The Engine Crackers (4)

[Glarean]

.

New Crackers have been added again:
https://glarean-magazin.ch/2020/06/12/c ... s-puzzles/

One of them is particularly interesting (Blitz-Game Lc0 0.25.1 t60-3972 vs Stockfish 240620 1-0):

WHITE TO MOVE

[d]rnbqk2r/pp1p1ppp/3P1n2/b1p5/8/2N2N2/PP2PPPP/R1BQKB1R w KQkq

8. e4 (!!)

[pgn][Event "LeelaChess - Stockfish, Comp 2020"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "The Engine Crackers 010"]
[Black "www.glarean-magazin.ch"]
[Result "*"]
[Annotator "Walter Eigenmann & Computer"]
[SetUp "1"]
[FEN "rnbqk2r/pp1p1ppp/3P1n2/b1p5/8/2N2N2/PP2PPPP/R1BQKB1R w KQkq - 0 8"]
[PlyCount "23"]
[SourceTitle "Glarean Magazin"]
[Source "W. Eigenmann"]
[SourceDate "2020.06.13"]
[SourceVersion "1"]
[SourceVersionDate "2020.06.13"]
[SourceQuality "1"]

8. e4 $1 (8. Bg5 $6 Qb6 9. Rc1 (9. Rb1 Ne4 10. Qd3 (10. Bd2 Bxc3 11. Bxc3 O-O
12. e3 Qxd6 $11) 10... f5 11. Bd2 Qxd6 12. Nxe4 Qxd3 13. exd3 fxe4 $11) 9...
Ne4 10. Qd5 Nxg5 11. Nxg5 O-O $14) (8. e3 $6 b6 $16) (8. Bf4 $6 b6 9. e3 O-O
$16) 8... Nxe4 9. Qc2 Nxc3 (9... Nf6 10. Bc4 O-O 11. O-O $18) 10. bxc3 Qf6 (
10... O-O 11. Ng5 Qe8+ 12. Be2 $18) 11. Bd2 Qe6+ (11... Nc6 12. O-O-O $18) 12.
Be2 h6 (12... Qxd6 13. O-O-O $18) 13. O-O-O Nc6 (13... O-O 14. Rde1 Qxd6 (14...
Nc6 15. h4 Qg6 16. Qxg6 fxg6 17. Bc4+ Kh8 18. Ng5 $18) 15. h4 $18) 14. Bc4 Qxc4
15. Rhe1+ Kd8 (15... Kf8 16. Re4 $18) 16. Re4 Nd4 17. cxd4 Qxc2+ 18. Kxc2 Bxd2
19. Kxd2 $18 *[/pgn]

Analysis by Brainfish 130720:

1. +/- (1.45): 8.e4 Sxe4 9.Dc2 Sxc3 10.bxc3 Df6 11.Ld2 De6+ 12.Le2 h6 13.0-0-0 0-0 14.Tde1 b5 15.h4 [...]
2. +/- (1.06): 8.Lg5 Db6 9.Tc1 Se4 10.Dd5 Sxg5 11.Sxg5 0-0 12.Dd3 g6 13.Dd2 Db4 14.h4 Dd4 15.h5 [...]
3. +/- (0.75): 8.Lf4 0-0 9.e3 Lxc3+ 10.bxc3 b6 11.Ld3 La6 12.Lxa6 Sxa6 13.0-0 De8 14.Te1 De4 15.Le5 [...]
(Deep 40 / Ryzen 7)

Analysis by Lc0 0.26.0 t60-4300:

1. +/- (0.77): 8.e4 Sxe4 9.De2 f5 10.Sg5 0-0 11.Dh5 Sxg5 12.Lxg5 De8+ 13.Dxe8 Txe8+ 14.Le7 b6 15.Lc4+ [...]
2. +/= (0.55): 8.g3 b6 9.Lg2 Lb7 10.0-0 Lxc3 11.bxc3 0-0 12.c4 Se4 13.Dd3 De8 14.Sh4 De6 15.Td1 [...]
3. +/= (0.46): 8.Lg5 Db6 9.Tc1 Dxb2 10.Ld2 Db6 11.e4 Dxd6 12.e5 De7 13.Le2 Lxc3 14.Lxc3 Se4 15.Lb2 [...]
(Deep 24 / Ryzen 7)

I think such opening novelties are the reason why Carlsen & Co. are so interested in Leela...

Regards: Walter

[peter]

.

New Crackers have been added again:
https://glarean-magazin.ch/2020/06/12/c ... s-puzzles/

One of them is particularly interesting (Blitz-Game Lc0 0.25.1 t60-3972 vs Stockfish 240620 1-0):

WHITE TO MOVE

[d]rnbqk2r/pp1p1ppp/3P1n2/b1p5/8/2N2N2/PP2PPPP/R1BQKB1R w KQkq

8. e4 (!!)

[pgn][Event "LeelaChess - Stockfish, Comp 2020"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "The Engine Crackers 010"]
[Black "www.glarean-magazin.ch"]
[Result "*"]
[Annotator "Walter Eigenmann & Computer"]
[SetUp "1"]
[FEN "rnbqk2r/pp1p1ppp/3P1n2/b1p5/8/2N2N2/PP2PPPP/R1BQKB1R w KQkq - 0 8"]
[PlyCount "23"]
[SourceTitle "Glarean Magazin"]
[Source "W. Eigenmann"]
[SourceDate "2020.06.13"]
[SourceVersion "1"]
[SourceVersionDate "2020.06.13"]
[SourceQuality "1"]

8. e4 $1 (8. Bg5 $6 Qb6 9. Rc1 (9. Rb1 Ne4 10. Qd3 (10. Bd2 Bxc3 11. Bxc3 O-O
12. e3 Qxd6 $11) 10... f5 11. Bd2 Qxd6 12. Nxe4 Qxd3 13. exd3 fxe4 $11) 9...
Ne4 10. Qd5 Nxg5 11. Nxg5 O-O $14) (8. e3 $6 b6 $16) (8. Bf4 $6 b6 9. e3 O-O
$16) 8... Nxe4 9. Qc2 Nxc3 (9... Nf6 10. Bc4 O-O 11. O-O $18) 10. bxc3 Qf6 (
10... O-O 11. Ng5 Qe8+ 12. Be2 $18) 11. Bd2 Qe6+ (11... Nc6 12. O-O-O $18) 12.
Be2 h6 (12... Qxd6 13. O-O-O $18) 13. O-O-O Nc6 (13... O-O 14. Rde1 Qxd6 (14...
Nc6 15. h4 Qg6 16. Qxg6 fxg6 17. Bc4+ Kh8 18. Ng5 $18) 15. h4 $18) 14. Bc4 Qxc4
15. Rhe1+ Kd8 (15... Kf8 16. Re4 $18) 16. Re4 Nd4 17. cxd4 Qxc2+ 18. Kxc2 Bxd2
19. Kxd2 $18 *[/pgn]

Analysis by Brainfish 130720:

1. +/- (1.45): 8.e4 Sxe4 9.Dc2 Sxc3 10.bxc3 Df6 11.Ld2 De6+ 12.Le2 h6 13.0-0-0 0-0 14.Tde1 b5 15.h4 [...]
2. +/- (1.06): 8.Lg5 Db6 9.Tc1 Se4 10.Dd5 Sxg5 11.Sxg5 0-0 12.Dd3 g6 13.Dd2 Db4 14.h4 Dd4 15.h5 [...]
3. +/- (0.75): 8.Lf4 0-0 9.e3 Lxc3+ 10.bxc3 b6 11.Ld3 La6 12.Lxa6 Sxa6 13.0-0 De8 14.Te1 De4 15.Le5 [...]
(Deep 40 / Ryzen 7)

Analysis by Lc0 0.26.0 t60-4300:

1. +/- (0.77): 8.e4 Sxe4 9.De2 f5 10.Sg5 0-0 11.Dh5 Sxg5 12.Lxg5 De8+ 13.Dxe8 Txe8+ 14.Le7 b6 15.Lc4+ [...]
2. +/= (0.55): 8.g3 b6 9.Lg2 Lb7 10.0-0 Lxc3 11.bxc3 0-0 12.c4 Se4 13.Dd3 De8 14.Sh4 De6 15.Td1 [...]
3. +/= (0.46): 8.Lg5 Db6 9.Tc1 Dxb2 10.Ld2 Db6 11.e4 Dxd6 12.e5 De7 13.Le2 Lxc3 14.Lxc3 Se4 15.Lb2 [...]
(Deep 24 / Ryzen 7)

I think such opening novelties are the reason why Carlsen & Co. are so interested in Leela...

Very interesting, Walter.
But 9.Qe2 in LC0- output deserves a ? I'd say.
9.Qc2 in Brainfish- output and in game is obviously much better. Maybe that's why evals in this one line are some higher and more correct, aren't they?

[Glarean]

Very interesting, Walter.
But 9.Qe2 in LC0- output deserves a ? I'd say.
9.Qc2 in Brainfish- output and in game is obviously much better. Maybe that's why evals in this one line are some higher and more correct, aren't they?

Of course, the output lines of engines should always be treated with caution. And furthermore, the networks of the game leela and my analysis leela generate different outputs.

Much more important is that all new Leela's keep the solution e4:

Analysis by Lc0 0.25.1 t60-3044:
8.e4 Sxe4 9.De2 f5 10.Lg5 Db6 11.0-0-0 Lxc3 12.bxc3 Da5 13.Db2 h6 14.Ld2 Db6 15.Le3 Dxb2+ 16.Kxb2 [..]
Weiss steht besser: +/- (0.80) Tiefe: 8/20 00:00:08 3216N
[...]
8.e4 Sxe4 9.De2 f5 10.Sg5 0-0 11.Dh5 Sxg5 12.Lxg5 De8+ 13.Dxe8 Txe8+ 14.Le7 b5 15.Lxb5 La6 16.a4 [...]
Weiss steht besser: +/- (0.79) Tiefe: 28/68 00:06:50 3337kN

Analysis by Lc0 0.25.1 t60-3972
8.e4 Sxe4 9.Dc2 Sxc3 10.bxc3 Df6 11.Ld2 De6+ 12.Le2 h6 13.0-0-0 Sc6 14.Lc4 Dxc4 15.The1+ Kf8 16.Te4 [...]
Weiss steht etwas besser: +/= (0.63) Tiefe: 8/19 00:00:01 2356N
Weiss steht besser: +/- (0.80) Tiefe: 35/76 00:05:03 2269kN, tb=11

AB engines have no chance here in a regular game...
This is a nice example of how important the new NN programs are for the development of the opening theory.

.

[zullil]

AB engines have no chance here in a regular game...
This is a nice example of how important the new NN programs are for the development of the opening theory.

.

I agree with your second sentence. The first one seems rather strong, since there's very little evidence that e4 is decisively better than alternatives---if it's better at all.

[peter]

AB engines have no chance here in a regular game...
This is a nice example of how important the new NN programs are for the development of the opening theory.

You mean AB engines don't have a chance to find e4, but e.g. komodo with raised King Safety (150 instead of 75) and raised Dynamism (200 instead of 122) is at it at once too. Doesn't stay stable on it, but 9.Qc2 is in output very soon too.

And then question is still, if AB- engine couldn't win that with White against LC0 with other moves instead of e4 yet still, depending on hardware- TC.
Don't get me wrong, I think e4 best too, but there were white wins in corr.- databases with Bg5 already also, which is one of the usual continuations here.

Nevertheless good found of Leela's.

[Glarean]

You mean AB engines don't have a chance to find e4, but e.g. komodo with raised King Safety (150 instead of 75) and raised Dynamism (200 instead of 122) is at it at once too.

Of course you can tinker with any good AB engine until it finds the solution move here. But we are talking about a "regular" game with the engine settings given by the programmer...

And then question is still, if AB- engine couldn't win that with White against LC0 with other moves instead of e4 yet still...

Find out by playing a theme engine tournament with this position...

[peter]

Find out by playing a theme engine tournament with this position...

That would be unfair with my weak hardware LC0- respective
And then if you say

AB engines have no chance here in a regular game...

, then a sinlge won game of an AB-engine against LC0 would be enough to prove you wrong, wouldn't it?
What if komodo with the quoted setting would chose e4, which it prbably would with short TC, and win, which it probably would after chosing the winning move?
Than both your premises would be proven wrong, that the setting was no good at game playing and that AB-engines wouldn't have a chance here in a (single) regular game.

[zullil]

AB engines have no chance here in a regular game...
This is a nice example of how important the new NN programs are for the development of the opening theory.

.

I agree with your second sentence. The first one seems rather strong, since there's very little evidence that e4 is decisively better than alternatives---if it's better at all.

OK, here's more evidence in favor of e4.

Cfish-dev:

info depth 60 seldepth 94 multipv 1 score cp 150 nodes 311328565797 nps 32770858 hashfull 1000 tbhits 20859972 time 9500165 pv e2e4 f6e4 d1e2 f7f5 c1g5 a5c3 b2c3 d8a5 a1c1 a5a6 e2e3 b7b5 f3d2 e8g8 d2e4 f5e4 g5e7 f8f7 e3c5 b8c6 f1b5 a6b6 c5b6 a7b6 b5c4 c6e5 c4f7 g8f7 e1g1 c8a6 f1d1 a6d3 e7g5 a8a2 g5e3 e5c4 d1e1 c4d6 e3b6 d6b5 b6c5 d7d6 c5e3 a2a4 h2h4 a4c4 e3d4 g7g6 f2f3 b5d4 c3d4 c4d4 c1c7 f7f6 f3e4 d3e4 e1f1 f6e5 c7h7 d4d2 f1f2 d2d4 g1h2 e4d3 h2g3 d3f5 h7c7 f5e6 f2e2 e5d5

info depth 60 seldepth 90 multipv 2 score cp 94 nodes 311328565797 nps 32770858 hashfull 1000 tbhits 20859972 time 9500165 pv c1g5 h7h6 g5h4 d8b6 a1b1 f6e4 d1d3 f7f5 f3d2 b6d6 d2e4 d6d3 e2d3 f5e4 d3e4 b8c6 h4g3 d7d5 e4d5 c6d4 e1d2 b7b5 b1e1 e8f7 d2c1 c8d7 e1e5 a8d8 f1d3 h8e8 d5d6 e8e5 g3e5 a7a6 h1d1 g7g5 c3e4 d8e8 e5d4 c5d4 d3c2 d7c6 d1d4 a5b6 d4d2 c6e4 d6d7 e8e5 c2e4 e5e4 d7d8r b6d8 d2d8 e4e1 d8d1 e1e2 d1d2 e2e1 c1c2 g5g4 c2c3 e1c1 c3d4 f7e6 d4e3 c1g1 f2f3 e6f5 e3f2 g1b1 b2b3 b1a1 f3g4 f5g4 h2h3 g4f5 d2d6 a1a2 f2f3

info depth 60 seldepth 79 multipv 3 score cp 74 nodes 311328565797 nps 32770858 hashfull 1000 tbhits 20859972 time 9500165 pv g2g4 f6e4 d1d3 e4c3 b2c3 d8f6 c1d2 b8c6 f1g2 b7b6 d3e3 f6e6 e3e6 f7e6 f3g5 c8b7 g2e4 c6d8 e1c1 b7d5 h2h4 d8b7 c3c4 d5e4 g5e4 a5d2 c1d2 e8c8 h4h5 h7h6 d1b1 h8f8 f2f3 c8b8 h1c1 f8f7 a2a4 d8c8 d2c3 b7a5 b1b5 a5b7 c1d1 b8a8 b5b1 b7a5 c3d3 c8f8 d1c1 f8c8 b1b5 a5b7 d3c2 f7f8 c1d1 a8b8 d1b1 b8a8 b5b2 f8f7 c2c3 b7a5 b2b5 a5b7 b1d1 a8b8 b5b1 b7a5 c3d3 f7f8 b1b5 b8b7

[Glarean]

AB engines have no chance here in a regular game...
This is a nice example of how important the new NN programs are for the development of the opening theory.

I agree with your second sentence. The first one seems rather strong, since there's very little evidence that e4 is decisively better than alternatives---if it's better at all.

You probably didn't analyze very well. 8.e4 is clearly the strongest move in this position.
Bg5 or Bf4 or g3 are good moves, but e4 is the strike.

Here is the output of Eman after 15 minutes and a little forward/backward analysis (Ryzen7-16Cores):

The Engine Crackers 010 - https://glarean-magazin.ch/2020/06/12/c ... s-puzzles/ , LeelaChess - Stockfish 1-0
rnbqk2r/pp1p1ppp/3P1n2/b1p5/8/2N2N2/PP2PPPP/R1BQKB1R w KQkq - 0 1

Analysis by Eman 5.70:

1. +- (1.69): 8.e4 Sxe4 9.Dc2 Sxc3 10.bxc3 Df6 11.Ld2 De6+ 12.Le2 h6 13.0-0-0 Sc6 14.Lc4 [...]
2. +/- (1.09): 8.Lg5 Db6 9.Tc1 Se4 10.Dd5 Sxg5 11.Sxg5 0-0 12.Dd3 g6 13.Dd2 f6 14.Sf3 [...]
3. +/- (0.91): 8.g3 b6 9.Lg2 Lb7 10.0-0 Lxc3 11.bxc3 0-0 12.Te1 Te8 13.c4 Se4 14.Lb2 [...]

.

[Glarean]

, then a sinlge won game of an AB-engine against LC0 would be enough to prove you wrong, wouldn't it?

But the task here is quite different, in fact it is:
Find an AB engine that can win this position against Stockfish 240620 without 8.e4!

.