Eman Chimera - hints and impressions

[AdminX]

PS: Also seems to work well in the Banksia GUI

[pohl4711]

I aborted my testrun. After 200 games, Chimera (first 52 moves Lc0 0.24.1 LS 14.3, then Komodo 14 for the rest of the games) had exactly the same score than Lc0 LS 14.3 had as final result. Perhaps Komodo is too weak?
Because of the release of Chimera 2.00, I decided to restart the testrun with Chimera (Lc0 0.24.1 LS 14.3 and switching to Stockfish 200418, when 16 or less pieces are on the board)(Stockfish 200418 is the same SF, which is the opponent in my testrun).
Hopefully, that will give a better result.

I have plenty of good results at short TC in Cutechess-CLI, the best seems to switch at move 36 to SF11 (4 threads) ||| + 60 Elo points.
Lc0 used is v25.1.
Strangely enough, material condition doesn't fare well, but needs to be investigated further. Both move and material conditions combined might work well.

Both move and material conditions are impossible in Chimera. You have to choose one option.

In my tests, cutechess-cli ends all games, when a 5man Syzygy is on the board. Perhaps, thats a reason for lc0 playing with more success (without Chimera)?!?

Not the main reason, but a reason. I don't adjudicate at all.

I think, it is a reason. Lc0 plays very simple endgames bad. Other endgames, lc0 plays ugly. That is an important point and a huge difference, even though both (bad and ugly) are looking quite similar for humans. Ugly does not mean, that the final game-result differs from playing not-ugly or elegant. Its the same reason, why tablebases do not give much Elo-gain for engines in their playing: With tablebases, the engines reach the same game-result less ugly and more straight. But the result is the same (most of the times).
And, when the GUI ends the game at 5man positions, the bad endgames of Lc0 are "filtered" out and only the ugly endgames remain...

[Laskos]

I aborted my testrun. After 200 games, Chimera (first 52 moves Lc0 0.24.1 LS 14.3, then Komodo 14 for the rest of the games) had exactly the same score than Lc0 LS 14.3 had as final result. Perhaps Komodo is too weak?
Because of the release of Chimera 2.00, I decided to restart the testrun with Chimera (Lc0 0.24.1 LS 14.3 and switching to Stockfish 200418, when 16 or less pieces are on the board)(Stockfish 200418 is the same SF, which is the opponent in my testrun).
Hopefully, that will give a better result.

I have plenty of good results at short TC in Cutechess-CLI, the best seems to switch at move 36 to SF11 (4 threads) ||| + 60 Elo points.
Lc0 used is v25.1.
Strangely enough, material condition doesn't fare well, but needs to be investigated further. Both move and material conditions combined might work well.

Both move and material conditions are impossible in Chimera. You have to choose one option.

In my tests, cutechess-cli ends all games, when a 5man Syzygy is on the board. Perhaps, thats a reason for lc0 playing with more success (without Chimera)?!?

Not the main reason, but a reason. I don't adjudicate at all.

I think, it is a reason. Lc0 plays very simple endgames bad. Other endgames, lc0 plays ugly. That is an important point and a huge difference, even though both (bad and ugly) are looking quite similar for humans. Ugly does not mean, that the final game-result differs from playing not-ugly or elegant. Its the same reason, why tablebases do not give much Elo-gain for engines in their playing: With tablebases, the engines reach the same game-result less ugly and more straight. But the result is the same (most of the times).
And, when the GUI ends the game at 5man positions, the bad endgames of Lc0 are "filtered" out and only the ugly endgames remain...

Maybe, but try the following:
Use short openings (3-4 movers or such), set Chimera at 40 moves to switch from Lc0 to SF_dev, and see the results compared to pure Lc0. Adjudication is 5 men TBs. Assuming your 2'+1'' TC, for ultra-short maybe the switch at 35 moves.

[pohl4711]

NN-testrun of Chimera (Lc0 0.24.1 LS 14.3 and switching to Stockfish 200418, when 16 or less pieces are on the board)(Stockfish 200418 is the same SF, which is the opponent in my testrun) aborted:
No progress after 200 games. It seems, that Lc0 is not as bad in endgames, as we all believed, as long as the games are adjucated by the GUI, when a 5man (or 6man) position is on the board, as in my (and the most other) testings.

[Ovyron]

I played against Eman Chimera on playchess with my "baby on tricycle" hardware and wasn't impressed. I could comfortably get to a +2.75 advantage (real or not.) Maybe the switch to Depth 60 to Stockfish was too late.

[pgn][Event "Rated game, 5 min"]
[Site "Machines A"]
[Date "2020.05.19"]
[Round "19"]
[White "Ulycrom, CFishR 281219"]
[Black "Ischy, Eman Chimera M"]
[Result "1/2-1/2"]
[ECO "E11"]
[WhiteElo "2463"]
[BlackElo "2551"]
[Annotator "0.42;0.09"]
[PlyCount "223"]
[EventDate "2020.05.19"]
[SourceTitle "playchess.com"]
[TimeControl "300"]

{Eman Chimera M (single): 28.6 plies; 1,857kN/s Intel(R) Core(TM) i7-4930K CPU
@ 3.40GHz 3400MHz, (6 cores, 12 threads), 347 MB} 1. d4 {0.05/1 0} Nf6 {
B 0 (d5)} 2. Nf3 {0.05/1 0} d5 {B 0} 3. c4 {0.05/1 0} e6 {B 0} 4. g3 {0.05/1 0}
Bb4+ {B 0} 5. Bd2 {0.05/1 0} Be7 {B 0} 6. Bg2 {0.05/1 0} O-O {B 0} 7. O-O {
0.05/1 0} Nbd7 {B 0 (c6)} 8. Qc2 {0.05/1 1} c6 {B 0} 9. Rc1 {0.05/1 0} a5 {B 0}
10. Bf4 {0.05/1 0} Nh5 {B 0} 11. Bd2 {0.05/1 0} Nhf6 {0.09/8 3} 12. Bg5 {
0.05/1 0 (e3)} b6 {B 0} 13. cxd5 {0.05/1 0} cxd5 {0.12/10 5} 14. Nc3 {0.05/1 0
(e3)} Ba6 {0.12/10 5} 15. Qb3 {0.05/1 0 (e3)} Bb4 {0.06/10 7 (a4)} 16. e4 {
0.42/25 7 (e3)} Bc4 {0.23/17 4} 17. Qd1 {0.40/25 0} h6 {0.23/17 5} 18. Bxf6 {
0.28/26 2} Nxf6 {0.24/18 1} 19. Ne5 {0.31/29 24} Bxc3 {0.22/18 6 (Rc8)} 20.
Rxc3 {0.31/30 7} Ba6 {0.21/18 2} 21. Nc6 {0.62/27 4} Qd7 {0.21/17 0} 22. exd5 {
0.46/30 21} Nxd5 {0.20/18 0} 23. Bxd5 {0.20/29 5} exd5 {0.21/13 1} 24. Qb3 {
0.21/27 11} Rfe8 {0.18/17 23} 25. Ne5 {0.00/29 0} Qb5 {0.17/18 3 (Qb7)} 26.
Qxb5 {0.54/22 2} Bxb5 {0.18/18 3} 27. Rb3 {0.66/26 3} Be2 {0.18/16 6} 28. Rxb6
{0.39/24 1} Reb8 {0.17/17 21 (f6)} 29. Rxb8+ {0.55/24 3} Rxb8 {0.18/18 0} 30.
b3 {0.37/30 16} Rc8 {0.16/13 0} 31. a4 {0.54/28 7 (Te1)} Rc2 {0.15/16 15} 32.
Rb1 {0.39/30 2 (Te1)} h5 {0.09/14 19 (f6)} 33. Nd7 {0.69/24 3 (Te1)} Rd2 {
0.08/13 9 (g5)} 34. Nc5 {1.41/22 3 (Se5)} Rxd4 {0.18/17 8} 35. Nb7 {1.31/28 0}
Rd2 {0.24/20 4} 36. Nxa5 {1.02/28 6} d4 {0.28/16 0 (Kf8)} 37. b4 {0.92/28 8} d3
{0.30/22 4 (Bd3)} 38. b5 {1.62/27 3} Rc2 {0.30/21 4 (Ra2)} 39. b6 {0.96/31 8
(Sb3)} d2 {0.29/20 5} 40. Nb3 {1.08/33 1} Ba6 {0.28/18 1} 41. Nxd2 {1.05/30 2
(a5)} Rxd2 {0.12/12 15} 42. b7 {1.01/37 0} Bxb7 {0.12/11 4} 43. Rxb7 {1.07/37 0
} Ra2 {0.11/9 4 (h4)} 44. Ra7 {1.88/27 3 (Tb4)} Kh7 {0.10/8 20 (g5)} 45. a5 {
1.75/31 8 (Kg2)} Kh6 {0.09/8 10 (Kg6)} 46. h4 {1.77/26 3 (Kg2)} f6 {0.08/7 5
(Kg6)} 47. a6 {2.03/30 5} g5 {0.09/8 7 (Kg6)} 48. Ra8 {0.66/34 7 (hxg5+)} gxh4
{0.05/8 2 (Kh7)} 49. a7 {2.75/35 3 (gxh4)} Kg7 {0.06/9 4 (Kh7)} 50. gxh4 {
2.45/46 3} Ra4 {0.07/9 5 (Kh7)} 51. Kg2 {2.40/46 3 (Kf1)} Ra3 {0.06/9 2 (Ra2)}
52. Kh2 {2.38/48 3 (Kf1)} Kh7 {0.06/8 4 (Ra2)} 53. Kg1 {2.28/44 6 (Kg2)} Kg7 {
0.06/9 3 (Ra1+)} 54. Kh2 {1.83/38 2 (Kg2)} Kh7 {0.06/8 3 (Ra2)} 55. Kg1 {
0.65/43 6 (f3)} Ra1+ {0.07/8 2 (Kg7)} 56. Kg2 {0.13/44 5} Ra3 {0.06/9 0 (Kg7)}
57. Rf8 {0.54/36 2 (f3)} Rxa7 {0.00/6 1} 58. Rxf6 {0.54/43 0} Ra4 {0.01/5 4
(Kg7)} 59. Kg3 {0.54/45 2} Rg4+ {0.00/41 3 (Kg7)} 60. Kh3 {0.54/44 2} Ra4 {
0.00/47 0 (Rg1)} 61. f3 {0.54/43 2 (Tf5)} Ra5 {0.00/43 0 (Ra7)} 62. Rb6 {
0.54/41 2 (Tc6)} Kg7 {0.00/44 1} 63. Kg3 {0.54/43 1} Rd5 {0.00/49 0 (Ra3)} 64.
Rb7+ {0.54/41 2} Kg6 {0.00/51 0} 65. Rb6+ {0.54/42 1 (Tb8)} Kg7 {0.00/47 1} 66.
Re6 {0.54/41 0} Kf7 {0.00/50 0 (Rf5)} 67. Rh6 {0.54/44 1 (Te4)} Kg7 {0.00/50 1
(Ra5)} 68. Rb6 {0.54/45 2} Kf7 {0.00/52 0 (Ra5)} 69. Rh6 {0.54/43 1 (Tb7+)} Kg7
{0.00/50 1 (Ra5)} 70. Re6 {0.54/45 1 (Tb6)} Kf7 {0.00/51 0} 71. Ra6 {0.54/45 0
(Te4)} Rb5 {0.00/41 0} 72. Rc6 {0.54/42 0 (Kf4)} Rb4 {0.00/48 0} 73. Ra6 {
0.54/44 1 (Tc3)} Kg7 {0.00/51 1 (Rb5)} 74. Re6 {0.54/43 2 (Ta5)} Rb3 {0.00/51 1
} 75. Kf4 {0.54/43 0} Rc3 {0.00/49 0 (Rb4+)} 76. Re3 {0.54/41 1 (Te5)} Rc4+ {
0.00/50 0 (Rc2)} 77. Re4 {0.54/42 2 (Kg3)} Rc8 {0.00/45 0 (Rc3)} 78. Re5 {
0.54/42 1 (Tb4)} Kg6 {0.00/46 0} 79. Rb5 {0.54/40 0 (Tg5+)} Rc4+ {0.00/48 0}
80. Kg3 {0.54/43 0} Rc3 {0.00/54 0 (Ra4)} 81. Rb6+ {0.54/42 1 (Tb4)} Kg7 {
0.00/52 0 (Kf7)} 82. Rb5 {0.54/45 1} Kg6 {0.00/55 0} 83. Re5 {0.54/43 1} Kh6 {
0.00/55 0 (Rc2)} 84. Re6+ {0.54/41 1} Kh7 {0.00/56 0 (Kg7)} 85. Re5 {0.54/42 1
(Td6)} Kg6 {0.00/50 0 (Kh6)} 86. Re6+ {0.54/42 1} Kg7 {0.00/50 1} 87. Rb6 {
0.54/44 0 (Ta6)} Rc5 {0.00/46 0 (Rc4)} 88. Rb7+ {0.54/42 1 (Td6)} Kf6 {0.00/40
0 (Kg6)} 89. Rd7 {0.54/42 1 (Tb8)} Rc4 {0.00/40 0} 90. Rd6+ {0.54/43 2 (Td5)}
Kg7 {0.00/40 1 (Kf7)} 91. Rd5 {0.54/43 6 (Td3)} Kg6 {0.00/41 0} 92. Re5 {
0.54/37 0 (Kh3)} Rb4 {0.00/37 0 (Rc2)} 93. Rc5 {0.54/36 1 (Te4)} Ra4 {0.00/35
0 (Rb1)} 94. Rc6+ {0.54/34 1 (Tg5+)} Kf7 {0.00/33 0 (Kh7)} 95. Rc5 {0.54/34 3}
Kg6 {0.00/37 0} 96. Rd5 {0.54/29 4 (Tg5+)} Ra3 {0.00/30 0} 97. Rd6+ {0.54/32 5
(Td4)} Kg7 {0.00/27 0} 98. Rc6 {0.54/31 1 (Td5)} Rb3 {0.00/30 1 (Ra2)} 99. Rc7+
{0.54/27 2} Kg6 {0.00/37 0 (Kh6)} 100. Re7 {0.54/31 1 (Tc1)} Ra3 {0.00/30 0
(Kf6)} 101. Rd7 {0.54/24 1 (Kf4)} Rb3 {0.00/28 0 (Kf6)} 102. Rd6+ {0.54/31 1
(Td5)} Kf7 {0.00/23 0 (Kg7)} 103. Rd4 {0.54/28 1 (Td5)} Kg6 {0.00/21 0 (Kg7)}
104. Rd8 {0.54/24 1 (Td6+)} Kf7 {0.00/22 0 (Kh6)} 105. Rd4 {0.35/24 2 (Td5)}
Kg6 {0.00/36 0 (Rb5)} 106. Rd8 {0.54/26 1 (Td5)} Kf7 {0.00/29 0 (Kh6)} 107. Kf4
{0.00/21 0 (Td5)} Rb4+ {0.00/27 0} 108. Kg5 {0.00/33 0} Rb5+ {0.00/37 0} 109.
Kh6 {0.00/32 0} Rf5 {0.00/37 0} 110. f4 {0.00/33 0} Rxf4 {0.00/36 0} 111. Kxh5
{0.00/33 0} Rf6 {0.00/36 0 Ulycrom offers a draw} 112. Kg5 {0.00/37 0 (Lag:
Av=0.30s, max=0.8s)} 1/2-1/2[/pgn]

Where's the 90ish elo advantage black's supposed to display?

[Werner]

I made an engine with Lc0 591226 + little Ender and started MEA testrun. Why is it so slow?
Look like the 50000 positions need 5h on CPU.

[corres]

I played on playchess with my "baby on tricycle" hardware...

What is it?

[Ovyron]

What is it?

Q8200 @2.33Ghz 3CPU 256MB RAM for the engines. Reaching around 2400 kn/s on average. 5men TBs.

I listen to videos while I sleep, which uses 1 of the cores, that's why I don't use 4 (and didn't notice any difference in performance between 3 and 4, 3 have performed better by luck .)

My loses have been mostly about poor opening selections in the book, I perform better as I patch them. I performed better with the bix book I was given at InfinityChess, but I've been too lazy to convert it.

This is the same hardware I was using 8 years ago, back when I was losing 90% of my games, which means software has caught up and now the hardware isn't as relevant. Currently it's all about books and the guys at the top play the best openings.

[Laskos]

What is it?

Q8200 @2.33Ghz 3CPU 256MB RAM for the engines. Reaching around 2400 kn/s on average. 5men TBs.

I listen to videos while I sleep, which uses 1 of the cores, that's why I don't use 4 (and didn't notice any difference in performance between 3 and 4, 3 have performed better by luck .)

My loses have been mostly about poor opening selections in the book, I perform better as I patch them. I performed better with the bix book I was given at InfinityChess, but I've been too lazy to convert it.

This is the same hardware I was using 8 years ago, back when I was losing 90% of my games, which means software has caught up and now the hardware isn't as relevant. Currently it's all about books and the guys at the top play the best openings.

If one play with an a3 book? Do many have anti- a3 books? Can't you exit prepared openings soon by making 2-3-4 weird moves? That's to avoid book battles which I wouldn't enter, these have no limit and no end.

[Ovyron]

If one play with an a3 book? Do many have anti- a3 books? Can't you exit prepared openings soon by making 2-3-4 weird moves? That's to avoid book battles which I wouldn't enter, these have no limit and no end.

There's freely available "draw books", that you can use and they go very deep into variations that aren't problematic, where the engine doesn't have much to do and games end in a draw. People don't use those because you basically give away winning chances from the get-go.

As for books that try to get others out of book ASAP with moves like a3, they're not as effective as those that go deep but have a trap and if the opponent falls for it you have great advantage. And avoiding book battles isn't recommended for people with my hardware (in a 5 minute game, thanks to my book, I can remain inside it playing instantly while my opponent thinks, and come up with 1min-2mins advantage, which is much better than any book avoidance.)