LCZero update

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

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Locked
Uri Blass
Posts: 8765
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Re: LCZero update

Post by Uri Blass » Fri Mar 23, 2018 5:31 pm

koedem wrote:I just played against someone who ran it on decently fast hardware (he said on average 10k playouts per move) and it actually played some nice chess. https://lichess.org/G7brl6KQ/white#0
Not only was it quite an effort to win, also the game looked very human like. Very impressive. :)
(but yeah, you need a decent number of playouts for it to play good chess)

Looking at the game I can easily detect it is a computer.

No weak human at the level of that program(not a beginner) is going to miss Bxf8 and play 12...Rb8

Weak humans who are not beginners may miss captures but no chance to miss Bxf8 after the opponent played Bh6.

The captures they have chances to miss are indirect threats or capture when they leave the defence of some pawn(like playing Bc1-f4 and missing Qxb2) or some long moves(Bh6xf8 is too short).

koedem
Posts: 91
Joined: Fri Mar 18, 2016 9:45 pm

Re: LCZero update

Post by koedem » Fri Mar 23, 2018 10:18 pm

Uri Blass wrote:
koedem wrote:I just played against someone who ran it on decently fast hardware (he said on average 10k playouts per move) and it actually played some nice chess. https://lichess.org/G7brl6KQ/white#0
Not only was it quite an effort to win, also the game looked very human like. Very impressive. :)
(but yeah, you need a decent number of playouts for it to play good chess)

Looking at the game I can easily detect it is a computer.

No weak human at the level of that program(not a beginner) is going to miss Bxf8 and play 12...Rb8

Weak humans who are not beginners may miss captures but no chance to miss Bxf8 after the opponent played Bh6.

The captures they have chances to miss are indirect threats or capture when they leave the defence of some pawn(like playing Bc1-f4 and missing Qxb2) or some long moves(Bh6xf8 is too short).
I can easily see a human being scared of Re8 Ng5 there and deciding it's a good practical decision to sac the exchange. (it's incorrect of course but I definitely know humans who totally would play that)

Nay Lin Tun
Posts: 599
Joined: Mon Jan 16, 2012 5:34 am

Re: LCZero update

Post by Nay Lin Tun » Sat Mar 24, 2018 2:38 am

Uri Blass wrote:
koedem wrote:I just played against someone who ran it on decently fast hardware (he said on average 10k playouts per move) and it actually played some nice chess. https://lichess.org/G7brl6KQ/white#0
Not only was it quite an effort to win, also the game looked very human like. Very impressive. :)
(but yeah, you need a decent number of playouts for it to play good chess)

Looking at the game I can easily detect it is a computer.

No weak human at the level of that program(not a beginner) is going to miss Bxf8 and play 12...Rb8

Weak humans who are not beginners may miss captures but no chance to miss Bxf8 after the opponent played Bh6.

The captures they have chances to miss are indirect threats or capture when they leave the defence of some pawn(like playing Bc1-f4 and missing Qxb2) or some long moves(Bh6xf8 is too short).
on move 12. Black cant avoid exchange. It is forced. If black play Re8, white will go Ng4 and Nxf7 is unstoppable. It is just 3 ply tactics that you missed.
It is strong human play. If I am in short of time like bullet/blitz, I wont consider moving rook at that position because, white bishop is position ally superior. And in reverse, if I were a white player, i will consider saving that powerful bishop first, search for possible mate threat and pressure , and only that options are failed, I will simply exchange vs stacked rook for material advantage.

https://ibb.co/nzLOUn

Uri Blass
Posts: 8765
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Re: LCZero update

Post by Uri Blass » Sat Mar 24, 2018 5:21 am

Nay Lin Tun wrote:
Uri Blass wrote:
koedem wrote:I just played against someone who ran it on decently fast hardware (he said on average 10k playouts per move) and it actually played some nice chess. https://lichess.org/G7brl6KQ/white#0
Not only was it quite an effort to win, also the game looked very human like. Very impressive. :)
(but yeah, you need a decent number of playouts for it to play good chess)

Looking at the game I can easily detect it is a computer.

No weak human at the level of that program(not a beginner) is going to miss Bxf8 and play 12...Rb8

Weak humans who are not beginners may miss captures but no chance to miss Bxf8 after the opponent played Bh6.

The captures they have chances to miss are indirect threats or capture when they leave the defence of some pawn(like playing Bc1-f4 and missing Qxb2) or some long moves(Bh6xf8 is too short).
on move 12. Black cant avoid exchange. It is forced. If black play Re8, white will go Ng4 and Nxf7 is unstoppable. It is just 3 ply tactics that you missed.
It is strong human play. If I am in short of time like bullet/blitz, I wont consider moving rook at that position because, white bishop is position ally superior. And in reverse, if I were a white player, i will consider saving that powerful bishop first, search for possible mate threat and pressure , and only that options are failed, I will simply exchange vs stacked rook for material advantage.

https://ibb.co/nzLOUn
You mean Ng5(Ng4 is not legal)

I can easily see Re8 Ng5 d5 but I agree that weak humans may miss d5 in their calculations.

In any case I think the logical human move if you miss d5 is Nc5 to prepare Ne6 or Be6.

Jhoravi
Posts: 255
Joined: Wed May 08, 2013 4:49 am

Re: LCZero update

Post by Jhoravi » Sat Mar 24, 2018 6:36 am

Hi,

I suggest a slight improvement on the Slow mode. Please allow us to make a move reply immediately during the 10 second pause. Then it automatically send our move to the server after 10 seconds. It fixes the unresponsive feel of the page while the result is just the same.

thanx

koedem
Posts: 91
Joined: Fri Mar 18, 2016 9:45 pm

Re: LCZero update

Post by koedem » Sat Mar 24, 2018 11:33 am

The slow mode got bug fixed and actually let's leela think for longer. (previously slow and fast mode gave leela same amount of time per move)
So it plays noticably stronger now on slow mode although not quite as strong as in the game I played on lichess. (imo)

Werewolf
Posts: 1272
Joined: Thu Sep 18, 2008 8:24 pm

Re: LCZero update

Post by Werewolf » Sat Mar 24, 2018 11:35 am

I'm currently running a match between LCZ and iPhone Magnus aged 10. It'll be interesting to see how they get on.

Incredible achievement you've made.

JJJ
Posts: 1304
Joined: Sat Apr 19, 2014 11:47 am

Re: LCZero update

Post by JJJ » Sat Mar 24, 2018 12:35 pm

[pgn]1. d4 d5
2. c4 dxc4
3. e3 e5
4. Bxc4 exd4
5. Qb3 Be6
6. Bxe6 fxe6
7. Qxe6+ Ne7
8. Nf3 Nc6
9. O-O Qd6
10. Qg4 h5
11. Qe4 dxe3
12. Bxe3 Rd8
13. Nc3 Qb4
14. Qc2 Qc4
15. Rad1 Nb4
16. Rxd8+ Kxd8
17. Qd2+ Ke8
18. a3 Nd3
19. Rd1 Rh7
20. Qxd3 Qxd3
21. Rxd3 a6
22. Nd5 Nxd5
23. Rxd5 Be7
24. Bg5 Bf8
25. Ne5 Be7
26. Bxe7 Kxe7
27. f4 h4
28. Rd7+ Ke6
29. Rxc7 Kf5
30. g3 b5
31. Ra7 hxg3
32. hxg3 Rh6
33. Kg2 g5
34. Nf7 Re6
35. Nxg5 Rd6
36. b4 Rd2+
37. Kf3 Rd3+
38. Ke2 Rxa3
39. Rf7+ Kg6
40. Rd7 Ra4
41. Rd6+ Kf5
42. Nf3 Rxb4
43. Rxa6 Rc4
44. Ra5 Rc5
45. Kd3 Kg4
46. Nd4 Kxg3
47. f5 Kf4
48. f6 Rc8
49. Rxb5 Rf8
50. Rb6 Ke5
51. Nf3+ Kd5
52. Ng5 Kc5
53. Ra6 Kb5
54. Re6 Kc5
55. f7 Kd5
56. Re8 Rxf7
57. Nxf7 Kc6
58. Re5 Kb7
59. Re6 Kc7
60. Kd4 Kd7
61. Ra6 Ke7
62. Ne5 Kd8
63. Ra7 Kc8
64. Kc5 Kb8
65. Rg7 Kc8
66. Kc6 Kb8
67. Rf7 Ka8
68. Kb6 Kb8
69. Rf8# [/pgn]

My game on http://play.lczero.org/ at 200ms per move for leela.

I m not a strong player, I m around 1500 / 1600 at blitz at chess.com. And of course I lost many games before winning :D

JJJ
Posts: 1304
Joined: Sat Apr 19, 2014 11:47 am

Re: LCZero update

Post by JJJ » Sat Mar 24, 2018 1:28 pm

And another one :p


[pgn]1. d4 d5
2. c4 dxc4
3. e3 Nf6
4. Bxc4 e6
5. Nf3 Be7
6. O-O O-O
7. a4 c5
8. a5 Nc6
9. Nc3 cxd4
10. exd4 Nxa5
11. Be2 a6
12. Qa4 b5
13. Qxa5 Qxa5
14. Rxa5 Bb7
15. Ra1 b4
16. Na4 Bd5
17. Nc5 a5
18. Bg5 Rfc8
19. Rfc1 a4
20. Nd7 Nxd7
21. Bxe7 Rxc1+
22. Rxc1 a3
23. bxa3 bxa3
24. Ra1 a2
25. Ne5 Nxe5
26. dxe5 Bb3
27. f4 g5
28. Bxg5 Bd5
29. Bf3 Bxf3
30. gxf3 Ra3
31. Kf2 Ra6
32. h4 Ra5
33. Ke2 h6
34. Bxh6 Kh7
35. Bg5 Kg6
36. Kd2 Ra6
37. Kc2 Kh5
38. Kb2 Kg6
39. Rxa2 Rxa2+
40. Kxa2 Kh5
41. Kb3 Kg6
42. Kc4 Kh5
43. Kc5 Kg6
44. Kd6 Kf5
45. Ke7 Kg6
46. Bf6 Kf5
47. Kxf7 Kxf4
48. Kxe6 Kxf3
49. h5 Ke3
50. h6 Kf3
51. h7 Kg4
52. h8=Q Kf3
53. Kd6 Ke4
54. e6 Kf5
55. Be5 Kg6
56. Qf6+ Kh7
57. Qg7# [/pgn]

And this time he missed mate in one

[pgn]1. d4 Nf6
2. Nf3 d5
3. c4 dxc4
4. e3 Ng8
5. Bxc4 e6
6. O-O Nf6
7. Nc3 Be7
8. e4 O-O
9. Bg5 h6
10. Bxf6 Bxf6
11. e5 Be7
12. Qe2 Nc6
13. Rad1 a6
14. Bd3 Nxd4
15. Qe4 c5
16. Qh7# [/pgn]

JJJ
Posts: 1304
Joined: Sat Apr 19, 2014 11:47 am

Re: LCZero update

Post by JJJ » Sat Mar 24, 2018 1:38 pm

Maybe I should try the slow mode after all

[pgn]1. d4 d5
2. c4 dxc4
3. e3 Nf6
4. Bxc4 e6
5. Nf3 Be7
6. O-O O-O
7. Nc3 c5
8. d5 exd5
9. Nxd5 Nxd5
10. Bxd5 Nd7
11. Be4 Nf6
12. Bc2 Qxd1
13. Rxd1 a5
14. b3 Rd8
15. Bb2 Nd5
16. Be4 Nf6
17. Rxd8+ Bxd8
18. Bxf6 Bxf6
19. Rd1 a4
20. Ne5 Bxe5
21. Rd8# [/pgn]

Locked