Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Game 2. Benjamin - Rybka (no c7 pawn) 1/2 - 1/2
- - By Victor Zakharov (*****) Date 2007-08-06 23:00
[Event ""]
[Site "chessplanet.ru"]
[Date "2007.8.7"]
[Round ""]
[White "GM Benjamin"]
[Black "Rybka"]
[Result "1/2-1/2"]
[Eco ""]
[Annotator ""]
[Source ""]

1.Na3 c5 2.Nf3 c4 3.Nxc4 Nh6 4.Na3 Ng4 5.Ng1 Nf6 6.Nb1 Ng8 7.e4
Qa5 8.c3 Nf6 9.d3 Nc6 10.Nf3 e5 11.Bg5 Be7 12.Nbd2 d6 13.Be2
O-O 14.O-O Qc7 15.h3 h6 16.Bh4 Re8 17.Re1 a6 18.Bf1 b5 19.d4
exd4 20.cxd4 Qb6 21.d5 Ne5 22.Rc1 Bb7 23.Nxe5 dxe5 24.Bg3 Bd6
25.Nf3 Bb4 26.Re2 Bd6 27.Qd3 Nh5 28.Bh2 Qd8 29.Rec2 f5 30.Nd2
Nf6 31.f3 fxe4 32.fxe4 Nh5 33.Nf3 Nf4 34.Qe3 Rf8 35.Kh1 Bc8 36.Rc6
Bb7 37.R6c2 Bc8 38.Rc6 Bb7 39.R6c2 { Draw by repetition} 1/2-1/2
Parent - - By FWCC (***) Date 2007-08-06 23:10
The Rybka team will do their homework tonight and come up with something.Rest assured they know where to do improvements.I'm glad she drew the second game which is a good sign.Is giving the GM a FULL pawn too much?I don't really think so,but we will see.Vas ,it's gonna be a long night hang tough and lets go get em!
Parent - - By lkaufman (*****) Date 2007-08-07 02:57
There is nothing wrong on our end (now that the game 1 bug is fixed). The only problem we may have is that Joel seriously prepared his openings for this match, whereas Ehlvest didn't spend much time on this. It may turn out that a full pawn is too much for a well-prepared GM. We will see.
Parent - - By Dominik Landau (*) Date 2007-08-07 07:05
Isn't it possible that Rybkas 3100 ELO are *MUCH* overrated ? When you take tactical issues apart (they don't appear if Joel does not allow), in my opintion Rybkas strength does not exceed 2500 ELO.
Parent - By Laszlo (***) Date 2007-08-07 07:36
Are you a german program supporter? Maybe a Loop fan??
Parent - By Linus (***) Date 2007-08-07 08:03
On which planet have you been living the last two years?
Parent - By Felix Kling (Gold) Date 2007-08-07 09:22
It's not possible, since Rybka is relativly (if compared to super tactical engines) weak in tactics and a much better program in strategy, so I don't know how you came to that oppinion. Have you looked at some Rybka games? You will soon see that Rybka wins with positional play.
Parent - - By lkaufman (*****) Date 2007-08-07 13:25
     It's very difficult to separate tactics from strategy. Compared to other programs, Rybka excels in positional play. Compared to humans, of course (like all programs) it is best at tactics. In terms of pure chess knowledge, Rybka is probably slightly weaker than I am (so maybe 2300 as I'm near 2400), since I'm still able to teach her things. But 2300 level chess knowledge applied to a very deep search produces 2800+ level positional play. Since tactics are unavoidable in most games, this must be combined with maybe a 3200 level for tactics to produce my overall estimate of 3000.
Parent - By SR (****) Date 2007-08-07 13:33
It is actually interesting that Rybka (that is considered "weak" in tactics) recently have played some of the most beautiful and impressive and tactical chess I ever have seen (Rybka - Diep in the world championship in Amsterdam 2007 is for example an amazing game). 

For me it was always striking how few "pearls" +2700 programs have been able to produce. With Rybka it is different and she has (as all the human masters) played at least a handful of truly brilliant games.
Parent - - By Dominik Landau (*) Date 2007-08-07 19:09
I dont want to give a formal definition of 'tacitic' and 'strategy'. When i say tactic i mean the following situation:

A human GM plays a move . After his move the engine evaluation rapidly increases significant. ... After seeing the increase, the human puts his handy on his Head and thinks 'oh, damn.. i was blind, i oversight this easy variant'...

My simple opionion is: I don't want to see this kind of mistakes in an Worldclass human-computer match. It's ok when i see it in a lokal tournament in my little village.

In my original post, i wantet to exclude this kind of human mistaces. (take back move. When i play with a friend, i usually allow him to take back moves when he oversee a simple variant).

You say, that Rybkas knowledge is about 2300 ELO, und you say that within the deep search the positional knowlege increases to 2800+ level. Are you really sure ? I am Not!
The tactical strength increases - sure (a lot of, more than 400+ ELO) . But the positional ELO increases about 100 ELO. Maybe 50 or 200. But not 500+!

Positional knowledge gives a hint about the future development of a position. The future is in the range of 30...80 full-moves.  It dont care a lot  if you search for some halfmoves...consider that the large searchdepth  (18 halfmoves or so) of rybka is not a full search, it is predominantly a tactical search. Thats why i dont believe that the positional ELO is increasing
that much.

I observed the matches vs Joel and Ehlvest, and i had not the feeling (regardless of pawn odd or time handicap)  that Rybka has 500+ ELO more !!

500+ Elo is .... a lot.  Garry Kasparaov wins against  a gaggle of  2250 Elo player  in a blind simultan, he overplays them in an execution manner.

Rybka does absolutely not execute Joel nor Ehlvest.  They seem to play at the same level, as long as there happen no tactical mistaces in my obove description.

@ Laszlo
And no, i am not involved in some other engine. The other engines positional play is about 300+ ELO less than rybka's. I did'nt verify all other Engines.
Rybka is considerably stronger than all others.
Parent - - By Uri Blass (*****) Date 2007-08-07 19:22 Edited 2007-08-07 19:29
I agree that the positional strength increase clearly more than 200 elo by searching deeper not only for rybka but for every program.

It is possible to make an experiment and tell program after depth 4 to change their choice only if they find something better than the move that they want to play by more than 0.5 pawn.

I expect programs rating to drop by more than 200 elo at tournament time control.
Maybe vasik can release personality of rybka that does it so you can see it for rybka.

I did not try to do it for my movei but I will be surprised if the drop in strength is not significantly more than 200 elo at long time control.

I expect rybka to beat every human when humans are allowed to take back tactical mistakes if we define tactical mistake as move that cause rybka's evaluation to increase by more than 0.5 pawn relative to the previous move(of course it may not be a tactical mistake and it is possible to have a situation that every move cause rybka's evaluation to go up by more than 0.5 pawn after rybka simply outplays the human positionally but the idea is that the human will be asked if he want to take back the move after something like that happens and we need to decide technical things about the time of the players and it is possible to decide that the time of both side go back to the situation before the blunder).

Uri
Parent - - By turbojuice1122 (Gold) Date 2007-08-07 20:36
I think a good test to your idea is to check out the difference between Rybka Winfinder 2.2 and Rybka 2.2 on the rating lists.  We have the CEGT blitz list as the only thing I know of that ranks both, and the rating difference is right around 200 points.
Parent - - By Uri Blass (*****) Date 2007-08-08 07:08
No

I think that it is a very bad test because winfinder is not blind positionally.

If Vasik writes an engine that never change it's mind unless it find a move that is at least 0.5 pawn better then it is going to be a good test.
winfinder does not behave in that way.

Uri
Parent - By Uri Blass (*****) Date 2007-08-08 07:52
I forgot to add that winfinder is only for finding wins and not for finding tactics.

Here is analysis of the opening position by winfinder
You can see that winfinder changes it's mind from Nc3 to d4 and later  for Nf3 for positional reasons(0.05 pawns is not a tactical reason).

New game - Glaurung Lodz 2007, 2880'/1
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka WinFinder 2.2 32-bit:

1.Nb1-c3
  =  (0.20)   Depth: 4   00:00:00
1.Nb1-c3
  =  (0.10)   Depth: 5   00:00:00
1.Nb1-c3 Nb8-c6
  =  (0.05)   Depth: 6   00:00:00  1kN
1.Nb1-c3 Nb8-c6 2.d2-d4
  =  (0.10)   Depth: 7   00:00:00  2kN
1.d2-d4 Nb8-c6 2.Nb1-c3
  =  (0.15)   Depth: 7   00:00:00  3kN
1.d2-d4 Nb8-c6 2.Nb1-c3 Ng8-f6
  =  (0.10)   Depth: 8   00:00:00  5kN
1.d2-d4 Ng8-f6 2.Nb1-c3 e7-e6 3.Ng1-f3
  =  (0.05)   Depth: 9   00:00:00  12kN
1.d2-d4 Ng8-f6 2.Nb1-c3 e7-e6 3.Ng1-f3 Nb8-c6
  =  (0.05)   Depth: 10   00:00:00  24kN
1.d2-d4 Ng8-f6 2.Nb1-c3 e7-e6 3.Ng1-f3 Nb8-c6 4.Bc1-g5
  =  (0.00)   Depth: 11   00:00:01  51kN
1.Nb1-c3 Nb8-c6 2.d2-d4 d7-d5 3.Bc1-f4 Bc8-f5 4.Ng1-f3
  =  (0.05)   Depth: 11   00:00:01  68kN
1.Nb1-c3 Ng8-f6 2.Ng1-f3 Nb8-c6 3.e2-e3 e7-e6 4.Bf1-e2 d7-d5
  =  (0.00)   Depth: 12   00:00:02  117kN
1.d2-d4 Ng8-f6 2.Nb1-c3 e7-e6 3.Ng1-f3 Nb8-c6 4.e2-e3 Bf8-b4
  =  (0.10)   Depth: 12   00:00:02  143kN
1.d2-d4 Ng8-f6 2.Nb1-c3 e7-e6 3.Ng1-f3 Nb8-c6 4.e2-e3 d7-d5 5.Bf1-d3
  =  (0.05)   Depth: 13   00:00:04  218kN
1.d2-d4 Ng8-f6 2.Nb1-c3 d7-d5 3.Bc1-f4 Nf6-h5 4.Qd1-d2 Nb8-c6 5.Ng1-f3 Nh5xf4
  =  (0.05)   Depth: 14   00:00:12  663kN
1.d2-d4 Ng8-f6 2.Nb1-c3 d7-d5 3.Bc1-f4 Nf6-h5 4.Qd1-d2 Nb8-c6 5.Ng1-f3 Nh5xf4 6.Qd2xf4 e7-e6
  =  (0.05)   Depth: 15   00:00:28  1521kN
1.Ng1-f3 Nb8-c6 2.Nb1-c3 Ng8-f6 3.d2-d4 d7-d5 4.g2-g3 g7-g6 5.Bf1-g2 Bf8-g7 6.Nf3-e5
  =  (0.10)   Depth: 15   00:00:55  2959kN
1.Ng1-f3 Nb8-c6 2.Nb1-c3 Ng8-f6 3.d2-d4 d7-d5 4.g2-g3 g7-g6 5.Bf1-g2 Bf8-g7 6.Nf3-e5 0-0
  =  (0.05)   Depth: 16   00:01:04  3434kN

(,  08.08.2007)

Uri
Parent - - By Vasik Rajlich (Silver) Date 2007-08-08 08:55
I have done some experiments with evaluation granularity. Lower granularity makes the search more efficient, for obvious reasons - it's one of those changes which is clearly good for tactics and clearly bad for positional play. As you can see, I have rejected any lower granularity.

Using a granularity of 1/2 pawn will cost way more than 200 Elo - unless you think that Rybka evaluating at half-pawn granularity can play on an equal level with Fritz. A better estimate would be 500 Elo.

Games between professional chess programs are 95% about positional play. The ratio is a bit different for human play, where tactical mistakes are more frequent and therefore where tactics are a bit more important.

Vas
Parent - - By Uri Blass (*****) Date 2007-08-08 10:12
I did not suggest something different in my post and my intuition tell me that it is clear that granularity of 1/2 pawn is going to cost more than 200 elo.

Note that there is no clear definition of positional play and tactical mistakes.
every positional mistake is tactical mistake if you search deep enough and what is considered by humans as positional mistake can be considered by chess programs as tactical mistake.

Uri
Parent - By Vasik Rajlich (Silver) Date 2007-08-09 08:40
Yes, that's true.

Already, with existing search depths, some "positional" concepts probably fit into this, related to king attacks and advanced passed pawns. The vast majority of positional concepts are miles beyond current horizons, though.

Vas
Parent - - By lkaufman (*****) Date 2007-08-07 19:53
500 elo points have a different meaning at 2800 level than at 2200 level. At high levels, chess is much more about subtleties. In the Ehlvest and Benjamin matches, there were hardly any moves at all that the GM would have taken back right away, and yet still they need a big handicap. I think that proves my point.
Parent - - By Dominik Landau (*) Date 2007-08-07 21:13
This pawn match is very interesting. because joel plays really good.

I would like to see Joel vs Rybka. Regular. I bet the result would be close.
Parent - By turbojuice1122 (Gold) Date 2007-08-07 21:18
I was just thinking about that earlier today--if Joel manages to draw or win this match, there should be a match with tournament time controls and perhaps 3-5 move book.
Parent - - By lkaufman (*****) Date 2007-08-08 04:23
Joel is indeed playing well, but despite that he is a game down after four (excluding the forfeit) starting a pawn up, so I don't see how you can imagine that a normal match would be anything but a massacre. However it is likely that we will have a second match at draw odds rather than pawn odds; each draw counts as a win for Joel. That could be close.
Parent - By Uri Blass (*****) Date 2007-08-08 08:02
I think that if Benjamin lose this match it may be better to give another GM a chance to try.

Uri
Parent - By Vasik Rajlich (Silver) Date 2007-08-08 08:59
Joel has an interesting and so far effective game plan. He's resisting all temptation to go into complications and to try too hard to win his better positions.

For example, in game 2, if he doesn't go for the Rc2-c6-c2 repetition, I am simply sure that his chances are much worse (although of course his position is better).

The ability for him to do this is one thing which won't show up in engine vs engine statistics at pawn odds.

In game 4, Rybka finally provoked him into a slugfest. I guess the passed pawn on e6 was too much to resist.

Vas
Parent - By Vasik Rajlich (Silver) Date 2007-08-08 09:02
It also depends on the audience. For most people, 2800 beating 2300 is already quite subtle. Probably, the place to look for something really crushing is games between 1500 players and 1000 players.

Vas
Parent - - By Uri Blass (*****) Date 2007-08-07 10:22
It seems that Benjamin is afraid to try to win and in better position he simply allowed repetition instead of trying to win.
Maybe his plan to win the match is simply to draw all the remaining games
Parent - By Laszlo (***) Date 2007-08-07 10:42
Rybka white game today could change Benjamin "chicken" tactic...
Parent - By lkaufman (*****) Date 2007-08-07 13:20
     It does seem that way, but it's partly an illusion. In the demonstration game, Joel repeated partly because he wanted a decent break before the next game, though he would have played to win if the plan to do so was more clear. In the second game, time pressure was the dominant factor in the decision to repeat.
Parent - - By sylvain.renard.27 (**) Date 2007-08-07 16:50
Game 2. Benjamin - Rybka (no c7 pawn)

1.Na3 c5 (??)

Sorry, but there is surely something someone did not explain to me.
How can you play c5 without c7 pawn? :)
  Best regards,
    Sylvain
Parent - - By Harvey Williamson (*****) Date 2007-08-07 16:51
Play the next few moves and you will understand :)
Parent - By sylvain.renard.27 (**) Date 2007-08-07 17:46
As a matter of fact, I am not interested in "pseudo-chess" games (960 or other form).
So, I did not go further than 2 half-moves :)
Nobody explained to me that the game was beginning in fact at move 7...
  Anyway, I am not very excited by "pseudo-chess".
Handicap games are not new! In 1891, the first Handicap Tournament of the Saint-Petersburg Chess Club took place.
But those kind of tournaments did not have much success in the history of chess...
  Best regards and thank you for taking me out of my natural lazyness :)
     Sylvain
Up Topic Rybka Support & Discussion / Rybka Discussion / Game 2. Benjamin - Rybka (no c7 pawn) 1/2 - 1/2

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill