Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / rybka without a knight against humans
1 2 Previous Next  
- - By Uri Blass (*****) Date 2007-08-24 16:02
I play 2 correspondence games of rybka and me without knight b1 against humans with no computer help to see if they can beat rybka with no computer help(I give here only one of them because the opponent asked me not to publish the second game without his agreement).

The following game is played in the israeli forum against aplayer with fide rating of 1998 and israeli rating of 2033

I decided to correct only one of rybka's moves and 3.c3 is move of me because I did not want to trade pieces but maybe that move was a practical mistake.

I cannot be sure that the opponent does not cheat but as there is no money prizes so I trust my opponents not to cheat.
I guess that rybka and me are going to lose
but it may be interesting to see if this is going to be the case.

game so far:

[Event "Edited game"]
[Site "URI-AMD"]
[Date "2007.08.24"]
[Round "-"]
[White "-"]
[Black "-"]
[Result "*"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/R1BQKBNR w KQkq - 0 1"]
[SetUp "1"]

r n b q k b n r
p p p p p p p p
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
R . B Q K B N R
white to play
512MB, RybkaII.ctg, URI-AMD
1. e4 c5 2. Nf3 Nc6 3. c3 Nf6 4. e5 Nd5 5. Bc4 Nb6 6. Bb3 d5 7. d4 cxd4 8.
cxd4 Bg4 9. Be3
Parent - By Gaмßito (****) Date 2007-08-24 18:30 Edited 2007-08-24 18:40
I see no reason to think that 3. c3 could be a mistake. Then, what other moves, do you suggest?

Mine are, 3. Bc4 or your move 3. c3. 

But definitively that play without a horse hurts enough, much more than what you believe. I think that play without a knight is too much advantage for your opponent and much more if you do not know if he can cheat at sometime, that can be enough to ruin the experiment. :-)

But, as an experiment it is fine and interesting.

Parent - - By lkaufman (*****) Date 2007-08-24 21:27
I almost always try to play an early c3 when giving knight odds, as it's main drawback is normally taking the square from the knight. Anyway, knight odds in correspondence chess to a 2000 level player sounds ridiculous, but I'm interested to see if you can even make a fight of it. At 40/2 I would bet on you.
Parent - By Gaмßito (****) Date 2007-08-24 22:12
Yes, in fact c3 is not a bad move.

Although it does not import too much that their opponent think all that he wants, rather it is important that he cannot cheat, what is difficult to trust in my opinion.

Parent - By Gaмßito (****) Date 2007-08-24 21:59 Edited 2007-08-24 22:35
2.3.2a prefers Bc4 after a lot of time. But even this way, many times programs do not play in the best way when these are losing.

So, we cannot trust too much in what they say. It is necessary to implement into the algorithms, something really good that does not allow to simplify when the program is clearly in inferior positions, unless of course this was strictly necessary and there are not other way to play. I remember a game some months ago when Rybka changed her queen in a clearly inferior endgame, and Vas said then that although their position was bad, Rybka would had to complicate or clearly try with another move.

r1bqkbnr/pp1ppppp/2n5/2p5/4P3/5N2/PPPP1PPP/R1BQKB1R w KQkq - 0 1

Analysis by Rybka 2.3.2a 32-bit :

3.Bd3 d6 4.0-0 Nf6 5.Re1 Bg4 6.h3 Be6
  -+  (-1.67)   Depth: 5   00:00:00
3.Bd3 d6 4.0-0 Nf6 5.Qe2 Bg4 6.Qe3 Qa5 7.Bc4
  -+  (-1.61)   Depth: 6   00:00:00  4kN
3.Bd3 d6 4.0-0 Nf6 5.Re1 Ng4 6.Bc4 Nce5 7.Nxe5 Nxe5
  -+  (-1.65)   Depth: 7   00:00:00  7kN
3.Bd3 d5 4.0-0 Nf6 5.e5 Ne4 6.Re1 Bf5 7.b3 Nd4 8.Nxd4 cxd4 9.Bb5+ Bd7
  -+  (-1.73)   Depth: 8   00:00:00  20kN
3.Bd3 d5 4.0-0 Nf6 5.exd5 Qxd5 6.b3 Bg4 7.Be2 0-0-0 8.h3 Bf5 9.Bb2
  -+  (-1.90)   Depth: 9   00:00:01  47kN
3.e5 Qc7 4.Bb5 Nxe5 5.0-0 Nf6 6.Nxe5 Qxe5 7.Re1 Qd5 8.d3 e6 9.Bc4 Qd4
  -+  (-1.89)   Depth: 9   00:00:02  81kN
3.d3 e6 4.Bf4 Nge7 5.Be2 Ng6 6.Bd2 Be7 7.0-0 0-0 8.h3 Nge5
  -+  (-1.80)   Depth: 9   00:00:03  123kN
3.Bc4 Nf6 4.e5 Ng4 5.Bxf7+ Kxf7 6.Ng5+ Kg8 7.Qxg4 h6 8.Qc4+ e6 9.Nf3 Qc7 10.Qe4 Nd4
  -+  (-1.68)   Depth: 9   00:00:04  132kN
3.Bc4 Nf6 4.e5 Ng4 5.Bxf7+ Kxf7 6.Ng5+ Kg8 7.Qxg4 h6 8.Qc4+ e6 9.Nf3 Qc7 10.Qe4 d5 11.exd6 Bxd6
  -+  (-1.69)   Depth: 10   00:00:04  152kN
3.Bc4 Nf6 4.Qe2 d5 5.exd5 Nxd5 6.0-0 Be6 7.Bb5 Qd6 8.Ne5 Bf5 9.d3 a6
  -+  (-1.72)   Depth: 11   00:00:06  217kN
3.Bb5 Qa5 4.Bxc6 dxc6 5.0-0 Nf6 6.d3 Be6 7.e5 Nd5 8.Bd2 Qa4 9.b3 Qg4 10.Qe2
  -+  (-1.65)   Depth: 11   00:00:10  367kN
3.Bb5 Qa5 4.Bxc6 dxc6 5.0-0 Nf6 6.d3 Be6 7.e5 Nd5 8.Bd2 Qa4 9.b3 Qa3 10.c4 Nb4
  -+  (-1.69)   Depth: 12   00:00:11  409kN
3.Bb5 d5 4.e5 Bd7 5.a4 Qc7 6.Qe2 Nh6 7.0-0 0-0-0 8.b3 Nf5 9.Bb2 Nb4 10.c3 Bxb5 11.axb5
  -+  (-1.69)   Depth: 13   00:00:29  1085kN
3.Bb5 d5 4.e5 Bd7 5.a4 Qc7 6.Qe2 Nh6 7.0-0 0-0-0 8.a5 a6 9.Bxc6 Bxc6 10.e6 Qd6 11.exf7 Nxf7 12.d4
  -+  (-1.73)   Depth: 14   00:00:47  1672kN
3.Bb5 d5 4.e5 Bd7 5.Qe2 Qb6 6.a4 0-0-0 7.0-0 Nh6 8.c3 a6 9.Bxc6 Qxc6 10.d4 Nf5 11.Bf4 Qb6 12.a5
  -+  (-1.72)   Depth: 15   00:01:16  2950kN
3.Bb5 d5 4.e5 Bd7 5.Qe2 Qb6 6.a4 0-0-0 7.0-0 Nh6 8.d3 Nf5 9.c3 a6 10.Bxc6 Qxc6 11.Bf4 h5 12.h3 Qb6
  -+  (-1.74)   Depth: 16   00:02:02  4952kN
3.Bb5 d5 4.e5 Bd7 5.Qe2 Nh6 6.0-0 a6 7.Bxc6 Bxc6 8.d3 Nf5 9.e6 fxe6 10.Qxe6 Qd7 11.Qe2 h5 12.Bf4 0-0-0 13.Qe5
  -+  (-1.78)   Depth: 17   00:04:27  11763kN
3.Bb5 d5 4.e5 Bd7 5.Qe2 Nh6 6.0-0 a6 7.Bxc6 Bxc6 8.d3 Nf5 9.e6 fxe6 10.Qxe6 Qd7 11.Qe2 g6 12.Bf4 Bg7 13.Ne5 Nd4
  -+  (-1.78)   Depth: 18   00:09:58  27176kN
3.Bb5 d5 4.e5 Bd7 5.Qe2 g6 6.0-0 Bg7 7.Bxc6 bxc6 8.d4 Nh6 9.dxc5 Qa5 10.Be3 Nf5 11.a3 0-0 12.b4 Qa4
  -+  (-1.82)   Depth: 19   00:20:38  58635kN
3.Bb5 g6 4.0-0 Bg7 5.c3 Nf6 6.d3 0-0 7.h3 d6 8.Qc2 a6 9.Ba4 b5 10.Bb3 Na5 11.Bf4
  -+  (-2.01)   Depth: 20   01:35:15  289mN
3.Bc4 d6 4.0-0 Nf6 5.d3 e6 6.a3 Be7 7.Bb3 0-0 8.h3 Bd7 9.Bd2 Qb6 10.Qe2 Nd4
  -+  (-1.87)   Depth: 20   02:19:26  418mN
3.Bc4 d6 4.0-0 Nf6 5.d3 e6 6.Bd2 Be7 7.Bb3 0-0 8.h3 Bd7 9.a4 Na5 10.Ba2 Qb6 11.Qe1
  -+  (-1.87)   Depth: 21   03:21:14  601mN
3.Bc4 d6 4.0-0 Nf6 5.d3 e6 6.Bd2 Be7 7.Bb3 0-0 8.h3 Bd7 9.a4 Na5 10.Ba2 Qb6 11.Bc3 Nc6
  -+  (-1.86)   Depth: 22   04:19:26  768mN

(, AMD 24.08.2007)

Parent - - By XmikeX (**) Date 2007-08-25 03:37 Edited 2007-08-25 03:45
32 or 64bit Rybka 2.3.2a running on relatively modern hardware in an OTB tournament game vs. ~2000 FIDE

-Straight Chess, no anti-computer tricks-

1) Rybka (white, any pawn) - ~99.999% Rybka wins :)

2) Rybka (white, any knight or bishop) - she wins 25%, draws 70%, loses 5%
3) Rybka (white, any rook) - she wins 10%, draws 60% and loses 30%

4) Rybka (white, queen) - ~99.999% Rybka loses :)

#2 and #3 are wild unsubstantiated guesses (too optimistic?), but I'd be willing to put money on #1 and #4. :)

Go Rybka! Destroy the competition!
Parent - - By lkaufman (*****) Date 2007-08-25 03:56
I'd love to bet against you on #1. We know that Rybka giving pawn odds at 40/2 is somewhere in the grandmaster range, but grandmasters only score about 97% against 2000 rated players, so that would be my expectation for Rybka (assuming a very high contempt factor). Your knight and rook odds estimates seem reasonable to me, but bishop handicap would be a larger one than knight because it breaks the bishop pair.
Parent - - By Roland Rösler (****) Date 2007-08-25 04:48
I´m always astonished about your remarks. You make your decisions about the evaluation of figure constellation by statistics of human games. But human games are full of mistakes and you can´t trust them. Engine play is more than 200 Elo better than the play of the best human. What´s the reason, you can trust in the statistics of human play? Do you believe, human evaluation (or the statistic about human evaluation, what´s the same I think) is the best? You are not in fear, you can adopt the faults of human play (in evaluation)?
Most human chess players believe in the force of the bishop pair. But are they right? If you do it in the evaluation of Rybka, we´ll never know (until another engine comes with negation of human evaluation and is best!).
Last questions: Do you believe, that human evaluation of chess positions is best? Will there be a time, where engines evaluate better? Is only the much better search of the engines (without evident faults) responsible for the better play of the engines?
Parent - - By lkaufman (*****) Date 2007-08-25 05:06
     I would trust statistics from engine play a bit more than statistics from human play; in fact I switched to engine database stats a few months ago, which resulted in some small changes to Rybka. One finding is that engines like major pieces more than humans do (relative to minor pieces).
     The value of the bishop pair is totally clear to me, no matter what database you use or how you refine the search. All good programs give a bonus to the bishop pair (I think), but probably Rybka gives more than most other programs. In my opinion, the bishop pair is worth nearly half a pawn, as I wrote 8 years ago and still believe. There is no disagreement here between engines and grandmasters. If we take away the bishop pair bonus, Rybka will play MUCH worse.
     Currently, human evaluation (top grandmaster level) is still better than engine evaluation, given equal search. I would say current Rybka evaluation is about the same level as my own. So yes, as of now only better search is the reason for engine superiority. But this is changing. Rybka's eval is getting better day by day, and there will come a time (maybe a few years) when it is as good as the best human player.
Parent - - By Roland Rösler (****) Date 2007-08-25 06:09
I´m yery pleased about your answer. First, bishop pair is totally clear for me (and you). I only mentioned it, because Vas alwawys announced, there are no great penalties in Rybka (as in Fritz or other engines). But the other side of penalty is to give much bonus to some figure constellation (like bishop pair).
That you go to engine database is wise, but not without risk. You always have in mind, that engine play in endgame is worse. That major pices are better than in human play may have the reason, that engines can make a mate with rook only, but not with bishop and knight (or with bishop and the false pawn).
But that human eval (top GM level) is best today, I like most. I´m also convinced about this. But why GMs (in evaluation not much worse than top-GMs, but with more mistakes in search) are so bad in freestyle? They can avoid opening traps (search issues) when they play their well known openings and then they can make efforts to win. But they don´t win! For me it´s a mystery, because all of the engines are so bad in endgame play (okay, Rybka 2.3.2a is not so bad as further releases). Is chess really so drawish? I can´t believe it, when I have a look to KRR vs KRN.
Parent - By lkaufman (*****) Date 2007-08-26 04:10
     Online engine play almost always uses tablebases, so bishop and knight mate (and other simplified endings) are no problem. As to why GMs draw a lot in freestyle, one reason might be that they normally play very safely, which might work against humans but just makes draws against Rybkas. To win they have to play risky openings, which they might not be so familiar with. This is just a guess, I don't really know much about freestyle chess.
Parent - By Vasik Rajlich (Silver) Date 2007-08-26 09:26
BTW - I was talking about speculative penalties. I like to have Rybka search 'up' to dynamic things, rather than guess about them and bounce up and down. Obviously, Rybka gives a fairly big penalty to being down a rook :)

Parent - By XmikeX (**) Date 2007-08-25 04:56
Possibly air-tight logic there.

Psychologically, however, I can not join you on that 97%.

.......especially after two GM's were found dead in a river and riddled with what appeared to be piranha bites.

Parent - - By gala.martin (**) Date 2007-08-25 06:13
IMHO Rook odd is by far too optimistic for rybka. She can win or draw only if major blunders happen, which should be the case in less than 70% of the games.
Parent - - By ologist (**) Date 2007-08-26 06:03 Edited 2007-08-26 06:28
Yes, you are right.  I just beat her 3 games in a row, with black every game, with rooks odds in my favor.  It was 5 minute games, and however, she was playing without any kind of book moves.  A rook is just simply too much material to give to almost anybody.  I don't have a USCF rating, but I have beat a GM in a simul once, drew with an IM, and went 60+ moves with Nakamura in a simul, which is lost, and I have been told I am about 1900-2000 USCF.  I will try her tomorrow with knight odds.  With the knight odds, my strategy will not necessarily be to trade pieces, but it will be trying to trade in a fashion that will double her pawns, so in an endgame my extra knight will be more powerful, and I will have good prospects for a pawn break through.  If anyone has some good starting positions for knight odds games, please let me know.  Thanks.

I also think GM analysis is better than a engine.  I think that most people that think engines are better at analysis don't realize that humans make small mistakes over the board analyzing, thinking only a few nodes per seconds, whereas an engines is thinking at 1000s of nodes per second. So in other words, when you factor in time, and/or just how the human is feeling on that day, he could make a mistake, even a very small one over the board in real time analysis, but given that same position just for the basis of analysis, a GM human is better.

I somewhat beg to differ about the bishop pair being near an 1/2 pawn advantage.  I think the strength of the bishop pair is relative to its position.  In some cases a bishop pair would be worth fives pawns, in some they could be just tall pawns.  In general, yeah sure have the bishop pair has been proven to be an advantage, but I think giving them nearly an 1/2 pawn advantage in general is a bit much.  I would think anywhere from 0.20 to 0.30 pawns advantage (in the general case of having the bishop pair) would be a more reasonable estimate. 
Parent - - By XmikeX (**) Date 2007-08-26 06:29
""It was 5 minute games, and however, she was playing without any kind of book moves.""

Parent - - By ologist (**) Date 2007-08-26 06:31

If more time, that would be better for me.  And really what kind of book would really make sense, as you could play just about any kind of opening and still win, unless you are a patzer.
Parent - - By XmikeX (**) Date 2007-08-26 06:42
......and perhaps you are a potential Grandmaster ! ! !  :)
Parent - - By ologist (**) Date 2007-08-26 06:50
Ha.  Thats some very good troll bait there.  Too bad that any decent chess player would know that you probably could not give rook odds to a 1400 player that has good opening knowledge with hopes to win.  Knight odds are a different story because Rybka herself would probably want to trade down pieces, to say a rook pair a two minor pieces. Going to into an endgame just up a knight against Rybka, in this condition, with hopes to win it, would take some serious skills.
Parent - - By XmikeX (**) Date 2007-08-26 07:07
Nonsense.  Proper trolling would involve firebombing your statement -> "1400 player that has good opening knowledge"


I'm just waiting for someone to utter "Grandmaster that has poor endgame skills". :)

Parent - - By Henrik Dinesen (***) Date 2007-08-26 18:34

I think it's time you check your system, it's responding illogically ;)
Parent - - By XmikeX (**) Date 2007-08-26 23:16 Edited 2007-08-26 23:21
Perhaps it is!

Meanwhile, in an alternate universe, I am setting up a 24-game match with certified FIDE1900-2000 human vs. Rykba team** (unlimited hw and preparation), rook odds.

If Rybka holds human to < 30% wins, team gets $100k USD + 10 minutes of open communion with Morphy's or Capablanca's Ghost.

If human >= 50%, he/she gets $1 million USD.

Have faith in human fallibility (and that includes this post)!


**Team decides which time controls are best suited for this match.
Parent - - By lkaufman (*****) Date 2007-08-26 23:39
If we can pick the time control, I'll bet on Rybka giving rook odds to most grandmasters! I pick 1 minute no increment. If I'm allowed to pick 30", then I'll bet on Rybka giving rook odds to any human.
Parent - By XmikeX (**) Date 2007-08-26 23:55
Right on... Now yer talking.. Death to the humans!
Parent - - By Gaмßito (****) Date 2007-08-27 00:41
I think that play games with time controls of 1 minute and no increment is something too ridiculous and really insane. There are not chance for anything, neither to think or create. 
Botvinnik said it clearly, blitz games are just for funny, nothing else. In this game you need really to think, but blitz games do not allow this in absolute. 
Parent - By lkaufman (*****) Date 2007-08-27 05:12
Of course, 1 min chess without increment is silly, it's all about hand-eye coordination and tricks, not chess skill. But normal blitz, especially with increment, does correlate reasonably well with skill in serious chess.
Parent - By Henrik Dinesen (***) Date 2007-08-28 09:51

> Have faith in human fallibility (and that includes this post)!

No doubt :-) It's great thing ;)
Parent - - By lkaufman (*****) Date 2007-08-26 21:39
I myself normally give rook odds to players rated up to about 1600, with good results. And these are not blitz games, either intermediate speed or in some cases 2 or 3 board simuls. But I would admit that 1900 or 2000 should be too high for rook odds from anyone, even Rybka, though there will be some upsets and draws.
Parent - - By Carl Bicknell (*****) Date 2007-08-26 21:42
I have just played a game between Rybka 2.3.2a and Hiarcs 6

Hiarcs had knight odds (b1 knight missing)

Rybka checkmated hiarcs on move 39. game to follow.
Parent - - By lkaufman (*****) Date 2007-08-26 22:45
     A few months ago I played several knight odds matches between Rybka MP (on a quad) and various programs of around the same age or level as Hiarcs 6, like Fritz 5.32 and Crafty 20.14, with fairly even results at various rapid levels. This was quite surprising to me, as these programs were considered to be at least of (lower) grandmaster strength. One explanation is that the programs aren't specifically programmed for handicap play, so they always try to find the "best" move rather than the safe one as a human would do.
Parent - - By Carl Bicknell (*****) Date 2007-08-26 23:54
Just for fun I decided to arrange a short match with NO handicap between Hiarcs 6 and Rybka 2.3.2a (which was on faster hardware) to see how they played out. Some people considered hiarcs 6 to be aggressive but Rybka won EVERY game with a king attack - the games nearly all ended in mate before move 30!

My conclusion is that "aggression" is naturally achieved as the elo of the prog increases. I think it is highly likely that the strongest program will also be the most aggressive, but that's a bit off topic
Parent - - By Uri Blass (*****) Date 2007-08-27 04:34
In this case I believe that I am clearly better than your hiarcs6

I am ready for a match against rybka when every game when rybka fails to force mate before move 30 against me is considered as a win for me.
I admit that I never tries to play seriously against rybka but
I believe that I can win that match.

Parent - By lkaufman (*****) Date 2007-08-27 05:33
Of course you would win that match, but to be fair Hiarcs6 was not told that the goal was to last 31 moves.
Parent - - By Vasik Rajlich (Silver) Date 2007-08-28 11:53
Depends on how you define aggressive. For many people, aggressive means stuff like 1. e4 c5 2. Qh5 or other 'direct attacks', so the most aggressive players will never be the strongest.

Parent - By Carl Bicknell (*****) Date 2007-08-28 11:57
er...well i suppose i mean it in the normal historical sense. maybe the most amount of sucessful king attacks = most aggressive player
Parent - - By XmikeX (**) Date 2007-08-26 22:53 Edited 2007-08-26 23:29
On 2007-08-25 03:37, I write:

3) Rybka (white, any rook) - she wins 10%, draws 60% and loses 30%

#2 and #3 are wild unsubstantiated guesses (too optimistic?)


On  2007-08-25 03:56, you write:

     ""Your knight and rook odds estimates seem reasonable to me, but bishop handicap would be a larger one than knight because it breaks the bishop pair.""

On 2007-08-26 21:39, you write:

     ""But I would admit that 1900 or 2000 should be too high for rook odds from anyone, even Rybka, though there will be some upsets and draws.""


I'll still root for the fish.
Parent - - By lkaufman (*****) Date 2007-08-26 23:36
     Probably your 40% estimated score (counting draws as 1/2) for Rybka is a bit too high for this rating range, maybe 30% is more realistic, and your estimate of draws is clearly too high, as it's not so easy for a rook odds game to end in a draw, but basically I think you were right that Rybka is an underdog but not without chances at rook odds.
Parent - - By XmikeX (**) Date 2007-08-27 10:38
Clearly, TDWRTS/God should share his/her/their 32-man Holybase.

I wonder if there are any games in the Holybase where white ( single piece odds, even rook or queen(!) ), manages to win and/or draw the game without a noticeable blunder by black (despite analysis by GM's and/or top engines).  If chess is a draw, there must be a "blunder" somewhere.. of course. :)
Parent - - By lkaufman (*****) Date 2007-08-27 13:54
I am sure that knight odds is an absolute forced win for Black, regardless of what non-silly defense is chosen against any move. The only blunder needed was the removal of the knight! Of course I cannot prove this, but I am more certain of this than I am of just about anything in life that is unproven.
Parent - - By XmikeX (**) Date 2007-08-27 18:31
""I am sure that knight odds is an absolute forced win for Black, regardless of what non-silly defense is chosen against any move.""

Clearly, we assume Black's play must include at least one blunder in order for White to win or draw, but could there be blunders which defy understanding?

Naturally, if the gap between perfection (the Holybase) and current understanding of the game accommodates at least one humanly/enginely-incomprehensible win/draw for white (minus a piece), the "fallible" FIDE2000 or GM player might as well never play again.

If no such games exist, then all is right with the Universe and a weak-solve of chess (?) should be no more than superengine horizon from one end meeting superextremetablebase from the other. (?)  On this weak solve, I ask the Rybka team to hurry, the years grow shorter and shorter. :)
Parent - By lkaufman (*****) Date 2007-08-27 19:04
Of course, there are countless errors ("blunders" if you will) in chess that defy human understanding, but to lose or draw a game in which you start with an extra knight you must either make a very serious and obvious error, or make many small ones. It is inconceivable to me that one error of the type that defies human understanding could be enough to cost the win. In other words, there may be drawn games at knight odds in which it is impossible to say for sure which was the final error that cost the win, but in such a case grandmasters will have no trouble identifying several errors that made the win more difficult.
Parent - - By lkaufman (*****) Date 2007-08-26 21:48
Regarding starting positions for knight odds, I have Rybka vary between 1e4, 1f4, 1b3, and 1Nf3 as probably the best four tries at the handicap; I made a tiny book for knight odds which I used when Joel Benjamin played some blitz games with it (he did poorly at straight 3' but won rather convincingly at 3'+1"). Of course this was on a fast quad, the latest (private) version, with the book, with contempt at 100, and played when Joel was tired from two serious battles with Rybka earlier. I've seen enough examples now to say that anyone who can hold his own with Rybka in straight 5' chess with this Rybka is near IM strength; if you do it on a sp machine without a book you're probably still close to FM strength.
     As for the bishop pair, I didn't think it was worth half a pawn either, but I have overwhelming statistical evidence from many sources that prove that it is quite close to that (at any rate not below 0.4 pawns). Saying that it depends on the position is like saying that about the value of the Exchange or almost anything in chess; sometimes a knight is better than a queen. But this is a highly reliable general principle, just as usually the Exchange is worth more than a pawn, but not always.
Parent - - By Uri Blass (*****) Date 2007-08-27 04:45
I think that you overestimate the bishop pairs.

1.e4 e5 2.Nf3 Nc6 3.Bb5 a6 4.Bxc6 is theory.
4.Bxc6 may not be the best move but it is at least equality for white.

If bishop pairs are worth 0.5 pawns then 4.Bxc6 is a blunder that make black better because having double pawns is not worth 0.5 pawns.

personally I give the bishop pair bonus based on the number of pawns in the board
I have the following code in my program in the function that evaluates the bishop pair

return ((numbishops[LIGHT]>1)-(numbishops[DARK]>1))*(20+32-2*(numpawns[LIGHT]+numpawns[DARK]));

This mean that the the bishop pair get bonus of 0.2-0.52 pawns.
0.52 pawns when there are no pawns in the board and 0.2 pawns when there are 16 pawns in the board.

Average value of bonus for the bishop pair is clearly less than 0.5 pawn.

Parent - - By lkaufman (*****) Date 2007-08-27 05:26 Edited 2007-08-27 05:43
     After 4Bxc6 dxc6 50-0 f6 (for example, this being the main line), White has huge positional advantages. Black's doubled pawn cripples his majority(after 6d4), he has a knight developed and is castled, while Black has no piece developed, and it is even White's move! Despite all this, the line is considered close to equal. I think this clearly demonstrates my point, not yours! Surely two tempi and a crippled majority are worth half a pawn!
     I am not sure that the bishop pair value depends on the number of pawns, rather it may be that the relative value of bishop to knight varies with the number of pawns, but the bishop pair itself may remain relatively constant in value. At least that is the position I took eight years ago in my article, and I have seen nothing to change my mind on this point, though it is quite possible that both the B/N value and the bishop pair value depend on the number of pawns (I haven't really researched this question much yet).
     Your program would give an average value of 0.32 for the pair if we assume five pawns each (which I think is most common), whereas Rybka is closer to my view. Perhaps we could settle the question by a match between Rybka and your program (just kidding)!
Parent - - By Felix Kling (Gold) Date 2007-08-27 11:02
OK, Larry, I think that your formulation bishoppair=0,5 pawnunits is quite good, but:

1.d4 Nf6 2.Bg5 d5 3.Bxf6 exf6 should then be better for black, since white has no other compensation but the doubled pawn - and white is still better according to the databases. Rybka gives black a slight advantage.

So, why is this line still good for white? Is a doubled c or f pawn much worse than any other doubled pawn? Is having a mobile pawn majority such a great advantage? Is the fact that white can keep the center closed and black can't open it (with the pawns on d4 and d5) enough to make the bishop pair less effective?
Parent - - By ologist (**) Date 2007-08-27 12:45 Edited 2007-08-27 12:52
I played Rybka with knight odds, on a 1.6gz duo core notebook on a 32 bit operating system, at five minutes and no book, and surprisingly I won way more games than I thought I would have. I will admit that she won way more than I did, but I would lose to some tactical trick in the middle game or the opening, and it got some draws because I blundered a pawn or two, or repetition; in other words I never lose to being out played because of the strategy that I was using.  I did not look at the score sheet, but I would say out of about 30 games I would say that I won seven, and drew probably seven or eight.  Rybka's style in a knights odds games is to create 'Marcozy-Bind Type' pawn structures so you will have somewhat backwards pawns so the bishops will have open diagonals; actually I learned something new about the bishop pair as a result of this match.  Anyway, she rarely traded a bishop for a knight, unless it was doubling my pawns, or unless not doing so would lose tempo.   She also rarely fiachettoed a bishop.  I would say to anyone that is daring to play Rybka at knight odds, forget about pinning knights with your bishops, and place the bishop on popular diagonals, e.g. g1-b2 diagonal, as to thwart any tricky attacks, and don't be scared to accept doubled pawns for a trade, especially if your queen is active, and it is the d or e pawn that is being doubled.  I do realize that is not a 'special' version of Rybka for handicap matches, but I feel this is a respectable result for a player of my caliber.
Parent - - By lkaufman (*****) Date 2007-08-27 14:02
     Your result sounds about like I would expect for an average 2200 player, not a 1900-2000. Two questions come to mind. First, do you have any blitz or internet rating (like ICC or Playchess)? Since you are playing blitz, it is your blitz level, not your tournament level, that is relevant. Second, did you do anything to avoid repeat openings? Without a book, this is a big problem unless you provide the variety yourself. Naturally if you just keep improving your play against the same opening you will do better and better with more games. In effect, you would be playing tournament chess vs. Rybka's blitz chess in the opening stages.
Parent - - By ologist (**) Date 2007-08-27 14:35 Edited 2007-08-27 14:42
Yeah I have a blitz rating at Playchess, and really its not relative to my blitz strength.  It hovers around 1800, although right now I am in the 1600s.  I have been as high as 2000 in blitz on the internet.  The only explanation that I can provide for this is that I pay much better OTB, in blitz or long games, than I do on a computer because I feel as though I am sitting right in front of the person and they see me that I have to play the best that I can play, and I focus with all of my energy.  I know you are wondering why this trait does not carry over to computer chess for me, and I really wish I could explain that, but its more or less that OTB I am much better in any case.  In this case with Rybka on my PC, I feel like I was on a mission, per se, and since I had an advantage in my favor and being that I work in the research field, I felt like this was more of an experiment than an actual match, so I put forth more effort than I normally would on a computer, and I also had a specific strategy in mind.  Also, I did not have any book whatsoever, and the opening were repeated, for the most part, but with variation, obviously.  I will say that my all of my wins were not late in the match either, they were pretty much at any point.  One thing though, I had a book loaded, or at least I thought I did, and it did not pull any moves from it; does there need to be a special book or setting for a handicap match to pull moves from a book, or just maybe I did not have it loaded when I thought I did?  If you have a sample book for this kind of match, I would gratefully appreciate a copy of it.
Parent - By lkaufman (*****) Date 2007-08-27 15:02
A normal opening book won't work in a handicap game, as the position on the board will not match any known opening. You need to have a specially made book. I made a small one myself, but to use it I have to remove the knight by entering a sequence of moves rather than using position setup. My advice would be to make your own tiny book with White opening 1e4, 1f4, 1Nf3, and 1b3 as his four choices (I like these four best at b1 knight odds). If you want more variety add in some second move choices against the obvious replies by Black. I think you will find that some of these first moves will give you more trouble than the default.
Parent - - By ologist (**) Date 2007-09-04 13:37
I have play the handicap on a quad, and let me tel you LK you are absolutely right when you say that that coupled with with a little book would make a huge difference.  However, I did beat Rybka once, and drew twice under this condition,  and i probably could have won/drew a few more, becasue I lost on time in a absolutely winning or drawn positions - its hard playing fast chess against Rybka, even without a knight, because pretty mush ANY mistake, no matter how frivolous, will pretty much be punished.  Anyway, I will show this one game here, the one I won, of course :) because I think I did a good job art neutralizing Rybka's bishops and and making it play 'my game', so to speak.

[Event "Friendly Game, 5m + 0s"]
[Site "Engine Room"]
[Date "2007.08.29"]
[Round "?"]
[White "Watchman, Rybka 2.3.2a mp"]
[Black "Ologist"]
[Result "0-1"]
[ECO "A00"]
[WhiteElo "2804"]
[BlackElo "1636"]
[Annotator "Osborne,Rob"]
[PlyCount "112"]
[EventDate "2007.08.30"]
[TimeControl "300"]

{Black_Knight_Odds.ctg,  n\} 1. Nc3 {[%eval 0,0] [%emt 0:00:00]} Nf6 {
[%emt 0:00:03]} 2. Ne4 {[%eval 0,0] [%emt 0:00:00]} Nxe4 {[%emt 0:00:00]} 3.
Nf3 {[%eval 0,0] [%emt 0:00:00]} Nf6 {[%emt 0:00:00]} 4. Ng1 {
[%eval 0,0] [%emt 0:00:00]} Ng8 {[%emt 0:00:00]} 5. e4 {
[%eval 0,0] [%emt 0:00:00]} c5 {[%emt 0:00:09]} 6. Nf3 {
[%eval 0,0] [%emt 0:00:00]} d6 {[%emt 0:00:01]} 7. d4 {
[%eval 0,0] [%emt 0:00:00]} cxd4 {[%emt 0:00:00]} 8. Nxd4 {
[%eval -182,14] [%emt 0:00:07]} Nc6 {(Nf6) [%emt 0:00:01]} 9. Bb5 {
[%eval -188,15] [%emt 0:00:11]} Bd7 {[%emt 0:00:01]} 10. Be3 {
[%eval -199,14] [%emt 0:00:10]} Nxd4 {(Nf6) [%emt 0:00:02]} 11. Bxd7+ {
[%eval -183,12] [%emt 0:00:00]} Qxd7 {[%emt 0:00:01]} 12. Qxd4 {
[%eval -198,13] [%emt 0:00:00]} Nf6 {[%emt 0:00:08]} 13. O-O {
[%eval -209,14] [%emt 0:00:03]} e5 {(Ng4) [%emt 0:00:01]} 14. Qd3 {
[%eval -241,15] [%emt 0:00:07]} Be7 {(Qc6) [%emt 0:00:01]} 15. c4 {
[%eval -209,14] [%emt 0:00:09]} O-O {[%emt 0:00:12]} 16. f3 {
[%eval -210,15] [%emt 0:00:06]} a6 {(Rfc8) [%emt 0:00:19]} 17. a4 {
[%eval -206,13] [%emt 0:00:06]} Rfb8 {(a5) [%emt 0:00:04]} 18. a5 {
[%eval -201,14] [%emt 0:00:09]} Qc6 {(Rc8) [%emt 0:00:06]} 19. Rfd1 {
[%eval -195,13] [%emt 0:00:08]} Nd7 {(h6) [%emt 0:00:03]} 20. b4 {
[%eval -194,13] [%emt 0:00:02]} b5 {(Nf6) [%emt 0:00:06]} 21. axb6 {
[%eval -185,15] [%emt 0:00:04]} Nxb6 {[%emt 0:00:01]} 22. Rdc1 {
[%eval -203,16] [%emt 0:00:08]} Rc8 {[%emt 0:00:21]} 23. Ra2 {
[%eval -212,17] [%emt 0:00:09]} h6 {(a5) [%emt 0:00:17]} 24. g3 {
[%eval -205,13] [%emt 0:00:11]} Bg5 {(a5) [%emt 0:00:08]} 25. Bxg5 {
[%eval -210,15] [%emt 0:00:07]} hxg5 {[%emt 0:00:01]} 26. c5 {
[%eval -227,16] [%emt 0:00:05]} Nd7 {(dxc5) [%emt 0:00:22]} 27. Rd2 {
[%eval -214,16] [%emt 0:00:06]} dxc5 {[%emt 0:00:20]} 28. Qxd7 {
[%eval -247,16] [%emt 0:00:07]} Qxd7 {[%emt 0:00:01]} 29. Rxd7 {
[%eval -274,17] [%emt 0:00:03]} cxb4 {[%emt 0:00:01]} 30. Rxc8+ {
[%eval -320,19] [%emt 0:00:02]} Rxc8 {[%emt 0:00:01]} 31. Rb7 {
[%eval -345,21] [%emt 0:00:05]} a5 {[%emt 0:00:01]} 32. Kg2 {
[%eval -330,20] [%emt 0:00:04]} Rc2+ {[%emt 0:00:02]} 33. Kh3 {
[%eval -340,21] [%emt 0:00:00]} Ra2 {(Rb2) [%emt 0:00:01]} 34. Kg4 {
[%eval -280,19] [%emt 0:00:27]} Ra3 {[%emt 0:00:05]} 35. Rb5 {
[%eval -309,17] [%emt 0:00:16]} b3 {(f6) [%emt 0:00:04]} 36. f4 {
[%eval -479,14] [%emt 0:00:13]} exf4 {(gxf4) [%emt 0:00:03]} 37. gxf4 {
[%eval -479,14] [%emt 0:00:14]} gxf4 {[%emt 0:00:05]} 38. e5 {
[%eval -559,13] [%emt 0:00:06]} a4 {[%emt 0:00:02]} 39. Kxf4 {
[%eval -559,12] [%emt 0:00:07]} Ra2 {[%emt 0:00:03]} 40. e6 {
[%eval -594,10] [%emt 0:00:05]} fxe6 {[%emt 0:00:03]} 41. Rb8+ {
[%eval -576,9] [%emt 0:00:02]} Kf7 {(Kh7) [%emt 0:00:02]} 42. Ke5 {
[%eval -606,9] [%emt 0:00:02]} Re2+ {[%emt 0:00:08]} 43. Kd4 {
[%eval -830,12] [%emt 0:00:00]} b2 {(Rxh2) [%emt 0:00:01]} 44. Kd3 {
[%eval -1062,11] [%emt 0:00:10]} Rxh2 {[%emt 0:00:02]} 45. Rb7+ {
[%eval -1355,7] [%emt 0:00:00]} Kf6 {[%emt 0:00:01]} 46. Rb4 {
[%eval -1829,7] [%emt 0:00:00]} a3 {[%emt 0:00:01]} 47. Rf4+ {
[%eval -2771,8] [%emt 0:00:06]} Ke7 {(Kg5) [%emt 0:00:01]} 48. Rb4 {
[%eval -2873,8] [%emt 0:00:05]} a2 {[%emt 0:00:06]} 49. Rb7+ {
[%eval -32755,9] [%emt 0:00:00]} Kf6 {(Kd6) [%emt 0:00:01]} 50. Kc4 {
[%eval -2788,5] [%emt 0:00:01]} b1=Q {(a1Q) [%emt 0:00:02]} 51. Rb3 {
[%eval -32757,4] [%emt 0:00:00]} Qxb3+ {(Qf1+) [%emt 0:00:02]} 52. Kxb3 {
[%eval -32762,3] [%emt 0:00:00]} a1=Q {[%emt 0:00:01]} 53. Kc4 {
[%eval -32763,3] [%emt 0:00:00]} Qd1 {(Qa4+) [%emt 0:00:01]} 54. Kb5 {
[%eval -32760,3] [%emt 0:00:00]} Rc2 {(Qb3+) [%emt 0:00:01]} 55. Kb6 {
[%eval -32762,3] [%emt 0:00:00]} Qb1+ {(Qd6+) [%emt 0:00:00]} 56. Ka6 {
[%eval -32766,3] [%emt 0:00:00]} Ra2# {(Lag: Av=0.53s, max=4.3s) [%emt 0:00:00]
} 0-1
Up Topic Rybka Support & Discussion / Rybka Discussion / rybka without a knight against humans
1 2 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill