Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Test Positions / Baffling
- - By h.g.muller (****) [nl] Date 2010-03-22 09:03 Edited 2010-03-22 09:05
The following position might make you rethink everything you thought you knew about piece values:

1q1qk1q1/pppppppp/8/8/8/8/PPPPPPPP/NNN1KNNN w - -


My own engine, with the evaluation especially modified for the purpose, (setting the value of N=500, in stead of the usual 325), can win this position with white against Stockfish, even when it has to face five-fold time odds, so that Stockfish searches 3 ply deeper on average. So I suppose this position is really very badly lost for the Queens.

Can Rybka handle positions like this? Not all engines can; in some there is a limit to the number of pieces of the same type.
Parent - By Vempele (Silver) [fi] Date 2010-03-22 10:20 Edited 2010-03-22 10:25

> Can Rybka handle positions like this?


info depth 13 score cp -1292 time 41934 nodes 2015015 nps 49205 pv a1b3 f7f5

Part of this seems to be caused by a bug: Rybka thinks 3 knights (c1,d1,f1) against 1 queen is about -2! 2N vs Q is -4.
Parent - - By daylen (**) Date 2010-03-22 21:46
Can't even paste it into the Fritz interface. It complains about it not being a valid position.
Parent - - By h.g.muller (****) [nl] Date 2010-03-22 23:13
Well, use a real GUI then! :-D WinBoard has no problems with this position...

The position, btw, might be lost for the Knights after all. Normal engines have grave difficulty to win it, though, against an engine that is especially tuned to play it. The trick is to set the Knight value to 500, if Q=950. If they don't know that, they are reluctant to trade a Queen for two Knights, until it is too late, and they then often have to trade it for only a single Knight or get checkmated... So they score only about 50% from this position. A good opportunity to see Rybka lose a game from a position where it starts at +12! :-)

When the side playing the Queens also knows that a Knight should be valued at 500, six Knights do not stand a chance. Add a seventh Knight (on d1), however, and even that knowledge does not help anymore. Seven Knights beat three Queens very badly. Even in blitz, where the occsional tactical error bungles a game, they score about 80%.
Parent - - By dragon49 (****) [us] Date 2010-03-25 17:18
Fritz is a real gui!  It is probably configured to not allow illegal positions.  This position is quite illegal, since both sides have all 8 pawns, it would not have been possible for White to have obtained the extra knights and for Black to have obtained the extra queens!!
Parent - - By h.g.muller (****) [nl] Date 2010-03-25 18:09
That is not illegal. Merely unreachable from the standard setup. Illegal is when the side not to move is in check.

But the point is, that a GUI should act as a graphical aid to facilitate doing what I want to do. Not to dictate me what I can and cannot do. If there are engines that can handle this position (and Rybka apparently can), and I am interested to analyze this position, a GUI that then says "sorry, I won't allow it" is an extremely poor GUI if you put it mildly, and an incredible piece of cr*p in a more honest qualification. I am sympathetic that some engines could not do it for technical reasons, because engines face efficiency constraints. But for GUIs I see no reason why it would be difficult to display a board with 64 Kings or anything else. The only reason they could not handle it is when they are intentionally sabotaged for this...
Parent - - By Vempele (Silver) [fi] Date 2010-03-25 18:22 Edited 2010-03-25 18:30

> Illegal is when the side not to move is in check.


I guess pawns on the first and last ranks are okay too?
Parent - - By h.g.muller (****) [nl] Date 2010-03-25 19:49 Edited 2010-03-25 19:53
I see no reason why a GUI should refuse that. Although Pawns on the last rank are not really useful. In fact there exists a Chess variant known as Lord Dunsany's game, where the initial position is

rnbqkbnr/pppppppp/8/8/PPPPPPPP/PPPPPPPP/PPPPPPPP/PPPPPPPP w - - 0 1


(Here we are obviously dealing with Cr*ppy forum software, that does refuse to render diagrams without King. So you cannot even discuss Giveaway / Suicide games here... Pawns on the first rank are no problem, however.)

rnbqkbnr/pppppppp/8/8/PPPPPPPP/PPPPPPPP/PPPPPPPP/PPPPKPPP w - - 0 1
Parent - - By dragon49 (****) [us] Date 2010-03-25 20:30
My Dutch Friend:

I need to further our discussion by sticking up for the Chessbase People.  The Fritz Products are designed to work with Fritz Engines and UCI Engines that Play Legal Games Of Chess.  Chess 960 may also be supported.  Somebody was nice enough to write wb2uci and share it with the world, so we can use Winboard engines with Fritz GUI's.  The Gui lets one set up positions that could arise from Legal games Of Chess.  Chessbase promises this and they deliver.  They did not promise to deliver a product that will analyze positions that could not have occurred from a legal game of chess.

Is it really appropriate to call the Fritz GUI Crappy?!
Parent - By Uly (Gold) [mx] Date 2010-03-25 21:19

>> Is it really appropriate to call the Fritz GUI Crappy?!


For him it is crappy, and he gave a reason, I don't see a problem.
Parent - - By h.g.muller (****) [nl] Date 2010-03-25 22:01
Well, I hate pedantic software, and would not want the Fritz GUI if I got it for free. Perhaps the Fritz engine has a limited design scope (and I would not want it either). So perhaps it is a perfect interface for Fritz. But if it restricts what other engines can do, it is not sutable as a general GUI for other engines. Even the WB2UCI adapter cannot repair that. It has intrinsic limitations, besides supporting only a limited selection of engine protocols. In my book a real GUI would support a wide range of engines without restricting their capabilities. This is just a toy made for running Fritz.

But all this distracts from the original point of the thread: 7 Knights seem to beat 3 Queens, and the strongest engines in the world can be made to lose from a position they evaluate as +8 for many moves in a row.
Parent - - By BrandonSi (***) [us] Date 2010-03-26 21:51
I'm not sure I would call requiring a valid piece configuration an intrinsic limitation, but to each their own. The Chessbase GUI is far from perfect, but it plays, and allows analysis of, a valid chess game. I guess I don't need any more than that.
Parent - - By h.g.muller (****) [nl] Date 2010-03-27 07:26
Apparently you do need more than that if you want to analyze the position of this thread, in order to contribute meaningfully...

Posts like "I have nothing to say on the subject, because I only tried it in a GUI that could not handle it, and I suffer from a mind-block that prevents me from using a GUI that can, even when I am given the advice to do so" are not very interesting.
Parent - By BrandonSi (***) [us] Date 2010-03-27 13:48
muller.. It's a chess GUI my friend.. Hardly worth an argument.. Go outside. :)
Parent - By Felix Kling (Gold) [de] Date 2010-03-27 14:07
I have to agree with the other posters. You can't expect a chess GUI or engine to support illegal chess positions. A position is illegal if one of the rules of chess is violated, in this case you have a position that is not possible to reach without violating the rules. An arbiter in a tournament seeing this position would ask the players to go back to the last legal position.

So you can neither blame GUIs nor engines, if they do weird things or don't accept that position. I don't like the Chessbase GUIs that much, but this is no good reason to blame them for.

I'm quite sure you are a clever guy, so I bet you already know that your argumentation is a bit artificial :)
Parent - By h.g.muller (****) [nl] Date 2010-03-27 15:38 Edited 2010-03-27 15:44
It is really a pity that what such a sensational revelation that could have provided deep insight in the matter of piece values degenerates into nitpicking on the intended capabilities of software, and competence in picking or using it.

I don't expect anything of the Fritz GUI (and apparently quite rightly so). I don't have it, I don't want it. All I was saying, is that if it cannot do what is needed, use something else that can, because almost any other GUI in existence has no problem handling this position at all, and there are plenty out there. I even named one, as a service to those who have never heard of anything but Fritz. And most engines can handle the position too. To post here that you fed the position to Fritz, and did not succeed, is not anymore useful (and does not look any smarter) than posting that you fed it to your vacuum cleaner, and it could not find a good move to play. Life is all about what you can do, rather than what you cannot.

Fact is that Fritz has a limitation, because without Fritz, in the bare engine, you can do something that you can no longer do when you run it under Fritz. That is an absolute fact, and not a "to each his own". Fritz does limit what the engine can do. Of course it could be that some people share that limitation, and thus cannot notice it in Fritz. That is a different matter, which says absolutely nothing about Fritz.

A carpenter that uses a screwdriver to hammer in nails, (and fails misarably because of it), is an incompetent carpenter. And advice not to use that screwdriver for this, is good advice. No amount of attestments how good a screwdriver it might be, and how well it woould work if one uses it on screws, can change that.
Parent - - By h.g.muller (****) [nl] Date 2010-03-28 12:06 Edited 2010-03-28 12:11
OK, out of consideration for cybernetically challenged GUIs and their worshippers, I give this other position:

1q1qk1q1/3ppp2/8/8/8/1NN2NN1/3PPP2/2NNKN2 w - -


Seven Knights against three Queens this time. Because of the lack of Pawn cover the Knights had to be re-arranged to defend each other to make the position tactically quiet. Some Knights thus get a free development move, but the Queens also start on open files or diagonals, and I guess this more or less balances out.

Normal engines count themselves at +7 or +8 as black. (3Q = 3 x 9.5 = 28.5; 7N = 7x3 = 21.) Yet they are totally crushed. My own engine, tuned for this, scores 87% in favor of the Knights in self-play (75+, 24=, 1-), most draws occurring because it does not know KNNK is a draw, and too hastily converts a won KNNNPKQ end-game to KNNK. Normal engines do not stand a chance at all against it with the Queens. (Even in the first position I gave, the tuned engine with 6N beats Daydreamer with 3Q with 60%, while it scores >95% in this position for the Queens in self-play. Stockfish is completely slaughtered when it has to play with the Queens.)
Parent - - By daylen (**) Date 2010-03-28 21:11

> My own engine, tuned for this, scores 87% in favor of the Knights in self-play


How does your engine work? Rybka vs. Rybka (15 minutes per side) results in a win for black.
Parent - By Uly (Gold) [mx] Date 2010-03-28 22:30

> How does your engine work?


His engine has a Knight value of 500. You can probably use a Komodo personality to test it.
Parent - - By h.g.muller (****) [nl] Date 2010-03-29 07:03 Edited 2010-03-29 07:06
I set the Knight value to 500, and increased the passer bonus by 50. I am not sure the latter is still relevant with the reduced number of Pawns, but with 8 Pawns creating a passer, and then pushing it until white had to give a Knight for it, was the dominant source of wins for black.

I put a version of the modified engine for download at http://hgm.nubati.net/QN.exe . (This is the bare executable; I did not even bother to zip it, as it is only 30KB. It can run anywhere, it does not need or create any files.) The modifications of course wreck its ability to play a normal game of Chess quite badly. It randomizes its moves, so no two games will be the same, even if you start from the same position.
Parent - - By dragon49 (****) [us] Date 2010-04-14 03:30
I'll take a look at it, play some games.

Felix - Help me with thread subscription options.  It appears that I am emailed and notified on the main forum page of replies to my posts.  Is it possible for me to automatically get emailed and notified for posts of all threads that I participate in, or maybe just emailed.  I had missed this thread continuation.

Thanks
Parent - - By h.g.muller (****) [nl] Date 2010-04-14 06:55 Edited 2010-04-14 06:57
OK, tat would be interesting. I tested with Crafty, Stockfish, Daydreamer and Gaviota. Most of them are able to win this, when playing the 7 Knights, against QN.exe as well as against eah other. Only Crafty is especialy inept at it: it tries to trade 3 Knights for a single Q, after which it resigns (and justly so, as 2Q vs 4N is usually an easy win for the Queens).

I could not think of a beter strategy for the side with the Queens. It is confronted wit an interesting dilemma: on the one hand it is favorable trading Q vs 2N: it relieves the pressure, as the 6th and 7th Knight are far more dangerous than the 3rd Queen. On the other hand, this violates the rules that one should not trade when behind, as it means progress for the opponent to convert to an easily won end-game. Q vs 2N is not on a schedule towards a draw here, as the opponent will be left with one Knight, which in the presence of Pawns woud be an easy win. The only chance for the Queens it to try to force trading of all Pawns. But white will decline trading of Pawns when it can also gain them without compensation.
Parent - - By dragon49 (****) [us] Date 2010-04-14 12:53
Last night, I had joker114w play jabba 1.0 2 games, changing sides (g-5) from this position.  Sorry to have to report that joker lost both games.  I'll attach then later, if it has any value to you.  I played the games with the Rybka Aquarium GUI pasting in the fen string for the starting position.  An odd thing happened along the way:  I had not used Joker in Aquarium, so I had to set up the engine.  I choose winboard as the type.  This did not work, as joker did not move, or show any analysis.  I then deleted the engine and re added it as a UCI engine, pointing to my existing wb2uci setup that works with the Fritz interface.  This worked!  I thought you would find that amusing.
Parent - By Vempele (Silver) [fi] Date 2010-04-14 13:07

> I choose winboard as the type.  This did not work, as joker did not move, or show any analysis.


Strange. Joker has no problems playing here (though it refuses to do infinite analysis).
Parent - By h.g.muller (****) [nl] Date 2010-04-14 14:36
Joker does not support the analyze command, so in that respect its behavior is not mysterious.

I am not sure Joker 1.1.14 even supports setting up a position. I think I implemented that first in Joker80, which was a Joker 1.1.14 derivative, because there I really needed it to play different opening setups of 10x8 Chess. Once I had the code, I ported that to the 8x8 version as well, but I am not sure I ever put a binary with that patch on my website. I always tested Joker in Nunn matches in PGN form, where the moves are fed to it in force mode, rather than  position. Upto version 1.1.14 (e) Joker was a WB v1 engines that neither supported setboard nor edit. I could not play the position with the Joker version I use in tournaments, because that is still a 1.1.14 (e) PGO compile made for me by Denis Mendoza, which is faster than the compiles I can make of 1.1.14 (g).
Up Topic The Rybka Lounge / Test Positions / Baffling

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill