Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / What's a Blind Bishop Position?
- - By awdunno Date 2008-08-02 18:04
Somebody here complained about how terrible Rybka is in "blind bishop positions." What's a blind bishop position?
Parent - - By Arkansaw (***) Date 2008-08-02 18:38
I did a search and found something related to "... the opponent may simply try to win on time in KBP vs K drawn endgame with blind bishop and push a pawn every 50 moves to avoid draw by the...". Maybe blind bishop just means making random moves with the bishop?
Parent - - By Permanent Brain (*****) Date 2008-08-02 18:52 Edited 2008-08-02 18:56
The blind bishop motif means, a pawn or h pawn, and a bishop which is not on the color of the promotion square. This is a draw if the defending king can reach the corner where that square is.

Example:

8/8/2B1b3/8/8/2k5/6PP/5K2 b - - 0 1


Only 1...Bh3!! draws.

I want to point to the fact that an engine's performance depends a lot if tablebases or other endgame databases are used or not, and which ones. My critizism was about if the related knowledge is contained in the engine itself, or not.

From a blitz game another computer chess fan has recently observed on Playchess.com:

8/3b4/8/4k1K1/6P1/7P/8/5B2 w - - 0 136


Here, an engine indicated as Rybka 3 played 136.h4?? which spoils the win: 136...Bxg4 draws due to White's blind bishop.

Another example similar to the first position, but showing that these situations are not always a draw: It depends on the defendig king, if he can reach the corner in time.

8/8/2B1b3/8/8/8/1k4PP/5K2 b - - 0 1


Here, 1...Bh3 does not draw because the bK's march to h8 is stopped (nevertheless it's a good practical "last hope" attempt :-) ).

(You can verify these three positions with 6-piece tablebases, for example on a web page which offers access to them using the FENs.)
Parent - - By Uri Blass (*****) Date 2008-08-02 19:37 Edited 2008-08-02 19:41
Your cannot decide about rybka3 based on blitz game when we do not know what depth it got.

In case of getting only depth 1 it is posible that rybka simply did not search deep enough.
I believe based on reading this forum that rybka3 clearly has the knowledge about the blind bishop if the weaker side has only a king
unlike the impression that you try to create.

Uri

Edit here is analysis by old toga that clearly has the relevant knowledge
You can see that toga needs depth 9 to see that h4 is bad.

New game
8/3b4/8/4k1K1/6P1/7P/8/5B2 w - - 0 1


Analysis by Toga II 1.2.1a:

