LCZERO GPU Id 211 A beautiful positional game

[Robert Flesher]

Hello folks, here is a game of LCZERO GPU Id 211 vs an old favourite of mine Gandalf 4.32. This is some amazingly, stunning chess! How many engine(s) play G4!!... and see they are winning!?

[d] 1r1n1rk1/pb2n1pp/1p2p3/3pPp2/2pP1P1N/P1P1BB2/2PK2PP/R6R w - - 0 21



Here is the full game. Enjoy!

[pgn] [Event "My Tournament"]
[Site "?"]
[Date "2018.04.28"]
[Round "1"]
[White "LCZERO"]
[Black "gandalfu"]
[Result "1-0"]
[ECO "C16"]
[PlyCount "129"]
[EventDate "2018.??.??"]

1. e4 e6 2. d4 d5 3. Nc3 Bb4 4. e5 b6 5. Bd3 c5 6. a3 Bxc3+ 7. bxc3 Ne7 8. Qg4
c4 9. Be2 O-O 10. Nf3 Nbc6 11. Bg5 Qe8 12. Nh4 Kh8 13. Qh5 Bb7 14. Be3 f6 15.
Qxe8 Raxe8 16. f4 Kg8 17. Bh5 Rd8 18. Bg4 f5 19. Bf3 Rb8 20. Kd2 Nd8 21. g4
fxg4 22. Bxg4 Bc8 23. a4 Ndc6 24. Rhg1 Bd7 25. Raf1 g6 26. Nf3 Nf5 27. h4 Kh8
28. h5 gxh5 29. Bxf5 Rxf5 30. Nh4 Rf7 31. f5 exf5 32. Bg5 Rbf8 33. Ng2 Nd8 34.
Nf4 Ne6 35. Nxd5 Nxg5 36. Rxg5 Rg8 37. Rxg8+ Kxg8 38. Nf4 Bxa4 39. e6 Rc7 40.
Nd5 Rg7 41. Rxf5 h4 42. Nf6+ Kh8 43. d5 Rg2+ 44. Ke3 Rg3+ 45. Kf4 Rg1 46. d6
Kg7 47. d7 Kg6 48. d8=Q Rf1+ 49. Ke3 Kxf5 50. Ng8 Kg4 51. Nh6+ Kh5 52. Nf7 Rf5
53. Qg8 h6 54. Qg7 Bxc2 55. e7 Ba4 56. Qxh6+ Kg4 57. Qg6+ Kh3 58. Qxf5+ Kg2 59.
Qg4+ Kf1 60. Qf3+ Kg1 61. Qf2+ Kh1 62. Qxh4+ Kg2 63. Qg4+ Kh2 64. Kf2 Be8 65.
Qg2# 1-0 [/pgn]

[jdart]

I am not sure g4 is winning. Looks to me like Black should not have played 23. .. Ndc6. Maybe Nf7 instead? Seems to be =.

--Jon

[carldaman]

I am not sure g4 is winning. Looks to me like Black should not have played 23. .. Ndc6. Maybe Nf7 instead? Seems to be =.

--Jon

It could be argued that a move that induces an error from an opponent that can't cope with the resulting position is quite strong and effective, despite not being "winning" in a pure sense.

CL

[Robert Flesher]

I am not sure g4 is winning. Looks to me like Black should not have played 23. .. Ndc6. Maybe Nf7 instead? Seems to be =.

--Jon

Hi Jon, yes it may not be winning, although it also could be. This is why I gave the !? However, with that in mind, what I have been seeing is that when Lczero gives a large winning evaluation, other engines, even stockfish, are often clueless until a few moves later. Lczero is growing to be a very strange beast and plays very different chess. Three weeks ago Gandalf 4.32 was walking all over Lczero, now it does not stand a chance.

In this game, black is better! ( black out playing white in the opening) But Lczero played the exchange sack Rxf4! and wins in a crushing fashion :

[d] 1rq2k2/4ppb1/3p4/pb1P1p2/N3nN1r/1Pp1P2P/Q1P2PB1/R3K1R1 b Q - 0 24

The game :
[pgn] [Event "My Tournament"]
[Site "?"]
[Date "2018.04.29"]
[Round "1"]
[White "gandalfu"]
[Black "LCZERO"]
[Result "*"]
[ECO "A45"]
[PlyCount "80"]
[EventDate "2018.??.??"]

1. d4 Nf6 2. Bg5 c5 3. Bxf6 gxf6 4. d5 Bg7 5. Nc3 a6 6. e3 d6 7. Qh5 f5 8. Nge2
b5 9. g3 Nd7 10. Bg2 Nf6 11. Qh4 Kf8 12. a3 h5 13. Qf4 Bd7 14. h3 a5 15. a4
bxa4 16. Nc1 h4 17. gxh4 Rb8 18. Nd3 Qc8 19. Qc4 Ne4 20. Nxa4 Rxh4 21. Rg1 Bb5
22. Qa2 c4 23. Nf4 c3 24. b3 Rxf4 25. exf4 Bd4 26. Bxe4 fxe4 27. Rg4 e3 28.
fxe3 Bxe3 29. Qb1 e5 30. Qd1 exf4 31. Rh4 Qf5 32. Rh8+ Kg7 33. Qh5 Qxh5 34.
Rxh5 Kg6 35. Nxc3 Kxh5 36. Nxb5 Rxb5 37. Rd1 f3 38. Rd3 f2+ 39. Kf1 Bc5 40. Rd1
Rb8 {0-1} [/pgn]

[epanek]

@ move 10 this is already a position Leela plays well. Its very common in her games.

[peter]

Hi Robert!

Hello folks, here is a game of LCZERO GPU Id 211 vs an old favourite of mine Gandalf 4.32. This is some amazingly, stunning chess! How many engine(s) play G4!!... and see they are winning!?

[d] 1r1n1rk1/pb2n1pp/1p2p3/3pPp2/2pP1P1N/P1P1BB2/2PK2PP/R6R w - - 0 21

Reason that other engines don't play 21.g4 could well be, it simply isn't the best move here, I guess.

Here's CFish after Backward of the game from move 25...g6(?) where the evals start climbing, and after 21...g4 with full hash:


[d]1r1n1rk1/pb2n1pp/1p2p3/3pPp2/2pP1PPN/P1P1BB2/2PK3P/R6R b - - 0 1

Analysis by Cfish 270418 64 POP N:

21...fxg4 22.Bxg4 Bc8 23.Ng2 Nf7 24.h4 Nh6 25.Be2 Bd7 26.h5 Nef5 27.Bf2 Ne7 28.a4 Rf7 29.a5 b5 30.Be3 Rff8 31.Rhd1 Nhf5 32.Bf2 b4 33.Ne3 Nxe3 34.Bxe3 Nf5 35.Bf2 Nh6 36.Be3 Nf5
= (0.00) Depth: 54/32 00:07:58 11062MN

And here after 21.g3 instead:

[d]1r1n1rk1/pb2n1pp/1p2p3/3pPp2/2pP1P1N/P1P1BBP1/2PK3P/R6R b - - 0 1

Analysis by Cfish 270418 64 POP N:

21...Ndc6 22.a4 Kf7 23.Ke2 Rh8 24.Bc1 g6 25.Kf2 h6 26.Ba3 Rbg8 27.Ng2 Na5 28.Ne3 Bc6 29.Bb4 Bd7 30.Rhb1 Nac6 31.Bd6 Na5 32.Rg1 h5 33.Ng2 Rd8 34.Nh4 Ng8 35.Bb4 Re8 36.h3 Kg7 37.Rgb1 Bc6 38.Rd1 Nh6 39.Rh1 Bd7 40.Ng2 Ng8 41.Ne3 Ne7 42.Rhb1 Kf7 43.Ng2 Nec6 44.Ba3 Ne7 45.Ne3 Kg7 46.Bd6
+/- (0.74) Depth: 47/49 00:05:20 7386MN


That might not be an at once winning move neither, and there seem to be other moves not much worse, but why do you think 21.g4 so much better?

[Robert Flesher]

Hi Robert!

Hello folks, here is a game of LCZERO GPU Id 211 vs an old favourite of mine Gandalf 4.32. This is some amazingly, stunning chess! How many engine(s) play G4!!... and see they are winning!?

[d] 1r1n1rk1/pb2n1pp/1p2p3/3pPp2/2pP1P1N/P1P1BB2/2PK2PP/R6R w - - 0 21

Reason that other engines don't play 21.g4 could well be, it simply isn't the best move here, I guess.

Here's CFish after Backward of the game from move 25...g6(?) where the evals start climbing, and after 21...g4 with full hash:


[d]1r1n1rk1/pb2n1pp/1p2p3/3pPp2/2pP1PPN/P1P1BB2/2PK3P/R6R b - - 0 1

Analysis by Cfish 270418 64 POP N:

21...fxg4 22.Bxg4 Bc8 23.Ng2 Nf7 24.h4 Nh6 25.Be2 Bd7 26.h5 Nef5 27.Bf2 Ne7 28.a4 Rf7 29.a5 b5 30.Be3 Rff8 31.Rhd1 Nhf5 32.Bf2 b4 33.Ne3 Nxe3 34.Bxe3 Nf5 35.Bf2 Nh6 36.Be3 Nf5
= (0.00) Depth: 54/32 00:07:58 11062MN

And here after 21.g3 instead:

[d]1r1n1rk1/pb2n1pp/1p2p3/3pPp2/2pP1P1N/P1P1BBP1/2PK3P/R6R b - - 0 1

Analysis by Cfish 270418 64 POP N:

21...Ndc6 22.a4 Kf7 23.Ke2 Rh8 24.Bc1 g6 25.Kf2 h6 26.Ba3 Rbg8 27.Ng2 Na5 28.Ne3 Bc6 29.Bb4 Bd7 30.Rhb1 Nac6 31.Bd6 Na5 32.Rg1 h5 33.Ng2 Rd8 34.Nh4 Ng8 35.Bb4 Re8 36.h3 Kg7 37.Rgb1 Bc6 38.Rd1 Nh6 39.Rh1 Bd7 40.Ng2 Ng8 41.Ne3 Ne7 42.Rhb1 Kf7 43.Ng2 Nec6 44.Ba3 Ne7 45.Ne3 Kg7 46.Bd6
+/- (0.74) Depth: 47/49 00:05:20 7386MN


That might not be an at once winning move neither, and there seem to be other moves not much worse, but why do you think 21.g4 so much better?

It opens the game up, allowing the bishops to come alive, and in the game the pasted pawns were crushing. Perhaps trusting Stockfish in positions like this is an error. Remember, AZ made Stockfish look stupid in many such positions. Even now, Lczero plays many of these types of positions MUCH better. Once the tactics level out, well.....let's let time proof me right, or wrong

[Robert Flesher]

Hi Robert!

Hello folks, here is a game of LCZERO GPU Id 211 vs an old favourite of mine Gandalf 4.32. This is some amazingly, stunning chess! How many engine(s) play G4!!... and see they are winning!?

[d] 1r1n1rk1/pb2n1pp/1p2p3/3pPp2/2pP1P1N/P1P1BB2/2PK2PP/R6R w - - 0 21

Reason that other engines don't play 21.g4 could well be, it simply isn't the best move here, I guess.

Here's CFish after Backward of the game from move 25...g6(?) where the evals start climbing, and after 21...g4 with full hash:


[d]1r1n1rk1/pb2n1pp/1p2p3/3pPp2/2pP1PPN/P1P1BB2/2PK3P/R6R b - - 0 1

Analysis by Cfish 270418 64 POP N:

21...fxg4 22.Bxg4 Bc8 23.Ng2 Nf7 24.h4 Nh6 25.Be2 Bd7 26.h5 Nef5 27.Bf2 Ne7 28.a4 Rf7 29.a5 b5 30.Be3 Rff8 31.Rhd1 Nhf5 32.Bf2 b4 33.Ne3 Nxe3 34.Bxe3 Nf5 35.Bf2 Nh6 36.Be3 Nf5
= (0.00) Depth: 54/32 00:07:58 11062MN

And here after 21.g3 instead:

[d]1r1n1rk1/pb2n1pp/1p2p3/3pPp2/2pP1P1N/P1P1BBP1/2PK3P/R6R b - - 0 1

Analysis by Cfish 270418 64 POP N:

21...Ndc6 22.a4 Kf7 23.Ke2 Rh8 24.Bc1 g6 25.Kf2 h6 26.Ba3 Rbg8 27.Ng2 Na5 28.Ne3 Bc6 29.Bb4 Bd7 30.Rhb1 Nac6 31.Bd6 Na5 32.Rg1 h5 33.Ng2 Rd8 34.Nh4 Ng8 35.Bb4 Re8 36.h3 Kg7 37.Rgb1 Bc6 38.Rd1 Nh6 39.Rh1 Bd7 40.Ng2 Ng8 41.Ne3 Ne7 42.Rhb1 Kf7 43.Ng2 Nec6 44.Ba3 Ne7 45.Ne3 Kg7 46.Bd6
+/- (0.74) Depth: 47/49 00:05:20 7386MN


That might not be an at once winning move neither, and there seem to be other moves not much worse, but why do you think 21.g4 so much better?

It opens the game up, allowing the bishops to come alive, and in the game the passed pawns were crushing. Perhaps trusting Stockfish in positions like this is an error. Remember, AZ made Stockfish look stupid in many such positions. Even now, Lczero plays many of these types of positions MUCH better. Once the tactics level out, well.....let's let time proof me right, or wrong

[Laskos]

Hi Robert!

Hello folks, here is a game of LCZERO GPU Id 211 vs an old favourite of mine Gandalf 4.32. This is some amazingly, stunning chess! How many engine(s) play G4!!... and see they are winning!?

[d] 1r1n1rk1/pb2n1pp/1p2p3/3pPp2/2pP1P1N/P1P1BB2/2PK2PP/R6R w - - 0 21

Reason that other engines don't play 21.g4 could well be, it simply isn't the best move here, I guess.

Here's CFish after Backward of the game from move 25...g6(?) where the evals start climbing, and after 21...g4 with full hash:


[d]1r1n1rk1/pb2n1pp/1p2p3/3pPp2/2pP1PPN/P1P1BB2/2PK3P/R6R b - - 0 1

Analysis by Cfish 270418 64 POP N:

21...fxg4 22.Bxg4 Bc8 23.Ng2 Nf7 24.h4 Nh6 25.Be2 Bd7 26.h5 Nef5 27.Bf2 Ne7 28.a4 Rf7 29.a5 b5 30.Be3 Rff8 31.Rhd1 Nhf5 32.Bf2 b4 33.Ne3 Nxe3 34.Bxe3 Nf5 35.Bf2 Nh6 36.Be3 Nf5
= (0.00) Depth: 54/32 00:07:58 11062MN

And here after 21.g3 instead:

[d]1r1n1rk1/pb2n1pp/1p2p3/3pPp2/2pP1P1N/P1P1BBP1/2PK3P/R6R b - - 0 1

Analysis by Cfish 270418 64 POP N:

21...Ndc6 22.a4 Kf7 23.Ke2 Rh8 24.Bc1 g6 25.Kf2 h6 26.Ba3 Rbg8 27.Ng2 Na5 28.Ne3 Bc6 29.Bb4 Bd7 30.Rhb1 Nac6 31.Bd6 Na5 32.Rg1 h5 33.Ng2 Rd8 34.Nh4 Ng8 35.Bb4 Re8 36.h3 Kg7 37.Rgb1 Bc6 38.Rd1 Nh6 39.Rh1 Bd7 40.Ng2 Ng8 41.Ne3 Ne7 42.Rhb1 Kf7 43.Ng2 Nec6 44.Ba3 Ne7 45.Ne3 Kg7 46.Bd6
+/- (0.74) Depth: 47/49 00:05:20 7386MN


That might not be an at once winning move neither, and there seem to be other moves not much worse, but why do you think 21.g4 so much better?

Funny!
You seemed to me to be advocating human mystical understanding of subtleties of Chess, subtleties machines, LC0 included, will never achieve. Now Robert comes with LC0 psychology and mystic of that g4 move played, and you come with Cfish backward analysis and depth 54. I seriously suggest you to study LC0 psychologically rather than using Cfish.

[peter]

Well, Robert, so is 21.g4 still a !! to you like given in your first posting in the thread, or is it a !? like you write here:

http://www.talkchess.com/forum/viewtopi ... 25&t=67270

yes it may not be winning, although it also could be. This is why I gave the !? However, with that in mind, what I have been seeing is that when Lczero gives a large winning Evaluation

Here is some LC0- output after about 15 minutes of computing at the position before 21st move with network 201.

I know, it's CPU only, but at least 24 CPU(12x3GHz)- threads should be comparable to an average GPU, shouldn't they?

[d]1r1n1rk1/pb2n1pp/1p2p3/3pPp2/2pP1P1N/P1P1BB2/2PK2PP/R6R w - - 0 1

Analysis by lczero v0.7:

21.Thb1 Sdc6 22.a4
-+ (-7.59) Tiefe: 7 00:00:00 0kN
...
21.a4 Sdc6 22.g4 fxg4 23.Lxg4 Lc8 24.Thg1 a6 25.Taf1 b5 26.axb5 axb5
+- (1.95) Tiefe: 20 00:00:07 5kN
21.a4 Sdc6 22.g4 fxg4 23.Lxg4 Lc8 24.Thg1 a6 25.Taf1 b5 26.axb5 axb5 27.f5 exf5 28.Lh3 b4
+- (2.02) Tiefe: 21 00:00:11 8kN
...
21.Taf1 Lc6 22.Thg1 Le8 23.g4 fxg4 24.Lxg4 Ld7 25.f5 exf5 26.Lh3 Se6 27.Sxf5 Sxf5 28.Lxf5 Txf5 29.Txf5 Tf8
+- (2.08) Tiefe: 24 00:00:51 48kN
...
21.Taf1 Sdc6 22.Le2 Lc8 23.Sf3 Te8 24.h4 h6 25.h5 b5 26.Ta1 a5 27.Thb1 Ld7 28.Sh4 b4 29.axb4 axb4 30.cxb4 Sxb4 31.c3
+- (1.88) Tiefe: 28 00:08:28 505kN
...
(Changed to English)
...
21.Raf1 Ndc6 22.Be2 Bc8 23.Nf3 h6 24.h4 Kh7 25.h5 Rg8 26.Nh4 b5 27.g4 a5 28.g5 b4 29.axb4 axb4 30.cxb4 Nxb4 31.Rb1 Bd7 32.c3
+- (1.81) Depth: 29 00:16:47 911kN

And it's 16 minutes, and the big winning eval is now for another move, it was even bigger for at least two other moves before, never seing 21.g4 among them.
Does output and eval mean anything as for engines like LC0 or is it only for fun and do the moves she suggests have to be judged psychologically only always, with or without output?


Can this engine be used for positions she didn't get on the board by herself in game- playing or may she be asked about positions she doesn't have any history for too?