Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Match agreement with FIDE 2700+ opponent.
1 2 3 Previous Next  
Poll Score of Milov vs. Rybka handicap match? (Closed)
Milov wins by 7-1 or more 2 2%
Milovwins by 6.5-1.5 1 1%
Milov wins by 6-2 0 0%
Milov wins by 5.5-2.5 3 3%
Milov wins by 5-3 7 7%
Milov wins by 4.5-3.5 10 10%
Tie match 4-4 11 11%
Rybka wins by 4.5-3.5 18 18%
Rybka wins by 5-3 16 16%
Rybka wins by 5.5-2.5 10 10%
Rybka wins by 6-2 12 12%
Rybka wins by 6.5-1.5 4 4%
Rybka wins by 7-1 or more 8 8%
- - By lkaufman (*****) Date 2008-08-08 20:16 Edited 2008-08-15 15:24
     Vadim Milov, World #28 at 2705 FIDE (2750 USCF), has agreed to play a handicap match with Rybka in mid-September. The plan is for an eight game match at FIDE time controls (90'+30"), to consist of two games of normal chess in which Milov gets the White pieces, and six handicap games (two games at pawn and move, meaning Rybka removes f7 pawn, and four games at odds of the Exchange (Rybka removes White rook from a1, Milov removes Black knight from b8). This handicap of the Exchange has not yet been tried in any of our matches, and was last seen in recorded games 150 years ago, when it was apparently considered to be more or less on a par with Pawn and move handicap. We have funding for the match, but if anyone would like to see two more games played under any particular conditions  and is willing to contribute to the prize fund, this might be possible, please message me.
Parent - - By Sesse (****) [no] Date 2008-08-08 20:21
Wow. 2700+ is really something. Will you be playing with Rybka 3, or something mildly improved?

/* Steinar */
Parent - - By lkaufman (*****) Date 2008-08-08 20:26
That isn't decided yet, but probably "something miildly improved" if only to make Rybka a little less predictable in the opening. I don't know if we will have any real improvement that soon.
Parent - - By oudheusa (*****) [dk] Date 2008-08-08 20:33
In these matches why is the GM not allowed at least to use his own opening book?

And imo a match whereby the GM is allowed to use a 'light' engine to check variations in a normal match set-up would be much more interesting than other handicaps.
Parent - By George Tsavdaris (****) Date 2008-08-08 20:49

>And imo a match whereby the GM is allowed to use a 'light' engine to check variations in a normal match set-up would be much more interesting than other handicaps.

In my opinion this is not so interesting. I prefer to compare human ability of playing Chess and computer's ability to play Chess.
And this can't happen with your above rule.

Of course if you are not interested in comparing human versus computer Chess playing ability, then a match such that you propose may indeed be more logical.

>In these matches why is the GM not allowed at least to use his own opening book?

He is allowed to use it.
He just have to use the normal method of doing it, as when he plays normal games against humans. I can't find a reason to change that.

What i want to compare is: Top human at Chess versus top computers at Chess with:
Top humans to play with the same way they play in human tournaments(that is without looking for example in any database with an opening book).
Top computers to play the same way they play in computer tournaments.

If you don't want to compare what i want, then indeed you should play with other rules.....
Parent - - By lkaufman (*****) Date 2008-08-08 21:06
Of course, for handicap games it makes no sense for him to have his own opening book. For the normal games, I probably would have agreed to that condition if he had asked for it, but he did not. I considered the option of having two games with "light" engine use allowed, but everything would then depend on the choice of engine (and hardware) and also on his centaur skill, not his chess skill. For a "centaur" handicap we can get a relatively weak player with good centaur skills for nominal or no prizes if we want to do this.
Parent - By George Tsavdaris (****) Date 2008-08-08 21:49 Edited 2008-08-08 21:52
Personally this time i'm interested more in human-computer games and not centaur-computer games.
Also i prefer normal Chess than handicap with piece handicaps.
I will create a poll to find out what other people prefer.

BTW you will be pleased to know that i wasn't able to win against Rybka 3 with me having a c-Bishop and a b-Knight less than Rybka(with me having a severe handicap that is) after about 70 games of trying. All loses for me. If you remember(here is the link) i have managed to do it against Rybka 1.2i but that was probably due to the bug you are saying in the end of the page of Rybka 1.2i.

Now i will focus on trying to find a winning sequence for white against Rybka 3 as black, that means playing with take backs.
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/R2QKBNR w KQkq - 0 1
Parent - By BB (****) [au] Date 2008-08-09 03:42

>For a "centaur" handicap we can get a relatively weak player with good centaur skills for nominal or no prizes if we want to do this.

I personally think that a better idea with centaurs/R3 would involve more entities (to use the nebulous Freestyle word) in action. I recently dredged up VR's old musing about a Scheveningen centaur match between GMs and Freestyle pros(?), but modified it to be centaur GMs versus stand-alone Rybkas (likely overclocked octals). I would envision here that the GMs would only have (say) quadcores, and with some edge from the voluminous books of the Rybka "operators", it is not clear whether the centauring edge would be sufficient. [The use of GMs rather than Freestylers for centaurs would be a publicity angle].
Parent - - By FWCC (***) [us] Date 2008-08-08 22:57 Edited 2008-08-08 23:01
Well we will see my exchange match which has not been played in a while.This set up is worth less than two pawns?Larry also based on ELO what are the predictions?In the future is there anyway we could use Monte Carlo to play the big dogs (+2700 and above) players,that might be interesting but so far it seems MC can not be implemented to success with a top player(though it has not been tried yet)I imagine you would have to learn a set up for this.
Parent - - By lkaufman (*****) Date 2008-08-09 04:47
The value of the rook/knight Exchange in the opening is roughly 1.5 pawns. Pawn and move handicap is also around 1.5 pawns, because the move is about 1/3 of a pawn and the missing pawn is a particularly bad one to be missing, due to the king's weakness. As for Elo value of the handicap, based on MC tests at 12 ply I got 311 Elo in 840 games, but this is like bullet chess speed between engines. At tournament level it is probably at least 400 Elo for engines, and we have repeatedly seen that handicaps are worth more in the hands of a human who knows enough to avoid unclear complications when well ahead. So in the given environment I would say that it should fully compensate for the difference between 3200 and 2700, and given that there is some suspicion that these 3200 ratings are a bit inflated in human terms, I think Rybka is the underdog in these Exchange odds games with Milov. 
Parent - - By tasdourian (**) [us] Date 2008-08-10 04:54
You have mentioned repeatedly, Larry, that the f7 pawn is the worst for black in the pawn and move handicap because of the king's exposed position. I certainly don't doubt this, but I do find it confusing.  Isn't part of the point of the King's Gambit for white that the f-pawn is removed so that white has a nice open file for his castled rook and thus pressure on black's kingside?  Wouldn't black have a similarly open f-file here? That is, doesn't black get some compensation as well for not having the f7 pawn?

Put another way, why isn't, say, the c7 or g7 pawn a bigger handicap?  Opening a single line for the queen or bishop doesn't seem to be much compensation for the lost central pawn or pawn protection for the castled position.  Again, I don't doubt f7 is the worst, I'm just curious as to why.
Parent - - By lkaufman (*****) Date 2008-08-10 05:17
Well, the g7 pawn is also pretty big, as it makes castling dubious, but c7 or any other pawn is much less than those two. Any missing pawn gives some potential compensation as long as you have a rook that might move to that half-open file. For White the f2 pawn is not so clearly the biggest handicap, as he will castle and the rook will be well-placed. But for Black it's a big problem, because after 1e4 the possible check on h5 prevents normal defense. Black can manage to castle, but not without conceding a large space and mobility advantage to White. Every normal defense has some serious drawback tactically. Black can castle by playing an early ...e6 and ...d6 and ...Nf6, ...Be7, and 0-0 but that's quite passive; he's hardly going to be in any position to use his half-open file aggressively.
Parent - By tasdourian (**) [us] Date 2008-08-10 05:24
Again, thanks for another helpful reply.
Parent - - By Graham Banks (*****) [nz] Date 2008-08-10 06:44
In club handicap competitions, we played forever with pawn and two moves as a handicap with the proviso that White wasn't allowed to play Qh5+ on the second move.
Because the queen check hasn't been played in any of your matches, I assume it's not really that fantastic then?

Parent - By lkaufman (*****) Date 2008-08-10 15:18
Why should it be a good move then? after 2...g6 3Qe5?! Nf6 White has just wasted time. But the threat of Qh5+ at any moment prevents lots of Black defenses, and if Black stops Qh5 by playing ...g6 first, that pawn becomes a target for h4-h5. I suppose you could check on move 2 and then retreat, to follow up with h4-h5, but that wastes too much time to be correct.
Parent - - By Lukas Cimiotti (Bronze) [de] Date 2008-08-10 16:19
My impression is that Rybka 3 Dynamic is best against weaker opponents.
In the meantime I made a match Rybka 3 - Rybka 3 Human using Jeroen's latest testset (my AC has been replaced - repair impossible - all of my computers can run now at full strength again). I made 60 games 30 min + 10 sec per move taking ~4 days. The result doesn't match predictions - the default version is stronger in longer games like it is in short games. The result was +14 =37 -9 54.17% +29 Elo. Maybe we should run a longer test. Here are the games:
Parent - By lkaufman (*****) Date 2008-08-10 17:24
Yes, 60 games is not enough to tell. Also, my data in blitz shows that default is strongest in heads-up play, but all three are about the same against other engines, so it is possible that Human or Dynamic could prove to be the strongest against other engines at the longer time controls.
Parent - By M ANSARI (*****) [kw] Date 2008-08-12 08:13
I have been testing at 120 minute a side time control and I get results that seem to show Rybka Human is much stronger.  Again very few games have been played so this could be just statistical noise ... I will let it run for at least another 40 games to see  how that goes.  The problem with these long time control games is that it takes forever to generate a decent number of games.
Parent - - By Quapsel (****) [de] Date 2008-08-25 06:49

> I don't know if we will have any real improvement that soon.

Do you think that Rybka has a significant danger to lose some half points only because she doesnt know some elementar Endgame-Facts ('wrong bishop' an so on), or do you believe that this is adequate absorbt by using tablebases?

Do you think that in critical Situations it is a good idea for the opponent to try to offer Rybka such a draw-position, she might evaluate as very good for her and takes it quick and very unhasitating? A human opponent might directly try to follow such a plan.

OK, ist is very hard to reach an endgame against Rybka, which isn't completely lost for the human. A 2700-human maybe will have success.
Will he then win some half points which Rybka in fact shouldn't give, which not would be given by most of other top engines?

Parent - - By lkaufman (*****) Date 2008-08-25 15:43
Of course tablebases cover only some but not all of the endgame problems. At the very least, I expect we will have one or two of the bugs fixed that affect wrong-bishop and bishops of opposite color endgames. But the chance of these bugs actually costing us half a point in an eight game match is pretty tiny. I think a human should not aim for such endings unless he finds himself in trouble with no other way out, not only because it is nearly impossible to aim for them early on, but also because he must assume that we will fix such bugs. However there will remain endgames that some other engines will play better than Rybka, simply because they have had many years to code these many special cases.
Parent - Date 2008-08-25 15:47
Parent - By plicocf (***) [br] Date 2008-08-08 20:42
IM Larry, congratulations ! Two normal games !
I will to see! Thanks.

Paulo Soares
Parent - By hal9000 (**) [no] Date 2008-08-08 20:43
Sounds great. It's very nice to have a player of Milov's stature in such a match. I particularly look forward to the normal games.
Parent - - By (**) Date 2008-08-08 21:07
WOW !!! I know Vadim Milov >>> He is an extreme Tourney player and knows how to prepare against opponents ! His style is in my opinion very risky and not good against Rybka. He is a very strong player, but because of his style, I believe he will lose all games against Rybka.

Well....we will see
Parent - - By lkaufman (*****) Date 2008-08-08 22:12
Regardless of style, a 2700+ player knows how to go about winning a won game, and I'm pretty certain that he will be starting with six won games in this match. He may play recklessly trying to get a won game, but he knows that this is not necessary after achieving a winning position.
Parent - - By Peter Hegger (*) [ca] Date 2008-08-20 05:26

> Regardless of style, a 2700+ player knows how to go about winning a won game, and I'm pretty certain that he will be starting with six won games in this match.

Does this mean that the non-handicap games are being played last?

Parent - - By lkaufman (*****) Date 2008-08-20 13:42
No, "starting" refers to the starting position of each game. I think Milov would prefer to play the non-handicap games first; if so that's fine with me.
Parent - - By Gaмßito (****) [cr] Date 2008-08-22 10:58
Concerning the handicap games, it seems to be a truly difficult challenge for Rybka, and I expect that Milov can be quite well prepared. I believe that Rybka will suffer here and I think that for each draw that Rybka can take, you should give it a point. This is hard to achieve against an opponent of their level and very well prepared. A draw would be really a very good result for Rybka in these handicap games.
Will have Rybka some small or any considerable change before this match or it will play exactly the same commercial version?
By the way, what's your prediction for these two non-handicap games Larry? Do you believe Rybka will squash Milov easily in these non-handicap games?

Parent - By lkaufman (*****) Date 2008-08-22 14:22
     I agree with your first paragraph, but there is no need for rewarding draws for Rybka, because the non-handicap games are the offsetting factor. I expect that Rybka will either win both of them or at least win one and draw the other. The handicaps are a bit much for a 2700+ player at a tournament time limit, I think.
     Rybka will be slightly changed, partly to minimize the likelihood of prepared lines in the handicap games based on what Rybka 3 would play. Probably the playing strength will be just a few Elo points stronger, partly from bug-fixes and partly from small eval changes.
Parent - - By JohnL (***) Date 2008-08-08 22:14
That's great Larry, this will be very interesting to follow, especially the exchange handicap!
But I am not convinced you will win, seems like you are stretching it to a 50-50 fight. I think Milov is underdog in the 2 normal games but slight favourite in the others.

BTW, nice article in chessbase today :-)
Parent - By lkaufman (*****) Date 2008-08-08 22:20
I agree completely with your assessment of the match. It is indeed intended to be a tossup; the two non-handicap games are there not only to please the spectators but to make the match result hard to predict. I'll be interested to hear if any of you master level players out there play some games with Rybka 3 at Exchange handicap; if so please let us know the results. If any of you score more than the occasional draw it would indicate that we are in big trouble with Milov at that handicap.
Parent - - By Arkansaw (***) Date 2008-08-08 22:26
For the two normal games, is Rybka using a book at all?
Parent - By lkaufman (*****) Date 2008-08-08 22:29
Parent - - By ubaldus (*) [us] Date 2008-08-08 23:54
Does your Handicap Book include lines with this Exchange handicap? I was trying to run Rybka3 sp w32 - Alaric707 match at this handicap in Aquarium, but Rybka is out of book on move 1.

Btw, in pawn and move blitz match (2+2'' time control), Rybka and Alaric seem to be evenly matched, +11=10-9 for 16:14 Rybka win. Since Alaric is 2700 CEGT, this is consistent with your estimates of 360 Elo for pawn and move handicap and R3 sp w32 3070 provisional CEGT rating. Interestingely, if you force Alaric to play on its own from move 1, it performes much worse. Did not expect this effect, since the book always seem to go with Nh6-Nf7-e6 line anyway.
Parent - - By lkaufman (*****) Date 2008-08-09 04:30
Unfortunately I did not cover the handicap of the Exchange in my handicap book, as we never  had any match that way. Now of course I will have to make a book for it. Perhaps it will be available as an update in the future. Regarding the match at pawn and move with Alaric, was there sufficient variety in the openings and games? If you were using MP this would be less vital, but with SP you are likely to get the same game repeated for a long time if the opening is the same. A better way to test would be to play maybe just five games at this time control, but then keep changing the time control for another five games each time, like 3'+2", 4'+3", 3'+3", etc. Variety is vital for the test to be valid. If Alaric plays on its own from move 1, the risk of repeat games is really high; maybe it just lost the same game (at least for many moves) again and again?
Parent - - By ubaldus (*) [us] Date 2008-08-09 05:25
When both Alaric and Rybka had an access to a handicap book, there seemed to be enough variety, there were some typical setups of course, but no repeated games. I played 10 games at exchange handicap as well, Alaric had no clue, losing 2:8. Perhaps in this cases it indeed falls into some setups it does not like, though there were no repeated games after move 12 or so. I have IE running in the background eating 5 to 10% of CPU, and this does introduce a randomizing effect.

I also played Rybka 1cpu - Glaurung 2.1 2cpu match at pawn and move, 2+2''. Glaurung is mp, and there was definitely enough randomization. Since Rybka w32 probably outranks Glaurung by "only" 200 points, it had hard time in the match, scoring 30% (21:49, +11=20-38). Probably, pawn and move with 1.e4 Nh6 setup is worth >350 points. I will try to let Glaurung to play on its own, I am sure it will not come close to 70% score.

Right now I have Rybka - Glaurung match running at Exchange odds. Even with no book, there seem to be a nice variety of openings - though Rybka seem to favor Reti setup, Glaurung tried a bunch of different replies. After 26 games it's 10:16 (+6=8-12), and Rybka seems to suffer less under this handicap compared to pawn and move suffocation.
Parent - - By lkaufman (*****) Date 2008-08-09 05:39
Thanks for the data, please keep posting results. I wonder though why you have Rybka running on 1 cpu while you have Glaurung running on 2; it's hard enough to give pawn and move or Exchange handicap without also giving two to one processor advantage as well! Also, you can introduce more variety by starting the pawn and move matches with each of the four serious Black replies to 1e4 (1...Nh6, 1...e6, 1...d6. 1...Nc6) and by starting the Exchange matches with the four normal White opening moves (1e4, 1d4, 1c4, 1Nf3). The more variety in the openings, the more valid your results will be.
Parent - By ubaldus (*) [us] Date 2008-08-09 05:53

>>  I wonder though why you have Rybka running on 1 cpu

Because I am cheap and did not pay for the mp version. All I have is a 2cpu machine anyway, and by the time I get a quad or an octal, I will get Rybka 4 to run on it.

And I wanted to see how Rybka manages to score an occasional win pawn and move down against 2850 opponent. It is impressive to see Rybka defending these terrible, terrible positions.

I will make a small Exchange handicap book and try to run matches with this book.
Btw, here is a typical Rybka-Alaric game. Rybka eval keeps climbing up and by move 40, Rybka thinks it's about even. All this time, Alaric has no clue. And Alaric can probably crush Milov in a match.

[Event "[#36] Rybka 3 1-cpu w32 - Alar"]
[Site "Local computer"]
[Date "2008.08.09"]
[Round "12"]
[White "Rybka 3 1-cpu w32"]
[Black "Alaric707"]
[Result "1-0"]
[FEN "r1bqkbnr/pppppppp/8/8/8/8/PPPPPPPP/1NBQKBNR w Kkq - 0 1"]
[SetUp "1"]
[TimeControl "2 min/game + 2 sec/move"]

{[%t Long] White checkmates.}{[%t bLon] Processor: Genuine Intel(R) CPU       
T2080  @ 1.73GHz^13 ^10 White: Rybka 3 1-cpu w32, hash size: 32M, opening book:
HandicapBook^13 ^10 Black: Alaric707, hash size: 32M, opening book:
HandicapBook^13 ^10 } 1. Nf3 {[%clk 0:01:52][%clko 0:02:00][%emt 0:00:07]}
{[%eval -89,11]}{[%t bLon] White out of book}{Black out of book} 1... Nf6
{[%clk 0:01:52][%clko 0:01:52][%emt 0:00:07]}{[%eval -185,13]} 2. g3 {[%clk
0:01:49][%clko 0:01:52][%emt 0:00:05]}{[%eval -94,11]} 2... c6 {[%clk
0:01:46][%clko 0:01:49][%emt 0:00:09]}{[%eval -181,12]} 3. Bg2 {[%clk
0:01:48][%clko 0:01:46][%emt 0:00:03]}{[%eval -100,11]} 3... g6 {[%clk
0:01:41][%clko 0:01:48][%emt 0:00:07]}{[%eval -178,12]} 4. O-O {[%clk
0:01:47][%clko 0:01:41][%emt 0:00:03]}{[%eval -102,11]} 4... Bh6 {[%clk
0:01:36][%clko 0:01:47][%emt 0:00:07]}{[%eval -196,14]} 5. c4 {[%clk
0:01:42][%clko 0:01:36][%emt 0:00:06]}{[%eval -89,12]} 5... d6 {[%clk
0:01:32][%clko 0:01:42][%emt 0:00:06]}{[%eval -198,13]} 6. h3 {[%clk
0:01:40][%clko 0:01:32][%emt 0:00:04]}{[%eval -91,12]} 6... Be6 {[%clk
0:01:27][%clko 0:01:40][%emt 0:00:06]}{[%eval -213,12]} 7. d3 {[%clk
0:01:38][%clko 0:01:27][%emt 0:00:04]}{[%eval -90,11]} 7... Bxc1 {[%clk
0:01:24][%clko 0:01:38][%emt 0:00:05]}{[%eval -204,14]} 8. Qxc1 {[%clk
0:01:40][%clko 0:01:24]}{[%eval -90,10]} 8... Qb6 {[%clk 0:01:19][%clko
0:01:40][%emt 0:00:08]}{[%eval -205,13]} 9. Nc3 {[%clk 0:01:38][%clko
0:01:19][%emt 0:00:04]}{[%eval -89,11]} 9... O-O {[%clk 0:01:16][%clko
0:01:38][%emt 0:00:05]}{[%eval -203,12]} 10. Qd2 {[%clk 0:01:32][%clko
0:01:16][%emt 0:00:08]}{[%eval -90,11]} 10... Kg7 {[%clk 0:01:13][%clko
0:01:32][%emt 0:00:05]}{[%eval -206,11]} 11. b4 {[%clk 0:01:26][%clko
0:01:13][%emt 0:00:07]}{[%eval -90,10]} 11... h6 {[%clk 0:01:09][%clko
0:01:26][%emt 0:00:06]}{[%eval -198,11]} 12. Rb1 {[%clk 0:01:26][%clko
0:01:09][%emt 0:00:02]}{[%eval -84,9]} 12... a5 {[%clk 0:01:04][%clko
0:01:26][%emt 0:00:06]}{[%eval -198,12]} 13. b5 {[%clk 0:01:23][%clko
0:01:04][%emt 0:00:05]}{[%eval -71,11]} 13... c5 {[%clk 0:01:02][%clko
0:01:23][%emt 0:00:04]}{[%eval -198,12]} 14. Qe3 {[%clk 0:01:18][%clko
0:01:02][%emt 0:00:07]}{[%eval -62,11]} 14... a4 {[%clk 0:00:57][%clko
0:01:18][%emt 0:00:06]}{[%eval -201,12]} 15. a3 {[%clk 0:01:17][%clko
0:00:57][%emt 0:00:04]}{[%eval -66,11]} 15... Ra7 {[%clk 0:00:54][%clko
0:01:17][%emt 0:00:06]}{[%eval -202,12]} 16. Qc1 {[%clk 0:01:15][%clko
0:00:54][%emt 0:00:03]}{[%eval -56,10]} 16... Rfa8 {[%clk 0:00:52][%clko
0:01:15][%emt 0:00:04]}{[%eval -210,12]} 17. Nd2 {[%clk 0:01:14][%clko
0:00:52][%emt 0:00:03]}{[%eval -51,10]} 17... Ra5 {[%clk 0:00:49][%clko
0:01:14][%emt 0:00:05]}{[%eval -204,12]} 18. Qb2 {[%clk 0:01:12][%clko
0:00:49][%emt 0:00:05]}{[%eval -49,10]} 18... Kh7 {[%clk 0:00:47][%clko
0:01:12][%emt 0:00:04]}{[%eval -192,13]} 19. Nf1 {[%clk 0:01:11][%clko
0:00:47][%emt 0:00:03]}{[%eval -50,11]} 19... Kg8 {[%clk 0:00:45][%clko
0:01:11][%emt 0:00:04]}{[%eval -187,12]} 20. Qd2 {[%clk 0:01:06][%clko
0:00:45][%emt 0:00:06]}{[%eval -47,12]} 20... Kg7 {[%clk 0:00:43][%clko
0:01:06][%emt 0:00:04]}{[%eval -188,13]} 21. Ne3 {[%clk 0:01:05][%clko
0:00:43][%emt 0:00:04]}{[%eval -46,12]} 21... Kh7 {[%clk 0:00:39][%clko
0:01:05][%emt 0:00:06]}{[%eval -185,13]} 22. Qc2 {[%clk 0:01:00][%clko
0:00:39][%emt 0:00:07]}{[%eval -46,10]} 22... Kh8 {[%clk 0:00:37][%clko
0:01:00][%emt 0:00:04]}{[%eval -188,12]} 23. Qd2 {[%clk 0:00:53][%clko
0:00:37][%emt 0:00:09]}{[%eval -46,11]} 23... Kg7 {[%clk 0:00:36][%clko
0:00:53][%emt 0:00:04]}{[%eval -187,12]} 24. Qb2 {[%clk 0:00:50][%clko
0:00:36][%emt 0:00:04]}{[%eval -46,10]} 24... Kg8 {[%clk 0:00:34][%clko
0:00:50][%emt 0:00:03]}{[%eval -185,12]} 25. Qc1 {[%clk 0:00:48][%clko
0:00:34][%emt 0:00:04]}{[%eval -47,11]} 25... R5a7 {[%clk 0:00:33][%clko
0:00:48][%emt 0:00:03]}{[%eval -184,11]} 26. Ned1 {[%clk 0:00:47][%clko
0:00:33][%emt 0:00:03]}{[%eval -46,11]} 26... Kh7 {[%clk 0:00:32][%clko
0:00:47][%emt 0:00:03]}{[%eval -185,12]} 27. Qd2 {[%clk 0:00:45][%clko
0:00:32][%emt 0:00:03]}{[%eval -46,11]} 27... h5 {[%clk 0:00:28][%clko
0:00:45][%emt 0:00:05]}{[%eval -183,12]} 28. Qg5 {[%clk 0:00:45][%clko
0:00:28][%emt 0:00:03]}{[%eval -42,9]} 28... Kg7 {[%clk 0:00:28][%clko
0:00:45][%emt 0:00:03]}{[%eval -178,12]} 29. Qd2 {[%clk 0:00:41][%clko
0:00:28][%emt 0:00:06]}{[%eval -40,10]} 29... Ra5 {[%clk 0:00:24][%clko
0:00:41][%emt 0:00:05]}{[%eval -182,12]} 30. Ne3 {[%clk 0:00:39][%clko
0:00:24][%emt 0:00:04]}{[%eval -42,10]} 30... Kg8 {[%clk 0:00:23][%clko
0:00:39][%emt 0:00:04]}{[%eval -176,12]} 31. Rf1 {[%clk 0:00:38][%clko
0:00:23][%emt 0:00:03]}{[%eval -39,9]} 31... Qc7 {[%clk 0:00:20][%clko
0:00:38][%emt 0:00:04]}{[%eval -182,12]} 32. Rd1 {[%clk 0:00:36][%clko
0:00:20][%emt 0:00:03]}{[%eval -42,10]} 32... R5a7 {[%clk 0:00:20][%clko
0:00:36][%emt 0:00:03]}{[%eval -181,10]} 33. Rb1 {[%clk 0:00:33][%clko
0:00:20][%emt 0:00:06]}{[%eval -43,11]} 33... Qd7 {[%clk 0:00:19][%clko
0:00:33][%emt 0:00:03]}{[%eval -179,12]} 34. Kh2 {[%clk 0:00:31][%clko
0:00:19][%emt 0:00:04]}{[%eval -45,10]} 34... Qc8 {[%clk 0:00:17][%clko
0:00:31][%emt 0:00:04]}{[%eval -177,11]} 35. Nc2 {[%clk 0:00:30][%clko
0:00:17][%emt 0:00:03]}{[%eval -34,10]} 35... Ra5 {[%clk 0:00:16][%clko
0:00:30][%emt 0:00:03]}{[%eval -174,11]} 36. Ne1 {[%clk 0:00:28][%clko
0:00:16][%emt 0:00:04]}{[%eval -37,10]} 36... Kg7 {[%clk 0:00:15][%clko
0:00:28][%emt 0:00:03]}{[%eval -189,11]} 37. Qf4 {[%clk 0:00:27][%clko
0:00:15][%emt 0:00:03]}{[%eval -25,10]} 37... Bf5 {[%clk 0:00:13][%clko
0:00:27][%emt 0:00:04]}{[%eval -179,11]} 38. Nf3 {[%clk 0:00:27][%clko
0:00:13][%emt 0:00:02]}{[%eval -24,10]} 38... Qd8 {[%clk 0:00:10][%clko
0:00:27][%emt 0:00:05]}{[%eval -181,11]} 39. Ng5 {[%clk 0:00:25][%clko
0:00:10][%emt 0:00:04]}{[%eval -20,9]} 39... Qb6 {[%clk 0:00:10][%clko
0:00:25][%emt 0:00:02]}{[%eval -182,10]} 40. Qe3 {[%clk 0:00:24][%clko
0:00:10][%emt 0:00:03]}{[%eval -17,8]} 40... Kf8 {[%clk 0:00:09][%clko
0:00:24][%emt 0:00:02]}{[%eval -176,11]} 41. Nd5 {[%clk 0:00:21][%clko
0:00:09][%emt 0:00:04]}{[%eval 69,8]} 41... Nxd5 {[%clk 0:00:09][%clko
0:00:21][%emt 0:00:02]}{[%eval -150,12]} 42. Bxd5 {[%clk 0:00:24][%clko
0:00:09]}{[%eval 85,7]} 42... e6 {[%clk 0:00:08][%clko 0:00:24][%emt 0:00:02]}
{[%eval -6,11]} 43. Bf3 {[%clk 0:00:21][%clko 0:00:08][%emt 0:00:05]}{[%eval
98,9]} 43... e5 {[%clk 0:00:08][%clko 0:00:21][%emt 0:00:02]}{[%eval -5,11]}
44. Bd5 {[%clk 0:00:20][%clko 0:00:08][%emt 0:00:03]}{[%eval 113,9]} 44... Re8
{[%clk 0:00:08][%clko 0:00:20][%emt 0:00:02]}{[%eval 0,11]} 45. Qf3 {[%clk
0:00:20][%clko 0:00:08][%emt 0:00:02]}{[%eval 186,9]} 45... Ke7 {[%clk
0:00:06][%clko 0:00:20][%emt 0:00:04]}{[%eval 117,12]} 46. e4 {[%clk
0:00:21][%clko 0:00:06]}{[%eval 186,7]} 46... Bd7 {[%clk 0:00:05][%clko
0:00:21][%emt 0:00:02]}{[%eval 157,11]} 47. Qxf7+ {[%clk 0:00:21][%clko
0:00:05][%emt 0:00:02]}{[%eval 230,8]} 47... Kd8 {[%clk 0:00:07][%clko
0:00:21]}{[%eval 157,11]} 48. Qxg6 {[%clk 0:00:21][%clko 0:00:07][%emt
0:00:02]}{[%eval 242,9]} 48... Kc8 {[%clk 0:00:06][%clko 0:00:21][%emt
0:00:04]}{[%eval 160,11]} 49. Nf7 {[%clk 0:00:21][%clko 0:00:06][%emt 0:00:03]}
{[%eval 278,10]} 49... Kc7 {[%clk 0:00:04][%clko 0:00:21][%emt 0:00:03]}{[%eval
284,12]} 50. Nxd6 {[%clk 0:00:21][%clko 0:00:04][%emt 0:00:02]}{[%eval 269,9]}
50... Rf8 {[%clk 0:00:04][%clko 0:00:21][%emt 0:00:02]}{[%eval 220,10]} 51. f4
{[%clk 0:00:21][%clko 0:00:04][%emt 0:00:01]}{[%eval 285,9]} 51... h4 {[%clk
0:00:04][%clko 0:00:21][%emt 0:00:02]}{[%eval 250,10]} 52. f5 {[%clk
0:00:20][%clko 0:00:04][%emt 0:00:04]}{[%eval 355,9]} 52... hxg3+ {[%clk
0:00:04][%clko 0:00:20][%emt 0:00:02]}{[%eval 288,10]} 53. Kxg3 {[%clk
0:00:21][%clko 0:00:04]}{[%eval 355,8]} 53... Rb8 {[%clk 0:00:04][%clko
0:00:21][%emt 0:00:02]}{[%eval 369,11]} 54. Qf6 {[%clk 0:00:18][%clko
0:00:04][%emt 0:00:06]}{[%eval 479,9]} 54... Ra6 {[%clk 0:00:04][%clko
0:00:18][%emt 0:00:02]}{[%eval 434,10]} 55. bxa6 {[%clk 0:00:18][%clko
0:00:04][%emt 0:00:01]}{[%eval 501,8]} 55... Qxb1 {[%clk 0:00:03][%clko
0:00:18][%emt 0:00:02]}{[%eval 452,9]} 56. Nb5+ {[%clk 0:00:20][%clko 0:00:03]}
{[%eval 501,7]} 56... Bxb5 {[%clk 0:00:03][%clko 0:00:20][%emt 0:00:02]}{[%eval
415,9]} 57. Qxe5+ {[%clk 0:00:17][%clko 0:00:03][%emt 0:00:06]}{[%eval 501,7]}
57... Kb6 {[%clk 0:00:03][%clko 0:00:17][%emt 0:00:02]}{[%eval 437,10]} 58.
Qxb8 {[%clk 0:00:14][%clko 0:00:03][%emt 0:00:04]}{[%eval 511,8]} 58... Qg1+
{[%clk 0:00:03][%clko 0:00:14][%emt 0:00:02]}{[%eval 465,10]} 59. Kh4 {[%clk
0:00:15][%clko 0:00:03][%emt 0:00:01]}{[%eval 511,8]} 59... Bxa6 {[%clk
0:00:02][%clko 0:00:15][%emt 0:00:02]}{[%eval 508,12]} 60. Qd6+ {[%clk
0:00:13][%clko 0:00:02][%emt 0:00:04]}{[%eval 562,9]} 60... Ka7 {[%clk
0:00:02][%clko 0:00:13][%emt 0:00:02]}{[%eval 482,11]} 61. f6 {[%clk
0:00:13][%clko 0:00:02][%emt 0:00:02]}{[%eval 604,9]} 61... Qf2+ {[%clk
0:00:02][%clko 0:00:13][%emt 0:00:02]}{[%eval 537,10]} 62. Kh5 {[%clk
0:00:13][%clko 0:00:02][%emt 0:00:01]}{[%eval 643,8]} 62... Qf3+ {[%clk
0:00:02][%clko 0:00:13][%emt 0:00:02]}{[%eval 529,10]} 63. Kg6 {[%clk
0:00:15][%clko 0:00:02]}{[%eval 511,7]} 63... Qg2+ {[%clk 0:00:02][%clko
0:00:15][%emt 0:00:02]}{[%eval 1030,10]} 64. Kf7 {[%clk 0:00:16][%clko
0:00:02][%emt 0:00:02]}{[%eval 707,8]} 64... Qxh3 {[%clk 0:00:02][%clko
0:00:16][%emt 0:00:02]}{[%eval 1187,10]} 65. Qxc5+ {[%clk 0:00:15][%clko
0:00:02][%emt 0:00:02]}{[%eval 991,9]} 65... Ka8 {[%clk 0:00:02][%clko
0:00:15][%emt 0:00:02]}{[%eval 1138,11]} 66. Ke7 {[%clk 0:00:14][%clko
0:00:02][%emt 0:00:04]}{[%eval 1124,9]} 66... Qh4 {[%clk 0:00:02][%clko
0:00:14][%emt 0:00:02]}{[%eval 1542,11]} 67. Qd6 {[%clk 0:00:12][%clko
0:00:02][%emt 0:00:04]}{[%eval 1124,8]} 67... Ka7 {[%clk 0:00:02][%clko
0:00:12][%emt 0:00:02]}{[%eval 1676,10]} 68. e5 {[%clk 0:00:12][%clko
0:00:02][%emt 0:00:02]}{[%eval 1401,8]} 68... Qg5 {[%clk 0:00:02][%clko
0:00:12][%emt 0:00:02]}{[%eval 1386,10]} 69. Ke8 {[%clk 0:00:10][%clko
0:00:02][%emt 0:00:03]}{[%eval 1386,8]} 69... Qf4 {[%clk 0:00:02][%clko
0:00:10][%emt 0:00:02]}{[%eval 1740,9]} 70. f7 {[%clk 0:00:10][%clko
0:00:02][%emt 0:00:02]}{[%eval 2076,9]} 70... Qe3 {[%clk 0:00:02][%clko
0:00:10][%emt 0:00:02]}{[%eval 29994,10]} 71. f8=Q {[%clk 0:00:13][%clko
0:00:02]}{[%eval 29987,8]} 71... Qb6 {[%clk 0:00:03][%clko 0:00:13]}{[%eval
29995,10]} 72. Qxb6+ {[%clk 0:00:14][%clko 0:00:03]}{[%eval 29988,6]} 72...
Kxb6 {[%clk 0:00:05][%clko 0:00:14]}{[%eval 29996,9]} 73. Kd7 {[%clk
0:00:16][%clko 0:00:05]}{[%eval 29995,8]} 73... Bb5+ {[%clk 0:00:07][%clko
0:00:16]}{[%eval 29996,7]} 74. cxb5 {[%clk 0:00:18][%clko 0:00:07]}{[%eval
29996,3]} 74... Kxb5 {[%clk 0:00:09][%clko 0:00:18]}{[%eval 29997,6]} 75. Bc4+
{[%clk 0:00:20][%clko 0:00:09]}{[%eval 29997,3]} 75... Kb6 {[%clk
0:00:10][%clko 0:00:20]}{[%eval 29998,5]} 76. Qf2+ {[%clk 0:00:21][%clko
0:00:10]}{[%eval 29998,3]} 76... Ka5 {[%clk 0:00:13][%clko 0:00:21]}{[%eval
29999,3]} 77. Qa7# {[%clk 0:00:24][%clko 0:00:13]}{[%eval 29999,3]} 1-0
Parent - - By ubaldus (*) [us] Date 2008-08-09 15:30
No matter what I try, Exchange handicap seems to be much easier for Rybka (1cpu-w32) than pawn and move. I use 2+2" and 3+2" controls.

It is destroying Alaric (CEGT 2706) in all openings, scoring >75% (pawn and move match was basically even), and is currently ahead of Spike 1.2 sp (CEGT 2780) 11.5:2.5 (+10=3-1 !!!). Rybka comes out of the opening with about -0.8 eval, and then outsearches Spike slowly, but surely. No identical games past move 9 or so.

I even tried a short match vs Toga II 1.4.2JD 2CPU (CEGT 2886). This was too much for Rybka, but it scored 35% so far, 10.5:19.5 (+6=9-15). Maybe you want to give sp Toga this handicap on your octal, and see whether Rybka comes out ahead?

Another interesting tidbit: Rybka 2 (1cpu-w32) - Adam 3.1 (CCRL 2251) at knight handicap, with your book - 67.5:12.5 (+60=15-5 !!!).  I will try some 2400 engine next.  It seems that your MC handicap estimates ovevalue some handicaps in engine games. Of course, we have seen what happens when you replace 2250 engine with a 2300 human.

I will post final results when I accumulate more data.
Parent - - By lkaufman (*****) Date 2008-08-10 00:11
Actually I think MC estimates handicaps pretty well for engines, the problem is that I get higher values with greater depth, so it's a question of picking the appropriate depth for the pairing in question. For example knight odds is only 576 Elo at five ply, 742 at 8 ply, and 1040 at 10 ply in my MC tests.
Parent - - By ubaldus (*) [us] Date 2008-08-10 00:34
Basically, you are saying that a program like Adam (900 pts below Rybka sp) gets wiped out in blitz at knight odds, but should be able to hold its own in a 40/120 match, since it's tactical weakness will matter less?

I took a CCRL 2500 engine (chose Gaia 3.5) and tried it at knight odds against Rybka at 3+2". It was really amusing, very few draws. If Rybka gets within -1.3 eval or so, Gaia then gets outsearched, and after the eval gets to 0, it's over. The result: 14.5:15.5 Rybka loss (+12=5-13), indicating about 600 handicap value for this matchup.

What are Elo values for exchange handicaps at different plies?
Parent - - By lkaufman (*****) Date 2008-08-10 02:22
Yes, the longer the time limit, the greater the rating value of a handicap. To win a lost game, you need mistakes, and shorter searching means more mistakes. For the Exchange, I got 192 at 5 ply, 252 at 8 ply, 278 at 10 ply, 325 at 11 ply, and 310 at 12 ply. Probably the last two are reversed just due to small sample size.
Parent - - By ubaldus (*) [us] Date 2008-08-10 06:33
I am really interested to see how Exchange handicap part of Milov match goes. It seems to be a somewhat smaller handicap than pawn and move, and white can strive for initiative in an almost normal way. Pawn and move handicap just leads black to a cramped position pawn down, not fun.

From what I have seen in my experiments against 2700-2900 engines, I think Rybka is a favorite in this part of the match, and pawn and move part is an even matchup. Of course, a lot will depend on the openings - both your prep and Milov's.

And thank you for your informative answers, this theoretical aspect of handicap chess is a lot of fun.
Parent - - By lkaufman (*****) Date 2008-08-10 15:43
I extended my Monte Carlo test of the Exchange handicap last night to 13 ply, and the result after nearly 500 games is an alarmingly high 393 Elo points. I agree that the evidence so far suggests that the Exchange is a smaller handicap than pawn and move (f7) (that was also Staunton's opinion back in the days when both handicaps were popular), and that the games should be more interesting as Rybka can play actively despite the material disadvantage. However I think Milov is a strong favorite at pawn and move in classical chess; the time limit makes a big difference in handicap play, and so may still be favored at the Exchange handicap.  The fact that I'm already showing nearly 400 Elo for the Exchange handicap at levels corresponding to blitz for Rybka is rather alarming to me.
Parent - - By ubaldus (*) [us] Date 2008-08-13 19:56
In 50 exchange handicap games against Bright 0.3a - 2cpu (CEGT 2800) my R3-sp-w32 scored 28:22 (25 games with 1.d4, 25 games with 1.c4, 8+2''), and against Alaric 707 (CEGT 2700) it was scoring in 65% range. These results indicate 2800+ ELO.

But against Toga 1.4.2JD - 2cpu (CEGT 2900) it "only" managed 18.5:41.5, and it scored 40% against Glaurung 2.1 w32 - 2cpu (CEGT 2850). These results indicate 2750 ELO, for 300 pts handicap value. Of course, the time control is quite fast, probably corresponding to 40/2' on your quad or even faster.

I suppose the ply effect explains why my sp Rybka 3 (CEGT 3060) outperforms in handicap tests against relatively weaker engines. The sample sizes are not large, but since the stronger engines effectively play on higher ply, the results seems to be intuitively correct.

Your octal Rybka is about CEGT 3300. I am quite sure it should at least hold its own against 2850 engines in blitz, and probably in rapid as well. Whether it translates into an even game against 2700 GM remains to be seen.
Parent - - By lkaufman (*****) Date 2008-08-13 20:39
"Your octal Rybka is about CEGT 3300. I am quite sure it should at least hold its own against 2850 engines in blitz, and probably in rapid as well. Whether it translates into an even game against 2700 GM remains to be seen. " I agree with the blitz part, but I have a lot of doubt about the "rapid" part. I'll run some tests soon. We already know that humans do better in these handicap matches than do similarly rated programs, so I think we'll have a tough time. Milov is already showing a serious intent to prepare well for this match.
Parent - - By ubaldus (*) [us] Date 2008-08-19 22:59
I ran some rapid exchange handicap matches (g/35+10'') of R3 sp w32 versus ProDeo 1.2 (CEGT 40/20 rating 2640) and Toga 1.4.2JD (2866). I chose ProDeo (=Rebel Uci) since it was playing Rybka quite well in my blitz tests, better than other 2600-2700 engines. I don't know why, maybe it was just a sampling error, or maybe it has a lot of chess knowledge to play relatively well in a won position.

In any case, Rybka beat ProDeo 29:21 (+26=6-18), and played unexpectedly well against Toga, scoring about 40% so far, 9.5:14.5. Toga managed to lose some dead drawn endgames by overpressing, any human master would certainly know better.

In any case, so far handicapped Rybka shows ~2740 performance, only 300 below its CEGT 40/20 rating. Bright 0.3a is next.

Oh, btw, I run Rybka - Toga blitz match with Toga giving exchange handicap. Toga lost 3.5:36.5 (+2=3-35). Rybka actually managed to lose 2 games by chasing some king attacks which never came through. It is quite funny to see 3000+ player losing when spotted an exchange.
Parent - - By lkaufman (*****) Date 2008-08-19 23:47
My Monte Carlo results for Exchange handicap gave a value of about 315 Elo for 11 and 12 ply searches on Rybka 3, but a whopping 395 Elo for 13 ply (though sample size was smaller). Based on this, I would put the handicap value for rapid games as somewhere in the mid 300s. Your value of 300 is not so far away considering the sample size and the difference in testing method. I tend to trust the MC results a bit more, just because your method does not produce much opening variety (I think you indicated you only input either 1d4 or 1c4 and no more moves). If you are using only single processor, what prevents you from duplicating the same game many times (or at least for many moves)? You might just be proving that the particular opening that each side likes is really the wrong choice for one side or the other. How much opening variety do you get?
Parent - - By ubaldus (*) [us] Date 2008-08-20 16:06
No long repeated sequences in these two matches. It looks to me that Rybka time management is not entirely deterministic, or maybe 3-7% of CPU taken by GUI and IE on my machine create this impression, by randomzing the search. Because of this, after at most 10 moves, the games always diverge, and Rybka eval is still in -0.7 to -1.0 range, so its position is still lousy.

Many engines simply cannot play on its own after 1.e4, it seems. Without the second knight, they want to play Alekhine or Scandinavian or Petroff, and then get outplayed. Right now, in Rybka-Bright match Rybka is ahead 12:2 (!!), mostly after 1.e4 (and Bright is CEGT 2725). Still, the eval after 10 moves the eval is in -0.8 range and all games are quite different. So, I stick to d4 and c4, makes it more competitive.
Parent - By lkaufman (*****) Date 2008-08-20 17:57
I don't think that there is a big difference in the merit of the four normal opening moves (e4, d4, c4, and Nf3) either at Exchange handicap or in normal chess. I would have used all four moves for maximum variety in your test, but the results would probably not be much different.
Up Topic Rybka Support & Discussion / Rybka Discussion / Match agreement with FIDE 2700+ opponent.
1 2 3 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill