Thermopilae 2011 (Live games!)

[hgm]

I am broadcasting again, at http://80.100.28.169/spartan/spartan.html .

Currently there are 40/5' games between ChessV and Fairy-Max there, playing Spartan Chess. After all engines have been checked out, we will have a tourney for that ('Thermopilae 2011').

[Vinvin]

Nice !
Do you have statistic with your engine playing against itself ? To know more accurately wich side is favor (Persians or Spartans).

Thanks,
Vincent.

[hgm]

I could not find a significant imbalance. The results over a few hundred games were as close to 50-50 as you could wish. Definitely closer than in normal Chess, where white has a 53-54% advantage.

This is why I am curious what other engines will do. It can of course never be excluded that there are some deep strategic insights the engines are too stupid to see that would favor one side as soon as you know how to exploit it. Like having a good evaluation of Pawn/Hoplite structures. Even for the Persians the concept of a passer would have to be revised to mean anything. Backward and isolated Pawns probably still apply, though.

[George Tsavdaris]

Nothing to do with the tournament but just now something odd happened with Fairy-Max.
And to report a bug in Winboard at the bottom of the post that seems a bit serious.

Initially and after i get used to the new pieces, Fairy-Max always won against me in Spartan Chess. After a short while the matches are pretty even with me winning equal time as Fairy-Max wins both with Spartans and Persians(black and white). There were many draws also.

But in the following 40/5 (5 minutes for 40 moves repeating) game it had just finished, Fairy-Max had a very easy win (i think, i mean it's obvious right?), we had just moved past 40 moves so it had plenty of time to think, it was reporting after 46...Hc1=W scores of 5.07/12, and after this +5.08/10 and yet, instead of playing correctly(i'm too lazy now to find the critical mistake i think 46...Hc1=W?? is obviously the mistake) to have the very easy win, it allowed itself to get into a repetition trap.

Here it played 46...Hc1=W?? giving me the draw with the perpetual.




A simple 46...Hc1=G would win on the spot or even a promotion to a King would suffice to win.


What do you think? It's a bug or just Fairy-Max was unlucky due to short time control? But i think the time control is no excuse because 3 crucial things are true:
-The win was super easy to see and calculate.
-The time was enough to calculate correctly.
-The perpetual was very easy also.

The game:

[Event "Computer Chess Game"]
[Site "??"]
[Date "2011.01.05"]
[Round "-"]
[White "George"]
[Black "Fairy-Max 4.8Q"]
[Result "1/2-1/2"]
[TimeControl "40/300"]
[Variant "spartan"]
[Annotator "1... +0.98"]

1. e4 Hef6 {+0.98/7 1.1} 2. d4 Lc6 {+0.90/7 2.5} 3. Nc3 We7 {+0.84/7 4} 4.
Nf3 Hdf5 {+0.76/7 6} 5. g3 Hhg6 {+0.62/6 2.3} 6. Bg2 Ce6 {+0.71/7 15} 7.
Nh4 Ld7 {+0.47/7 2.8} 8. O-O Lg8 {+0.71/7 10} 9. d5 Ce5 {+0.66/8 1.0} 10.
f4 Wc5+ {+0.75/8 11} 11. Kh1 Hxf4 {+0.44/8 3} 12. Bxf4 He6 {+0.37/8 23} 13.
a3 Le7 {+0.78/8 5} 14. b4 Wd7 {+0.77/8 8} 15. Nf3 Ld6 {+0.88/9 12} 16. Bxe5
Hxe5 {+1.19/9 7} 17. Nd4 Hc5 {+1.11/9 2.6} 18. Ne6 Ke7 {+0.43/9 2.7} 19.
Nxg7 Cf8 {+0.45/9 4} 20. Bf3 Cf7 {+0.52/9 11} 21. Nh5 Hf5 {+0.37/9 8} 22.
Bg4 Lh7 {+0.47/7 3} 23. Qe2 Hcd4 {+0.52/7 8} 24. Nd1 Lg6 {+0.82/8 2.7} 25.
exf5 Lxf5 {+0.75/9 1.9} 26. Bf3 Lh3 {+0.69/9 7} 27. Rg1 Cf5 {+0.84/9 1.3}
28. g4 Hf4 {+0.73/10 2.0} 29. Bg2 Cxh5 {+0.81/9 5} 30. gxh5 Lg4
{+1.10/10 0.9} 31. Qf2 Lxh5 {+0.98/9 2.4} 32. Re1 Wg4 {+1.31/10 2.4} 33.
Qg3 Lf3+ {+1.25/9 15} 34. Bxf3 Wxf3+ {+1.26/12 1.2} 35. Qxf3 Hxf3
{+1.24/11 1.4} 36. Re3 Hc6 {+1.22/11 14} 37. Rxf3 Gb5 {+1.12/10 3} 38. Nc3
Gc4 {+1.15/9 4} 39. Rg1 Hb5 {+1.19/8 3} 40. Rg8 Kb7 {+1.44/7 8} 41. Rg4
Hxb4 {+1.68/8 4} 42. Ne4 Gxc2 {+2.02/10 1.0} 43. Nxd6 Kxd6 {+2.39/11 0.9}
44. Rxf6 Hbc3 {+3.23/11 6} 45. Rxd6 Hb2 {+4.88/13 4} 46. Rgg6 Hc1=W
{+5.07/11 2.5} 47. Rb6+ Ka7 {+5.08/10 0.8} 48. Ra6+ Kb8 {+4.72/11 0.9} 49.
Rab6+ Kc8 {+4.26/11 16} 50. Rg8+ Kd7 {+4.00/11 2.1} 51. Rg7+ Kc8
{+0.01/12 1.1} 52. Rg8+ Kd7 {+0.00/28 1.1} 53. Rg7+ Kc8 {+0.00/17 0.9} 54.
Rg8+
{XBoard adjudication: repetition draw} 1/2-1/2



And for last but more important a bug in Winboard which probably is related with the one i reported in the other thread about Spartan Chess, about having to change the variant to something else than Spartan and then again to Spartan Chess, in order the graphics for Spartan to be correct:
When i played the game and replayed it later everything worked fine with the black promotion on move 46.

But when i closed Winboard and opened it later via the Spartan shortcut (this is irrelevant and it happens when i open it with normal Winboard also and go to Spartan variant-with the usual way of going to another variant first in order the graphics to load correctly) and then pasted the game, when the promotion occurred in 46th move, the promoted piece, a black Warlord with an Archbishop icon in this case, appeared as a white Pawn!!

And i can't fix that by doing the trick of first loading another variant etc.... When i'm pasting a Spartan Chess game(i didn't checked about other games-i will do shortly) where there is a promotion (black only for white is OK) to any piece(i checked that), the promoted piece either disappears or appears as a white Pawn or as a white King in a negative image and other odd behavior.

[George Tsavdaris]

Great( great because you will be able to spot the bug easier-not that it is great this bug exists )! The bug occurs with your original configuration also.

When you paste** the above game in Winboard, the promotion piece appears incorrectly:


**I say paste only, because the piece in real time, i.e when the game was being played, the piece appeared correctly.

The game:

[Event "Computer Chess Game"]
[Site "??"]
[Date "2011.01.05"]
[Round "-"]
[White "George"]
[Black "Fairy-Max 4.8Q"]
[Result "1/2-1/2"]
[TimeControl "40/300"]
[Variant "spartan"]
[Annotator "1... +0.98"]

1. e4 Hef6 {+0.98/7 1.1} 2. d4 Lc6 {+0.90/7 2.5} 3. Nc3 We7 {+0.84/7 4} 4.
Nf3 Hdf5 {+0.76/7 6} 5. g3 Hhg6 {+0.62/6 2.3} 6. Bg2 Ce6 {+0.71/7 15} 7.
Nh4 Ld7 {+0.47/7 2.8} 8. O-O Lg8 {+0.71/7 10} 9. d5 Ce5 {+0.66/8 1.0} 10.
f4 Wc5+ {+0.75/8 11} 11. Kh1 Hxf4 {+0.44/8 3} 12. Bxf4 He6 {+0.37/8 23} 13.
a3 Le7 {+0.78/8 5} 14. b4 Wd7 {+0.77/8 8} 15. Nf3 Ld6 {+0.88/9 12} 16. Bxe5
Hxe5 {+1.19/9 7} 17. Nd4 Hc5 {+1.11/9 2.6} 18. Ne6 Ke7 {+0.43/9 2.7} 19.
Nxg7 Cf8 {+0.45/9 4} 20. Bf3 Cf7 {+0.52/9 11} 21. Nh5 Hf5 {+0.37/9 8} 22.
Bg4 Lh7 {+0.47/7 3} 23. Qe2 Hcd4 {+0.52/7 8} 24. Nd1 Lg6 {+0.82/8 2.7} 25.
exf5 Lxf5 {+0.75/9 1.9} 26. Bf3 Lh3 {+0.69/9 7} 27. Rg1 Cf5 {+0.84/9 1.3}
28. g4 Hf4 {+0.73/10 2.0} 29. Bg2 Cxh5 {+0.81/9 5} 30. gxh5 Lg4
{+1.10/10 0.9} 31. Qf2 Lxh5 {+0.98/9 2.4} 32. Re1 Wg4 {+1.31/10 2.4} 33.
Qg3 Lf3+ {+1.25/9 15} 34. Bxf3 Wxf3+ {+1.26/12 1.2} 35. Qxf3 Hxf3
{+1.24/11 1.4} 36. Re3 Hc6 {+1.22/11 14} 37. Rxf3 Gb5 {+1.12/10 3} 38. Nc3
Gc4 {+1.15/9 4} 39. Rg1 Hb5 {+1.19/8 3} 40. Rg8 Kb7 {+1.44/7 8} 41. Rg4
Hxb4 {+1.68/8 4} 42. Ne4 Gxc2 {+2.02/10 1.0} 43. Nxd6 Kxd6 {+2.39/11 0.9}
44. Rxf6 Hbc3 {+3.23/11 6} 45. Rxd6 Hb2 {+4.88/13 4} 46. Rgg6 Hc1=W
{+5.07/11 2.5} 47. Rb6+ Ka7 {+5.08/10 0.8} 48. Ra6+ Kb8 {+4.72/11 0.9} 49.
Rab6+ Kc8 {+4.26/11 16} 50. Rg8+ Kd7 {+4.00/11 2.1} 51. Rg7+ Kc8
{+0.01/12 1.1} 52. Rg8+ Kd7 {+0.00/28 1.1} 53. Rg7+ Kc8 {+0.00/17 0.9} 54.
Rg8+
{XBoard adjudication: repetition draw} 1/2-1/2

[hgm]

OK, thanks for spotting this. There are actually several issues here.

That the white Pawn appears was simply a bug in the PGN parser. It was not looking promotion suffixes on piece moves (i.e. with leading upper-case piece designator) that required disambiguation. So the move was treated by the back-end as one with a missing promotion character, which defaults to Queen. But in this variant black has no Queen, so an attempt to find the character for it resulted in a period, as you can see in the move displayed above the board. Apparently the routine that performs the moves interprets the period as a white Pawn (which is encoded by 0 internally).

This problem is fixed now in the installer download.

That Fairy-Max lets itself be tricked into a rep-draw is a completely unrelated problem. And even that is due to a combination of reasons:

1) Fairy-Max does not consider under-promotions in its search. In Spartan Chess all promotions are to Warlord.
2) Fairy-Max has no true repetition check in its search. It uses a work-around that freezes all positions it moves out of at game level in the hash table with score 0. So after a position has been actually played in the game, it will avoid it in the search (when ahead). But it cannot see forced repetitions coming, or plan for them.

You cannot expect too much of a program with an AI of a mere 100 lines of C code. It is really not much more than a collection of clever kludges, replacing a correct analysis by a course simplification that happens to be much smaller but works 90% of the time...

[George Tsavdaris]

1) Fairy-Max does not consider under-promotions in its search. In Spartan Chess all promotions are to Warlord.

Oh, so here i have my reason for the bad play. Fair enough. So i got beat actually.

You cannot expect too much of a program with an AI of a mere 100 lines of C code. It is really not much more than a collection of clever kludges, replacing a correct analysis by a course simplification that happens to be much smaller but works 90% of the time...

I keep forgetting that. About the super extra small size of it.
And i'm currently getting beat by it in Spartan Chess, ugh.