Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka 3 and opposite color bishops
- - By buffos (Silver) Date 2008-08-03 21:24
Has the issue with opposite collored bishops been fixed?
Rybka falls easily in plain opposite bishop endings , evaluating the positions with high score ,  but as usual a pawn is not enough to win such positions.
Parent - - By George Tsavdaris (****) Date 2008-08-03 21:44
Larry said that Rybka 3 has this knowledge.
Parent - By buffos (Silver) Date 2008-08-03 21:46
I know but i ask for some reason :)
Parent - - By lkaufman (*****) Date 2008-08-03 23:16
This is not a simple question. Rybka 3 has a much better understanding about such endgames being drawish compared to Rybka 2.32a, but it is quite difficult for a program to determine whether a pawn up (or two pawns up) bishops of opposite color is a win or not without a very deep search. I'm pretty sure that Rybka 3 would not have allowed the exchange of rooks which led to a draw in Mexico city despite three extra pawns. In general, if Rybka 3 is up a pawn or two in a bishops of opposite color endgame she will consider that she is only up about half that amount or even less, although of course the actual score depends on a great many positional factors. She also now understands that bishops of opposite color tend to make the game more drawish even with other pieces present, though this depends on how many and which pieces are present.
Parent - - By buffos (Silver) Date 2008-08-04 03:58
For example this position

5rk1/5ppp/3p1b2/p7/PpB1P3/7P/1PP2PP1/1R4K1 w - -


Rybka thinks Rd1 Re8 Rd5 Re5 b3 allowing the exchange of Rooks , stepping in an endgame with practically no chances to win.
On the other hand Hiarcs correctly estimates that white should keep Rooks in the game for an advantage.

Also in this position 

8/7p/2pBkp2/2Pp3P/p4PP1/PbP1K3/8/8 b - -


Hiarcs knows its a draw. Rybka on the other hand thinks its +1 which is really disappointing since the position is a very simple draw
Parent - - By Roland Rösler (****) Date 2008-08-04 04:28
On the other hand Hiarcs correctly estimates that white should keep Rooks in the game for an advantage.

And what? I can say that Hiarcs is right here and I´m a patzer! You have to win this! Does Hiarcs win it against R2?
Parent - - By buffos (Silver) Date 2008-08-04 04:42
In this game Hiarcs was black. The problem is big. Rybka loses wins by stepping into op.colour bishop endings.
Parent - - By Roland Rösler (****) Date 2008-08-04 04:48
In this game Hiarcs was black.

What´s the problem? Give Hiarcs white here and let it win against R2 with black. I´m curious ...
Parent - - By buffos (Silver) Date 2008-08-04 04:50
If you don't see a problem ok.
Parent - - By Roland Rösler (****) Date 2008-08-04 05:03
I see many problems in Rybkas endgame play. But I´m not sure, that this position is really won (for other engines)?
Parent - - By buffos (Silver) Date 2008-08-04 05:09
The position does not have to be won.

In the first position white just spoils his chances to play for a win. There is not reason to allow swapping rooks falling in a drawn ending.

The second one is just the tip of the iceberg. White there too spoiled a winning position by exchanging rooks.
Parent - By Roland Rösler (****) Date 2008-08-04 05:22
We only speak about first position! Second position is only bullshit! But I like the sequence, because many people think, tbs are good for endgame play. Okay here every patzer sees what happens (and tbs are not on horizont). The same with blind bishop! R3 eval is decisive at every position! But here, I´m not sure, she really lost a half point. But she loses the potential to win!
Parent - - By lkaufman (*****) Date 2008-08-04 04:38
In both of these positions, Rybka 3 does as you say, she does not score them properly. The extra pawn should be scored as only about half a pawn, but for some reason this is not happening. I strongly suspect a bug here. I'll ask Vas to check the code to see what is wrong. Thank you for the examples.
Parent - - By Bill Forster (**) Date 2008-08-04 05:45
Hi Larry reading this makes me wonder (not for the first time) how you communicate your work on chess knowledge to Vas. I am guessing you have some custom in-house software tool that lets you build encoded rules that are interpreted (or maybe compiled) for runtime evaluation. Have you ever written about this subject in this forum (or elsewhere) ?

While I'm here I'll take the opportunity to thank you for the insightful, scholarly, calm way you interact with the community here. Despite the presence of some odd individuals :-(this seems to be an unfortunate and unintended negative side-effect of the whole internet communications revolution). I for one find this journey you are on hugely interesting and impressive. Have you considered the possibility that you have the best job in the world :-) ?
Parent - By lkaufman (*****) Date 2008-08-04 13:58
Well, it's definitely not the best-paying job in the world, but it is very interesting. My work basically consists of three things: 1. Proposing new terms (or features or time algorithms etc.), which Vas codes. 2. Setting values for the terms (or parameters or anything with a numerical value) in a configurable version of Rybka. 3. Testing.
Parent - - By Maitra (**) Date 2008-08-04 16:27
Hi- I think in this position too Rybka has a problem. It gives a completely wrong eval in a drawn position. Other engines see the draw quite easily.

8/8/p1b5/5k2/8/2K5/P5P1/8 w - - 0 44
Parent - - By buffos (Silver) Date 2008-08-04 17:26
This is a special case. It does not have to do with general knowledge.
Parent - - By Maitra (**) Date 2008-08-07 14:08
Hi- Can you please explain why you are considering this position as a special case. Shouldn't Rybka see the draw in this position?

8/8/p1b5/5k2/8/2K5/P5P1/8 w - - 0 44
Parent - - By buffos (Silver) Date 2008-08-07 14:32
Actually there is a study (Klings i think but i am not sure) with the wrong bishop pawn that black wins.
This position you gave, yes its simple but for example

8/8/2b5/8/p1k5/P7/2K5/8 w - - 0 44


its not clear. I dont remember the position from the study but its like that. You move the king to the other side and then give the bishop to get back in time and take the pawn and win.
This is why i said its different.

The first one is based on general considerations about opposite colored bishops and the other one is more about rules and endgame technique
Parent - By Apples (*) Date 2008-08-07 20:31
4b3/8/8/p7/8/P7/1K6/3k4 w - - 0 9
Parent - - By Maitra (**) Date 2008-08-09 16:22
I agree but other engines see the draw here. Does tbs's here help Rybka?
Parent - - By buffos (Silver) Date 2008-08-09 16:25
but this is not supposed to be a draw.

> Does tbs's here help Rybka.


Yes
Parent - - By Maitra (**) Date 2008-08-09 16:28
This position is a draw. Just analyse this with Fritz or even with Rybka 2.1 mp and it will see the draw.
8/8/p1b5/5k2/8/2K5/P5P1/8 w - - 0 44
Parent - - By buffos (Silver) Date 2008-08-09 19:45
what does fritz say about the position Apples posted before?
Parent - By Maitra (**) Date 2008-08-10 08:06
Fritz says its a draw.
Parent - - By herakleitos (***) Date 2008-08-09 06:09
Rybka 1.2f sees the draw very quickly. The later versions do not.

1.2f is indispensable for rook-pawn situations like this one.
Parent - By Maitra (**) Date 2008-08-09 07:24
Yes this knowledge was removed in the later Rybka versions.
Parent - - By Roland Rösler (****) Date 2008-08-04 04:59
... and which pieces are present.

Maybe you can admit, that rooks are the best pieces for the stronger side with different coloured bishops (assumed you have only one same extra piece on both sides). They are also the most frequent pairings in this type of endgame.
Parent - - By chessorpoker (**) Date 2008-08-04 05:22
are we sure #2 is a draw? f5 Kf7 Kf4 Kg7 g5 Kf7 Bc7 Kg7 Bd8 fg5 Kg5 and white gains entry into the black position... i guess also you can factor in the bishop tempo h5 pawn to whites' kingside pawn advance. f5+ Kf7 Kf4 Bd1 h6 Bc2 Bc7 Bd3 Bd8 Bc2 g5 fg5 Kg5 Bd3 Kf4 Ba6 Ke5... about as far as i can calculate in my head... but its looks like white has made some progress? is this what rybka says?
Parent - - By Roland Rösler (****) Date 2008-08-04 05:29
I have to check later. Up to now I have no look to R2 eval of position 1 or 2. Maybe you are right and position 2 is a winning position for white (but I don´t believe in it).
Parent - By chessorpoker (**) Date 2008-08-04 05:31
for some reason in my tournament games ive won every single opposite bish endgame... strange but drawing them is tougher than it looks
Parent - - By buffos (Silver) Date 2008-08-04 06:04

>are we sure #2 is a draw? f5 Kf7 Kf4 Kg7


f5 Kf7 Kf4 h6 = simply and draw
Parent - - By chessorpoker (**) Date 2008-08-04 06:15
maybe then f5+ Kf7 h6! preventing the g5 blockade and the tempo Bd1 i think its difficult for black here
Parent - - By buffos (Silver) Date 2008-08-04 06:18
Black just plays his bishop. No improvement. On g5 i ll take. No way to make progress
Parent - - By Maitra (**) Date 2008-08-10 11:46 Edited 2008-08-10 11:49
Hi buffos- Here's another example where Rybka's clueless of what to play in these situations:
8/3b4/8/4k1K1/6P1/7P/8/5B2 w - - 0 136


You can refer this post :
http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?tid=6000

Here Rybka 3 wants to play 1.h4?? and even after 1...Bxg4! Rybka 3 gives a positive evals for White. Here's what Fritz thinks:

Analysis by Fritz 11:

1. +-  (#57): 1.Bc4 Kd4 2.h4 Ke3 3.h5 Bc6 4.h6 Be4 5.Bd3 Bc6 6.h7 Bd5 7.h8Q Bf3 8.Qe5+ Kd2 9.Bc4 Bc6 10.Bd5 Bd7 11.Qb2+ Ke3 12.Qc3+ Kf2 13.Qd2+ Kf1 14.Bc4+ Kg1 15.Qxd7 Kf2
2. +-  (17.33): 1.Be2 Be6 2.h4 Bd5 3.h5 Ke6 4.Kg6 Be4+ 5.Kh6 Ke5 6.Bc4 Kd6 7.g5 Kc5 8.g6 Bc6 9.g7 Kd6 10.Bb5 Bf3 11.g8Q Bd1 12.Bc4 Kc7 13.Qg1 Kd6 14.Qxd1+ Kc7 15.Qd6+ Kb7

And when 1...Bxg4 is shown to Fritz:

Analysis by Fritz 11:

1. =  (0.00): 2.Kxg4
2. =  (0.00): 2.Bh3
3. =  (0.00): 2.Be2
4. =  (0.00): 2.Bd3
5. =  (0.00): 2.Bc4
6. =  (0.00): 2.Bb5
7. =  (0.00): 2.Bg2
8. =  (0.00): 2.Ba6
9. =  (0.00): 2.Kg6
10. =  (0.00): 2.Kh6
11. =  (0.00): 2.h5

(Personal 10.08.2008)
Parent - By George Tsavdaris (****) Date 2008-08-10 11:54

>Here Rybka 3 wants to play 1.h4?? and even after 1...Bxg4! Rybka 3 gives a positive evals for White. Here's what Fritz thinks:


I guess all these WITHOUT 3,4,5 piece tablebases(obviously i'm not speaking about 6 piece tbs) because Rybka 3 with 3,4,5 piece tablebases wants to play 1.Bd3 immediately.
Parent - By MoKy (**) Date 2008-08-07 20:05 Edited 2008-08-07 20:10
1.f5+ Kf7 2.h6 seems winning, but d4! (sacrifice freeing d5 square) suffice the draw.
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka 3 and opposite color bishops

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill