chessjoker

[Pippo]

Do it when you want!!

And always thanks, P.

[hgm]

OK, the fixed version can now be downloaded from http://hgm.nubati.net/jokermax.zip . I used the method that defines the Joker in the initial setup as 47 (i.e. with the non-virgin bit set), rather than 15. That a non-virgin piece in the initial setup led to nonsense characters in the FEN that Fairy-Max sent to WinBoard for the initial position was really a bug in itself, that had to be fixed anyway.

Defining extra 'shadow Pawns' that would lack the double-move, and setting the imitation value to those in stead of real Pawns, seemed a less elegant solution. At least, as long as I haven't found a more general solution for asymmetric pieces in Fairy-Max. The current Joker implement would not work in variants where Pawns were not the only asymmetric pieces (e.g. if the Joker would also have to imitate, say, Gold Generals). Because currently the only way to use Gold Generals in Fairy-Max is to define them as different pieces for black and white.

I guess at some point I will change Fairy-Max so that it does not use the piece type to index the list of moves, but the colored piece type. So that corresponding white and black pieces would automatically have different lists of moves. Where the list of the black piece would then be constructed at game initialization from the list of the white type (as written in the fmax.ini file) by inverting all the move steps. (For symmetric pieces this would not make the list any different, of course. but just change the order of moves in the list.) In that design, the imitation value would never have to swap any piece types, but could simply use the moves of the imitated piece of its own color.

[Pippo]

Good sunday, Harm!

OK, the fixed version can now be downloaded from http://hgm.nubati.net/jokermax.zip . I used the method that defines the Joker in the initial setup as 47 (i.e. with the non-virgin bit set), rather than 15. That a non-virgin piece in the initial setup led to nonsense characters in the FEN that Fairy-Max sent to WinBoard for the initial position was really a bug in itself, that had to be fixed anyway.

Thank you so much!

I guess at some point I will change Fairy-Max so that it does not use the piece type to index the list of moves, but the colored piece type. So that corresponding white and black pieces would automatically have different lists of moves. Where the list of the black piece would then be constructed at game initialization from the list of the white type (as written in the fmax.ini file) by inverting all the move steps. (For symmetric pieces this would not make the list any different, of course. but just change the order of moves in the list.) In that design, the imitation value would never have to swap any piece types, but could simply use the moves of the imitated piece of its own color.

In fact I was asking myself: why it moves wrongly if itself knows that is wrong!?

Ok, to test again.
Regards, Pippo

[Pippo]

Dear Harm, your program plays very well and feels comfortable with the jokers (it plays as if it always known them...).
If I think that it's only a first draft...

I absolutely love this new game: chess professionals (which I unfortunately am not) should try it!!

I have not found other bugs. I will continue to try.

Best regards, Pippo

[hgm]

Dear Harm, your program plays very well and feels comfortable with the jokers (it plays as if it always known them...).

That is the advantage of a program that knows nothing: it isvery easy to get it up to its original level when you learn it something new!

The major flaw now should be that it doesn't consider under-promotions.

[Pippo]

Hello Harm,

I have won my first game!! Incredible!

# WinBoard game file -- Wed Apr 02 19:40:11 2014
pc vs. Fairy-Max 4.8joker

[--------------
r n b q k b j n r
p p p p p p p p p
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
P P P P P P P P P
R N B Q K B J N R
white to play
--------------]
1. Ng3 d5
2. e3 f5
3. Nc3 Nc6
4. Bb5 Nf7
5. Bxc6+ bxc6
6. b3 g5
7. Ba3 Be6
8. Qe2 Jg6
9. O-O f4
10. exf4 gxf4
11. Nh5 Ng5
12. h4 Ne4
13. Nxe4 dxe4
14. f3 a5
15. fxe4 Bg4
16. Qc4 Bxh5
17. Jc5 Bi4
18. Rcf1 Bxg2
19. Rxf4 h5
20. Rg1 i5
21. Rxg2 Bh6
22. Jxe7 Qxe7
23. Bxe7 Bxf4
24. Rxg6 Bd6
25. Bxd6 O-O
26. Be7 Rcf8
27. Bxf8 Rg8
28. Rxg8 Kb7
29. hxi5 h4
30. Rh8 h3
31. ixh3 a4
32. Qxa4 Kb6
33. Qb4+ Ka6
34. Be7 Ka7
35. Bd8 c5
36. Qxc5+ Ka6
37. Bxc7 Kb7
38. Qb6#
1-0 (White mates)

Last movements (starting by 24) of Fairy are strange, I suspect for understanding no possibilities... it is so? In fact I repeated the game starting by 24 in "two machines" mode and White wins in very few movements.

I will study deeply the game before 24 to check if Fairy has played "good" or some kind of "performance bug" has occurred.

Saluti, Pippo

[hgm]

Congratulations with your win!

I don't think this is the manifestation of a bug. There seems to be some very deep tactics here, which is way behind its horizon at the momamet it plays 19... h5?. Which turns out to be a decisive blunder, as it leaves its Joker unprotected, so that it can be skewered by 20. Rg1!

I suppose it counted on the Bishop being protected by J-R against Rxg2. But after 21. Rxg2 Jxg2 it needs 10 ply to recognize 22. Jxc6! as very dangerous. At 8 ply it still evaluates this as -2.38 (white POV), and at 9 ply ~0. So when it played 20. i5 at 11 ply it would still have looked OK.

[Pippo]

Hi, Harm

Congratulations with your win!

I don't think this is the manifestation of a bug. There seems to be some very deep tactics here, which is way behind its horizon at the momamet it plays 19... h5?. Which turns out to be a decisive blunder, as it leaves its Joker unprotected, so that it can be skewered by 20. Rg1!

I suppose it counted on the Bishop being protected by J-R against Rxg2. But after 21. Rxg2 Jxg2 it needs 10 ply to recognize 22. Jxc6! as very dangerous. At 8 ply it still evaluates this as -2.38 (white POV), and at 9 ply ~0. So when it played 20. i5 at 11 ply it would still have looked OK.

yeah, those are exactly my conclusions. I have observed that only putting 10 min instead of 5, Fairy has enough deep to notice those dangerous results and play differently. It seems just that it has more difficult to recognize joker's movement than any other piece: as it was predictable.
Probably in this game I have won just to predicte joker's playing better than him...
Is it a symptom that the joker can represent a new weapon for human player against the machine?
Yes, I know, I know: I don't have to make too many illusions...

[hgm]

I don't think that this is something specific for the presence of Joker pieces. It is more a matter of strategic vs tactical thinking, which often occurs in orthodox Chess as well. Fairy-Max King Safety is strongly compromised in a way that is not easy to solve after the Joker appears on c6 (paralyzing the Queen). But there is no penalty in the evaluation for this strategic feature, until it really gets checkmated. So as long as the checkmate is beyond its tactical horizon, it remains blissfully unaware of the trouble it is in. Orthodox pieces can also cause such patterns.

[Pippo]

Ok, but Jxc6 already threats checkmate at next turn: Qf7#. In fact, black joker, in g6, cannot eat the white Queen: if does that, it assigns Q value to white joker making check.

Joker playing is not easy to predicte. Also for humans, sure...