Topic Rybka Support & Discussion / Rybka Discussion / Rybka refuses to resign a hopeless position. Why?
Here's the FEN position that was posted today at Susan Polgar's blog. It's taken from the match between China's Ju Wenjun against Russia's Tatiana Kosintseva. I wanted to see how Rybka would defend...and more importantly I wanted to know if Rybka would see the hopelessness of continuing the match when I played this FEN setup. Alas...that's not the case here.
[Event "[01] Hagen vs Rybka 3 960 1-cpu w32 (24/08/2009)"]
[Site "Local computer"]
[Date "2009.08.24"]
[Round "1"]
[White "Hagen"]
[Black "Rybka 3 960 1-cpu w32"]
[Result "1-0"]
[FEN "r5n1/ppp2R2/4Q1pk/3p2q1/3P4/2PB4/P5P1/6K1 w - - 0 1"]
[SetUp "1"]
[TimeControl "Fixed Depth: 12"]
1. Qh3+ Qh5 2. Rh7+ Kh7 3. Qxh5+ Kg7 4. Qxg6+ Kf8 5. Bf5 Re8 6. Be6 Rxe6 7.
Qxe6 Ne7 8. g4 Ke8 9. g5 c5 10. g6 Kd8 11. g7 b6 12. g8=Q+ Nxg8 13. Qxg8+ Kc7
14. Qxd5 cxd4 15. cxd4 Kc8 16. Qf7 Kd8 17. d5 Kc8 18. d6 Kd8 19. d7 Kc7 20. Qe7
Kb7 21. d8=Q+ Ka6 22. Qa8 Kb5 23. Qaxa7 Kc4 24. Qxb6 Kc3 25. Qec5+ Kd3 26. Qd4+
Ke2 27. Qb3 Ke1 28. Qbd1# 1-0
My question to anybody who looks at this...why didn't Rybka resign this hopeless position before I checkmated her?
[Event "[01] Hagen vs Rybka 3 960 1-cpu w32 (24/08/2009)"]
[Site "Local computer"]
[Date "2009.08.24"]
[Round "1"]
[White "Hagen"]
[Black "Rybka 3 960 1-cpu w32"]
[Result "1-0"]
[FEN "r5n1/ppp2R2/4Q1pk/3p2q1/3P4/2PB4/P5P1/6K1 w - - 0 1"]
[SetUp "1"]
[TimeControl "Fixed Depth: 12"]
1. Qh3+ Qh5 2. Rh7+ Kh7 3. Qxh5+ Kg7 4. Qxg6+ Kf8 5. Bf5 Re8 6. Be6 Rxe6 7.
Qxe6 Ne7 8. g4 Ke8 9. g5 c5 10. g6 Kd8 11. g7 b6 12. g8=Q+ Nxg8 13. Qxg8+ Kc7
14. Qxd5 cxd4 15. cxd4 Kc8 16. Qf7 Kd8 17. d5 Kc8 18. d6 Kd8 19. d7 Kc7 20. Qe7
Kb7 21. d8=Q+ Ka6 22. Qa8 Kb5 23. Qaxa7 Kc4 24. Qxb6 Kc3 25. Qec5+ Kd3 26. Qd4+
Ke2 27. Qb3 Ke1 28. Qbd1# 1-0
My question to anybody who looks at this...why didn't Rybka resign this hopeless position before I checkmated her?
![[fi]](/mwf/v12/flags/fi.png "Finland")
Because Rybka can't resign. UCI engines can't resign.
Many GUIs can resign for the engine, though. It's just a matter of adjusting the settings.
Many GUIs can resign for the engine, though. It's just a matter of adjusting the settings.
Are you saying Aquarium can't help Rybka resign hopeless positions?!
Just that the default is to play until the end.
Check if Aquarium offers a configuration setting for resignation (I would guess it has that).
Arena has that setting for UCI engines, allowing to choose the eval (e.g. -700 or lower) and for how many moves. The Fritz interface offers "never, late, or early" for resigning and for draw agreements.
Arena has that setting for UCI engines, allowing to choose the eval (e.g. -700 or lower) and for how many moves. The Fritz interface offers "never, late, or early" for resigning and for draw agreements.
Maybe the UCI spec could be modified to allow the engine to make inciteful comments such as "Thanks for letting me play this shitty position!".
But first, they'd have to extend the English language to include the word 'inciteful'.
(And they could make it legal to start a sentence with a word like 'and' or 'or'.)
(And they could make it legal to start a sentence with a word like 'and' or 'or'.)
My bad. I didn't realize this option wasn't checked before playing out this position. Thanks. Oh...any news on that Aquarium 3.2.1 patch fix yet?
![[us]](/mwf/v12/flags/us.png "United States")
nope!
![[nz]](/mwf/v12/flags/nz.png "New Zealand")
This post has been educational for me. I am making a little effort to improve my Tarrasch Toy Engine. So just in the last few days I've experienced the whole world of engine versus engine play for the very first time. I must say it is a lot of fun and I finally understand why there are so many enthusiasts for this hobby on this board. Anyway, I have been using Arena and running my new version of the engine (initially Tarrasch 2, then T2, then as a fan of the movie, Arnie) versus the old engine. What frustrated me was the opposite problem to the OP's. I was annoyed that Arena did resign on behalf of my engine(s). I want to see blood! No prisoners! No merciful deaths! Mate only! Part of this is that I want to improve my engine's ability to mop up in winning positions.
I fiddled around with Arena's options, but for some reason decided what I wanted was impossible, and so concluded it was not even worth asking a question here. I obviously didn't look closely enough! In the end I solved my problem by adding an internal cap to both engines - my development versions now never report a score of better or worse than 8 pawns.
I fiddled around with Arena's options, but for some reason decided what I wanted was impossible, and so concluded it was not even worth asking a question here. I obviously didn't look closely enough! In the end I solved my problem by adding an internal cap to both engines - my development versions now never report a score of better or worse than 8 pawns.
bill, are you going to develop Tarrasch toy into an expert (or above) -level engine if possible ? i sort of liked it as a computer foe i could actually beat.
Don't worry. Obviously I am going to try and make Arnie as strong as I possibly can, but the Tarrasch Toy engine will still be available.
To get this thing really strong I have a lot to learn. Right now Arnie is beating up Tarrasch Toy around 8-2 in 30 minute games. So far I've almost doubled the speed of the move generator and improved the time management but that's all. Both engines do a full width (no pruning at all) search with alpha-beta. The evaluation function is fairly simple except that it does an enprise evaluation, as described by David Levy in an old Batsford book. Basically this means that although the engines cannot see very deep, at least they don't foolishly conclude that they are ahead (or behind) in material when they are actually just cutting off in the middle of a sequence of exchanges.
I think the Tarrasch Toy engine is fun to play because it moves quickly and yet the Levy function means that it doesn't do anything really stupid (except in endgames where it's usual problem is not understanding the value of a running passed pawn). It moves quickly because I made a rash decision to cut off the search at 5 ply (three moves from the engine, two from the opponent). Arnie takes a much greater interest in how much time it has, and as it searches it tries to predict how much time deeper search will take. If it decides it can afford the time it uses it. In endings it sometimes now searches to 10 ply, using a minute or two in these 30 minute games (on an ancient test machine). It's not much fun waiting for it to make an obviously forced recapture in an ending while it does a full width 10 ply search. Still, in general the deeper search makes Arnie seem reasonably capable - it still screws up endings with passed pawns - but often it doesn't need to play an ending as it seems to have a penchant for rather vigourous attacks.
But as I say, I have a lot to learn - the big advances will come with a deeper, pruned search tree I am sure.
To get this thing really strong I have a lot to learn. Right now Arnie is beating up Tarrasch Toy around 8-2 in 30 minute games. So far I've almost doubled the speed of the move generator and improved the time management but that's all. Both engines do a full width (no pruning at all) search with alpha-beta. The evaluation function is fairly simple except that it does an enprise evaluation, as described by David Levy in an old Batsford book. Basically this means that although the engines cannot see very deep, at least they don't foolishly conclude that they are ahead (or behind) in material when they are actually just cutting off in the middle of a sequence of exchanges.
I think the Tarrasch Toy engine is fun to play because it moves quickly and yet the Levy function means that it doesn't do anything really stupid (except in endgames where it's usual problem is not understanding the value of a running passed pawn). It moves quickly because I made a rash decision to cut off the search at 5 ply (three moves from the engine, two from the opponent). Arnie takes a much greater interest in how much time it has, and as it searches it tries to predict how much time deeper search will take. If it decides it can afford the time it uses it. In endings it sometimes now searches to 10 ply, using a minute or two in these 30 minute games (on an ancient test machine). It's not much fun waiting for it to make an obviously forced recapture in an ending while it does a full width 10 ply search. Still, in general the deeper search makes Arnie seem reasonably capable - it still screws up endings with passed pawns - but often it doesn't need to play an ending as it seems to have a penchant for rather vigourous attacks.
But as I say, I have a lot to learn - the big advances will come with a deeper, pruned search tree I am sure.
> my development versions now never report a score of better or worse than 8 pawns.
It's adjustable, and also possible to switch off resign in Arena. But it has two similar settings:
1. In "Engines/Manage..." (F11), and there under the UCI tab, as a general setting for UCI engines.
2. In a tournament configuration (F9), under the Options tab, as an option to adjudicate games as losses below a certain evaluation.
Maybe you adjusted or disabled only one of them. But I think your workaround is not required if disable both.
You are quite right, I got one of the options but couldn't find the other. I am glad to find this new information because obviously my workaround is only good for engines I build myself and soon Arnie will be much too good for the Tarrasch Toy engine and he will need to fight some more capable engines.
Topic Rybka Support & Discussion / Rybka Discussion / Rybka refuses to resign a hopeless position. Why?
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill