The Knightmate Challenge 2010

[hgm]

I am preparing a new edition of the 'Knightmate Challenge' tournament. I already opened up my Chess PC for viewing of the games over the internet, and will start a quick test tourney there somewhere this afternoon. (Currently, it still displays the last game of last time, between a strongly time-handicapped Joker and MSCP.)

Jim Ablett has provided many new engines that can play Knightmate. The current list is:

Code: Select all

JokerKM 1.1.14
Lime KM 62
Faile KM 1.4.4
CCCP Knightmate
Fairy-Max 4.8v
Dababba 2.62g
Gerbil-KM 2.0
MSKCP
Vanilla Knightmate 1.4.4
Fimbulwinter KM 5.00

Last time we did this, there were only five engines now we have 10!

The strength range is rather large: the ChessWar ratings of the normal Chess versions of Joker and Fimbulwinter, for instance, differ by about 1000 Elo points. So to make the games more interesting to watch, I will handicap the stronger engines by facing them with time odds.

I will assign the time-odds factors in the following way:
Class A (factor 16): Joker
Class B (factor 8): Lime
Class C (factor 4): Faile, CCCP, Fairy-Max, Dababba, Gerbil
Class D (factor 1): MSKCP, Vanilla, Fimbulwinter

This means that when (e.g.) Lime plays CCCP, it gets only half the time CCCP gets, as its time-odds factor is double that of CCCP's. One of the purpose of the test tourney is to test if these assignments are realistic.

After the test tourney, the handcaps might be adjusted, and the real event will start. This will involve having each participant play each other participant 6 times, at three different time controls: 40/16, 40/24 and 40/36. (Where the time given to the engine with the highest time-odds factor will be reduced accordingly.)

The games can be viewed live (once they start!) at

http://80.100.28.169/gothic/battle.html

[hgm]

The test games have started now (at a nominal TC of 40/8'). Based on the experience from a previos (closed) test tourney that just fiished, I lowered Lime's time-odds factor from 8 to 7, and increased Faile's factor from 4 to 5. That means the number 1, 2 and 3 seeds have factors 16, 7 and 5, while most other engines have 4.

[hgm]

After a small pause, needed to make repairs to Lime (which did not understand time controls of non-integer minutes), the main event has now started.

This will be a ponder-off tournament (2.4Gz Core 2 Duo, 128MB hash) with three cycles of full round-robin: 40/24', 40/16' and 14/36'. So every participant will play 6 games against every other participant, 54 games in total.

The time-odds factors that were finally decided upon are:

Code: Select all

16: Joker
 7: Lime
 5: Faile
 4: Gerbil, CCCP, Dababba, Fairy-Max
 1: VanillaChess, MSKCP, Fimbulwinter

It might be that for VanillaChess this time odds is a bit low, as it seems to get comparatively stronger at longer TC. We will see if it can overtake one of the engines with factor 4...

[hgm]

The first cycle has been completed. The results are more or less what could be expected. Gerbil-KM is doing a little better than I expected; at faster TC it did not nearly as well, so I did not give it an extra time-odds handicap compared to the main field. But so far Gerbil could stay ahead of Faile, which is lightly handicapped compared to it. (It gets 20% less time.) It could just be luck, though, and things can still easily reverse.

Despite the time odds more than double that of the remaining field, JokerKM is leading the pack:

Code: Select all

Cross table, sorted by score percentage, Buchholz, SB

                              Jo Li Ge Fa Fa CC Da Va MS Fi
 1. Joker KM / 16             ## 10 01 == 11 11 11 11 11 11   83%  15.0 (150.0, 111.5)
 2. Lime KM 62 / 7            01 ## 10 11 =1 11 01 1= 11 11   78%  14.0 (152.0, 109.0)
 3. Gerbil-KM 0.2 / 4         10 01 ## =0 11 11 10 =1 11 11   72%  13.0 (154.0,  97.0)
 4. Faile KM 1.4.4 / 5        == 00 =1 ## 01 == 11 1= 11 =1   64%  11.5 (157.0,  86.0)
 5. Fairy-Max 4.8v / 4        00 =0 00 10 ## =1 0= 11 10 11   47%   8.5 (163.0,  56.0)
 6. CCCP new / 4              00 00 00 == =0 ## 11 10 11 11   47%   8.5 (163.0,  48.3)
 7. Dabbabba 2.62 wb / 4      00 10 01 00 1= 00 ## 01 10 11   42%   7.5 (165.0,  54.3)
 8. Vanilla Knightmate 2.6g   00 0= =0 0= 00 01 10 ## 01 11   36%   6.5 (167.0,  43.3)
 9. MSKCP 1.4.4               00 00 00 00 01 00 01 10 ## 00   17%   3.0 (174.0,  22.5)
10. Fimbulwinter KM 5.00      00 00 00 =0 00 00 00 00 11 ##   14%   2.5 (175.0,  11.8)

[hgm]

After cycle #2, the standings are:

Code: Select all

Cross table, sorted by score percentage, Buchholz, SB

                              Joke Lime Gerb Fail Fair Vani Dabb CCCP MSKC Fimb
 1. Joker KM / 16             #### 1010 0111 ==10 1100 1111 1111 1111 1111 1111   81%  29.0 (604.0, 443.5)
 2. Lime KM 62 / 7            0101 #### 1011 1111 =111 1=01 0111 1101 1101 1111   78%  28.0 (608.0, 454.0)
 3. Gerbil-KM 0.2 / 4         1000 0100 #### =00= 1111 =111 1001 11=1 111= 1111   65%  23.5 (626.0, 334.3)
 4. Faile KM 1.4.4 / 5        ==01 0000 =11= #### 0110 1=11 111= ==1= 1110 =111   64%  23.0 (628.0, 351.3)
 5. Fairy-Max 4.8v / 4        0011 =000 0000 1001 #### 1100 0=01 =110 1011 1111   49%  17.5 (650.0, 250.8)
 6. Vanilla Knightmate 2.6g   0000 0=10 =000 0=00 0011 #### 1011 0111 0101 1111   46%  16.5 (654.0, 222.8)
 7. Dabbabba 2.62 wb / 4      0000 1000 0110 000= 1=10 0100 #### 0011 1011 1111   44%  16.0 (656.0, 215.3)
 8. CCCP new / 4              0000 0010 00=0 ==0= =001 1000 1100 #### 11== 1111   40%  14.5 (662.0, 188.5)
 9. MSKCP 1.4.4               0000 0010 000= 0001 0100 1010 0100 00== #### 0001   24%   8.5 (686.0, 147.3)
10. Fimbulwinter KM 5.00      0000 0000 0000 =000 0000 0000 0000 0000 1110 ####   10%   3.5 (706.0,  37.0)

Gerbil was less lucky as in the previous cycle, and lost most of its lead on Faile. It still has half a point advantage, though. In the bottom half of the list, Vanilla Chess performed surprisingly well. It beat most of its competitors, and was even able to score a point against the current number 2, Lime. As a result, it has risen to well above two higher seeded (and thus higher handicapped) engines, CCCP and Dababba.

Next and last cyce will be 40/36'. The battle for first place is very close!

[Jim Ablett]

Joker not having it all it's own way for once.
Vanilla's performance is pleasing
It's gonna be a tight finish.
Will the victory smell of citrus fill the air?

Jim.

[Richard Allbert]

Hi HG,

Just checking, knightmate has a knight in the place of the king and vice versa?

Plus, the FEN will have and "H" sent for the kings, otherwise same as normal chess?

If it's just simple differences, then I'll try to make a Knightmate version of Jabba this weekend.

Regards

Richard

[hgm]

The FEN caharacters WinBoard uses for Knightmate are:

M for the non-royal King (.k.a. Man or Commoner) that replaces the Knights
K for the Royal Knight (depicted as Unicorn)

So the initial setup would be

[d] rmbqkbmr/pppppppp/8/8/8/8/PPPPPPPP/RMBQKBMR w

The Royal Knight castles like a normal King, so the differences with normal Chess from the engine perspective are tiny. (In converting Joker I did encounter n unexpeced problem, though: Joker's move generator first looks for pinned pieces, and then generates moves along the pin ray separately. But, unlike a Knight, a Commoner can have moves along the pin ray!)

If your eval recognizes certain combinations of material as draw (which Joker doesn't), you would have to make changes there too: KRK is a draw! (Just like KBK and KMK.) I have no idea what that does for the Rook value. (Never tested it!)

[Richard Allbert]

*np answered my own Q

[Richard Allbert]

Jabba's playing its first testgame vs Fmax as I write

Bizarre looking!!!

But castling has worked, as have the knight moves.

If it passes two test games, I'll load up the file for download.

I've redone the psqt and valued the King (or M) as 200cp.

Richard