Crazyhouse tournaments and rating list

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

User avatar
hgm
Posts: 27808
Joined: Fri Mar 10, 2006 10:06 am
Location: Amsterdam
Full name: H G Muller

Re: Crazyhouse tournaments and rating list

Post by hgm »

Sunsetter 7g sometimes also does weird things. It can show a mate score (move 21 and 22), and then still lose: :!:

Code: Select all

[Event "Computer Chess Game"]
[Site "MAKRO-PC"]
[Date "2016.11.02"]
[Round "22"]
[White "CrazyWa 0.0.9qs9"]
[Black "sunsetter"]
[Result "1-0"]
[TimeControl "40/60"]
[Variant "crazyhouse"]

1. Nh3 {+0.24/10} e5 {+0.28/13 1.5} 2. e4 {-0.13/8 0.9} d6 {+0.26/12 1.5}
3. Qf3 {+0.29/9 2.3} Nf6 {+0.42/11 1.5} 4. Bc4 {+0.36/8 0.9} Bxh3
{+0.76/11 1.5} 5. Qxh3 {+1.07/7 0.9} N@f4 {+0.41/12 6} 6. Qf3 {+2.05/8 2.8}
d5 {-0.13/11 1.4} 7. exd5 {+3.26/8 3} N6xd5 {-0.23/10 1.4} 8. P@g3
{+4.25/7 1.6} P@g4 {-0.18/11 1.4} 9. Qe4 {+3.97/8 2.1} Nxg2+ {-0.25/11 2.6}
10. Qxg2 {+4.61/8 1.6} P@f3 {-0.32/11 1.3} 11. Qf1 {+4.32/7 1.1} Nb4
{-0.42/10 1.3} 12. Bxf7+ {+4.99/6 1.3} Kxf7 {+0.32/9 1.3} 13. P@e6+
{+1.88/6 2.7} Kxe6 {+1.08/9 1.2} 14. Qc4+ {+2.19/6 1.2} B@d5 {+1.85/10 1.2}
15. Qxg4+ {+2.20/6 1.9} P@f5 {+1.66/9 1.2} 16. N@g5+ {+0.04/6 1.0} Qxg5
{+2.23/10 2.4} 17. Qxg5 {+2.75/7 1.2} Nxc2+ {+2.57/10 2.3} 18. Kd1
{+2.74/7 1.2} P@e2+ {+2.38/9 1.2} 19. Kxc2 {+0.08/9 1.6} N@d4+
{+2.23/9 2.2} 20. Kd3 {+2.43/8 1.5} Nbc6 {+3.88/9 2.2} 21. B@a4
{+4.93/5 2.3} Be4+ {+299.20/10 1.7} 22. Ke3 {+5.65/7 2.9} Nc2+
{+299.60/33 0.1} 23. Bxc2 {+9.27/8 2.0} Bc5+ {+299.70/33 0.1} 24. d4
{+10.41/6 0.8} exd4+ {+2.11/11 4} 25. Kd2 {+9.25/6 0.8} P@e3+
{+2.16/10 0.4} 26. fxe3 {+3.54/7 1.6} dxe3+ {+2.11/9 0.4} 27. Kc3
{+4.75/7 2.3} Bd4+ 28. Kb3 {+7.13/6 0.6} P@c4+ {+2.53/9 1.4} 29. Ka3
{+9.82/8 2.8} Bc5+ 30. b4 {+8.80/6 0.6} Bxb4+ {+1.89/9 0.6} 31. Kb2
{+8.98/7 0.9} P@a3+ 32. Nxa3 {+9.75/6 2.5} Bxa3+ 33. Kxa3 {+5.58/6 0.5}
N@b5+ {+0.65/8 0.3} 34. Kb2 {+10.38/7 2.9} c3+ {+0.20/8 0.3} 35. Kb1
{+10.37/6 1.8} Bxc2+ {-0.98/8 0.6} 36. Kxc2 {+10.37/5 0.2} B@e4+ 37. Kb3
{+11.59/5 0.2} Ncd4+ {-5.21/8 0.5} 38. Kc4 {+16.59/5 0.5} Bd5+ 39. Kd3
{+16.60/5 0.2} Be4+ {-10.97/9 3} 40. Kxe3 {+21.15/5 0.2} Rae8 41. B@c4+
{+22.77/3 2.6} Kd7 {-28.83/9 25} 42. Qxg7+ {+28.20/3 1.5} Kc6 43. B@d7+
{+29.10/3 1.2} Kb6 44. Qxd4+ {+1000.05/5 1.5} Ka5 {-299.50/5 0.1} 45. N@b3+
{+1000.02/8 0.1} Ka6 46. Bdxb5# {+1000.01/3 0.1}
{Xboard adjudication: Checkmate} 1-0
BTW, I just uploaded CrazyWa 0.0.9. This now also implement repetition detection. I also fixed a bug in the LMR, that it sometimes reduced directly into QS, because the limiting of the reduction did not always work. But fixig the bug made it weaker, so I figured I should let the LMR kick in at lower depth already. This version does that, and although it is too early to have a statistically significant conclusion, this put it in the lead in my 40/1 test matches against both Imortal 1.0 and Sunsetter 7g after 25 games each. (While before the bug fix it was doing at best 36% aganist Sunsetter 7g.)

This version still clears the hash table before every move. I wanted to do that during development to make errors reproducable. (Unfortunately this did not help, though.) It does crash a lot less often than previous versions; I located one cause of crashes (the randomization during the first 5 moves), and fixed another suspect (execution of the check-detection code after null move).
Top
Carlos777
Posts: 1736
Joined: Sun Dec 13, 2009 6:09 pm

Re: Crazyhouse tournaments and rating list

Post by Carlos777 »

hgm wrote:Sunsetter 7g sometimes also does weird things. It can show a mate score (move 21 and 22), and then still lose: :!:
Interesting. The SF Variant's author said it was caused by hash collisions and it is fixed in the new version.

Same conditions (also same SF version as previous match)

Stockfish 011116 - TJChess 1.3: +53 -13 =0

Both matches could be downloaded here:

http://www.mediafire.com/file/vdav2835s ... 1116zh.rar

PS: Thanks for the new version of CrazyWa btw. It is getting stronger very quick.
Top
Ferdy
Posts: 4833
Joined: Sun Aug 10, 2008 3:15 pm
Location: Philippines

Re: Crazyhouse tournaments and rating list

Post by Ferdy »

Carlos777 wrote:
Ferdy wrote:
TonyJH wrote:I just saw a post on the lichess forum (from the Sunsetter author) that the crazyhouse version of Stockfish is playing very well now:

https://github.com/ddugovic/Stockfish/issues/88

"Result of Stockfish vs. Sunsetter 9: 44 - 5 - 1 (W - L - D)"

"Score of Stockfish 301016 64 vs TJchess 1.3-x64: 44 - 6 - 0"
Amazing results :!: and this is from TC 40moves/10s. I will include it in the rating list and the upcoming champs tour.
I ran a little match at 15'+3" with my own test suite of 33 positions:

Stockfish 011116 - Imortal 2.9001: +43 -22 =1

Very strong engine, although its eval is sometimes inconsistent.

For example:

Code: Select all

[Event "My Tournament"]
[Site "PhenomII"]
[Date "2016.11.01"]
[Round "1"]
[White "Stockfish 011116"]
[Black "Imortal 2.9001"]
[Result "0-1"]
[PlyCount "65"]
[Termination "adjudication"]
[TimeControl "900+3"]
[Variant "crazyhouse"]

1. e4 {book} e5 {book} 2. Nf3 {book} Nc6 {-0.40/15 45s} 3. Nc3 {+0.37/20 44s}
Nf6 {-0.46/14 36s} 4. Bb5 {+0.59/20 20s} Bc5 {-0.06/13 41s}
5. O-O {+0.79/18 8.8s} d6 {-0.04/13 39s} 6. d3 {+0.83/20 21s} a6 {+0.09/14 26s}
7. Ba4 {+1.12/21 39s} Ng4 {-0.29/13 36s} 8. Nd5 {+1.45/18 9.8s}
Be6 {-0.08/12 24s} 9. Bg5 {+1.69/18 32s} Qd7 {-0.44/12 34s}
10. h3 {+2.64/17 9.4s} Nxf2 {+1.02/13 32s} 11. Rxf2 {+2.93/16 5.3s}
Bxd5 {+0.45/11 31s} 12. exd5 {+3.35/19 14s} P@g3 {+0.08/11 29s}
13. dxc6 {+5.01/18 18s} gxf2+ {-1.36/12 28s} 14. Kh2 {+5.37/20 44s}
bxc6 {-1.38/12 19s} 15. N@h5 {+5.82/18 34s} P@g3+ {-1.57/11 18s}
16. Nxg3 {+6.42/18 14s} h6 {-1.60/10 25s} 17. P@b7 {+6.93/18 16s}
N@g4+ {-2.04/10 15s} 18. hxg4 {+6.65/19 43s} hxg5+ {-6.70/9 24s}
19. N@h5 {+7.70/17 24s} Rb8 {-5.03/10 22s} 20. B@c8 {+8.01/17 35s}
B@f4 {-7.48/8 22s} 21. N@f5 {+8.52/17 26s} Qxf5 {-5.39/9 21s}
22. Bxf5 {+9.50/17 18s} N@e7 {-6.02/10 13s} 23. Q@c8+ {+7.39/17 32s}
Nxc8 {-1.96/10 19s} 24. bxc8=Q+ {+0.64/16 60s} Rxc8 {-1.96/10 18s}
25. Bxc6+ {+1.47/17 10s} Kf8 {-2.50/11 17s} 26. N@e2 {0.00/17 28s}
P@e6 {+1.72/10 11s} 27. Bxe6 {-0.75/16 82s} fxe6 {+4.55/9 11s}
28. P@h3 {0.00/16 23s} B@e1 {+7.21/10 15s} 29. Qxe1 {-3.04/17 33s}
fxe1=Q {+7.22/11 11s} 30. Rxe1 {-2.85/16 2.3s} Rxh5 {+6.93/10 15s}
31. P@e7+ {-9.56/18 90s} Kxe7 {+9.28/9 12s} 32. B@e3 {-12.72/18 67s}
Rxh3+ {+17.61/8 11s} 33. Kxh3 {-17.83/16 16s, Black wins by adjudication} 0-1
At move 22, it seems Stockfish has a win, but it got worse from it and finally lost the game.

Another example:

Code: Select all

[Event "My Tournament"]
[Site "PhenomII"]
[Date "2016.11.01"]
[Round "1"]
[White "Stockfish 011116"]
[Black "Imortal 2.9001"]
[Result "0-1"]
[PlyCount "67"]
[Termination "adjudication"]
[TimeControl "900+3"]
[Variant "crazyhouse"]

1. d4 {book} e6 {-0.43/16 45s} 2. Nf3 {+0.15/22 47s} Nc6 {-0.39/15 43s}
3. Nbd2 {+0.33/20 31s} Nf6 {-0.22/14 41s} 4. e4 {+0.59/18 12s} d6 {0.00/14 39s}
5. Bb5 {+0.66/19 29s} Bd7 {-0.40/15 37s} 6. O-O {+0.80/18 8.6s}
Be7 {-0.02/14 35s} 7. c3 {+1.20/19 16s} Rb8 {-0.47/13 34s} 8. Re1 {+1.25/20 39s}
a6 {-0.21/14 27s} 9. Bd3 {+1.35/20 16s} Ng4 {-0.42/15 29s}
10. h3 {+1.59/19 8.5s} Nxf2 {+0.04/14 19s} 11. Kxf2 {+2.90/16 8.8s}
h5 {-0.62/12 28s} 12. Rh1 {+2.27/20 83s} Bh4+ {-0.25/12 22s}
13. g3 {+3.09/16 8.0s} P@e3+ {-0.56/13 27s} 14. Kxe3 {+3.57/18 8.1s}
Bxg3 {-1.57/14 26s} 15. Qg1 {+3.64/21 55s} Qf6 {-1.53/12 24s}
16. Nf1 {+3.83/20 45s} Bh4 {-1.11/11 23s} 17. Nxh4 {+3.75/20 22s}
Qxh4 {-2.85/10 21s} 18. N@f3 {+4.43/19 11s} Qe7 {-3.53/11 22s}
19. Qxg7 {+5.64/17 12s} Qf8 {-5.04/10 21s} 20. Qxf8+ {+5.81/16 8.1s}
Rxf8 {-2.73/11 20s} 21. P@g2 {+5.91/17 43s} e5 {-1.27/9 19s}
22. Q@e2 {+6.00/16 24s} N@f4 {+0.78/9 12s} 23. dxe5 {+5.74/17 29s}
Nxe2 {+0.19/9 18s} 24. Kxe2 {+5.90/17 32s} P@g4 {-0.02/9 13s}
25. Ne1 {+6.88/16 23s} Nxe5 {-2.65/9 16s} 26. N@g7+ {+M21/17 49s}
Kd8 {-0.57/10 14s} 27. B@f6+ {+4.87/19 45s} Kc8 {-1.49/11 15s}
28. Bxe5 {+5.91/17 25s} dxe5 {+0.90/9 15s} 29. P@e7 {+7.23/17 38s}
P@f3+ {+1.03/7 14s} 30. Kd1 {+6.14/16 35s} B@e2+ {+9.34/8 11s}
31. Bxe2 {+1.01/16 22s} fxe2+ {+9.64/8 13s} 32. Kxe2 {-M26/16 15s}
Bb5+ {+7.59/8 13s} 33. c4 {-M20/20 16s} Bxc4+ {+11.45/8 10s}
34. N@d3 {-M12/20 12s, Black wins by adjudication} 0-1
Stockfish announces mate in 21 at move 26th, next move the eval is +4.87 only and keeps going down.
The program is not fully developed yet, but is already very strong. It does not panic under severe attacks, defending it well with its high search depths.
Top
Ferdy
Posts: 4833
Joined: Sun Aug 10, 2008 3:15 pm
Location: Philippines

Re: Crazyhouse tournaments and rating list

Post by Ferdy »

hgm wrote:Sunsetter 7g sometimes also does weird things. It can show a mate score (move 21 and 22), and then still lose: :!:

Code: Select all

[Event "Computer Chess Game"]
[Site "MAKRO-PC"]
[Date "2016.11.02"]
[Round "22"]
[White "CrazyWa 0.0.9qs9"]
[Black "sunsetter"]
[Result "1-0"]
[TimeControl "40/60"]
[Variant "crazyhouse"]

1. Nh3 {+0.24/10} e5 {+0.28/13 1.5} 2. e4 {-0.13/8 0.9} d6 {+0.26/12 1.5}
3. Qf3 {+0.29/9 2.3} Nf6 {+0.42/11 1.5} 4. Bc4 {+0.36/8 0.9} Bxh3
{+0.76/11 1.5} 5. Qxh3 {+1.07/7 0.9} N@f4 {+0.41/12 6} 6. Qf3 {+2.05/8 2.8}
d5 {-0.13/11 1.4} 7. exd5 {+3.26/8 3} N6xd5 {-0.23/10 1.4} 8. P@g3
{+4.25/7 1.6} P@g4 {-0.18/11 1.4} 9. Qe4 {+3.97/8 2.1} Nxg2+ {-0.25/11 2.6}
10. Qxg2 {+4.61/8 1.6} P@f3 {-0.32/11 1.3} 11. Qf1 {+4.32/7 1.1} Nb4
{-0.42/10 1.3} 12. Bxf7+ {+4.99/6 1.3} Kxf7 {+0.32/9 1.3} 13. P@e6+
{+1.88/6 2.7} Kxe6 {+1.08/9 1.2} 14. Qc4+ {+2.19/6 1.2} B@d5 {+1.85/10 1.2}
15. Qxg4+ {+2.20/6 1.9} P@f5 {+1.66/9 1.2} 16. N@g5+ {+0.04/6 1.0} Qxg5
{+2.23/10 2.4} 17. Qxg5 {+2.75/7 1.2} Nxc2+ {+2.57/10 2.3} 18. Kd1
{+2.74/7 1.2} P@e2+ {+2.38/9 1.2} 19. Kxc2 {+0.08/9 1.6} N@d4+
{+2.23/9 2.2} 20. Kd3 {+2.43/8 1.5} Nbc6 {+3.88/9 2.2} 21. B@a4
{+4.93/5 2.3} Be4+ {+299.20/10 1.7} 22. Ke3 {+5.65/7 2.9} Nc2+
{+299.60/33 0.1} 23. Bxc2 {+9.27/8 2.0} Bc5+ {+299.70/33 0.1} 24. d4
{+10.41/6 0.8} exd4+ {+2.11/11 4} 25. Kd2 {+9.25/6 0.8} P@e3+
{+2.16/10 0.4} 26. fxe3 {+3.54/7 1.6} dxe3+ {+2.11/9 0.4} 27. Kc3
{+4.75/7 2.3} Bd4+ 28. Kb3 {+7.13/6 0.6} P@c4+ {+2.53/9 1.4} 29. Ka3
{+9.82/8 2.8} Bc5+ 30. b4 {+8.80/6 0.6} Bxb4+ {+1.89/9 0.6} 31. Kb2
{+8.98/7 0.9} P@a3+ 32. Nxa3 {+9.75/6 2.5} Bxa3+ 33. Kxa3 {+5.58/6 0.5}
N@b5+ {+0.65/8 0.3} 34. Kb2 {+10.38/7 2.9} c3+ {+0.20/8 0.3} 35. Kb1
{+10.37/6 1.8} Bxc2+ {-0.98/8 0.6} 36. Kxc2 {+10.37/5 0.2} B@e4+ 37. Kb3
{+11.59/5 0.2} Ncd4+ {-5.21/8 0.5} 38. Kc4 {+16.59/5 0.5} Bd5+ 39. Kd3
{+16.60/5 0.2} Be4+ {-10.97/9 3} 40. Kxe3 {+21.15/5 0.2} Rae8 41. B@c4+
{+22.77/3 2.6} Kd7 {-28.83/9 25} 42. Qxg7+ {+28.20/3 1.5} Kc6 43. B@d7+
{+29.10/3 1.2} Kb6 44. Qxd4+ {+1000.05/5 1.5} Ka5 {-299.50/5 0.1} 45. N@b3+
{+1000.02/8 0.1} Ka6 46. Bdxb5# {+1000.01/3 0.1}
{Xboard adjudication: Checkmate} 1-0
BTW, I just uploaded CrazyWa 0.0.9. This now also implement repetition detection. I also fixed a bug in the LMR, that it sometimes reduced directly into QS, because the limiting of the reduction did not always work. But fixig the bug made it weaker, so I figured I should let the LMR kick in at lower depth already. This version does that, and although it is too early to have a statistically significant conclusion, this put it in the lead in my 40/1 test matches against both Imortal 1.0 and Sunsetter 7g after 25 games each. (While before the bug fix it was doing at best 36% aganist Sunsetter 7g.)

This version still clears the hash table before every move. I wanted to do that during development to make errors reproducable. (Unfortunately this did not help, though.) It does crash a lot less often than previous versions; I located one cause of crashes (the randomization during the first 5 moves), and fixed another suspect (execution of the check-detection code after null move).
Thanks for the update.
Top
Ferdy
Posts: 4833
Joined: Sun Aug 10, 2008 3:15 pm
Location: Philippines

Re: Crazyhouse tournaments and rating list

Post by Ferdy »

Carlos777 wrote:
hgm wrote:Sunsetter 7g sometimes also does weird things. It can show a mate score (move 21 and 22), and then still lose: :!:
Interesting. The SF Variant's author said it was caused by hash collisions and it is fixed in the new version.

Same conditions (also same SF version as previous match)

Stockfish 011116 - TJChess 1.3: +53 -13 =0

Both matches could be downloaded here:

http://www.mediafire.com/file/vdav2835s ... 1116zh.rar

PS: Thanks for the new version of CrazyWa btw. It is getting stronger very quick.
Thanks I will use these games in the rating list calculation.
Top
TonyJH
Posts: 183
Joined: Tue Jun 20, 2006 4:41 am
Location: USA

Re: Crazyhouse tournaments and rating list

Post by TonyJH »

Carlos777 wrote:
hgm wrote:Sunsetter 7g sometimes also does weird things. It can show a mate score (move 21 and 22), and then still lose: :!:
Interesting. The SF Variant's author said it was caused by hash collisions and it is fixed in the new version.

Same conditions (also same SF version as previous match)

Stockfish 011116 - TJChess 1.3: +53 -13 =0

Both matches could be downloaded here:

http://www.mediafire.com/file/vdav2835s ... 1116zh.rar

PS: Thanks for the new version of CrazyWa btw. It is getting stronger very quick.
Stockfish is using one core?
Top
Ferdy
Posts: 4833
Joined: Sun Aug 10, 2008 3:15 pm
Location: Philippines

Stockfish zh in rating list

Post by Ferdy »

Rapid rating list all versions:

Code: Select all

   # PLAYER                          :  RATING    +/-   GAMES  PTS(%)
   1 Stockfish zh 01Nov2016 64bit    :  2979.5   90.6     132      73
   2 Imortal 2.9001 32bit            :  2852.8   71.2     294      66
   3 TJchess 1.3 64bit               :  2732.6   62.8     366      55
   4 Sunsetter 9 32bit               :  2700.2   66.9     244      53
   5 Imortal 2.0 32bit               :  2660.9   76.5     151      74
   6 Sunsetter 7g 64bit              :  2526.5   68.3     166      69
   7 Sjeng 11.2 32bit                :  2300.0   61.7     213      45
   8 Sjaak II 1.3.1a 64bit           :  2250.0   59.3     236      41
   9 NebiyuAlien 1.45a 32bit         :  2236.6   67.9     176      39
  10 Feuerstein 0.4.6.1 32bit        :  2230.3   69.1     176      38
  11 Pulsar 2009 9b 32bit            :  2035.5   69.2     159      34
  12 NebiyuAlien 1.45 32bit          :  1967.9  119.2      49      31
  13 TSCP zh 1.1 32bit               :  1937.9   93.0      96      26
  14 KKFChess 2.6.7 beta 32bit       :  1891.9   77.9     140      21

White advantage = 73.11 +/- 12.01
Draw rate (equal opponents) = 2.04 % +/- 0.45
Using Ordo, Sjeng 11.2 32bit is set to 2300
Average depth and time comparison:

Code: Select all

 nr                       player  games  pts%   sumTime aveDep    aveTime
  1 Stockfish zh 01Nov2016 64bit    132  73.1  24:31:29  18.03  00:24:997
  2       NebiyuAlien 1.45 32bit     49  30.6  04:56:11  16.95  00:15:280
  3      NebiyuAlien 1.45a 32bit    176  38.9  23:14:56  16.87  00:19:946
  4            Sunsetter 9 32bit    244  53.3  42:19:28  16.51  00:21:545
  5           Sunsetter 7g 64bit    166  69.0  12:45:07  12.27  00:10:953
  6         Imortal 2.9001 32bit    294  66.3  49:54:57  11.30  00:20:757
  7        Sjaak II 1.3.1a 64bit    236  41.3  32:42:28  10.56  00:19:347
  8            TJchess 1.3 64bit    366  54.8  64:47:50  10.53  00:21:361
  9            Imortal 2.0 32bit    151  74.2  21:45:42   9.47  00:19:334
 10    KKFChess 2.6.7 beta 32bit    140  20.7  21:00:59   8.12  00:20:273
 11     Feuerstein 0.4.6.1 32bit    176  38.4  19:03:32   7.88  00:17:440
 12             Sjeng 11.2 32bit    213  44.6  33:25:12   7.85  00:21:074
 13            TSCP zh 1.1 32bit     96  25.5  11:57:42   5.91  00:17:399
 14         Pulsar 2009 9b 32bit    159  33.6  24:26:11   5.79  00:21:091

file         : Rapid.pgn
score window : [-1000000.0, +1000000.0]
notes        : 1. games and pts% are not affected by score window.
               2. Table is sorted by aveDep in descending order.
               3. aveTime is the average time/move in m:s:ms
elapsed      : 00:00:40
Site:
https://sites.google.com/site/zhassocia ... list/rapid
Top
Vinvin
Posts: 5228
Joined: Thu Mar 09, 2006 9:40 am
Full name: Vincent Lejeune

Re: Crazyhouse tournaments and rating list

Post by Vinvin »

Carlos777 wrote:
hgm wrote:Sunsetter 7g sometimes also does weird things. It can show a mate score (move 21 and 22), and then still lose: :!:
Interesting. The SF Variant's author said it was caused by hash collisions and it is fixed in the new version.

Same conditions (also same SF version as previous match)

Stockfish 011116 - TJChess 1.3: +53 -13 =0

Both matches could be downloaded here:

http://www.mediafire.com/file/vdav2835s ... 1116zh.rar

PS: Thanks for the new version of CrazyWa btw. It is getting stronger very quick.
Stockfish zh 2016-11-03 is out now. Hash value changed with pieces in hand.
Rating improvement on 100 games : ELO: 92.46 +-69.2 ( Total: 100 W: 60 L: 34 D: 6 )
https://github.com/ddugovic/Stockfish/pull/99
Top
User avatar
hgm
Posts: 27808
Joined: Fri Mar 10, 2006 10:06 am
Location: Amsterdam
Full name: H G Muller

Re: Stockfish zh in rating list

Post by hgm »

Ferdy wrote:Rapid rating list all versions:
I only see 2, 1 or 0 versions. :?

BTW, Sunsetter 7g seems a very temperamental opponent for testing, and I wonder if I should pay much attention to how my engine scores against it. Considering the horrendously complex checkmates it finds, it seems to have proof-number search for that, and alpha-beta could not compete with that anyway, no matter how much I improve it. In the games it loses, Sunsetter mainly seems to do itself in, starting with a wildly speculative optimistic score, tossing all its material in checks, until it eventually is left with nothing in hand. Only then it sees it will be mated in 2 or 3.

Changes that improve the result against Imortal 1.0 usually deteriorate the result against Sunsetter 7g. Perhaps the dominant source of wins against Sunsetter 7g currently is enticing it to make rash attacks by flaunting weakness in King Safety...

I also wonder of Sunsetter could be learning. I have seen several times now that the match goes equal upto 20-20 or so, and after that Sunsetter just stops losing games, and the score goes to 55-25 in the next 40 games. There is a file in its directory that it is writing, while it is playing.
Top
Carlos777
Posts: 1736
Joined: Sun Dec 13, 2009 6:09 pm

Re: Crazyhouse tournaments and rating list

Post by Carlos777 »

TonyJH wrote:
Carlos777 wrote:
hgm wrote:Sunsetter 7g sometimes also does weird things. It can show a mate score (move 21 and 22), and then still lose: :!:
Interesting. The SF Variant's author said it was caused by hash collisions and it is fixed in the new version.

Same conditions (also same SF version as previous match)

Stockfish 011116 - TJChess 1.3: +53 -13 =0

Both matches could be downloaded here:

http://www.mediafire.com/file/vdav2835s ... 1116zh.rar

PS: Thanks for the new version of CrazyWa btw. It is getting stronger very quick.
Stockfish is using one core?
Yes, like all the engines tested.
Top

Return to “Computer Chess Club: General Topics”