Battle of the Goths 10x8 Championship 2009

[hgm]

Another question: is as "Fairy-Max 4.8 v" really the same engine playing as last year? This time it is earning about 50%.

Yes, it really is the same binary. Last time the fast compile was playing as well. (It was made for me by Denis Mendoza.)

It could be that the shorter time control we are playing this year (25'+0" in stead of 55'+5") works in its favor. Fairy-Max has a poor move ordering compared to other engines, and thus a higher effective branching ratio. But even more important might be that the hsh table is more effective at smaller size of the tree. Fairy-Max has a very primitive hashing scheme (always replace), and the efficiency of its internal iterative deepening is very much dependent on the moves of the previous iteration still being stored. It might even search the best move in the root twice, if the entry gets lost and there is no cutoff the second time.

For Smirf I have the impression that having sudden-death in stead of an increment of 5"/move works to its disadvatage. Smirf allocates the time very aggressively towards the beginning of the game, and I have seen some games where it was in an equal position, but under so much time pressure (e.g. four times less time than the opponent) that it was outsearched, and lost the game. With an increment this does not happen so easily, because Smirf itself does not really expect an increment, and it comes to it as a pleasant surprise.

[smrf]

For Smirf I have the impression that having sudden-death in stead of an increment of 5"/move works to its disadvatage. Smirf allocates the time very aggressively towards the beginning of the game, and I have seen some games where it was in an equal position, but under so much time pressure (e.g. four times less time than the opponent) that it was outsearched, and lost the game. With an increment this does not happen so easily, because Smirf itself does not really expect an increment, and it comes to it as a pleasant surprise.

Well, I made a comparable observation. Maybe I have to adapt the time management a little bit ...

[smrf]

... It could be that the shorter time control we are playing this year (25'+0" in stead of 55'+5") works in its favor. ...

Maybe a follow-up event with a timing equal to that of last year could raise some additional results, which might give a chance to better detect engine progresses during this year.

[George Tsavdaris]

I just started this tournament on my server. Live viewing of the games is possible at:

http://80.100.28.169/gothic/battle.html

Code: Select all

    name                       perc.  pnt games  Buchh    S-B

 1. Joker80 n                   94%  51.5 / 55  (1316.0, 1216.5)
 2. Smirf 1.75t                 73%  40.0 / 55  (1396.5, 820.8)
 3. TJchess10x8 0.121           69%  38.5 / 56  (1460.0, 755.0)
 4. Fairy-Max 4.8 v             50%  28.0 / 56  (1544.0, 500.0)
 5. TSCP Gothic                 48%  26.0 / 54  (1547.0, 534.0)
 6. ChessV 0.94                 40%  22.5 / 56  (1588.0, 329.0)
 7. ArcBishop80 1.00            14%   8.0 / 56  (1704.0, 154.8)
 8. BigLion80 2.23x             12%   6.5 / 54  (1664.0, 102.5)

Thanks for another very interesting tournament.
This new(?) Joker 80 n seems to be a lot stronger than the one last year and it's demolishing its opponents. I wonder if you can set it(when the tournament finishes) online for us to play against it. Or is it available for download?

[hgm]

Well, the tourney is not finished yet! After the 4yh cycle (Modern Carrera), the standings are:

Code: Select all

    name                       perc.  pnt games  Buchh    S-B

 1. Joker80 n                   94%  52.5 / 56  (1372.0, 1271.5)
 2. Smirf 1.75t                 71%  40.0 / 56  (1472.0, 833.8)
 3. TJchess10x8 0.121           69%  38.5 / 56  (1484.0, 768.0)
 4. TSCP Gothic                 50%  28.0 / 56  (1568.0, 547.5)
 5. Fairy-Max 4.8 v             50%  28.0 / 56  (1568.0, 504.0)
 6. ChessV 0.94                 40%  22.5 / 56  (1612.0, 336.0)
 7. ArcBishop80 1.00            14%   8.0 / 56  (1728.0, 157.8)
 8. BigLion80 2.23x             12%   6.5 / 56  (1740.0, 106.5)

ChessV had a very good cycle, performing third, even better than TJchess. This, however, was not enough to bridge the gap with the engines above it, Fairy-Max and TSCP Gothic. The latter two ar engaged in a fierce battle for the 4th place. In the previous cycle, Fairy-Max overtook TSCP for the first time, creating a gap of 3 points. But this time Fairy-Max had a very bad cycle, losing its lead completely, so that the programs now are exactly tied.

Code: Select all

Cross table, sorted by score percentage, Buchholz, SB

                              Joke Smir TJch TSCP Fair Ches ArcB BigL
 1. Joker80 n                 #### 1110 1111 11=1 1111 1111 1101 1111
                              #### 1111 1011 1111 1111 1111 1111 1111   94%  52.5 (1372.0, 1271.5)

 2. Smirf 1.75t               0001 #### 1010 0101 1010 1111 111= 1111
                              0000 #### 1011 1111 0111 11=1 1111 1111   71%  40.0 (1472.0, 833.8)

 3. TJchess10x8 0.121         0000 0101 #### 1=11 =011 1111 1111 1111
                              0100 0100 #### 1=10 1011 1110 1111 1111   69%  38.5 (1484.0, 768.0)

 4. TSCP Gothic               00=0 1010 0=00 #### 11=1 11=0 1111 1101
                              0000 0000 0=01 #### 10=1 0110 1011 0111   50%  28.0 (1568.0, 547.5)

 5. Fairy-Max 4.8 v           0000 0101 =100 00=0 #### 1=11 1111 1111
                              0000 1000 0100 01=0 #### 1100 1011 1111   50%  28.0 (1568.0, 504.0)

 6. ChessV 0.94               0000 0000 0000 00=1 0=00 #### 1111 =111
                              0000 00=0 0001 1001 0011 #### 1111 1=11   40%  22.5 (1612.0, 336.0)

 7. ArcBishop80 1.00          0010 000= 0000 0000 0000 0000 #### =110
                              0000 0000 0000 0100 0100 0000 #### 1100   14%   8.0 (1728.0, 157.8)

 8. BigLion80 2.23x           0000 0000 0000 0010 0000 =000 =001 ####
                              0000 0000 0000 1000 0000 0=00 0011 ####   12%   6.5 (1740.0, 106.5)

Now the 5th and final cycle of the tourney has started, where the engines have to play a mirrored version of the Embassy Chess setup. (Mirrored, so that the King starts on its usual f-file, and normal Capablanca Chess castling rules can be applied.

It seems that first place is pretty much decided in favor of Joker; the battle for 2nd/3rd place between Smirf and TJchess, and 4th/5th place between the earliere-mentioned Fairy-Max and TSCP-G is completely open, however.

[hgm]

Thanks for another very interesting tournament.
This new(?) Joker 80 n seems to be a lot stronger than the one last year and it's demolishing its opponents. I wonder if you can set it(when the tournament finishes) online for us to play against it. Or is it available for download?

The mysterious (and perhaps disappointing) thing is that there is no new Joker80. This is exactly the same binary as I was using last year. I see no reason why playing 25' would make it do better compared to its opponents than at 55'+5". but the only other explanation is that it would have simply been more lucky, and this is even less likely, considering the number of games. I also cannot believe that the opponents would be collectively weaker.

Perhaps it would indeed be interesting to do a follow-up event at 55'+5", as Reinhard suggests, but the problem is that that would take twice as much time as the current event. And in January I wanted to have another tourney first, for Knightmate. There are currently quite a few WinBoard-compatible Knightmate engines that I never tried, and I think it is high time we should give them a chance to show their abilities.

[George Tsavdaris]

Code: Select all

    name                       perc.  pnt games  Buchh    S-B

 1. Joker80 n                   92%  62.5 / 68  (1941.0, 1724.0)
 2. Smirf 1.75t                 71%  47.0 / 66  (2004.0, 1162.3)
 3. TJchess10x8 0.121           71%  48.0 / 68  (2108.0, 1153.5)
 4. TSCP Gothic                 55%  36.0 / 66  (2151.0, 810.5)
 5. Fairy-Max 4.8 v             46%  30.5 / 66  (2246.0, 624.0)
 6. ChessV 0.94                 40%  26.5 / 66  (2307.0, 475.0)
 7. ArcBishop80 1.00            12%   8.0 / 66  (2512.0, 186.3)
 8. BigLion80 2.23x             11%   7.5 / 66  (2508.0, 173.5)

Since Joker80 n is about to win the tournament i decided to have a match with it. I was winning last year against it(meaning i had a >50% score) at the Gothic Chess game at short time controls, and i wanted to play against it in all variants your tournament is running.
Bird, Capablanca, Gothic, Carrera and Embassy.

But the problem is:

•1)I don't know how to make Joker80 play all these variants. OK for Gothic Chess i use for example joker80 /boardWidth=10 /variant="gothic"
But what i should use for Carrera? Also how do i force Winboard (i use version 4.3.15k for some small graphical issues i have with the latest 4.4) to recognize the Carrera variant? Should i put a FEN="...." with the Carrera setup before i open it?

•2)Does Joker80 i have form last year recognize castling in all these 5 variants? I know that for Capablanca and Gothic it does, but what about for others? What should i do to make it recognize it? Start it with Capablanca setup (joker80 /boardWidth=10 /variant="capablanca") and setup a FEN with the position?
But how i do that and setup the right FEN? Just pasting the right FEN and start playing seems fine, but what about a more elegant way?

•3)How do you make the tournament to run by setting the different variants? I mean how do you load each game with the different variant? What parameters are you using?
Also how do you handle the situation that most of the engines(especially TSCP) does not know any other variant and for example don't recognize the castle in all variants except Gothic Chess and Capablanca?

•4)And finally what Chess variants does Joker recognizes with their names? "gothic" and "capablanca" are supported but what others?


Until i find a way to play the other 4 variants here are my 2 games(black/white) against the tournament winner Joker80 in Gothic Chess. Time control was 40/10.

I won the first game with the classic style i used to beat Gothic Vortex, that is in a wonderful style(keep the branching factor huge, keep as many pieces as possible in the board and create a hidden attack) and i lost the second.


[Event "Computer Chess Game"]
[Site "????"]
[Date "2009.12.28"]
[Round "1"]
[White "Tsavdaris"]
[Black "Joker80"]
[Result "1-0"]
[Variant "gothic"]
[TimeControl "40/600"]
[Annotator "1... +0.06"]

1. f4 d5 {+0.06/10} 2. d4 Nc6 {+0.12/10} 3. c3 g6 {+0.02/10} 4. h3 f5
{+0.13/10} 5. Be3 Bf6 {+0.18/10} 6. i3 Nh6 {+0.30/9} 7. g3 Cd6 {+0.72/11}
8. b3 Af7 {+0.91/11} 9. Bf3 g5 {+0.91/11} 10. Ah2 g4 {+1.48/11} 11. Bg2 O-O
{+1.38/10} 12. h4 Qe8 {+1.45/10} 13. j3 Bd7 {+1.43/10} 14. i4 Qf8
{+1.36/10} 15. i5 Ng8 {+1.20/11} 16. j4 Ae8 {+1.18/9} 17. Kf2 Ah5 {+1.56/9}
18. Ch1 e6 {+1.53/9} 19. Nd2 Cb5 {+1.62/9} 20. Rc1 Ca3 {+1.50/9} 21. Bi4
Ag7 {+1.34/11} 22. Cj2 Cxa2 {+2.21/11} 23. j5 Qa3 {+2.21/10} 24. Ng2 Qb2
{+2.21/10} 25. i6 hxi6 {+1.51/11 31} 26. Bg6+ Kj8 {-1.10/12 1:44} 27. jxi6
j6 {-1.10/9 33} 28. Cxj6+ ixj6 {-100.06/12 12} 29. Rxj6+ Ki7
{-100.05/10 0.1} 30. Rj7+ Kh6 {-100.04/8 0.1} 31. Ai4+ Kxi6 {-100.03/6 0.1}
32. Qi1 Ai5 {-100.04/8 2.5} 33. Bh5+ Rxh5 {-100.03/6 0.1} 34. Axh5+ Kxh5
{-100.02/4 0.1} 35. Qxi5+ Kg6 {-100.01/2 0.1} 36. h5#
{White mates} 1-0


[Event "Computer Chess Game"]
[Site "????"]
[Date "2009.12.28"]
[Round "2"]
[White "Joker80"]
[Black "Tsavdaris"]
[Result "1-0"]
[Variant "gothic"]
[TimeControl "40/600"]
[Annotator "1. +0.07"]

1. g4 {+0.07/10} g6 2. h3 {+0.02/10} d5 3. d4 {+0.00/9} c6 4. Bg2 {+0.15/9}
f6 5. Bf4 {+0.26/9} Na6 6. Bg3 {+0.30/9} h6 7. Ah2 {+0.49/9} Ah7 8. Nd2
{+0.37/10} Nc7 9. Nj3 {+0.52/10} Bg7 10. c4 {+0.49/10} e6 11. cxd5
{+0.72/8} cxd5 12. Qc2 {+0.91/10} Na6 13. Nh4 {+0.91/10} f5 14. Qa4
{+0.86/9} Bd7 15. Qb3 {+1.31/9} Nj6 16. Qxb7 {+1.77/8} O-O 17. gxf5
{+1.84/9} gxf5 18. Qxa6 {+1.88/9} f4 19. Bxf4 {+1.84/11} Qxh4 20. Nf3
{+1.80/10} Qe7 21. O-O {+1.89/10} Bf6 22. h4 {+1.96/9} Nh5 23. Bi4
{+2.14/10} Aj6 24. Rc1 {+2.25/9} Cg7 25. Bd6 {+3.09/10} Qe8 26. Rc7
{+3.29/9} Ci6 27. Bj5 {+3.63/10} Cj4 28. Axj4 {+3.85/10} Nxj4 29. Rxa7
{+3.92/10} Rxa7 30. Qxa7 {+3.75/10} e5 31. dxe5 {+3.99/10} Bxi2 32. exf6
{+3.62/12 1:27} Bxh1 33. f7 {+3.51/12 1:25} Qe6 34. Cg1 {+3.13/11 37} Qxd6+
35. Kxh1 {+4.17/12 31} Qf8 36. Cg8 {+5.17/12 31} i6 37. Cxf8 {+4.21/13 31}
Rxf8 38. Qe7 {+4.71/13 31} Ah7 39. Bi4 {+4.94/13 1:09} Rh8 40. Ne5
{+8.14/13 25} h5 41. Bg2 {+8.58/13 19}
1-0

[hgm]

OK, let me try to answer that point by point. But first, I must admit that despite its good performance in this tourney, Joker80 is certainly not unbeatable. It is strategicaly weak, and has many blind spots there. One could clearly be recognized in the game it managed to lose against ArcBishop: for a Human it was immediately clear it let its King be locked in in an unsolvable nd completely fatal way. Fortunately Capablanca variants are mostly tactical.

In the tourney I played all variants (even gothic) as /variant=capablanca, and then played from a setup position. In the automated tournament the tournament manager (PSWBTM) has an option to do this, automatically passing the name of a file with FENs to WinBoard. You could also do that by hand, by adding the option /lpf=start.fen (lpf is short for loadPositionFile), where the file start.fen then contains the FEN. If you do that when using WinBard interactively, it should continue to use the position in that file as start position until you use the New Variant menu to select another (or the same) variant.

An alternative is to copy-paste a FEN with an initial position, after setting the time control. You can do that through a menu item (File -> Paste Position), when the position is on the clipboard as FEN. (You can also use Control-V, but this might not work in the older version you are using, and you might need Alt+Shift+V or something like that.)

Gothic and Capablanca are the only variants Joker80 recognizes, and some other engines only recognize capablanca. In WB protocol the only other names that are defined for this type of Chess are caparandom (for CRC) and janus, but Joker80 cannot do the required castlings.

Because it is really all Capablanca with different start positions, the castling rules are always those of Capablanca Chess. Of course one can suppress castling by feeding WinBoard a FEN without castling rights. But when I played from the Carrera position, I did not do that. I think the game is more interesting with castling, and this is why I called what was actually played "Modern Carrera". WinBoard (and most engines) would also not allow "variant wildcastle" type castling, i.e. with a King from the e-file. This is why I played "Mirrored Embassy" in stead of Embassy.

The /boardWidth=10 option is no longer needed, btw, even in the WB version you are using, provided you set /boardWidth=-1 (which can also be done in the New Variant menu) once and for all. The -1 makes it use the default board width for the variant you are using, and WB knows gothic and capablanca are played on a board with 10 files.

I cannot play a fully automated tournament cycling through different variants. What I in fact did was play five different tournaments (each round-robins with 2 games / pairing), giving the same name for the PGN file to append the games to. The utility to convert PGN to standings and a cross table always uses the entire PGN file, so this in effect merges the tourneys. But to start a new cycle required human intervention, where I would instruct PSWBTM to use a different file with the start position (or alter the FEN in the file it was using before). And I also had to change the html page through which external spectators were following the tourney to use the applicable setup and other info.

As you noticed, some engines needed special treatment, because they do not fully support WinBoard protocol, and in particular do not support any WinBoard commands to set up a position. These were TSCP Gothic and ChessV. ChessV can be programmed to play a different setup through its command line. It must be programmed this way for the variant it plays too, as it ignores the WinBoard /variant command. In stead of using the engine command 'ChessV "Capablanca Chess"' with its complicated nested quoting needed because the variant name contains a space, I always used "ChessV start.SGF", where the start.SGF file in the ChessV folder described the variant as a sub-variant of Capblanca Chess with a different setup. (The ChessV docs describe how to do this.)

TSCP was a harder nut to crack. I had to hack its binary, to alter the initial position. Having started by printing all ASCII strings of 8 or more characters in the binary, I immediately struck pay-dirt, as at adress 0x13120 I found a FEN-string for the gothic setup. So I created 4 new versions of it, by altering this string to the positions I wanted to play, and put all these binaries in a separate folder from the original tscpgothic.exe, so that they would not be able to find the opening book. (I was not sure what they would do with a book for the wrong start posiiton.) In each cycle, I used another binary for TSCP. If you are interested to play TSCP with other positions, this is the required program (hack.c):

Code: Select all

#include <stdio.h>

char a&#91;100&#93;;
char FEN&#91;&#93; = "rnbackqbnr/pppppppppp/10/10/10/10/PPPPPPPPPP/RNBACKQBNR w KQkq -";
main&#40;)
&#123;
  FILE *f, *g;
  char c;
  int i = 0, j = 0, k;
  f = fopen&#40;"tscpgothic.exe", "rb");
  g = fopen&#40;"TSCPembassy.exe", "wb");
  while&#40;&#40;k = fgetc&#40;f&#41;) != EOF&#41; &#123;
    if&#40;j < 0x13120 || j >= 0x13120+64&#41; fputc&#40;k, g&#41;; else fputc&#40;FEN&#91;j-0x13120&#93;, g&#41;;
    j++;
  &#125;
  fclose&#40;f&#41;;
  fclose&#40;g&#41;;
  printf&#40;"total %d bytes", j&#41;;
&#125;

[hgm]

The last round contained a large number of surprises. Joker lost to TJchess twice, the first time because it seemed completely blind to a promotion, suggesting a bug somewhere. TSCP again had a very strong cycle, beating Smirf twice and TJchess once. Smirf lost and drew against Fairy-Max, while Fairy-max lost and drew against ChessV.

Result of all this upheaval is that the battle for second place was decided in favor of TJchess10x8, with a difference of only 1.5 point. TSCP Gothic beat Fairy-Max to the fourth place, with a lead of 4 points.

This led to the final standings:

Code: Select all

    name                       perc.  pnt games  Buchh    S-B

 1. Joker80 n                   92%  64.5 / 70  (2155.0, 1932.0)
 2. TJchess10x8 0.121           70%  49.0 / 70  (2310.0, 1284.3)
 3. Smirf 1.75t                 68%  47.5 / 70  (2325.0, 1264.5)
 4. TSCP Gothic                 54%  38.0 / 70  (2420.0, 955.5)
 5. Fairy-Max 4.8 v             49%  34.0 / 70  (2460.0, 758.8)
 6. ChessV 0.94                 41%  28.5 / 70  (2515.0, 570.8)
 7. ArcBishop80 1.00            14%   9.5 / 70  (2705.0, 233.8)
 8. BigLion80 2.23x             13%   9.0 / 70  (2710.0, 199.5)

Code: Select all

Cross table, sorted by score percentage, Buchholz, SB

                              Joker TJche Smirf TSCP  Fairy Chess ArcBi BigLi
 1. Joker80 n                 ##### 11111 11101 11=11 11111 11111 11011 11111
                              ##### 01100 11111 11111 11111 11111 11111 11111   92%  64.5 (2155.0, 1932.0)

 2. TJchess10x8 0.121         00000 ##### 01010 1=111 =0111 11111 11111 11111
                              10011 ##### 10000 =1010 01111 110=1 11111 11111   70%  49.0 (2310.0, 1284.3)

 3. Smirf 1.75t               00010 10101 ##### 01011 10100 11111 111=1 11111
                              00000 01111 ##### 11100 1110= 1=111 11111 11101   68%  47.5 (2325.0, 1264.5)

 4. TSCP Gothic               00=00 0=000 10100 ##### 11=11 11=00 11111 11010
                              00000 =0101 00011 ##### 0=111 11001 01111 11111   54%  38.0 (2420.0, 955.5)

 5. Fairy-Max 4.8 v           00000 =1000 01011 00=00 ##### 1=111 11111 11111
                              00000 10000 0001= 1=000 ##### 100=0 01111 11111   49%  34.0 (2460.0, 758.8)

 6. ChessV 0.94               00000 00000 00000 00=11 0=000 ##### 11111 =1111
                              00000 001=0 0=000 00110 011=1 ##### 11110 =1111   41%  28.5 (2515.0, 570.8)

 7. ArcBishop80 1.00          00100 00000 000=0 00000 00000 00000 ##### =1101
                              00000 00000 00000 10000 10000 00001 ##### 1000=   14%   9.5 (2705.0, 233.8)

 8. BigLion80 2.23x           00000 00000 00000 00101 00000 =0000 =0010 #####
                              00000 00000 00010 00000 00000 =0000 0111= #####   13%   9.0 (2710.0, 199.5)

[hgm]

The performance ratings from this tournaments are:

Code: Select all

Rank Name                Elo    +    - games score oppo. draws
   1 Joker80 n          2435  125   96    70   92%  1938    1%
   2 TJchess10x8 0.121  2172   81   76    70   70%  1975    6%
   3 Smirf 1.75t        2156   81   77    70   68%  1978    4%
   4 TSCP Gothic        2047   75   75    70   54%  1993    9%
   5 Fairy-Max 4.8 v    1990   74   74    70   49%  2001    9%
   6 ChessV 0.94        1921   73   75    70   41%  2011   10%
   7 ArcBishop80 1.00   1642   88  103    70   14%  2051    4%
   8 BigLion80 2.23x    1636   88  103    70   13%  2052    6%

To obtain them I used BayesElo, with an offset of 2000. There were four losses on time, one by Fairy-Max where it was just to be mated, and three by ChessV after the opponent castled. As castling usually happens in the beginning, the position was about equal in those three cases. But they were all against stronger opponents (Joker, SMIRF, TJchess), so they were not ery disruptive to the ratings of the opponents, which were expected to wn anyway. So I did not bother to delete those results before calculating the rating.