Chess variant tournament: Capablanca and similar

[hgm]

Yes, I understand. The range of strengths is furthermore very large, making it difficult to measure the effect of an improvement: there are only a few games where the results can go either way, and the rest is a certain 1 or a certain 0. In my self-test the score at normal is now 71%, and at Capablanca 77%. The win/loss ratio is in fact very similar, but in normal Chess there are many more draws, pulling the result towards 50%.

I am very satisfied that this is an great improvement. (As it should be, when done without introducing cripling bugs.) But in the end it remains an engine with virtually no evaluation. It does not even have decent piece-square tables. Just a very light attraction towards the center for some piece types (PNBK), and a push bonus for Pawns. (This morning I discovered I had flipped the sign of the centralization bonus in the new version, and this was enough to make it lose every game, though! It could not even win KRNK!)

[enhorning]

I have a new version here https://sites.google.com/site/dshawul/N ... ects=0&d=1
I did not work on the evaluation but only some changes in the search. Bihasa is just too strong for nebiyu right now. Nebiyu has a hard time getting a win out of 10 games. I have to think of cheaper ways to include some king safety terms because that seems to be its major problem.

Okay, I'll use that new version then!

Code: Select all

Cross table, sorted by score percentage, Buchholz, SB

                              Bi Sp Ne TJ Ne Jo Fa Ma Sm TS He Pa Sj Ch Bi Ar
 1. Bihasa v3.2               ## 10 11 01 11 11 11 11 11 11 11 11 11 11 11 11   93%  28.0 (424.0, 373.0)
 2. Spartacus 1.02            01 ## 00 11 10 11 11 11 11 11 11 11 11 11 11 11   87%  26.0 (428.0, 326.5)
 3. NebiyuChess_1.43          00 11 ## 01 10 11 11 11 11 11 11 11 11 11 11 11   87%  26.0 (428.0, 325.5)
 4. TJchess10x8 1.1-x64       10 00 10 ## 1= 11 =1 11 11 11 11 11 11 11 11 11   83%  25.0 (430.0, 306.3)
 5. NebiyuChess_1.44          00 01 01 0= ## 00 11 11 11 10 11 11 11 11 11 11   72%  21.5 (437.0, 247.0)
 6. Joker80.np                00 00 00 00 11 ## 1= 00 11 0= 01 01 11 11 11 11   53%  16.0 (448.0, 161.8)
 7. Fairy-Max 4.8S            00 00 00 =0 00 0= ## == 10 11 01 11 01 01 11 11   47%  14.0 (452.0, 132.5)
 8. Max-Plus 4.8S             00 00 00 00 00 11 == ## 01 01 10 0= =1 11 11 11   47%  14.0 (452.0, 125.8)
 9. Smirf BC-173g-X           00 00 00 00 00 00 01 10 ## 01 11 11 01 =1 11 11   45%  13.5 (453.0, 114.3)
10. TSCPtest                  00 00 00 00 01 1= 00 10 10 ## 10 00 11 11 11 01   42%  12.5 (455.0, 124.5)
11. Heretic 0.3               00 00 00 00 00 10 10 01 00 01 ## =1 00 11 11 11   38%  11.5 (457.0,  94.0)
12. Pair-o-Max 4.8S           00 00 00 00 00 10 00 1= 00 11 =0 ## 10 11 10 11   37%  11.0 (458.0,  96.3)
13. Sjaak 524                 00 00 00 00 00 00 10 =0 10 00 11 01 ## 01 11 11   35%  10.5 (459.0,  84.0)
14. ChessV_WinBoard           00 00 00 00 00 00 10 00 =0 00 00 00 10 ## 11 10   18%   5.5 (469.0,  39.3)
15. BigLion80 2.23x WB        00 00 00 00 00 00 00 00 00 00 00 01 00 00 ## 11   10%   3.0 (474.0,  15.0)
16. ArcBishop80 1.01 WB       00 00 00 00 00 00 00 00 00 10 00 00 00 01 00 ##    7%   2.0 (476.0,  18.0)

... proving again the perils in looking at low number of games at blitz time controls, where 1.44 scored worse than 1.43.

[Daniel Shawul]

Oops this is definately wrong. If it is not much of a problem, I will just keep on working on it for a couple of hours and you can download it right before you start your tournament. Otherwise please use 1.43. I will post here when i have a newer 1.44 uploaded.
Daniel

[Daniel Shawul]

I have updated it now. It will probably fix the regression but there will not be new significant elo gain. It simply needs an evaluation to cope with bihasa.

[enhorning]

Okay, downloaded it and running it now.

This as a particularly unambitious game:

[Event "Computer Chess Game"]
[Site "ENHORNING"]
[Date "2013.03.29"]
[Round "6"]
[White "Pair-o-Max 4.8S"]
[Black "NebiyuChess_1.44"]
[Result "1/2-1/2"]
[TimeControl "40/60"]
[Variant "capablanca"]
[FEN "rbanbkncqr/pppppppppp/10/10/10/10/PPPPPPPPPP/RBANBKNCQR w KQkq - 0 1"]
[SetUp "1"]

{--------------
r b a n b k n c q r
p p p p p p p p p p
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
P P P P P P P P P P
R B A N B K N C Q R
white to play
--------------}
1. Ne3 {+0.23/7} Ne6 {-0.10/14 1.1} 2. c4 {+0.14/8 1.2} Nf6 {-0.10/14 1.4}
3. Bd3 {+0.17/7 0.8} Ad6 {-0.10/13 1.4} 4. Nf5 {-0.06/6 0.8} Ac8
{-0.10/15 1.4} 5. Ne3 {-0.01/8 0.8} Ad6 {+0.00/16 1.2} 6. Nf5 {+0.00/8 1.4}
Ac8 {+0.00/18 1.2} 7. Ne3 {-0.01/9 1.0}
{XBoard adjudication: repetition draw} 1/2-1/2

[Ferdy]

If anybody wants me to use a newer version of their program, or one tuned more for Capa variants, I need to have it within 24 hours - I plan to start the tourney around this time tomorrow morning!

I have sent a new version of Bihasa in your email. I hope it is not too late to enter this version .
Changes are bug fixes in rook and pawn combo attack and also on move ordering.
Also added chancellor and archbishop mobility with king attack and added bonus for own minors
that defend squares attacked by opp chancellor and archbishop to own king.

It gets a nice result against v3.2 scoring 33.5/60 at TC 40 moves in 3 minutes even when debug is on where all asserts are active.

[enhorning]

If anybody wants me to use a newer version of their program, or one tuned more for Capa variants, I need to have it within 24 hours - I plan to start the tourney around this time tomorrow morning!

I have sent a new version of Bihasa in your email. I hope it is not too late to enter this version .

I can use it - but I haven't received the email. Gmail blocks .zip and .exe files - so you need to either pack it up as .rar (which works), or rename it (.zif or .axe or such). Sorry for the trouble.

[Ferdy]

Just sent again, renaming the .exe to .png. I think it gets in because I did not receive a failure notice.

[enhorning]

It has started! Will take a while to run - this many programs and version causes a round robin to have a lot of games... but I expect to post some intermediate results tonight - that should be enough time for a few games for each engine, I hope.

With this many programs, I am not convinced I'll run all programs RR for all the setups - I might run only a couple of setups with everybody in, and then cut to the top programs (and no more than one per author) for the remaining setups, in order to cut down on the number of games.

Depends on how long the games end up being as well - I don't really have a good feeling for how long Capablanca games typically last.

[Ferdy]

It has started! Will take a while to run - this many programs and version causes a round robin to have a lot of games... but I expect to post some intermediate results tonight - that should be enough time for a few games for each engine, I hope.

With this many programs, I am not convinced I'll run all programs RR for all the setups - I might run only a couple of setups with everybody in, and then cut to the top programs (and no more than one per author) for the remaining setups, in order to cut down on the number of games.

Depends on how long the games end up being as well - I don't really have a good feeling for how long Capablanca games typically last.

With the presence of additional 2 powerful pieces on both sides, games generally last shorter than the normal variant.

Winboard is capable of swiss tournament. Why not do a couple of swiss rounds say 9 then take the top 6 for final round-robin games.
However I don't know if a swiss can be done using a pre-defined setup positions.

An alternative is a 4-game pairing on a 9-round swiss that is every pairing is a 4-game match. As rounds progresses engines will be evenly matched.