Nh4 is hard to find, but the stats will tell you it is right

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

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
mwyoung
Posts: 1598
Joined: Wed May 12, 2010 8:00 pm

Re: Nh4 is hard to find, but the stats will tell you it is right

Post by mwyoung » Sat Aug 24, 2019 4:35 pm

Using Deep Position Analysis Stockfish 230819 thinks Bc1 is the best move score +.69
Lc0 42859 thinks Bxc6 is best with a score of +.88.

Settings
Settings.png
Settings.png (44.79 KiB) Viewed 822 times


[Event "Stockfish"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "New game"]
[Black "?"]
[Result "*"]
[SetUp "1"]
[FEN "r3k2r/ppqbnppp/2n1p3/1B1pP3/P1pP4/2P2N2/2PB1PPP/R2Q1RK1 w kq - 0 1"]
[PlyCount "21"]

{[#]} 1. Bc1 (1. Bg5 h6 (1... O-O 2. Qd2 (2. Bxc6 Bxc6 3. a5 b6 4. a6 Bb5 5. h3
h6 6. Bxe7 Qxe7 7. Nh2 Bd7 8. f4 b5 9. Qb1 Rab8 10. Qb4 Qd8 11. Ng4 $11 {
0.21/29}) 2... b6 3. Rfb1 Be8 4. Bxc6 Bxc6 5. Qc1 h6 6. Bxe7 Qxe7 7. a5 Qc7 8.
a6 Kh7 9. Qd2 Be8 10. Qe3 Bd7 11. Re1 $11 {0.23/25}) (1... a6 2. Bxc6 (2. Bxe7
Nxe7 3. Bxd7+ Qxd7 4. Qc1 O-O 5. h4 Rfb8 6. h5 h6 7. g4 b5 8. axb5 Qxb5 9. Kh2
a5 10. Rg1 Kh7 11. Nh4 $11 {0.00/35}) 2... Nxc6 3. h3 h6 4. Bh4 Na7 5. Re1 Qa5
6. Qd2 Nc8 7. Nh2 Nb6 8. f4 Nxa4 9. Re3 Bb5 10. Rg3 Rg8 11. Bf6 $16 {0.74/29})
2. Bc1 (2. Bxc6 Nxc6 3. Bc1 Qa5 4. Ba3 Qxc3 5. Bd6 Qa5 6. Qc1 Qd8 7. Qf4 g5 8.
Qd2 g4 9. Ne1 Ne7 10. Bxe7 Qxe7 11. g3 $11 {0.00/34}) 2... a5 3. Ba3 Na7 4.
Bxd7+ Qxd7 5. Qb1 O-O 6. Qb6 Nac8 7. Qb2 Ra6 8. h4 Re8 9. Rfb1 Rb6 10. Qc1 Rxb1
11. Qxb1 $11 {0.18/28}) (1. Bxc6 Bxc6 (1... Nxc6 2. Bc1 (2. h4 h6 3. h5 O-O-O
4. Bf4 Rdg8 5. Qc1 b6 6. Bg3 Be8 7. Bh4 Qd7 8. Nh2 f6 9. exf6 Bxh5 10. Qe3 Kb7
11. Rfb1 $14 {0.36/30}) 2... Qa5 3. Qd2 Ne7 4. Nh4 Bxa4 5. f4 Nf5 6. Nxf5 exf5
7. Qf2 Qa6 8. Qh4 Qc6 9. Ba3 Qe6 10. Bd6 Bc6 11. g4 $16 {0.90/30}) (1... Qxc6
2. a5 (2. Bg5 a5 3. Bxe7 Kxe7 4. Nh4 Qc8 5. g3 h6 6. f4 g6 7. Qc1 Kf8 8. g4 Ra6
9. Kh1 Ke8 10. Qa3 Kd8 11. Rab1 $11 {0.00/28}) 2... Qc7 3. Bc1 O-O 4. Ba3 Rfe8
5. h4 Nc6 6. Bd6 Qd8 7. g4 Nxa5 8. Qd2 Nc6 9. Kh2 Ne7 10. h5 Rf8 11. g5 $16 {
1.22/27}) 2. Nh4 (2. a5 b6 3. a6 Bb5 4. Bc1 Nf5 5. h3 Qc8 6. g4 Ne7 7. Nh4 Qd7
8. Kh2 O-O-O 9. f4 f5 10. Ba3 fxg4 11. hxg4 $11 {0.00/30}) 2... Ng6 3. Nxg6
hxg6 4. a5 b6 5. axb6 axb6 6. Rxa8+ Bxa8 7. Bg5 Bc6 8. h4 Qa7 9. Qc1 Qa2 10.
Qb1 Qxb1 11. Rxb1 $14 {0.57/31}) (1. h4 h6 (1... Qd8 2. h5 (2. Bc1 h6 3. Bxc6
Bxc6 4. a5 Bb5 5. h5 Qd7 6. Ba3 O-O-O 7. Nh2 Kb8 8. Bc5 Ka8 9. Qf3 Qe8 10. Ng4
Rg8 11. Rfe1 $14 {0.57/31}) 2... h6 3. Bc1 Nf5 4. Bxc6 Bxc6 5. a5 b6 6. g4 Ne7
7. axb6 axb6 8. Ba3 Nc8 9. Ne1 g6 10. hxg6 fxg6 11. f4 $16 {0.93/26}) (1... O-O
2. Bc1 (2. Bxc6 Nxc6 3. Bc1 f6 4. Ba3 Rf7 5. Bd6 Qd8 6. exf6 gxf6 7. Nh2 Rg7 8.
Re1 Ne7 9. Bf4 Nf5 10. h5 Qe7 11. g4 $16 {0.84/27}) 2... f6 3. Bxc6 Nxc6 4. Ba3
Rf7 5. Bd6 Qd8 6. exf6 Rxf6 7. Re1 Rf5 8. g4 Rf6 9. Bg3 h5 10. gxh5 Qe7 11. Ne5
$16 {0.92/27}) 2. Bc1 (2. h5 a6 3. Bxc6 Bxc6 4. a5 O-O-O 5. Bc1 Rdg8 6. Nh4 Be8
7. Ba3 Nc6 8. Rb1 Na7 9. Rb6 Qd8 10. Qg4 Nb5 11. Bd6 $11 {0.00/30}) 2... Qa5 3.
h5 g6 4. hxg6 Nxg6 5. Ba3 O-O-O 6. Kh1 Rdg8 7. Bc5 Rg7 8. Qb1 Qd8 9. Bxc6 Bxc6
10. Bxa7 Nh4 11. Bb6 $11 {0.00/41}) (1. Nh4 Qa5 (1... O-O 2. Qg4 (2. Bxc6 Nxc6
3. Bg5 h6 4. Bc1 Qa5 5. Qf3 Ne7 6. Qg3 Kh7 7. Ba3 Rfe8 8. f4 Nf5 9. Nxf5 exf5
10. Rfb1 Rg8 11. Bd6 $14 {0.59/30}) 2... Rfb8 3. Bg5 Qd8 4. f4 Qf8 5. Rf3 a6 6.
Bxc6 bxc6 7. Bxe7 Qxe7 8. f5 exf5 9. Nxf5 Bxf5 10. Rxf5 Ra7 11. Raf1 $16 {
1.14/25}) (1... Ng6 2. Nxg6 (2. Nf3 Qd8 3. Bg5 Nge7 4. Bxc6 Bxc6 5. a5 h6 6.
Bc1 b6 7. a6 Bb5 8. Nh4 Ng6 9. Nxg6 fxg6 10. Qg4 Kf7 11. h4 $11 {0.09/31}) 2...
hxg6 3. Re1 b6 4. Bg5 a5 5. Ra2 Rb8 6. Qf3 Na7 7. Bxd7+ Qxd7 8. g3 b5 9. axb5
Rxb5 10. h4 Nc6 11. Kg2 $16 {0.91/28}) 2. Qg4 (2. Bxc6 Bxc6 3. Qg4 Ng6 4. f4 h5
5. Qg5 Nxh4 6. Qxh4 Qd8 7. Qf2 g6 8. a5 b6 9. axb6 axb6 10. Rfb1 Kd7 11. Rxa8
$14 {0.58/33}) 2... Ng6 3. Bxc6 Bxc6 4. f4 h5 5. Qg3 Nxh4 6. Qxh4 Qd8 7. Qe1 f5
8. exf6 Qxf6 9. Rf3 O-O 10. Bc1 Rf7 11. a5 $14 {0.61/24}) 1... b6 (1... O-O 2.
Bxc6 (2. h4 f6 3. Ba3 Rf7 4. h5 Be8 5. Bd6 Qd8 6. Qe2 a6 7. Bxc6 Nxc6 8. h6 Rd7
9. Qe3 Bh5 10. Nh2 f5 11. Qg3 $14 {0.58/26}) 2... Bxc6 3. a5 h6 4. h4 b6 5. a6
Be8 6. Qe2 Kh7 7. h5 f6 8. exf6 Rxf6 9. Ne5 Nc6 10. f4 Nxe5 11. dxe5 $14 {
0.69/28}) (1... Qa5 2. Qd2 (2. h4 Nf5 3. h5 a6 4. Bxc6 Bxc6 5. Qd2 Bxa4 6. g4
Ne7 7. Nh2 Nc6 8. f4 O-O-O 9. f5 Rd7 10. Nf3 Nd8 11. Ba3 $14 {0.65/26}) 2... a6
3. Bb2 h6 4. Rfb1 Qc7 5. Bxc6 Bxc6 6. a5 Bb5 7. Ba3 Nc6 8. Bd6 Qd8 9. Bb4 Rc8
10. Qe3 Kd7 11. h3 $16 {0.80/27}) 2. h4 (2. Ba3 a6 3. Bxc6 Bxc6 4. Qd2 h6 5.
Ra2 a5 6. g3 Nc8 7. Rfa1 Qd8 8. Qf4 Bxa4 9. Rb1 Bc6 10. Qg4 Rg8 11. Qh5 $15 {
-0.40/32}) 2... a6 3. Bxc6 Bxc6 4. h5 a5 5. Ba3 Nf5 6. g4 Nh6 7. Nh2 Bxa4 8.
Bc1 b5 9. f4 Ra6 10. Qe2 Qd7 11. Rf2 $14 {0.69/26} *

[Event "Lc0"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "New game"]
[Black "?"]
[Result "*"]
[SetUp "1"]
[FEN "r3k2r/ppqbnppp/2n1p3/1B1pP3/P1pP4/2P2N2/2PB1PPP/R2Q1RK1 w kq - 0 1"]
[PlyCount "21"]

{[#]} 1. Bxc6 (1. Bc1 a6 (1... Nf5 2. h3 (2. Qd2 a6 3. Bxc6 Bxc6 4. a5 O-O-O 5.
Ba3 h5 6. Bc5 Rdg8 7. Rfb1 Bb5 8. g3 f6 9. h3 Rh6 10. Qe2 Kb8 11. Nh4 $16 {
0.78/14}) 2... a6 3. Bxc6 Bxc6 4. a5 O-O-O 5. Ba3 Rdg8 6. Qd2 h6 7. Rfb1 Bb5 8.
Qc1 h5 9. Bc5 Rh6 10. g3 Kb8 11. g4 $16 {0.81/16}) (1... a5 2. Ba3 (2. Ng5 f6
3. exf6 gxf6 4. Nh3 O-O-O 5. Nf4 Rhg8 6. g3 e5 7. dxe5 Nxe5 8. Qe2 h5 9. Bxd7+
Qxd7 10. Rb1 Qf5 11. f3 $14 {0.57/16}) 2... O-O 3. Qe2 f6 4. Rfe1 Rae8 5. h4
Rf7 6. h5 h6 7. exf6 Rxf6 8. Rab1 Rf4 9. g3 Re4 10. Qf1 Rxe1 11. Qxe1 $16 {
0.82/14}) 2. Bxc6 Bxc6 3. a5 O-O-O 4. Ba3 Rdg8 5. Rb1 Qd7 6. g3 Kb8 7. Ng5 Nf5
8. h4 h6 9. Nh3 Ka8 10. h5 g6 11. Nf4 $11 {0.18/14}) (1. Ng5 h6 (1... O-O 2.
Bxc6 (2. Qg4 f5 3. exf6 Rxf6 4. Qg3 Qc8 5. Bxc6 Nxc6 6. a5 Rg6 7. Qd6 Qe8 8. f4
Rd8 9. Qc5 Rf6 10. Nf3 h6 11. h3 $14 {0.57/11}) 2... Nxc6 3. Qg4 Qd8 4. Nf3 f6
5. exf6 Qxf6 6. a5 Rae8 7. Rfe1 h6 8. Qg3 Kh7 9. h3 Qf5 10. Ra2 Rf7 11. Bc1 $14
{0.67/14}) (1... Nd8 2. Nh3 (2. Qh5 h6 3. Nh3 Ndc6 4. Bc1 Qa5 5. Bxc6 Bxc6 6.
Ng5 g6 7. Qh3 Nf5 8. g4 Ng7 9. Qh4 Qd8 10. a5 Kd7 11. Nf3 $16 {0.72/15}) 2...
O-O 3. Bxd7 Qxd7 4. Bg5 h6 5. Bxe7 Qxe7 6. Nf4 Rb8 7. Qe2 g6 8. g3 Nc6 9. h4
Rbe8 10. Kg2 Qd7 11. Nh3 $16 {0.92/10}) 2. Qh5 (2. Bxc6 Bxc6 3. Nh3 a5 4. Bc1
O-O-O 5. Qf3 Be8 6. Ba3 b6 7. Qg4 Rh7 8. Nf4 Kb7 9. Bd6 Qd7 10. Ba3 Nc8 11.
Rfb1 $11 {0.00/11}) 2... O-O 3. Bxc6 Bxc6 4. Nf3 a5 5. Rfe1 Be8 6. Qh3 Ng6 7.
Qg4 Ra6 8. h4 f5 9. exf6 Rxf6 10. Bc1 Nf8 11. Qh3 $14 {0.39/12}) (1. h4 a6 (
1... h6 2. Bc1 (2. h5 O-O-O 3. Nh2 Qa5 4. Ng4 Kb8 5. Qf3 Nxd4 6. cxd4 Qxd2 7.
Qxf7 Nc6 8. Qxg7 Rdg8 9. Qxd7 Rxg4 10. Qd6+ Ka8 11. Bxc6 $11 {0.29/23}) 2... f5
3. Ba3 a6 4. Bd6 Qd8 5. Bxc6 Bxc6 6. a5 O-O 7. g3 Rf7 8. Kg2 b5 9. axb6 Qxb6
10. Rh1 Qd8 11. Ng1 $16 {0.80/11}) (1... O-O-O 2. Bc1 (2. Bxc6 Nxc6 3. Nh2 Rdf8
4. Qf3 Qd8 5. Bg5 f6 6. exf6 gxf6 7. Bh6 Rfg8 8. Rfb1 Rg6 9. Bf4 Ne7 10. Nf1
Nf5 11. Ne3 $14 {0.37/14}) 2... h6 3. Ba3 Rdg8 4. Bc5 Qd8 5. Qb1 g5 6. h5 g4 7.
Nh4 Rg5 8. Qb2 Rxh5 9. Rfb1 Na5 10. g3 b6 11. Qa3 $16 {0.75/17}) 2. Bxc6 Bxc6
3. Bc1 O-O-O 4. a5 Rdg8 5. Re1 h6 6. Nh2 f5 7. exf6 gxf6 8. Qf3 Nf5 9. Qh3 Qf7
10. h5 Ng7 11. Ng4 $11 {0.21/18}) (1. Nh4 Ng6 (1... Qa5 2. Bxc6 (2. Qg4 Rg8 3.
Bxc6 Bxc6 4. Qh5 Rh8 5. Qg5 Ng6 6. Qg3 O-O-O 7. Nxg6 hxg6 8. Ra2 Bxa4 9. h3 Rd7
10. Bc1 b5 11. Qe3 $14 {0.41/8}) 2... Nxc6 3. f4 Ne7 4. f5 Nxf5 5. Qh5 h6 6. g4
Nxh4 7. Qxf7+ Kd8 8. Qxg7 Kc7 9. Rf7 Rad8 10. Kf2 Rhe8 11. Qf6 $16 {1.04/19}) (
1... O-O-O 2. Qf3 (2. Qh5 Rdf8 3. Bc1 Qa5 4. Qf3 Rhg8 5. g3 f5 6. exf6 Rxf6 7.
Qe3 Qc7 8. Bxc6 Nxc6 9. Nf3 Rf5 10. Ne5 Nxe5 11. dxe5 $14 {0.33/14}) 2... Rdg8
3. g3 Be8 4. Bc1 g5 5. Ng2 h5 6. Qf6 Qa5 7. Ne3 Rh7 8. Ba3 Rg6 9. Qf3 Qd8 10.
Rfb1 Na5 11. Bxe8 $14 {0.43/14}) 2. Qg4 (2. Bxc6 Bxc6 3. Qg4 a5 4. Bc1 Kd7 5.
Ba3 Kc8 6. Rfb1 Ra6 7. Nf3 Bxa4 8. Ra2 f5 9. Qg3 Qd7 10. h4 Rg8 11. Kf1 $11 {
-0.13/10}) 2... Nxh4 3. Bxc6 Nf5 4. Bxd7+ Qxd7 5. Qh5 O-O 6. g4 Ne7 7. Bg5 Kh8
8. f4 f6 9. exf6 gxf6 10. Bh6 Rg8 11. Qf7 $11 {0.14/17}) 1... Nxc6 (1... Bxc6
2. a5 (2. Bc1 a5 3. Ba3 Qd7 4. Nh4 Bxa4 5. g3 Kd8 6. Ra2 Bc6 7. Qg4 Ng6 8. Rb2
b5 9. f4 Nxh4 10. Qxh4+ Kc8 11. Qg5 $11 {0.01/10}) 2... O-O-O 3. Bc1 Rdg8 4.
Ba3 g5 5. Qc1 h6 6. Bc5 a6 7. Qa3 Nf5 8. g4 Ng7 9. Bd6 Qd7 10. Be7 Ne8 11. h3
$16 {1.14/11}) (1... Qxc6 2. a5 (2. Bc1 a5 3. Nh4 Qc7 4. Qg4 Ng6 5. Qg3 Kd8 6.
Nf3 Kc8 7. h4 h5 8. Ba3 Bxa4 9. Ra2 Qd8 10. Rb2 Bc6 11. Rfb1 $11 {0.30/9}) 2...
Qc7 3. Bc1 f5 4. Ng5 O-O 5. h4 h6 6. Nh3 Kh7 7. Ba3 Be8 8. Nf4 Bf7 9. Bd6 Qd7
10. Qf3 b6 11. h5 $16 {1.49/11}) 2. Bc1 (2. Ng5 h6 3. Qh5 O-O-O 4. Nxf7 Be8 5.
Nd6+ Qxd6 6. Qxe8 Rhxe8 7. exd6 Rxd6 8. h4 h5 9. Bg5 Rd7 10. Rfe1 Rf7 11. Re3
$11 {0.15/9}) 2... h6 3. Ba3 a5 4. Nh4 Ne7 5. g3 Bc6 6. Ng2 Qd7 7. Nf4 Kd8 8.
Nh5 Rg8 9. Qg4 Ng6 10. Rfb1 Bxa4 11. Rb6 $16 {0.88/13} *
Professing themselves to be wise, they became fools,
Take on me. foes 0

User avatar
Ovyron
Posts: 2325
Joined: Tue Jul 03, 2007 2:30 am

Re: Nh4 is hard to find, but the stats will tell you it is right

Post by Ovyron » Mon Aug 26, 2019 10:43 am

zullil wrote:
Sat Aug 24, 2019 11:25 am
Sure, we could "evaluate" any position using game play. Play Nh4, and then let two engines play a long match, and then do the same with Bc1 (and then all other root moves). Sure, such a scheme could be automated, streamlined and made statistically sensible. But then the "best" move would still be dependent on the "rollout" system you've chosen, i.e, on the engines, the time controls, etc.
No, the best move is the one that beats the strongest thing available. So unless you claim that the strongest thing available plays already perfect chess, just because you don't have access to it, it doesn't mean it doesn't exist.

And the move that beats it the fastest is the best (without quotes.)

Putting quotes around the word is admitting you don't know, not showing it doesn't exist. If you can't produce a string of moves that beats your Depth 56, you need to try harder.
Make someone happy today.

zullil
Posts: 5552
Joined: Mon Jan 08, 2007 11:31 pm
Location: PA USA
Full name: Louis Zulli

Re: Nh4 is hard to find, but the stats will tell you it is right

Post by zullil » Mon Aug 26, 2019 11:05 am

Ovyron wrote:
Mon Aug 26, 2019 10:43 am
zullil wrote:
Sat Aug 24, 2019 11:25 am
Sure, we could "evaluate" any position using game play. Play Nh4, and then let two engines play a long match, and then do the same with Bc1 (and then all other root moves). Sure, such a scheme could be automated, streamlined and made statistically sensible. But then the "best" move would still be dependent on the "rollout" system you've chosen, i.e, on the engines, the time controls, etc.
No, the best move is the one that beats the strongest thing available. So unless you claim that the strongest thing available plays already perfect chess, just because you don't have access to it, it doesn't mean it doesn't exist.

And the move that beats it the fastest is the best (without quotes.)

Putting quotes around the word is admitting you don't know, not showing it doesn't exist. If you can't produce a string of moves that beats your Depth 56, you need to try harder.
Your definition of "best" is not one.

Dann Corbit
Posts: 9894
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: Nh4 is hard to find, but the stats will tell you it is right

Post by Dann Corbit » Mon Aug 26, 2019 5:55 pm



Some say the best move here is 1.e4
Some say the best move here is 1.d4
Some say the best move here is 1.c4
Some say the best move here is something else.
Chances are very good with perfect play all of the first 3 end in a draw.
So which one is best?

I am of the opinion that "best move" is very subjective until the game is driven all the way to 1-0, 1/2, 0-1 from the given position and even then it is not definite if any pruning beyond alpha-beta has been applied.

But there is nothing wrong with calling something the best move when what we really mean is, "I think this move is best."
On the other hand, let's not be dogmatic and say that our choice really is (in truth) best because we don't know something is factual until it is proven.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

User avatar
Ovyron
Posts: 2325
Joined: Tue Jul 03, 2007 2:30 am

Re: Nh4 is hard to find, but the stats will tell you it is right

Post by Ovyron » Wed Aug 28, 2019 4:26 am

Dann Corbit wrote:
Mon Aug 26, 2019 5:55 pm
Chances are very good with perfect play all of the first 3 end in a draw.
So which one is best?
My claim is perfect play is irrelevant in these discussion, because nobody can play perfectly.

All these discussions don't make any sense in isolation, they only make sense when black is actually someone, and a game is being played.

Then, this someone will not play perfectly, so there's a best move (1.e4/1.d4/1.c4/other move) that produces the shortest string of moves that beat this player. This is objective.

And 1.f3 is just like that, 1.f3 is a draw with perfect players, but it doesn't get lumped with 1.e4/1.d4/1.c4 because it's objectively worse, so the same method used to compare 1.f3 can be used to figure out which one from 1.e4/1.d4/1.c4 is worse than the others, and then what is the best from the remaining two. And it'll be objectively the best move. And the same procedure can be used in the OP to figure out if Nh4 is objectively best or not.

There's just no incentive to do it.
Make someone happy today.

mmt
Posts: 8
Joined: Sun Aug 25, 2019 6:33 am
Full name: Jim Fields

Re: Nh4 is hard to find, but the stats will tell you it is right

Post by mmt » Tue Sep 10, 2019 7:40 am

After over 1 billion nodes total, LC0:
61.1% Ng5 Nf5 Nh3 O-O-O Nf4 g6 Bxc6 Bxc6 a5 h6 g4 Ng7 Ng2 Rh7 h4 a6 f3 Rg8 Qc1 Rgh8 Qe1 Rg8 Nf4 Qd8 Nh3 Ne8 Bc1 Nc7 Ba3 Ne8 Bb4 Nc7 Kg2 Nb5 Bc5 Nc7 Nf4 Ne8 (N: 212M, P: 6.42%)

60.5% Bxc6 Bxc6 a5 O-O-O Nh4 Rdg8 f4 Qd7 Qe2 Re8 h3 Bb5 Qh5 Ba4 Ra2 g6 Qe2 Bb5 Nf3 a6 Qf2 Qd8 Rfa1 Reg8 Bc1 Nf5 Nh2 h5 Nf3 h4 Ng5 Rg7 Qd2 Rhg8 Ba3 Qxa5 Bf8 Qc7 Bxg7 (N: 300M, P: 21.01%)

60.4% Bc1 a6 Bxc6 Bxc6 a5 h6 Ba3 O-O-O Bc5 Qd7 Qb1 Kb8 g3 Ka8 Qb4 Rde8 Nh4 Rhg8 Rae1 h5 Ng2 Bb5 Bxe7 Rxe7 Nf4 Rh8 h4 Qd8 Kg2 Rc7 Rh1 Qe8 Kf3 Ba4 Re2 Bb5 Ke3 Qd8 Kd2 Bd7 Ree1 Rc6 Rb1 Bc8 f3 Rh6 Qa3 (N: 675M, P: 16.17%)

zullil
Posts: 5552
Joined: Mon Jan 08, 2007 11:31 pm
Location: PA USA
Full name: Louis Zulli

Re: Nh4 is hard to find, but the stats will tell you it is right

Post by zullil » Tue Sep 10, 2019 9:41 am

mmt wrote:
Tue Sep 10, 2019 7:40 am
After over 1 billion nodes total, LC0:
61.1% Ng5 Nf5 Nh3 O-O-O Nf4 g6 Bxc6 Bxc6 a5 h6 g4 Ng7 Ng2 Rh7 h4 a6 f3 Rg8 Qc1 Rgh8 Qe1 Rg8 Nf4 Qd8 Nh3 Ne8 Bc1 Nc7 Ba3 Ne8 Bb4 Nc7 Kg2 Nb5 Bc5 Nc7 Nf4 Ne8 (N: 212M, P: 6.42%)

60.5% Bxc6 Bxc6 a5 O-O-O Nh4 Rdg8 f4 Qd7 Qe2 Re8 h3 Bb5 Qh5 Ba4 Ra2 g6 Qe2 Bb5 Nf3 a6 Qf2 Qd8 Rfa1 Reg8 Bc1 Nf5 Nh2 h5 Nf3 h4 Ng5 Rg7 Qd2 Rhg8 Ba3 Qxa5 Bf8 Qc7 Bxg7 (N: 300M, P: 21.01%)

60.4% Bc1 a6 Bxc6 Bxc6 a5 h6 Ba3 O-O-O Bc5 Qd7 Qb1 Kb8 g3 Ka8 Qb4 Rde8 Nh4 Rhg8 Rae1 h5 Ng2 Bb5 Bxe7 Rxe7 Nf4 Rh8 h4 Qd8 Kg2 Rc7 Rh1 Qe8 Kf3 Ba4 Re2 Bb5 Ke3 Qd8 Kd2 Bd7 Ree1 Rc6 Rb1 Bc8 f3 Rh6 Qa3 (N: 675M, P: 16.17%)
So it's not really clear what is "best', but Nh4 isn't in Lc0's top three. What network did you use? You must have quite a bit of RAM!

mmt
Posts: 8
Joined: Sun Aug 25, 2019 6:33 am
Full name: Jim Fields

Re: Nh4 is hard to find, but the stats will tell you it is right

Post by mmt » Tue Sep 10, 2019 12:13 pm

zullil wrote:
Tue Sep 10, 2019 9:41 am
So it's not really clear what is "best', but Nh4 isn't in Lc0's top three. What network did you use? You must have quite a bit of RAM!
Network 42850. 512 GB :)

User avatar
Ovyron
Posts: 2325
Joined: Tue Jul 03, 2007 2:30 am

Re: Nh4 is hard to find, but the stats will tell you it is right

Post by Ovyron » Wed Sep 11, 2019 6:18 am

You know, in another thread lkaufman asked about what was the best NN for a CPU powered Leela, and Ferdy came up with this very interesting graph...

Image

What is interesting about it? Well, that that's just how chess moves behave!

So you can see in the graph how different networks perform at different time controls. My claim is that chess moves, and their quality, can be measured in the same way NNs were tested here.

There would be a chess move, say, the red line, that would be THE BEST move, but at fast time controls engines don't know how to play it, so it performs worse than the other 2. As you improve the level, the engines catch on about how to play it, and it rises to second best, and with a time control long enough, the engines know what to do and it's clearly best. It'd take at least some 5 +4 time control to see how the move initially preferred at blitz is actually third best.

My claim is that beyond this level the blue line will never be again over the red line, which makes the BEST move the red one.

I know this is about Leela Networks and not chess moves, but they behave very similarly.

If someone wants to know if Nh4 is truly the best on this position, or if it's 4th best or worse, all they'd need to do is running this test, but instead of different networks they'd have the same chess entity try the different moves, and compare its performance with them as time control becomes slower. There will be a point at which it'll be clear the BEST move will remain performing better than the others, and it'll be enough to know the BEST move objectively.
Make someone happy today.

Post Reply