1.Bf1-e2
  +-  (1.93)   Depth: 1/1   00:00:00
1.Bf1-d3
  +-  (2.16)   Depth: 1/1   00:00:00
1.Bf1-d3 Bd7-c6
  +-  (2.22)   Depth: 2/4   00:00:00
1.h3-h4 Bd7-c6 2.h4-h5 Bc6-d5
  +-  (2.40)   Depth: 2/5   00:00:00
1.h3-h4 Bd7-c6 2.h4-h5 Bc6-d5 3.h5-h6 Bd5-e4
  +-  (2.60)   Depth: 3/7   00:00:00
1.h3-h4 Bd7-c6 2.h4-h5 Ke5-d4 3.h5-h6 Bc6-e4 4.Bf1-b5
  +-  (2.71)   Depth: 4/8   00:00:00
1.h3-h4 Bd7-c6 2.h4-h5 Bc6-e4 3.h5-h6 Ke5-d5 4.Kg5-f4 Kd5-d4
  +-  (2.86)   Depth: 5/11   00:00:00
1.h3-h4 Bd7-c6 2.h4-h5 Bc6-e4 3.h5-h6 Be4-c2 4.Bf1-g2 Bc2-d3 5.Bg2-c6
  +-  (2.81)   Depth: 6/12   00:00:00  6kN
1.h3-h4 Bd7-c6 2.h4-h5 Bc6-d5 3.h5-h6 Bd5-e4 4.Bf1-g2 Be4-c2 5.Bg2-c6 Bc2-d3
  +-  (3.03)   Depth: 7/12   00:00:00  13kN
1.h3-h4 Bd7-c6 2.h4-h5 Bc6-e4 3.h5-h6 Be4-h7 4.Bf1-b5 Bh7-e4 5.Bb5-c4 Be4-c2 6.Bc4-g8
  +-  (2.98)   Depth: 8/18   00:00:00  24kN
1.h3-h4 Bd7xg4 2.Bf1-g2 Bg4-e2 3.Bg2-h3 Be2-f3 4.Bh3-f5 Ke5-d4 5.Bf5-g4 Bf3-d5 6.h4-h5 Kd4-e5
  ²  (0.28)   Depth: 9/18   00:00:01  75kN
1.h3-h4 Bd7xg4 2.Bf1-d3 Bg4-f3 3.Bd3-f5 Bf3-h5 4.Bf5-d7 Bh5-f7 5.Bd7-c6 Ke5-d4 6.Kg5-f4 Bf7-e6
  ²  (0.27)   Depth: 10/18   00:00:01  145kN
1.Bf1-d3 Bd7-c6 2.Kg5-h5 Bc6-d7 3.g4-g5
  ±  (0.73)   Depth: 10/18   00:00:01  148kN
1.Bf1-d3 Ke5-d4 2.Bd3-f5 Bd7-b5 3.Kg5-f4 Kd4-c5 4.h3-h4 Kc5-d6 5.g4-g5 Bb5-e2 6.g5-g6 Kd6-e7 7.Kf4-e5 Ke7-f8
  +-  (3.56)   Depth: 11/22   00:00:01  457kN
1.Bf1-d3 Ke5-d4 2.Bd3-f5 Bd7-a4 3.Kg5-f4 Kd4-c5 4.h3-h4 Kc5-d6 5.g4-g5 Ba4-d1 6.g5-g6 Kd6-e7 7.Kf4-g5 Ke7-f8 8.h4-h5 Bd1-f3
  +-  (3.83)   Depth: 12/24   00:00:01  556kN
1.Bf1-d3 Ke5-d6 2.Bd3-f5 Bd7-b5 3.h3-h4 Kd6-e7 4.h4-h5 Ke7-f7 5.h5-h6 Bb5-e2 6.Bf5-e6+ Kf7-f8 7.Kg5-f4 Be2-d3 8.g4-g5 Kf8-e7 9.Be6-f5 Bd3-c4
  +-  (4.20)   Depth: 13/26   00:00:01  752kN
1.Bf1-d3 Ke5-d6 2.Bd3-f5 Bd7-b5 3.h3-h4 Kd6-e7 4.h4-h5 Ke7-f7 5.h5-h6 Kf7-g8 6.Kg5-f4 Bb5-e2 7.g4-g5 Be2-h5 8.Bf5-e4 Kg8-f7 9.Kf4-e5 Bh5-g6
  +-  (4.54)   Depth: 14/28   00:00:01  981kN
1.Bf1-d3 Ke5-d6 2.Bd3-f5 Bd7-b5 3.h3-h4 Kd6-e7 4.h4-h5 Ke7-f7 5.h5-h6 Bb5-e2 6.Bf5-e6+ Kf7-f8 7.Kg5-f4 Be2-d3 8.Be6-f5 Bd3-e2 9.g4-g5 Be2-h5 10.Bf5-e4 Kf8-f7
  +-  (4.48)   Depth: 15/28   00:00:01  1032kN
1.Bf1-d3 Ke5-d6 2.Bd3-f5 Bd7-b5 3.h3-h4 Kd6-e7 4.h4-h5 Ke7-f7 5.h5-h6 Bb5-c4 6.Kg5-f4 Bc4-e2 7.g4-g5 Be2-h5 8.Kf4-e5 Kf7-g8 9.Bf5-c2 Kg8-f7 10.Bc2-e4 Bh5-g6
  +-  (4.54)   Depth: 16/29   00:00:02  1098kN
1.Bf1-d3 Ke5-d6 2.Bd3-f5 Bd7-b5 3.h3-h4 Kd6-e7 4.h4-h5 Ke7-f7 5.h5-h6 Bb5-c4 6.Kg5-f4 Bc4-e2 7.g4-g5 Be2-h5 8.Kf4-e5 Kf7-g8 9.Bf5-c2 Kg8-f7 10.Bc2-e4 Kf7-f8 11.Ke5-d4
  +-  (4.66)   Depth: 17/30   00:00:02  1318kN
1.Bf1-d3 Ke5-d6 2.Bd3-f5 Bd7-b5 3.h3-h4 Kd6-e7 4.h4-h5 Ke7-f7 5.h5-h6 Bb5-c4 6.Kg5-f4 Bc4-e2 7.g4-g5 Be2-h5 8.Kf4-e5 Kf7-g8 9.Ke5-f6 Bh5-e8 10.Bf5-e4 Be8-f7 11.Kf6-e5 Bf7-c4
  +-  (4.65)   Depth: 18/30   00:00:02  1449kN
1.Bf1-d3 Ke5-d6 2.Bd3-f5 Bd7-b5 3.h3-h4 Kd6-e7 4.h4-h5 Ke7-f7 5.h5-h6 Bb5-c4 6.Kg5-f4 Kf7-f8 7.g4-g5 Bc4-f7 8.Bf5-e4 Kf8-g8 9.Kf4-e5 Bf7-e8 10.Ke5-f6 Be8-h5 11.Be4-f5 Bh5-e8 12.h6-h7+ Kg8-h8
  +-  (4.65)   Depth: 19/32   00:00:03  1754kN
1.Bf1-d3 Ke5-d6 2.Bd3-f5 Bd7-b5 3.h3-h4 Kd6-e7 4.h4-h5 Ke7-f7 5.h5-h6 Bb5-c4 6.Kg5-f4 Kf7-g8 7.g4-g5 Bc4-f7 8.Bf5-e4 Kg8-f8 9.Kf4-e5 Kf8-g8 10.Ke5-f6 Bf7-e8 11.Be4-d3 Be8-f7 12.Kf6-e5 Kg8-f8
  +-  (4.64)   Depth: 20/34   00:00:03  2209kN
1.Bf1-d3 Ke5-d6 2.Bd3-f5 Bd7-b5 3.h3-h4 Kd6-e7 4.h4-h5 Ke7-f7 5.h5-h6 Bb5-c4 6.Kg5-f4 Kf7-g8 7.g4-g5 Bc4-f7 8.Bf5-e4 Kg8-f8 9.Kf4-e5 Kf8-g8 10.Ke5-f6 Bf7-e8 11.Be4-d3 Be8-h5 12.Bd3-f5 Bh5-e8 13.h6-h7+ Kg8-h8
  +-  (4.65)   Depth: 21/34   00:00:04  2684kN
1.Bf1-d3 Ke5-d6 2.Bd3-f5 Bd7-b5 3.h3-h4 Kd6-e7 4.h4-h5 Ke7-f7 5.h5-h6 Bb5-c4 6.Kg5-f4 Kf7-g8 7.g4-g5 Bc4-f7 8.Bf5-e4 Kg8-f8 9.Kf4-e5 Kf8-g8 10.Ke5-f6 Kg8-h8 11.Be4-d3 Kh8-g8 12.Kf6-e5 Kg8-h8 13.Ke5-d4 Kh8-g8
  +-  (4.65)   Depth: 22/34   00:00:04  3451kN
1.Bf1-d3 Ke5-d6 2.Bd3-f5 Bd7-b5 3.h3-h4 Kd6-e7 4.h4-h5 Ke7-f7 5.h5-h6 Bb5-c4 6.Kg5-f4 Kf7-g8 7.g4-g5 Bc4-f7 8.Bf5-e4 Kg8-h8 9.Kf4-e5 Bf7-h5 10.Ke5-d6 Kh8-g8 11.Kd6-e6 Bh5-e8 12.Ke6-f6 Kg8-h8 13.Be4-d3 Kh8-g8 14.h6-h7+ Kg8-h8
  +-  (4.65)   Depth: 23/36   00:00:05  4571kN
1.Bf1-d3 Ke5-d6 2.Bd3-f5 Bd7-b5 3.h3-h4 Kd6-e7 4.h4-h5 Ke7-f7 5.h5-h6 Bb5-c4 6.Kg5-f4 Kf7-g8 7.g4-g5 Bc4-f7 8.Bf5-e4 Kg8-h8 9.Kf4-e5 Bf7-h5 10.Ke5-d6 Kh8-g8 11.Kd6-e6 Bh5-e8 12.Ke6-f6 Kg8-h8 13.Kf6-f5 Kh8-h7 14.Kf5-e5+ Kh7-g8 15.Ke5-d4
  +-  (4.65)   Depth: 24/36   00:00:06  6252kN

(so k,  02.08.2008)
Parent - - By Permanent Brain (*****) Date 2008-08-02 19:46

> Your cannot decide about rybka3 based on blitz game when we do not know what depth it got.


The depth was 10:

135. Kg5 {5.09/11 0} Bd7 {(Bf7) 5.07/20 0}
136. h4 {(Kh6) 5.09/10 0} Bxg4 {(Ke4) 1.36/12 0} 137. Kxg4 {0} Kf6 {0.00/1 0}
138. h5 {0} Kf7 {0.00/1 0} 139. Kg5 {0} Kf8 {0.00/1 0} 140. h6 {0} Kg8 {
0.00/1 0} 141. Bb5 {0} Kh7 {0.00/1 0} 142. Bc4 {
(Lag: Av=0.38s, max=1.1s) 0.00/0 0} 1/2-1/2

We also see from the 137th move that 5-piece tablebases were present, although it is not sure if the white engine accessed them, or the GUI only. The only remaining doubt is if this was really Rybka 3, or which beta version. The PGN had [White "Spearfish, Rybka 3_DEFAULT"]. For the full game score, see where I have it from:

http://f23.parsimony.net/forum50826/messages/186060.htm
Parent - - By lkaufman (*****) Date 2008-08-02 21:41 Edited 2008-08-02 22:22
I just tried it out. Without endgame tablebases, Rybka 3 likes h4 initially but switches off to a good move after just 7 ply. If I then ask her to play Black she immediately spots ...Bxg4 with a draw score even on a one ply search. With five man tablebases, she keeps h4 until ply 12, so this game could be by Rybka 3. I don't know why it takes more ply with tablebases than without. However I would say that the choice of h4 on short searches (with or without tablebases) is related to selectivity; we have to go deep enough for the bishop sacrifice not to be pruned out. Anyway clearly the basic "blind bishop" knowledge is in Rybka 3. I constructed another position where the defender had an extra pawn on the other side of the board, and Rybka 3 found the drawing sacrifice and the draw score in seven plies. So it seems to me that there is no problem at all on this topic, it's just that selectivity can cause the sacrifice to be overlooked if the search is too short, but in any normal situation even a second should be enough to see it on a good machine or maybe a few seconds on 32 bit sp.
Parent - - By George Tsavdaris (****) Date 2008-08-02 22:07

>I just tried it out. Without hash tables, Rybka 3 likes h4 initially but switches off to a good move after just 7 ply. If I then ask her to play Black she immediately spots ...Bxg4 with a draw score even on a one ply search. With five man hash tables, she keeps h4 until ply 12, so this game could be by Rybka 3. I don't know why it takes more ply with hash tables than without. However I would say that the choice of h4 on short searches (with or without hash tables) is related to selectivity;


I'm a bit confused. By saying "hash tables" and "5 man hash tables" do you mean endgame tables(tablebases) and 5 man endgame tablebases?
Or do you mean hash tables(for example 256 MB, 2GB, etc.)?

As far as i know the endgame tables are not hash tables, right?
Parent - By lkaufman (*****) Date 2008-08-02 22:21
Sorry, that was just a "senior moment". I meant endgame tablebases, not hash tables.
Parent - - By Uri Blass (*****) Date 2008-08-02 22:22
note  that it is five piece tablebases and not five man hash tables.
Programs always use hash tables but they can avoid using tablebases(edit I see that it is already mentioned).

I guess that with tablebases  when capture lead to tablebase position rybka choose to do only one of the following:

1)searching the capture and calling tablebases(if the depth is big enough)
2)not searching and calling evaluation.

I guess that rybka has not the option to search the capture and only call the evaluation without calling tablebases and the result is that tablebases can be counter productive at fixed depth.

Maybe Vas chose to do it to avoid problems of having wrong information in the hash and to do things more simple but the result is that I doubt if tablebases help to increase playing strength.

Uri
Parent - By lkaufman (*****) Date 2008-08-02 22:27
I did run some fixed-depth tests on the value of five man TBs in Rybka 3, using SP with reversed openings so the games would be identical until TBs kicked in. The result was +3 Elo with the default "rarely" for TBs, and +8 Elo for "normal". These numbers are in line with other estimates of the value of 5 man TBs, although they do not take into account any slowdown.
Parent - - By Maitra (**) Date 2008-08-10 13:11
Hi-I'd just like to know if the following position qualifies as a blind bishop endgame problem. I think it does becos if Black bishop takes the g2 pawn or even if it doesn't then its a draw as White's king can reach a8 to queen his pawn as the bishop controls the queening square. Please correct me if I'm wrong.

8/8/p1b5/5k2/8/2K5/P5P1/8 w - - 0 44


This position is objectively a draw and other engines know its a draw but apparently Rybka 2.3.2a or Rybka 3 don't see it.
Parent - By Permanent Brain (*****) Date 2008-08-10 15:02

> I'd just like to know if the following position qualifies as a blind bishop endgame problem.


Yes, it is a blind bishop position. Neither side can win here. If we put the bB on the other square color, for example on c7, then Black wins.
Parent - By awdunno Date 2008-08-02 21:35
I don't have 6-man tablebases, but it's interesting to note that stupid old Deep Shredder 10 (running on a much slower computer than the one I have Deep Rybka running on) finds 1...Bh3 immediately. At 24 ply, Deep Rybka 2.3.2a hasn't a clue. After giving Rybka 8 minutes to think, I lost interest.
Up Topic Rybka Support & Discussion / Rybka Discussion / What's a Blind Bishop Position?

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill