The Fried Fish Attack

[mohzus]

Hi Robert.

When you modified the king value, probably you also changed/scaled down a whole bunch of king safety terms, linked to this value. King safety alway seems to bring sufficient elo increase, as you can turn almost any position into attacking/king safety position. That is why it is so important.

Something I do not understand is why Ralph's version would sacrifice on g4 when it sees a negative score for itself? Is this the first choice of the engine?

Hi Lyudmil,
I did not modify the king value but the bonus awarded to the king in Ralph's patch. What I noticed is that even with an extremely small bonus (200 times smaller than Ralph's), the evaluation of stockfish is still completely different from the original stockfish.
About the knight sacrifice on g4, yes Ralph's version would still sacrifice on g4 even though it sees that white is better. It's the first choice of the engine, it does not see any better move.

[Ralph Stoesser]

What I noticed is that even with an extremely small bonus (200 times smaller than Ralph's), the evaluation of stockfish is still completely different from the original stockfish.

It is expected because when you do not award the bonus (or only a very little one), you still avoid the storm pawn and shelter pawn penalties for the h file. The total compensational bonus of the patch is more than 400 value units. (Pawn is about 256 value units).

[mohzus]

What I noticed is that even with an extremely small bonus (200 times smaller than Ralph's), the evaluation of stockfish is still completely different from the original stockfish.

It is expected because when you do not award the bonus (or only a very little one), you still avoid the storm pawn and shelter pawn penalties for the h file. The total compensational bonus of the patch is more than 400 value units. (Pawn is about 256 value units).

I see Ralph, thanks for the information

I've just done a multi pv (3) analysis on the position. My cpu is very slow so I somehow stopped early (only depth 28) but the result might interest Tsvetkov:

Code: Select all

info depth 28 seldepth 46 score cp 0 nodes 2854327685 nps 1385314 time 2060418 multipv 1 pv f6g4 h3g4 c8g4 g1h2 d8f6 b1c3 c5f2 c3d5 f2g3 h2h1 f6g6 d5c7 e8f8 g2f3 a8c8 c7d5 g4d7 c1e3 b7b5 f1g1 c6e7 d5e7 f8e7 e3a7 g6e6 f3g2 b5b4 g1f1

Code: Select all

info depth 28 seldepth 46 score cp -53 nodes 2854327685 nps 1385314 time 2060418 multipv 2 pv c8d7 c1g5 d8c8 g5f6 g7f6 b1d2 a7a5 e2c3 c6e7 d1f3 h8h6 a2a4 h6g6 d2b3 c5b6 f1d1 d7e6 b3d2 b6c5 d2c4 c8d8 c4e3 e7c6 c3b5 c6d4 b5d4

Code: Select all

info depth 28 seldepth 46 score cp -85 nodes 2854327685 nps 1385314 time 2060418 multipv 3 pv c8e6 e2c3 d8d7 c3a4 e8f8 a4c5 d6c5 c1e3 b7b6 b1d2 a8e8 a2a4 g7g6 a4a5 f8g7 a5b6 c7b6 f1e1 c6b4 a1a3 e8c8 d2f3 b4c6

In other words, Nxg4 gives an evaluation of 0 cp.
Bd7 gives 53 cp advantage for white.
Be6 gives a 85 cp advantage for white.

These are the 3 best moves according to SF (Ralph's version).

[Lyudmil Tsvetkov]

Hi Robert.

When you modified the king value, probably you also changed/scaled down a whole bunch of king safety terms, linked to this value. King safety alway seems to bring sufficient elo increase, as you can turn almost any position into attacking/king safety position. That is why it is so important.

Something I do not understand is why Ralph's version would sacrifice on g4 when it sees a negative score for itself? Is this the first choice of the engine?

Hi Lyudmil,
I did not modify the king value but the bonus awarded to the king in Ralph's patch. What I noticed is that even with an extremely small bonus (200 times smaller than Ralph's), the evaluation of stockfish is still completely different from the original stockfish.
About the knight sacrifice on g4, yes Ralph's version would still sacrifice on g4 even though it sees that white is better. It's the first choice of the engine, it does not see any better move.

That is strange, Tournevisse , as the Ng4 sacrifice is probably the worse move in the position: it leads to a straightforward loss, while at the same time there are many other moves, including the 2 mentioned by you, that still hold the balance.

[phenri]

Hi, I want to show you a game that I find interesting because it gives a slight advantage for black, without a series of inaccurate moves from the white, I think it would eventually draw. Because it is particularly a closed position when looking at the variant with 20.Qe1

What do you think? Particularly 1...f4 instead of g6. And do you think there may be improvements to gaming skills of Stockfish?

Code: Select all

[Event "?"]
[Site "?"]
[Date "2014.01.12"]
[Round "?"]
[White "Infamous"]
[Black "Hero, Stockfish 090114"]
[Result "0-1"]
[SetUp "1"]
[FEN "r2q1rk1/4b1pp/2npb3/p2Npp2/1pP1P2P/3QN1P1/PP2BP2/R4RK1 b - - 0 1"]
[PlyCount "67"]
[EventDate "2014.01.12"]

1... f4 {!!} 2. Nf5 Kh8 3. Kg2 Bf6 4. h5 Bg5 5. g4 a4 6. Bf3 a3 7. b3 Qb8 8. Rfc1
Ra5 9. Rh1 Rf7 10. Rad1 Rb7 11. Be2 Nd8 12. Qc2 Rc5 13. Rd3 Bf7 14. Bf3 Ne6 15.
Qe2 Rc8 16. Rd2 Rd8 17. Qd3 Qa7 18. Qf1 Rdd7 19. Rd3 Bd8 20. Qe2 {?!} (20. Qe1
{!!} Qb8 21. Bd1 Qa8 22. f3 Qa5 23. Be2 Bg5 24. Rd1) 20... Ng5 21. Kh2 Be6 22.
Bg2 Rb8 23. Rdd1 g6 24. Nh4 Rg7 25. hxg6 hxg6 26. Kg1 Kg8 27. Rd3 Rh7 28. Qe1
Bxg4 29. Nf3 Nxf3+ 30. Bxf3 Rxh1+ 31. Kxh1 Qh7+ 32. Kg1 Bxf3 33. Rxf3 g5 34.
Rd3 g4 0-1

[pgn][Event "?"]
[Site "?"]
[Date "2014.01.12"]
[Round "?"]
[White "Infamous"]
[Black "Hero, Stockfish 090114"]
[Result "0-1"]
[SetUp "1"]
[FEN "r2q1rk1/4b1pp/2npb3/p2Npp2/1pP1P2P/3QN1P1/PP2BP2/R4RK1 b - - 0 1"]
[PlyCount "67"]
[EventDate "2014.01.12"]

1... f4 {!!} 2. Nf5 Kh8 3. Kg2 Bf6 4. h5 Bg5 5. g4 a4 6. Bf3 a3 7. b3 Qb8 8. Rfc1
Ra5 9. Rh1 Rf7 10. Rad1 Rb7 11. Be2 Nd8 12. Qc2 Rc5 13. Rd3 Bf7 14. Bf3 Ne6 15.
Qe2 Rc8 16. Rd2 Rd8 17. Qd3 Qa7 18. Qf1 Rdd7 19. Rd3 Bd8 20. Qe2 {?!} (20. Qe1
{!!} Qb8 21. Bd1 Qa8 22. f3 Qa5 23. Be2 Bg5 24. Rd1) 20... Ng5 21. Kh2 Be6 22.
Bg2 Rb8 23. Rdd1 g6 24. Nh4 Rg7 25. hxg6 hxg6 26. Kg1 Kg8 27. Rd3 Rh7 28. Qe1
Bxg4 29. Nf3 Nxf3+ 30. Bxf3 Rxh1+ 31. Kxh1 Qh7+ 32. Kg1 Bxf3 33. Rxf3 g5 34.
Rd3 g4 0-1[/pgn]
[D]r2q1rk1/4b1pp/2npb3/p2Npp2/1pP1P2P/3QN1P1/PP2BP2/R4RK1 b - - 0 1

1...f4 2.Nxe7+ Nxe7 3.Nd5 Bxd5 4.cxd5 fxg3 5.fxg3
= (-0.28) Depth: 7/8 00:00:00 16kN
1...f4 2.Nf5 Bxf5 3.exf5 Nd4 4.Nxf4 Rxf5 5.Bg4
=/+ (-0.50) Depth: 8/10 00:00:00 28kN
1...f4 2.Nf5 a4 3.Bg4 Bxf5 4.exf5 fxg3 5.fxg3 a3 6.b3
= (-0.07) Depth: 9/12 00:00:00 91kN
1...f4 2.Nf5 fxg3 3.fxg3 Bxf5 4.Rxf5 Rxf5 5.exf5 Nd4 6.Rd1 Rc8
= (0.00) Depth: 10/12 00:00:00 143kN
1...f4 2.Nf5 g6 3.Ndxe7+ Nxe7 4.Nxd6 Qb6 5.g4 f3 6.Bxf3 Rxf3 7.Qxf3 Qxd6 8.Rad1
= (-0.14) Depth: 11/15 00:00:00 278kN
1...g6 2.f4 Nd4 3.Bf3 Rc8 4.Rad1 Kh8 5.exf5 gxf5 6.Nxe7 Qxe7 7.Bd5
=/+ (-0.41) Depth: 12/19 00:00:00 835kN
1...g6 2.f4 Nd4 3.Bf3 Rc8 4.Rad1 Kh8 5.exf5 gxf5 6.Bg2 Bf6 7.Nxf6 Qxf6 8.b3 Qg7
=/+ (-0.39) Depth: 13/19 00:00:00 1073kN
1...f4 2.Nf5 Bf6 3.Bf3 fxg3 4.fxg3 Nd4 5.a3 bxa3 6.Rxa3 Rb8 7.Nxd4 exd4 8.Rb3 Be5 9.Rxb8 Qxb8
= (-0.29) Depth: 14/19 00:00:00 1704kN
1...g6 2.f4 Nd4 3.Bf3 Rc8 4.Rad1 Kh8 5.b3 Rg8 6.Bg2 Rg7 7.Qd2 Bf8 8.Qd3 Kg8 9.fxe5 dxe5 10.exf5 gxf5 11.Nxf5
=/+ (-0.32) Depth: 15/20 00:00:01 2300kN

(...)

1...g6 2.f4 Nd4 3.Bf3 Rc8 4.Bg2 Qe8 5.Nxe7+ Qxe7 6.Kh2 fxe4 7.Qxe4 exf4 8.Qxd4 fxe3 9.Qxe3
= (-0.17 --) Depth: 28/43 00:01:42 489MN
1...g6 2.f4 Nd4 3.Bf3 Rc8 4.Bg2 Kh8 5.b3 Bf6 6.Nxf6 Qxf6 7.Rad1 Qg7 8.Kh2 Rcd8 9.Rf2 Rg8 10.Rff1 Rge8 11.Rfe1 Qf6 12.Rf1 Rg8 13.Nd5 Qg7 14.Ne3
= (-0.17) Depth: 28/43 00:01:57 561MN
1...f4 2.Nf5 Kh8 3.Ndxe7 Nxe7 4.Nxd6 Nc6 5.c5
= (-0.25 ++) Depth: 29/43 00:02:43 800MN
1...f4 2.Bg4 Bxg4 3.Nxg4 Kh8 4.Rad1 Qc8 5.Nh2 Bd8 6.Nf3 Qh3 7.Nxf4
=/+ (-0.33 ++) Depth: 29/43 00:02:52 847MN
1...f4 2.Nf5 Kh8 3.Ndxe7 Nxe7 4.Nxd6 Nc6
=/+ (-0.45 ++) Depth: 29/43 00:03:05 910MN

[Ralph Stoesser]

Hi Robert.

When you modified the king value, probably you also changed/scaled down a whole bunch of king safety terms, linked to this value. King safety alway seems to bring sufficient elo increase, as you can turn almost any position into attacking/king safety position. That is why it is so important.

Something I do not understand is why Ralph's version would sacrifice on g4 when it sees a negative score for itself? Is this the first choice of the engine?

Hi Lyudmil,
I did not modify the king value but the bonus awarded to the king in Ralph's patch. What I noticed is that even with an extremely small bonus (200 times smaller than Ralph's), the evaluation of stockfish is still completely different from the original stockfish.
About the knight sacrifice on g4, yes Ralph's version would still sacrifice on g4 even though it sees that white is better. It's the first choice of the engine, it does not see any better move.

That is strange, Tournevisse , as the Ng4 sacrifice is probably the worse move in the position: it leads to a straightforward loss, while at the same time there are many other moves, including the 2 mentioned by you, that still hold the balance.

There is a lot of difference with the new version.
(1) it recognizes early that the recapture hxg4 is a good move
(2) it seems to slightly prefer another move than Nxg4 from about depth 30 on. That move is Be6.

Since the new version does understand that blocking a storming rook pawn with king does help it's king safety, you may want to try to play that game (after Nxg4) again with the new version. I bet you won't win that game.