Swiss Tour Nr 19 is started here.
http://ccva.challonge.com/tour_nr_19
Crazyhouse tournaments and rating list
Moderators: hgm, Rebel, chrisw
-
- Posts: 4833
- Joined: Sun Aug 10, 2008 3:15 pm
- Location: Philippines
Re: CCVA Swiss Tour Nr 19
A disconnect from CrazyWa 0.0.1
Code: Select all
[Event "My Tournament"]
[Site "i7-2600K"]
[Date "2016.10.22"]
[Round "1"]
[White "CrazyWa 0.0.1 32bit"]
[Black "TJchess 1.3 64bit"]
[Result "0-1"]
[PlyCount "46"]
[Termination "abandoned"]
[TimeControl "180+2"]
[Variant "crazyhouse"]
1. d3 {book} e6 {book} 2. Nf3 {book} Nc6 {book} 3. e4 {book} Nf6 {book}
4. Nc3 {book} Bb4 {book} 5. Bg5 {book} h6 {+0.35/11 2.4s} 6. Bc1 {-0.28/10 3.2s}
Qe7 {+0.76/11 2.7s} 7. Bd2 {-0.09/11 6.0s} Bxc3 {+0.81/11 2.8s}
8. Bxc3 {-0.01/10 3.5s} N@a4 {+0.62/12 6.0s} 9. Be2 {+0.11/11 5.3s}
Nxc3 {+0.64/11 4.2s} 10. bxc3 {+0.13/11 8.1s} Qc5 {+0.46/12 12s}
11. B@d4 {+0.16/11 8.3s} Nxd4 {+1.57/12 3.1s} 12. cxd4 {+0.15/10 3.3s}
B@c3+ {+1.83/11 2.3s} 13. Kf1 {-1.01/12 6.7s} Bxd4 {+2.09/12 13s}
14. N@b3 {-1.70/11 16s} Bxf2 {+3.43/11 6.8s} 15. Qb1 {-2.38/9 8.8s}
Qb6 {+4.38/11 9.1s} 16. N@c4 {-4.02/10 6.8s} Ng4 {+4.15/11 22s}
17. Nxb6 {-4.22/9 6.6s} Bxb6 {+7.65/9 3.3s} 18. Q@b4 {-7.35/8 11s}
B@f2 {+9.04/9 4.6s} 19. Bd1 {-11.17/9 4.3s} Nxh2+ {+10.12/9 4.3s}
20. Ke2 {-11.66/9 8.3s} N@g3+ {+12.86/9 7.5s} 21. Kd2 {-11.67/9 8.2s}
Bfe3+ {+14.40/10 17s} 22. Kc3 {-13.12/9 7.5s} a5 {+15.46/10 5.7s}
23. Nxa5 {-12.90/8 3.2s} Bxa5 {+15.46/8 8.4s, White disconnects} 0-1
-
- Posts: 27790
- Joined: Fri Mar 10, 2006 10:06 am
- Location: Amsterdam
- Full name: H G Muller
Re: CCVA Swiss Tour Nr 19
I fed it the position, but I cannot reproduce the problem. It plays a move (with a mated-in-7 score at 7 ply). Even when I start it a move earlier.
This is reason for concern, as this version is supposed to be fully deterministic: I clear the hash table before every move, the killers are self-clearing, and history is not yet implemented.
Do you happen to have a debug file for this game? There still is a fair number of asserts in CrazyWa, and I expect one of those got triggered. This would then have printed the reason (e.g. exceeding MAXPLY, or an empty from-square of a hash move), plus the path to the node where it happened.
This is reason for concern, as this version is supposed to be fully deterministic: I clear the hash table before every move, the killers are self-clearing, and history is not yet implemented.
Do you happen to have a debug file for this game? There still is a fair number of asserts in CrazyWa, and I expect one of those got triggered. This would then have printed the reason (e.g. exceeding MAXPLY, or an empty from-square of a hash move), plus the path to the node where it happened.
-
- Posts: 4833
- Joined: Sun Aug 10, 2008 3:15 pm
- Location: Philippines
Re: CCVA Swiss Tour Nr 19
Sorry I did not save the debug.hgm wrote:I fed it the position, but I cannot reproduce the problem. It plays a move (with a mated-in-7 score at 7 ply). Even when I start it a move earlier.
This is reason for concern, as this version is supposed to be fully deterministic: I clear the hash table before every move, the killers are self-clearing, and history is not yet implemented.
Do you happen to have a debug file for this game? There still is a fair number of asserts in CrazyWa, and I expect one of those got triggered. This would then have printed the reason (e.g. exceeding MAXPLY, or an empty from-square of a hash move), plus the path to the node where it happened.
I will try to save it in later matches.
-
- Posts: 27790
- Joined: Fri Mar 10, 2006 10:06 am
- Location: Amsterdam
- Full name: H G Muller
Re: Crazyhouse tournaments and rating list
Apparently CrazyWa was lucky in these first 20 games against Sjaak II; with some 200 games Sjaak beat it by 55% also in my tests. (First CrazyWa even did much worse, until I discovered that pondering was on, while CrazyWa cannot ponder, and Sjaak can. So CrazyWa had to share the CPU with Sjaak while thinking.)
The improvements from 0.0.0 to 0.0.1 were all search; probably null move contributed the most. Evaluation can do a lot too, though: I noticed that CrazyWa was very passive, so I added two very simple evaluation terms to cure it, by encouraging it to develop its pieces. These were penalties for a Rook trapped in a corner (e.g. Ra1,with own pieces on a2 and b1), and a penalty for a Bishop trapped on its original square (e.g. Bc1 with own pieces on b2 and d2). Those terms alone increase the result against Sjaak from 45% to 63%! (measured in 360 games 40 moves/min).
Tomorrow I hope to add some King safety; it shouldn't be very expensive to calculate attacked squares next to the enemy King, as a side effect of move generation, and then add some bonus for it. And fix this problem that it does not know check with a Pawn can be evaded by e.p. capture. Then I will release a new version.
The improvements from 0.0.0 to 0.0.1 were all search; probably null move contributed the most. Evaluation can do a lot too, though: I noticed that CrazyWa was very passive, so I added two very simple evaluation terms to cure it, by encouraging it to develop its pieces. These were penalties for a Rook trapped in a corner (e.g. Ra1,with own pieces on a2 and b1), and a penalty for a Bishop trapped on its original square (e.g. Bc1 with own pieces on b2 and d2). Those terms alone increase the result against Sjaak from 45% to 63%! (measured in 360 games 40 moves/min).
Tomorrow I hope to add some King safety; it shouldn't be very expensive to calculate attacked squares next to the enemy King, as a side effect of move generation, and then add some bonus for it. And fix this problem that it does not know check with a Pawn can be evaded by e.p. capture. Then I will release a new version.
-
- Posts: 27790
- Joined: Fri Mar 10, 2006 10:06 am
- Location: Amsterdam
- Full name: H G Muller
Re: Crazyhouse tournaments and rating list
Unfortunately I wasted yesterday by trying to tune a version that later turned out to be quite buggy. I seem to have fixed those bugs now, and implemented the history heuristic, which seems to do some good against Sjaak II. So I uploaded a new version (0.0.4).
Compared to 0.0.2 it has some eval patterns for undoing the weak spots in the opening position (Rooks and Bishops trapped on their starting square), a King Safety term based on attacked empty squares near the King, and history for sorting non-captures. And it should also have fixed the bug that e.p. captures would never be considered check evasions (which actually did make in bungle games rather often, sacrificing a lot for something that after alldid not turn out to be a mate.)
Compared to 0.0.2 it has some eval patterns for undoing the weak spots in the opening position (Rooks and Bishops trapped on their starting square), a King Safety term based on attacked empty squares near the King, and history for sorting non-captures. And it should also have fixed the bug that e.p. captures would never be considered check evasions (which actually did make in bungle games rather often, sacrificing a lot for something that after alldid not turn out to be a mate.)
-
- Posts: 4833
- Joined: Sun Aug 10, 2008 3:15 pm
- Location: Philippines
CCVA CrazyWa 0.0.4 Gauntlet Tour Nr 21
https://sites.google.com/site/zhassocia ... tour-nr-21
CrazyWa 0.0.4 has abandoned two games based from Cute Chess GUI.
I don't have debug file since I forgot again to specify the stderrFile before running the tour.
CrazyWa 0.0.4 has abandoned two games based from Cute Chess GUI.
I don't have debug file since I forgot again to specify the stderrFile before running the tour.
Code: Select all
[Event "CCVA Tour Nr 21"]
[Site "i7-2600K"]
[Date "2016.10.26"]
[Round "1"]
[White "Sunsetter 9 32bit"]
[Black "CrazyWa 0.0.4 32bit"]
[Result "1-0"]
[PlyCount "41"]
[Termination "abandoned"]
[TimeControl "180+2"]
[Variant "crazyhouse"]
1. d4 {book} d5 {book} 2. Qd3 {book} Nf6 {book} 3. Nf3 {book} Nc6 {book}
4. Bf4 {book} e6 {+0.19/10 3.9s} 5. e3 {-0.09/21 7.4s} Bb4+ {+0.34/10 4.0s}
6. Nbd2 {-0.08/19 7.3s} Nh5 {+0.26/10 6.3s} 7. Bg5 {-0.04/19 7.1s}
f6 {+0.46/10 3.8s} 8. Bh4 {-0.02/20 6.9s} g5 {+0.24/10 3.5s}
9. c3 {+0.07/20 6.7s} Bd6 {+0.19/11 9.8s} 10. g4 {+0.06/20 6.5s}
Ng7 {+0.14/11 4.8s} 11. Bg3 {+0.15/20 6.3s} e5 {-0.02/10 6.0s}
12. dxe5 {+0.21/18 6.2s} Bxe5 {-0.19/9 6.4s} 13. Bxe5 {+0.33/16 6.0s}
Nxe5 {-0.67/8 4.2s} 14. Nxe5 {+0.56/15 5.8s} fxe5 {-0.88/8 8.2s}
15. P@h6 {+0.94/14 5.7s} Ne6 {-1.40/8 7.3s} 16. Qxd5 {+0.89/14 8.3s}
N@c2+ {+1.43/8 21s} 17. Kd1 {+1.43/15 5.3s} Nxa1 {+3.27/6 3.4s}
18. Bb5+ {+0.51/12 5.2s} P@c6 {+2.68/7 13s} 19. Qxd8+ {+1.94/12 5.0s}
Nxd8 {+2.00/7 13s} 20. N@f6+ {+M5/8 0.038s} Ke7 {+0.96/7 8.6s}
21. Nd5+ {+M3/6 0.001s, Black disconnects} 1-0
Code: Select all
[Event "CCVA Tour Nr 21"]
[Site "i7-2600K"]
[Date "2016.10.25"]
[Round "1"]
[White "Sjeng 11.2 32bit"]
[Black "CrazyWa 0.0.4 32bit"]
[Result "1-0"]
[PlyCount "53"]
[Termination "abandoned"]
[TimeControl "180+2"]
[Variant "crazyhouse"]
1. e3 {book} d5 {book} 2. Nc3 {+0.42/9 9.6s} Nf6 {-0.21/11 6.2s}
3. d4 {+1.00/8 8.8s} g5 {-0.09/10 2.9s} 4. Nf3 {+1.71/9 8.1s}
Bh6 {-0.01/11 4.2s} 5. h4 {+2.35/8 8.8s} g4 {+0.11/11 5.3s}
6. Ne5 {+2.16/8 7.7s} Be6 {+0.15/10 4.2s} 7. Be2 {+2.51/8 7.8s}
Rg8 {+0.17/10 4.4s} 8. Bb5+ {+2.36/7 4.5s} Nbd7 {+0.15/10 3.9s}
9. Nxd7 {+2.98/8 4.2s} Bxd7 {-0.37/10 10s} 10. Bxd7+ {+2.27/7 4.9s}
Qxd7 {-0.37/9 5.4s} 11. N@c5 {+3.07/7 4.6s} Qf5 {-1.39/9 8.6s}
12. B@a4+ {+3.38/6 4.5s} B@c6 {-1.29/9 3.6s} 13. Nxb7 {+2.52/7 4.3s}
Bxa4 {-0.97/9 4.8s} 14. Nxa4 {+2.41/6 5.1s} g3 {-0.61/8 8.9s}
15. B@c6+ {+1.03/6 3.9s} B@d7 {+1.37/8 3.9s} 16. Bxd7+ {+0.25/6 4.2s}
Qxd7 {+2.18/8 5.7s} 17. fxg3 {+0.13/6 4.2s} Qxa4 {+2.13/7 4.0s}
18. Nc5 {-0.19/7 9.0s} B@b4+ {+1.86/7 9.8s} 19. B@d2 {0.00/6 3.8s}
Bxc5 {+4.18/7 3.7s} 20. P@b7 {-0.87/8 8.6s} Rb8 {+4.47/9 9.2s}
21. dxc5 {-0.65/7 2.9s} Rxg3 {+5.65/8 10s} 22. b3 {-3.07/6 4.4s}
Qe4 {+6.17/7 9.6s} 23. B@a4+ {-2.32/7 10s} Kf8 {+7.82/7 3.0s}
24. P@f3 {-3.14/7 4.4s} Qe5 {+8.68/7 6.2s} 25. e4 {-5.60/7 8.4s}
Bxd2+ {+9.13/6 3.8s} 26. Bxd2 {-6.75/6 3.2s} Rxg2 {+9.61/6 6.5s}
27. B@h6+ {-9.45/6 8.5s, Black disconnects} 1-0
-
- Posts: 4833
- Joined: Sun Aug 10, 2008 3:15 pm
- Location: Philippines
Re: Crazyhouse tournaments and rating list
In blitz this is +35 rating points after 216 games so far over previous version 0.0.1.hgm wrote:Unfortunately I wasted yesterday by trying to tune a version that later turned out to be quite buggy. I seem to have fixed those bugs now, and implemented the history heuristic, which seems to do some good against Sjaak II. So I uploaded a new version (0.0.4).
Compared to 0.0.2 it has some eval patterns for undoing the weak spots in the opening position (Rooks and Bishops trapped on their starting square), a King Safety term based on attacked empty squares near the King, and history for sorting non-captures. And it should also have fixed the bug that e.p. captures would never be considered check evasions (which actually did make in bungle games rather often, sacrificing a lot for something that after alldid not turn out to be a mate.)
Code: Select all
# PLAYER : RATING +/- GAMES PTS(%)
1 Imortal 2.9001 32bit : 2857.4 40.3 1075 75
2 TJchess 1.3 64bit : 2792.8 38.3 1105 66
3 Imortal 2.0 32bit : 2670.6 52.6 295 66
4 Sunsetter 9 32bit : 2667.4 37.9 915 43
5 TJchess 1.1 64bit : 2596.3 68.2 152 54
6 Sunsetter 7g 64bit : 2544.8 47.6 296 52
7 Imortal 1.0 32bit : 2520.1 88.0 94 38
8 Sjeng 11.2 32bit : 2300.0 29.3 792 49
9 NebiyuAlien 1.45a 32bit : 2289.1 31.4 754 43
10 Sjaak II 1.3.1a 64bit : 2278.9 29.6 859 45
11 CrazyWa 0.0.4 32bit : 2259.8 58.4 216 43
12 CrazyWa 0.0.1 32bit : 2224.1 66.8 129 56
13 Pulsar 2009 9b 32bit : 2122.2 39.7 451 39
14 NebiyuAlien 1.45 32bit : 2065.8 56.4 204 37
15 CrazyWa 0.0.0 32bit : 2010.0 78.3 96 32
16 TSCP zh 1.1 32bit : 2009.7 41.5 431 28
17 KKFChess 2.6.7 beta 32bit : 1962.4 44.0 452 22
White advantage = 58.62 +/- 6.43
Draw rate (equal opponents) = 1.03 % +/- 0.18
Using Ordo, Sjeng 11.2 32bit is set to 2300
https://sites.google.com/site/zhassocia ... ingdetails
-
- Posts: 4833
- Joined: Sun Aug 10, 2008 3:15 pm
- Location: Philippines
Re: CCVA CrazyWa 0.0.4 Gauntlet Tour Nr 21
Average depth on this blitz (TC 3'+2") tour nr 21, when scores are within the window of [-10.0, +10.0] pawn unit.
Code: Select all
nr player aveDepth games
1 Sunsetter 9 32bit 12.92 24
2 NebiyuAlien 1.45a 32bit 11.40 24
3 TJchess 1.3 64bit 9.60 24
4 Imortal 2.9001 32bit 9.35 24
5 CrazyWa 0.0.4 32bit 9.18 216
6 Sjaak II 1.3.1a 64bit 8.85 24
7 KKFChess 2.6.7 beta 32bit 8.20 24
8 Sjeng 11.2 32bit 7.70 24
9 TSCP zh 1.1 32bit 7.42 24
10 Pulsar 2009 9b 32bit 7.07 24
-
- Posts: 2929
- Joined: Sat Jan 22, 2011 12:42 am
- Location: NL
Re: CCVA CrazyWa 0.0.4 Gauntlet Tour Nr 21
Interesting that Sunsetter reaches such a large depth.
It seems to not consider drops during quiescence search, but if I try that in SjaakII it loses some 170 Elo points...
It seems to not consider drops during quiescence search, but if I try that in SjaakII it loses some 170 Elo points...