually engaged in these fantastically complex and obscure tests and computer operations to systematically squeeze ELO points, or fractions of ELO points, out of the world's leading chess program!  What a fabulously eccentric life!  With the recent passing of E. Gary Gygax (creator of Dungeons & Dragons) you have become the uncrowned King of Geeks!  Rock on, Larry!
Parent - By lkaufman (*****) Date 2008-07-09 19:56
I'm hoping for a cure to aging, so I can do this for another century or two.
Parent - By Kapaun (****) Date 2008-07-09 22:34
Gygax is dead? Oh boy! During the last two or three years I certainly happened to enter an age where it doesn't go without saying anymore that the companions of your earlier years are still alive.

Good job, Gary... 
Parent - - By 8lrr8 (***) Date 2008-07-10 00:09
rybka is what it is today (basically a chess monster) because:

1. they have 3 chess experts on the team (jeroen most definitely can not be counted out).  how many other engines can say that?
2. these experts love what they do.
3. these experts all have something in common: they're always looking for the next level.

in a nutshell, u have 3 experts who are deathly afraid to (either that or hating to) losing.  and love what they do.  add all of it together, and it's really not surprising rybka's #1 (and will likely remain so indefinitely).
Parent - - By Uly (Gold) Date 2008-07-10 02:17

> jeroen most definitely can not be counted out


The testers count him out by using generic books.
Parent - - By 8lrr8 (***) Date 2008-07-10 02:21
good pt.  however isnt there some rating list that uses an engine's default book?
Parent - - By turbojuice1122 (Gold) Date 2008-07-10 02:22
Yes, the SSDF rating list does this.
Parent - - By 8lrr8 (***) Date 2008-07-10 02:26
thx.  but what's going on w/ that list?  it hasnt been updated in 8 months, and they use archaic hardware.
Parent - - By turbojuice1122 (Gold) Date 2008-07-10 02:55
Well, their "archaic hardware" goes at tournament time controls.  Also, they have upgraded to Q6600 machines and have been doing a lot of tests lately, but they are waiting for Rybka 3 to come out before publishing the next update.  It takes a long time to test at tournament time controls.
Parent - By Uly (Gold) Date 2008-07-10 03:05
Well, I think we need an alternative to SSDF.
Parent - By 8lrr8 (***) Date 2008-07-10 03:10
"Well, their "archaic hardware" goes at tournament time controls"

which is equivalent to something like 40/5 on today's hardware, no?

"Also, they have upgraded to Q6600 machines and have been doing a lot of tests lately, but they are waiting for Rybka 3 to come out before publishing the next update."

good to hear.  unfortunately it's not listed on their website.  looking fwd to their update.
Parent - By Fulcrum2000 (****) Date 2008-07-10 07:46
Also don't forget Jeroen has a fide ELO of about 2200 or so. He knows something about chess...
Parent - By 8lrr8 (***) Date 2008-07-11 06:00
well it's not rybka loses its #1 ranking when no books or generic books are used, so...
Parent - - By Banned for Life (Gold) Date 2008-07-10 10:43
Watch out. Iweta will probably be coming to kick your ass! :-)
Parent - - By 8lrr8 (***) Date 2008-07-10 11:31
she's definitely part of the team, yes.  however i get the feeling her contribution probably has the smallest effect on rybka's (current) strength and its strength increasing rate compared to the effect the other 3 people have on rybka.
Parent - - By Banned for Life (Gold) Date 2008-07-11 05:40
No, I am quite sure she has a larger influence than Jeroen on the engine development. The book is, and should be, independent from the engine.
Parent - - By 8lrr8 (***) Date 2008-07-11 05:51
"No, I am quite sure she has a larger influence than Jeroen on the engine development."

i was referring to her contribution relative to those who work on engine development, namely vas and larry.  unless u believe she contributes more than at least one of those two?

jeroen is the book author, obviously.  that is never in question.
Parent - - By Banned for Life (Gold) Date 2008-07-11 06:03
Your statement 1 implied that you were giving most credit to Vas, Larry, and Jeroen for Rybka's development. The order is clearly Vas, Larry, and Iweta (actually, its not really clear to us mortals at this point if the search improvements are worth more than the eval improvements).

Alan
Parent - - By 8lrr8 (***) Date 2008-07-11 06:07
"The order is clearly Vas, Larry, and Iweta..."

why is it "clearly" the case?  can u elaborate?  i know iweta tests rybka for vas and gives him feedback, but does that mean vas does zero testing himself?

as for jeroen's contribution, i'm actually comparing the rybka3 book's contribution to rybka, if that makes any difference.
Parent - - By Banned for Life (Gold) Date 2008-07-11 06:10
Jeroen's book will contribute to other engine's too, maybe almost as much as to Rybka (or maybe even more).

A developer is handicapped when he tests his own code and most can't do this effectively. So Iweta might very well be a better tester than Vas.
Parent - - By 8lrr8 (***) Date 2008-07-11 06:16
"...(or maybe even more)."

i dont see this as likely because w/ a different engine other than rybka being used for the calculations once it leaves book, it may not find the correct continuation.  so what would've been a win for rybka may turn into just a draw for say, fritz.

"A developer is handicapped when he tests his own code and most can't do this effectively. So Iweta might very well be a better tester than Vas."

1. the question now is if u think this applies to vas?
2. even if iweta were a better tester (just assuming), is it the case she spends more time testing it than vas does himself?
Parent - - By Banned for Life (Gold) Date 2008-07-11 06:26
Sure, some of the lines will only be useful for Rybka, but other engines come with such crappy books, that going to a top book could make an enormous difference.

I consider Vas to be a premier developer, but not a premier tester. Rybka has certainly had it's share of bugs over the years, including the worst bug of all - instability. I have no clue how much time Vas or Iweta spends working on Rybka. Who knows, maybe only Larry is really working! :-)

Alan
Parent - - By Harvey Williamson (*****) Date 2008-07-11 16:30
yep all engines that don't have Rybka in the title automatically have crappy books.
Parent - - By Banned for Life (Gold) Date 2008-07-12 01:36
In my opinion, any book that lists 1.b3?! as a first move is in the the crappy category. :-)

Alan
Parent - - By Venator (Silver) Date 2008-07-12 07:58
In my opinion, any book that lists 1.b3?! as a first move is in the the crappy category.

Hm, then RybkaII.ctg belongs into that category. I am glad I made 1.b3 a red move in Rybka3.ctg :-).
Parent - - By Banned for Life (Gold) Date 2008-07-12 17:17
I am hoping that the R3 book will be more competitive against 1.b3 than the RII book. I will know very quickly after the release. :-)
Parent - - By Venator (Silver) Date 2008-07-12 18:44
Ha ha, let me know :-). FYI I imported all your losses in the 1.b3 games in the R3 book! Well, up to somewhere in May, if I recall correctly.
Parent - By Banned for Life (Gold) Date 2008-07-12 21:09
Recent games would be mostly Zappa on white vs Rybka on black so for your sake I hope the losses were due to bad opening lines and not Rybka outplaying Zappa in the late middle game! :-)

There are actually two things I will be looking for immediately after release:

1) My current system is based on a set of identified positions that Rybka plays poorly. This list will certainly change due to differences between R2.3.2a and R3. Its an open question how much it will change though.
2) Assuming some, most, or all of the weak Rybka positions remain, it will be interesting to see if your book still allows me to reach these positions.

Whatever the outcome, I'll post some early games along with R2.3.2a games from the same opening line to either tweak the Rybka team or congratulate them...

Regards,
Alan
Parent - By Kapaun (****) Date 2008-07-12 21:09
But the fish scores quite well with b3. As well as with Nc3...
Parent - By Roland Rösler (****) Date 2008-07-10 00:33
where you could sit on a rocking chair

He, Larry is only 60, not 80 or 90! If I remember right, Harry Schnapp is older than 70 and his books are very fine!
Parent - - By George Tsavdaris (****) Date 2008-07-09 18:37 Edited 2008-07-09 18:39
OK with Rybka 3. It will obviously be 70-150 ELO points stronger than its predecessor and you needed around a year for that.
It wasn't easy of course.
Let's say it was 100 ELO / year.

But do you think that following the same procedure of working on Rybka, the same ratio of ELO per year will be able to remain as it is(100 ELO per year)- hardware's advances excluded-, or this procedure has reached its limits and for Rybka 4 you should make other things for improving Rybka so to have again +100 ELO next year?

And specifically to the area of your expertise: Do you think Rybka as it now, will be able to continue accepting the "lessons" you give her and improve by another 100 ELO for the next year, or there are not so much things to teach her left with the current situation. And does the whole structure of her needs a drastic redesign?

And to go it a little further:
With the current way of working on Rybka how much is the potential improve? 100, 200, 300 ELO? Or more?
Or to have more it needs a whole new design and way of computing her moves?

The last is valid also to other programs. While i think there are plenty of space left for further improvements working this way, the moment that the improvements will be tooo difficult to come is not far away. Hardware excluded.
And that programs and programmers will have to find new ways for the thinking process of engines.....
Parent - By estragone Date 2008-07-09 19:49
Interesting questions!

I my opinion, developing software is deeply affected by the dynamic parameters. Excluding hardware e.g., is not possible, because the parameters of your search efficiency depends on that. You always have to adapt your program by searching the current peaks of the multidimensional circumstances (cpu, ram, bus-speed, endgame databases, opening books or total changes like quantum computing).
The solutions and assumptions of yesterday doesn't reflect the current situation and possibilities. At the moment Rybka could be near in finding the current maximum but that is not a guarantee for being the overall best in the future. Maybe other programs can benefit of the coming changes more than Rybka. So the prediction of ELO rises based on static assumptions is very difficult.
Parent - - By lkaufman (*****) Date 2008-07-09 20:07
If we made no changes in our procedures, we would indeed face diminishing returns, maybe only fifty Elo in the next year, thirty in the year after, etc. But we do and will make changes, partly because we learn more, partly because there is more incentive to try radical ideas when incremental gains dwindle. Also, as pointed out here by another respondent, hardware affects what we do. Just to give one example, at some point the "Monte Carlo" approach will be the way to go, when it becomes difficult to get big speedups out of more processors, because with Monte Carlo the benefit of doubling the number of processors remains nearly constant. I think Rybka 4 will be a comparable jump over Rybka 3 as Rybka 3 is over 2.3.2a. I believe that in a decade or less the best program will be rated over 4000 on the rating lists, assuming no major change in rating methodology.
Parent - - By Nelson Hernandez (Gold) Date 2008-07-09 20:40
4000--!  This reminds me of Kennedy calling for man to land on the moon in 1961.
Parent - By Pia (****) Date 2008-07-09 20:54
I think computer chess is "Moon-landed" long time ago. Next frontier - Mars, where we can play with another civilizations in chess :)
Parent - - By FWCC (***) Date 2008-07-09 20:50
That is amazing over 4000 ELO.Now we are approaching theoreticals,perfect chess for instance.Is perfect chess relative,is it achievable?When I ask if it is relative I mean by what standards can we compare "Perfect Chess " too?Do we have anything to compare it to?And another thing,with "Perfect"chess we are also talking about chess being solved eventually.So then Larry when do you see chess being solved as this will be ultimately the result of "Perfect Chess".The result in effect will be zero.Two perfectly playing entities playing together will result in a draw a day I do not really look forward to.But until then there is plenty of chess to be played.
Parent - - By lkaufman (*****) Date 2008-07-09 21:06
I don't speak of chess being solved, at least not in the sense of a 32 man tablebase. That seems impossible, like uncooking an egg (but maybe someday that will be possible). Chess may be solved in a few decades in the sense of a proof that the correct result is a draw, or at least a "near-proof". But a 4000 rating on CCRL or CEGT won't mean anything like perfect chess, though it will probably appear to be so to human players. Maybe all the games between such programs might end up in draws with huge, conservative books, but remember these rating lists use varied, short books so someone can still win games. That's why I speak of 4000 only assuming present rating methods; probably 4000 is impossible with both sides using identical, really good and deep books.
Parent - By Roland Rösler (****) Date 2008-07-10 01:48
But a 4000 rating on CCRL or CEGT won't mean anything like perfect chess, ...

Let´s assume, Rybka 3 has 3200 Elo. Then she will make 1 point out of 100 games against the leader in 2018. Not bad I think.
I´m sure, it will be 2 draws. :-)
Parent - - By duncan (**) Date 2008-07-09 23:38
Just to give one example, at some point the "Monte Carlo" approach will be the way to go, when it becomes difficult to get big speedups out of more processors,

....

how many procesors would that be  16, 32 ?
Parent - By lkaufman (*****) Date 2008-07-09 23:42
Depends on what "difficult" means. I'm not an expert on mp scaling, but I think the speedup for each doubling will decline very gradually. Personally I think that the time when Monte Carlo will be practical for chess programs is not so far away, but I don't have a solid basis for saying this.
Parent - By Vempele (Silver) Date 2008-07-09 19:43
They're trying to get it to link.

(Damn! Still can't come up with a good "some assembly required" kind of joke :-()
Parent - By Nick (*****) Date 2008-07-09 20:59

> we will reach the goal of a 100 Elo gain


Simply amazing and my congratulations to you.
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka 3 -- a hundred elo!
1 2 3 4 5 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill