The Fried Fish Attack

[Lyudmil Tsvetkov]

Another thing that comes to my mind when the pawn shelter is concerned is that as a rule the most important pawn of the immediate king shelter would not be the pawn vertically in front of the king (for example, Kg1 and g2 pawn, or Kh2 and h3 pawn), but rather, both with Kg1 and Kh1/h2, the pawn on the g file, i.e. either g2 or g3. This is a centrally placed pawn of the shelter, it shelters the king from different sides.

[Lyudmil Tsvetkov]

[d]r7/q2nk2r/2ppbp2/p3p3/N1PbP1P1/3PBPPp/P3B2K/1RQ2R2 w - - 0 25

Look again at this diagram. Another salient feature of the position is the difference in performance of the 2 larger groups of pawns/islands. The white island of 6 pawns (although with a pair of doubled pawns) is much stronger than the black island of 4 pawns. It closes the center successfully, shelters the own king, etc. There are islands and islands. Just like on a geographical map, larger islands have bigger significance.

And here vone other game on the same theme, I think more interesting, however this time against Stockfish 4. This time Stockfish sacrificed a full piece with similar positional connotations, so this might help clear the picture a bit. I know latest versions are better, but I would not be surprised if they also would go for the dubious sacrifice Stockfish 4 made. Well, as you see, I lost the game, but I think I was winning at least 5 or 6 times. In the end, in a still not very clear endgame with a piece more, I got so despaired by my dull play and the many missed opportunities, that I simply resigned.

[pgn][PlyCount "86"]
[Event "Blitz 2m+2s"]
[Site "Sofia"]
[Date "2013.11.29"]
[White "Tsvetkov, Lyudmil"]
[Black "Stockfish 4 64 SSE4.2"]
[Result "1-0"]
[ECO "C20"]
[TimeControl "120+2"]
[Annotator "Tsvetkov,Lyudmil"]
[MLNrOfMoves "43"]
[MLFlags "000100"]

{1024MB, Dell XPS 4Cores} 1. g3 {[%emt 0:00:00]} 1... e5 {[%emt 0:00:06]} 2. e4
{[%emt 0:00:04]} 2... Nf6 {[%emt 0:00:07]} 3. Bg2 {[%emt 0:00:02]} 3... Bc5
{[%emt 0:00: 05]} 4. Ne2 {[%emt 0:00:13]} 4... Nc6 {[%emt 0:00:07]} 5. O-O
{[%emt 0:00:03]} 5... h5 {[%emt 0:00:18]} 6. h3 {[%emt 0:00:05]} 6... h4
{[%emt 0:00:08]} 7. g4 {[%emt 0:00: 07]} 7... d6 {[%emt 0:00:03]} 8. d3
{[%emt 0:00:09]} 8... Nxg4 {[%emt 0:00:08]} 9. hxg4 {[%emt 0:00:02]} 9... Bxg4
{[%emt 0:00:04]} 10. Kh2 {[%emt 0:00:09]} 10... h3 {[%emt 0:00: 00]} 11. Bh1
{[%emt 0:00:02]} 11... Nd4 {[%emt 0:00:18]} 12. f3 {[%emt 0:00:14]} 12... Be6
{[%emt 0:00:09]} 13. Nxd4 {[%emt 0:00:09]} 13... Bxd4 {[%emt 0:00:04]} 14. Nc3
{[%emt 0:00:11]} 14... c6 {[%emt 0:00:10]} 15. Ne2 {[%emt 0:00:08]} 15... Bb6
{[%emt 0:00: 04]} 16. f4 {[%emt 0:00:13]} 16... Bd7 {[%emt 0:00:08]} 17. f5
{[%emt 0:00:10]} 17... Qh4 {[%emt 0:00:00]} 18. Qe1 {[%emt 0:00:08]} 18... Qh7
{[%emt 0:00:03]} 19. Ng3 {[%emt 0: 00:11]} 19... O-O-O {[%emt 0:00:00]} 20. Be3
{[%emt 0:00:12]} 20... Bc7 {[%emt 0:00:10]} 21. Bxa7 {[%emt 0:01:06]} 21... Rdg8
{[%emt 0:00:06]} 22. a4 {[%emt 0:00:09]} 22... g6 {[%emt 0:00:03]} 23. f6
{[%emt 0:00:07]} 23... Bb8 {[%emt 0:00:10]} 24. Bxb8 {[%emt 0: 00:32]} 24...
Kxb8 {[%emt 0:00:00]} 25. a5 {[%emt 0:00:16]} 25... Rc8 {[%emt 0:00:05]} 26. a6
{[%emt 0:00:14]} 26... b6 {[%emt 0:00:00]} 27. c4 {[%emt 0:00:26]} 27... Qh4
{[%emt 0: 00:04]} 28. b4 {[%emt 0:00:12]} 28... Qg5 {[%emt 0:00:04]} 29. a7+
{[%emt 0:00:24]} 29... Kb7 {[%emt 0:00:00]} 30. c5 {[%emt 0:00:20]} 30... dxc5
{[%emt 0:00:05]} 31. bxc5 {[%emt 0:00:02]} 31... b5 {[%emt 0:00:01]} 32. d4
{[%emt 0:00:00]} 32... Ka8 {[%emt 0:00:03]} 33. d5 {[%emt 0:00:15]} 33... Rh4
{[%emt 0:00:00]} 34. Qc1 {[%emt 0:00:54]} 34... Qxc1 {[%emt 0:00:05]} 35. Rfxc1
{[%emt 0:00:03]} 35... Rc7 {[%emt 0:00:02]} 36. d6 {[%emt 0:00:08]} 36... Rxa7
{[%emt 0:00:07]} 37. Rxa7+ {[%emt 0:00:02]} 37... Kxa7 {[%emt 0:00:01]} 38. Ne2
{[%emt 0:00:47]} 38... Kb7 {[%emt 0:00:00]} 39. Bf3 {[%emt 0:00:00]} 39... Kc8
{[%emt 0:00:04]} 40. Kg3 {[%emt 0:00:03]} 40... g5 {[%emt 0:00:00]} 41. Rb1
{[%emt 0: 00:31]} 41... Rh6 {[%emt 0:00:04]} 42. Nc1 {[%emt 0:00:11]} 42... g4
{[%emt 0:00:02]} 43. Bh1 {[%emt 0:00:04]} 43... h2 {[%emt 0:00:01]} 1-0
[/pgn]

[d]r1bqk2r/pppp1pp1/2n2n2/2b1p2p/4P3/6P1/PPPPNPBP/RNBQ1RK1 w kq - 0 6
White plays again a strange opening, and Stockfish counters with h5.

[d]r1bqk2r/ppp2pp1/2np4/2b1p3/4P1np/3P3P/PPP1NPB1/RNBQ1RK1 w kq - 0 9
Stockfish has just sacrificed a piece with Ng4. For me, black does not have any compensation. Do latest versions also play like that?

[d]r2qk2r/ppp2pp1/3p4/2b1p3/3nP1b1/3P1P1p/PPP1N2K/RNBQ1R1B b kq - 0 12
I like the white position. Would be surprised if Stockfish thinks it is better here.

[d]2kr3r/pp1b1ppq/1bpp4/4pP2/4P3/3PB1Np/PPP4K/R3QR1B b - - 0 20
Where is the black compensation? Black has a strong storming pawn/passer on h3, but that is all. The g black pawn can advance maximum to g4, after which it is firmly blocked by the white knight on g3. So both black h3 and g4 passers/storming pawns would be firmly blocked by the white king and knight. The center is more or less closed, so black can not hope to do anything more and the white piece more will gradually tell.

[d]1bk3rr/Bp1b1p1q/2pp1Pp1/4p3/P3P3/3P2Np/1PP4K/R3QR1B w - - 0 24
Now, I can not win that. I exchange bishops, but of course Bb6 would have been the right move.

[d]k1r4r/P2b1p2/2p2Pp1/1pP1p1q1/3PP3/6Np/7K/R3QR1B w - - 0 33
Now, I can not win that. I am sure engines quickly see some mating solutions, but I need more time.

[d]8/k2b1p2/2pP1Pp1/1pP1p3/4P2r/6Np/7K/2R4B w - - 0 38
Unable to win in the middlegame, unable to win in the endgame. I think this is still won for white, although not that easy.

[d]2k5/3b1p2/2pP1P1r/1pP1p3/4P1p1/6K1/7p/1RN4B w - - 0 44
Here I got so despaired by my impossible game, that I simply resigned.

I think there is one takeback in this game (but this is the first game against Stockfish that I post with a takeback), I think I took back some move in already clearly winning position after getting distracted and playing some very weak move, but that did not help me either, and probably that is one of the reasons (feeling guilty that I took back a move) that I resigned so eagerly in the end position. Nothing really worked for me in that game.

But I think that is not the most important thing, the game was very winning for white and I could have converted with more time, the interesting thing is why Stockfish sometimes readily considers dubious sacrifices.

[Lyudmil Tsvetkov]

Well, actually I remembered wrongly, I see now I have saved the game as my win, which is about right in terms of the ensuing positions, but fully wrong, as I took back a move and even the end positions is not very clear which side favours.

I must have been very angry at getting an enormous advantage and then making mistakes every single move, probably someone also distracted me, so I decided to adjudicate a win for myself (an important rule is never leave Stockfish adjudicate games).

[mohzus]

Thank you immensily Tsvetkov for this review of stockfish.
Ralph successfully created a patch that passed both stages of sprt.
Now Stockfish's evaluation of the fen [d]r2qk2r/ppp2p2/2npbn2/2b1p3/2P1P1P1/2NB1PPp/PPNP3K/R1BQ1R2 b kq - 0 13 is COMPLETELY different from the one of the previous dev version.
Here's the eval of the "old" SF:

Code: Select all

info depth 26 seldepth 39 score cp 53 nodes 1039146588 nps 1303870 tbhits 0 time 796971 multipv 1 pv e6d7 h2h1 c6d4 c2e3 f6h7 c3a4 h7g5 a4c5 d6c5 f3f4 g5e6 d1e1 c7c6 b2b3 d8c7 c1b2 e8c8 e1f2 c7d6 g4g5 h8e8 h1g1

, so an advantage of 53 cp for black.
Now the new SF's eval is :

Code: Select all

info depth 29 seldepth 42 score cp -162 nodes 1136784845 nps 1325336 time 857733 multipv 1 pv e6d7 c2e3 c6b4 d3e2 b4c6 b2b3 c6d4 c1b2 a7a6 e2d3 h8g8 a2a3 c5a7 b3b4 c7c6 a3a4 a6a5 b4b5 a7c5 b2a3 c5a3 a1a3 e8e7 a3a1

So an advantage of 162 cp for WHITE. A totally different SF.
And it's been proven that it gained strength in general compared to the old SF.
So a big thank you to you Tsvetkov to ignite the idea in Ralph to create a patch and a big thank you to Ralph for coding it

[Lyudmil Tsvetkov]

Thank you immensily Tsvetkov for this review of stockfish.
Ralph successfully created a patch that passed both stages of sprt.
Now Stockfish's evaluation of the fen [d]r2qk2r/ppp2p2/2npbn2/2b1p3/2P1P1P1/2NB1PPp/PPNP3K/R1BQ1R2 b kq - 0 13 is COMPLETELY different from the one of the previous dev version.
Here's the eval of the "old" SF:

Code: Select all

info depth 26 seldepth 39 score cp 53 nodes 1039146588 nps 1303870 tbhits 0 time 796971 multipv 1 pv e6d7 h2h1 c6d4 c2e3 f6h7 c3a4 h7g5 a4c5 d6c5 f3f4 g5e6 d1e1 c7c6 b2b3 d8c7 c1b2 e8c8 e1f2 c7d6 g4g5 h8e8 h1g1

, so an advantage of 53 cp for black.
Now the new SF's eval is :

Code: Select all

info depth 29 seldepth 42 score cp -162 nodes 1136784845 nps 1325336 time 857733 multipv 1 pv e6d7 c2e3 c6b4 d3e2 b4c6 b2b3 c6d4 c1b2 a7a6 e2d3 h8g8 a2a3 c5a7 b3b4 c7c6 a3a4 a6a5 b4b5 a7c5 b2a3 c5a3 a1a3 e8e7 a3a1

So an advantage of 162 cp for WHITE. A totally different SF.
And it's been proven that it gained strength in general compared to the old SF.
So a big thank you to you Tsvetkov to ignite the idea in Ralph to create a patch and a big thank you to Ralph for coding it

All thanks go to Ralph, in his place I would not have known what to do, but he knew.

We will see if Stockfish got better when I download the next official release and play some games.

An advantage of 162cps for white, I thought I was better, but so much better...

The interesting thing for me is if latest Stockfish would still sacrifice its knight on g4 in the position from the second game I posted.

[jmartus]

Houdini 4.0 Pro

move 8 for black
[d]r1bqk2r/ppp2pp1/2np1n2/2b1p3/4P1Pp/3P3P/PPP1NPB1/RNBQ1RK1 b kq - 0 8

Houdini_4_Pro_x64A-v1x:
21/58+ 00:22 117,940,606 5,273,000 +0.72 Nf6xg4
21/58 00:27 144,449,211 5,298,000 +0.82 Nf6xg4 Nb1-c3 Ng4-f6 Bc1-g5 Rh8-h5 Bg5-d2 Bc5-b6 Nc3-a4 Bc8-e6 Na4xb6 a7xb6 Ne2-c3 Rh5-h8 Nc3-b5 Nf6-d7 c2-c4 Nd7-c5 Qd1-e2 Be6-d7 Bd2-e3 g7-g5 b2-b3 f7-f6 d3-d4 e5xd4 Nb5xd4 Qd8-e7 Nd4xc6 b7xc6 Be3xc5 b6xc5
22/58 00:32 173,840,871 5,332,000 +0.80 Nf6xg4 Nb1-c3 Ng4-f6 Bc1-g5 Rh8-h5 Bg5-d2 Bc5-b6 Nc3-a4 Bc8-e6 Na4xb6 a7xb6 Ne2-c3 Rh5-h8 Nc3-b5 Nf6-d7 a2-a4 g7-g5 Kg1-h1 Rh8-g8 Rf1-g1 Nd7-c5 b2-b4 Nc5-d7 Qd1-f3 g5-g4 h3xg4 Rg8xg4 Bg2-h3 Rg4xg1+ Kh1xg1
23/58 00:46 253,236,348 5,432,000 +0.79 Nf6xg4 Nb1-c3 Ng4-f6 Bc1-g5 Rh8-h5 Bg5-d2 Bc5-b6 Nc3-a4 Bc8-d7 a2-a3 Rh5-h8 b2-b4 Qd8-c8 Kg1-h2 Nf6-h5 Na4xb6 a7xb6 c2-c4 Nc6-e7 a3-a4 Ne7-g6 a4-a5 b6xa5 Ra1xa5 Ra8xa5 b4xa5 Ke8-f8 Rf1-g1 Ng6-f4 Ne2xf4 Nh5xf4 Bd2xf4 e5xf4
24/63 01:21 451,750,100 5,574,000 +0.74 Nf6xg4 Nb1-c3 Ng4-f6 Bc1-g5 Rh8-h5 Bg5-d2 Bc5-b6 Nc3-a4 Bc8-d7 a2-a3 Rh5-h8 b2-b4 Qd8-c8 Kg1-h2 Ke8-f8 Na4xb6 a7xb6 b4-b5 Nc6-e7 a3-a4 Nf6-h5 Rf1-g1 Kf8-g8 Qd1-f1 c7-c6 Bg2-f3 c6xb5 Qf1-g2 g7-g6
25/63+ 02:06 713,107,060 5,631,000 +0.85 Nf6xg4
25/69 02:44 930,514,904 5,649,000 +0.67 Nf6xg4 Nb1-c3 Ng4-f6 Bc1-g5 Rh8-h5 Bg5-d2 Bc5-b6 Nc3-d5 Nf6xd5 e4xd5 Nc6-e7 d3-d4 Ke8-f8 c2-c4 Rh5-f5 Bd2-e3 a7-a5 Qd1-d3 Kf8-g8 Ra1-e1 Bc8-d7 Bg2-e4 Rf5-h5 Be4-f3 Rh5-h8 Bf3-g2 Bd7-f5 Qd3-c3 f7-f6 d4xe5
26/69+ 03:46 1,295,325,927 5,716,000 +0.78 Nf6xg4
26/69 03:58 1,359,178,570 5,702,000 +0.78 Nf6xg4 Nb1-c3 Ng4-f6 Bc1-g5 Rh8-h5 Bg5-d2 Bc8-d7 Nc3-d5 Nc6-e7 b2-b4 Bc5-b6 Nd5xb6 a7xb6 f2-f4 Ke8-f8 f4xe5 d6xe5 Ne2-c3 Rh5-h7 Nc3-d5 Nf6xd5 e4xd5 Kf8-g8 Bg2-e4 Rh7-h8 Qd1-f3 f7-f5 d5-d6 c7xd6 Be4xb7 Ra8-a3 Bb7-d5+ Ne7xd5 Qf3xd5+ Kg8-h7 Qd5xd6 Rh8-e8 b4-b5 Ra3-a8


FEN: r1bqk2r/ppp2pp1/2np1n2/2b1p3/4P1Pp/3P3P/PPP1NPB1/RNBQ1RK1 b kq - 0 8

Stockfish 192014:

26/32 00:22 82,942,118 3,680,754 +1.58 Nf6xg4 Ne2-c3 Ng4-f6 Nc3-d5 Nf6xd5 e4xd5 Nc6-e7 Nb1-d2 Bc8-f5 Nd2-b3 Bc5-b6 a2-a4 a7-a6 a4-a5 Bb6-a7 Bc1-e3 Qd8-d7 Qd1-f3 Ba7xe3 f2xe3 O-O-O e3-e4 Bf5-g6 d3-d4 f7-f6 c2-c4
27/36- 00:44 164,894,449 3,744,452 +1.50 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Qd8-d7 Bc1-g5 f7-f6 Bg5-d2 O-O-O d3-d4 Bc5xd4 Ne2xd4 e5xd4 h3xg4 Be6xg4 f2-f3 h4-h3 f3xg4 h3xg2 Kg1xg2 Rh8-h4 Qd1-f3 Rh4xg4+ Kg2-f2 Nc6-e5 Qf3-f5
27/37+ 00:50 188,885,280 3,754,577 +1.66 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Qd8-d7 Bc1-g5 a7-a6 Ne2-c3 f7-f6 Bg5-d2 Ng4-h6 Qd1-h5+ Nh6-f7 Qh5-g6 O-O-O Qg6xg7 Rd8-g8
27/41- 01:04 243,933,142 3,802,780 +1.42 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Qd8-d7 Bc1-g5 a7-a6 h3xg4 Be6xg4 Kg1-h2 h4-h3 Bg2-h1 Rh8-h5 Bg5-d2 f7-f6 f2-f3 Bg4-e6 b2-b4 Bc5-a7 Bd2-e3 Ba7xe3 Nd5xe3 Nc6xb4 Ne2-g3 Rh5-h4 Ng3-f5
27/41- 01:15 287,573,130 3,822,941 +1.24 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Qd8-d7 Bc1-g5 a7-a6 h3xg4 Be6xg4 Kg1-h2 h4-h3 Bg2-h1 Rh8-h5 Bg5-d2 f7-f6 f2-f3 Bg4-e6 b2-b4 Bc5-a7 Bd2-e3 Ba7xe3 Nd5xe3 Nc6xb4 Ne2-g3 Rh5-h4 Ng3-f5
27/41 01:30 344,461,922 3,825,696 +1.38 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Qd8-d7 Bc1-g5 Ng4-h6 Kg1-h2 f7-f5 e4xf5 Nh6xf5 b2-b4 Bc5xb4 Nd5xb4 Nc6xb4 Bg2xb7 Ra8-b8 Bb7-g2 O-O c2-c3 Nb4-d5 d3-d4 e5xd4 c3xd4 Rb8-b2 Bg5-c1 Rb2-b6 Ne2-f4 Nd5xf4
28/41+ 01:34 361,956,084 3,826,093 +1.46 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Qd8-d7 Bc1-g5 Ng4-h6 Kg1-h2 f7-f5 e4xf5 Nh6xf5
28/41 01:42 394,024,540 3,829,980 +1.45 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Qd8-d7 Bc1-g5 Ng4-h6 Kg1-h2 f7-f5 b2-b4 Bc5xb4 Bg5xh6 Rh8xh6 Qd1-b1 Bb4-a5 e4xf5 Be6xf5 Qb1xb7 Ra8-b8 Qb7-a6 Ba5-b6 Qa6-c4 Nc6-e7 Ne2-c3 Bf5-e6 Ra1-b1 c7-c6
29/41- 01:54 438,077,798 3,838,242 +1.37 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Qd8-d7 Bc1-g5 Ng4-h6 Kg1-h2 f7-f5 b2-b4 Bc5xb4 Ra1-b1 Nh6-f7 Bg5-c1 a7-a5 e4xf5 Be6xd5 Bg2xd5 Nf7-d8 a2-a3 Bb4-c5 Ne2-c3 Qd7xf5 Nc3-e4
29/41+ 02:07 488,667,046 3,846,984 +1.54 Nf6xg4 Nb1-c3 Bc8-e6
29/41- 02:26 565,353,992 3,860,810 +1.29 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Qd8-d7 h3xg4 Be6xg4 Kg1-h2 Qd7-d8 Bg2-h1 Nc6-d4 f2-f3 Nd4xe2 f3xg4 Ne2-g3 Rf1-f3
29/41- 03:11 744,426,891 3,896,890 +1.11 Nf6xg4 Nb1-c3
29/48 05:34 1,299,718,988 3,883,548 +1.23 Nf6xg4 Nb1-c3 Qd8-d7 Nc3-d5 Ng4-f6 Kg1-h1 Nf6xd5 e4xd5 Nc6-e7 d3-d4 e5xd4 Ne2xd4 O-O c2-c4 Ne7-f5 Nd4-b3 Bc5-b6 a2-a4 a7-a5 Bc1-d2 Qd7-d8 Bd2-c3 Qd8-g5 Bc3-d2 Qg5-g6
30/48- 06:14 1,455,304,054 3,889,502 +1.15 Nf6xg4 Nb1-c3 Qd8-d7 Nc3-d5 Ng4-f6 Kg1-h1 Nf6xd5 e4xd5 Nc6-e7 d3-d4 e5xd4
30/48- 07:29 1,747,937,906 3,892,263 +1.07 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Ng4-f6 c2-c3 Bc5-b6 d3-d4 Qd8-d7 Qd1-d3 e5xd4 Nd5xb6 a7xb6 Ne2xd4 Nc6xd4 c3xd4 d6-d5 e4-e5 Nf6-e4 Rf1-e1 Be6-f5 Qd3-f3 c7-c5 Bc1-e3 O-O
30/48 10:12 2,400,851,013 3,922,305 +1.07 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Ng4-f6 c2-c3 Bc5-b6 d3-d4 Qd8-d7 Qd1-d3 e5xd4 Nd5xb6 a7xb6 Ne2xd4 Nc6xd4 c3xd4 d6-d5 e4-e5 Nf6-e4 Rf1-e1 Be6-f5 Qd3-f3 c7-c5 Bc1-e3 O-O

[Ralph Stoesser]

A big thanks also from me again for this game. I haven't yet tried to analyze the games with the new SF, but I'm rather confident that SF will not sacrifice the knight anymore. It also should not play g5 in the opening of the first game. The patch itself looks a little crude, but seems to work and most probably will be improved soon. It tries to handle advanced rook passers that are being blocked by the king as if they would be part of the own pawn shelter. A eval correction term is applied for that purpose, in king safety eval only. Passed pawn eval looks ok for the h3 passer as you can see from the eval command output. The patch will become applied to SF master branch tomorrow and then we can pick the "official" executables and analyze both games. Perhaps that will trigger the next ideas for further improvements?!:-)

[mohzus]

All thanks go to Ralph, in his place I would not have known what to do, but he knew.

I understand you perfectly since I am in the same boat as you with respect to programming. But you still have credit to report the interesting positions and your thoughts on them.

We will see if Stockfish got better when I download the next official release and play some games.

Hmm why would you wait until the next "official" release when the dev releases are also available from Stockfish's website without the need to compile?

An advantage of 162cps for white, I thought I was better, but so much better...

Me too! Using my brain only (around average chess player), I would have thought that the evaluation should return a number close to 0. With my very limited chess knowledge, I could not see any advantage for either white (despite being 1 pawn up) nor black.
So I had an idea: modify a value in the Ralph's patch. He gave a bonus to the king of 200 (cp I guess) or something like that. Hard for me to understand really what's going on when I don't know programming. The thing is, I reduced this bonus to 35 and let stockfish analyze the position. It gave an advantage for white, too big to my taste. So I lowered the value again and again, until finally I reached a bonus of "1". With this seemingly small value, SF would still favor white (by around 35 cp which looked more reasonable to me). So I ran a test in the fishtest with that value. It failed at 15+0.05 time control, although it didn't seem to be harmful overall; it just didn't bring any strength gain.

The interesting thing for me is if latest Stockfish would still sacrifice its knight on g4 in the position from the second game I posted.

I cannot say with 100% accuracy because Marco did not commit yet Ralph's patch and I do not know exactly how the next SF dev release will be, but using Ralph's version of SF I get the following:

Code: Select all

info depth 29 seldepth 44 score cp -34 nodes 755144879 nps 1354269 time 557603 multipv 1 pv f6g4 h3g4 c8g4 g1h2 d8f6 d1e1 h4h3 g2h1 g4f3 c1g5 f6g5 h1f3 c6d4 e2d4 g5f4 h2h1 c5d4 b1d2 d4b2 a1d1 h8h4 e1e3 e8e7 f1g1 a8h8 e3f4

So yes, SF would still sacrifice its knight on g4. On the other hand, the evaluation is totally different from the "old SF", if you look at Joel Martus analysis... 107 cp (with respect to black's viewpoint?), while Ralph's version gives an advantage to white of 34 cp.

[Lyudmil Tsvetkov]

Houdini 4.0 Pro

move 8 for black
[d]r1bqk2r/ppp2pp1/2np1n2/2b1p3/4P1Pp/3P3P/PPP1NPB1/RNBQ1RK1 b kq - 0 8

Houdini_4_Pro_x64A-v1x:
21/58+ 00:22 117,940,606 5,273,000 +0.72 Nf6xg4
21/58 00:27 144,449,211 5,298,000 +0.82 Nf6xg4 Nb1-c3 Ng4-f6 Bc1-g5 Rh8-h5 Bg5-d2 Bc5-b6 Nc3-a4 Bc8-e6 Na4xb6 a7xb6 Ne2-c3 Rh5-h8 Nc3-b5 Nf6-d7 c2-c4 Nd7-c5 Qd1-e2 Be6-d7 Bd2-e3 g7-g5 b2-b3 f7-f6 d3-d4 e5xd4 Nb5xd4 Qd8-e7 Nd4xc6 b7xc6 Be3xc5 b6xc5
22/58 00:32 173,840,871 5,332,000 +0.80 Nf6xg4 Nb1-c3 Ng4-f6 Bc1-g5 Rh8-h5 Bg5-d2 Bc5-b6 Nc3-a4 Bc8-e6 Na4xb6 a7xb6 Ne2-c3 Rh5-h8 Nc3-b5 Nf6-d7 a2-a4 g7-g5 Kg1-h1 Rh8-g8 Rf1-g1 Nd7-c5 b2-b4 Nc5-d7 Qd1-f3 g5-g4 h3xg4 Rg8xg4 Bg2-h3 Rg4xg1+ Kh1xg1
23/58 00:46 253,236,348 5,432,000 +0.79 Nf6xg4 Nb1-c3 Ng4-f6 Bc1-g5 Rh8-h5 Bg5-d2 Bc5-b6 Nc3-a4 Bc8-d7 a2-a3 Rh5-h8 b2-b4 Qd8-c8 Kg1-h2 Nf6-h5 Na4xb6 a7xb6 c2-c4 Nc6-e7 a3-a4 Ne7-g6 a4-a5 b6xa5 Ra1xa5 Ra8xa5 b4xa5 Ke8-f8 Rf1-g1 Ng6-f4 Ne2xf4 Nh5xf4 Bd2xf4 e5xf4
24/63 01:21 451,750,100 5,574,000 +0.74 Nf6xg4 Nb1-c3 Ng4-f6 Bc1-g5 Rh8-h5 Bg5-d2 Bc5-b6 Nc3-a4 Bc8-d7 a2-a3 Rh5-h8 b2-b4 Qd8-c8 Kg1-h2 Ke8-f8 Na4xb6 a7xb6 b4-b5 Nc6-e7 a3-a4 Nf6-h5 Rf1-g1 Kf8-g8 Qd1-f1 c7-c6 Bg2-f3 c6xb5 Qf1-g2 g7-g6
25/63+ 02:06 713,107,060 5,631,000 +0.85 Nf6xg4
25/69 02:44 930,514,904 5,649,000 +0.67 Nf6xg4 Nb1-c3 Ng4-f6 Bc1-g5 Rh8-h5 Bg5-d2 Bc5-b6 Nc3-d5 Nf6xd5 e4xd5 Nc6-e7 d3-d4 Ke8-f8 c2-c4 Rh5-f5 Bd2-e3 a7-a5 Qd1-d3 Kf8-g8 Ra1-e1 Bc8-d7 Bg2-e4 Rf5-h5 Be4-f3 Rh5-h8 Bf3-g2 Bd7-f5 Qd3-c3 f7-f6 d4xe5
26/69+ 03:46 1,295,325,927 5,716,000 +0.78 Nf6xg4
26/69 03:58 1,359,178,570 5,702,000 +0.78 Nf6xg4 Nb1-c3 Ng4-f6 Bc1-g5 Rh8-h5 Bg5-d2 Bc8-d7 Nc3-d5 Nc6-e7 b2-b4 Bc5-b6 Nd5xb6 a7xb6 f2-f4 Ke8-f8 f4xe5 d6xe5 Ne2-c3 Rh5-h7 Nc3-d5 Nf6xd5 e4xd5 Kf8-g8 Bg2-e4 Rh7-h8 Qd1-f3 f7-f5 d5-d6 c7xd6 Be4xb7 Ra8-a3 Bb7-d5+ Ne7xd5 Qf3xd5+ Kg8-h7 Qd5xd6 Rh8-e8 b4-b5 Ra3-a8


FEN: r1bqk2r/ppp2pp1/2np1n2/2b1p3/4P1Pp/3P3P/PPP1NPB1/RNBQ1RK1 b kq - 0 8

Stockfish 192014:

26/32 00:22 82,942,118 3,680,754 +1.58 Nf6xg4 Ne2-c3 Ng4-f6 Nc3-d5 Nf6xd5 e4xd5 Nc6-e7 Nb1-d2 Bc8-f5 Nd2-b3 Bc5-b6 a2-a4 a7-a6 a4-a5 Bb6-a7 Bc1-e3 Qd8-d7 Qd1-f3 Ba7xe3 f2xe3 O-O-O e3-e4 Bf5-g6 d3-d4 f7-f6 c2-c4
27/36- 00:44 164,894,449 3,744,452 +1.50 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Qd8-d7 Bc1-g5 f7-f6 Bg5-d2 O-O-O d3-d4 Bc5xd4 Ne2xd4 e5xd4 h3xg4 Be6xg4 f2-f3 h4-h3 f3xg4 h3xg2 Kg1xg2 Rh8-h4 Qd1-f3 Rh4xg4+ Kg2-f2 Nc6-e5 Qf3-f5
27/37+ 00:50 188,885,280 3,754,577 +1.66 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Qd8-d7 Bc1-g5 a7-a6 Ne2-c3 f7-f6 Bg5-d2 Ng4-h6 Qd1-h5+ Nh6-f7 Qh5-g6 O-O-O Qg6xg7 Rd8-g8
27/41- 01:04 243,933,142 3,802,780 +1.42 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Qd8-d7 Bc1-g5 a7-a6 h3xg4 Be6xg4 Kg1-h2 h4-h3 Bg2-h1 Rh8-h5 Bg5-d2 f7-f6 f2-f3 Bg4-e6 b2-b4 Bc5-a7 Bd2-e3 Ba7xe3 Nd5xe3 Nc6xb4 Ne2-g3 Rh5-h4 Ng3-f5
27/41- 01:15 287,573,130 3,822,941 +1.24 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Qd8-d7 Bc1-g5 a7-a6 h3xg4 Be6xg4 Kg1-h2 h4-h3 Bg2-h1 Rh8-h5 Bg5-d2 f7-f6 f2-f3 Bg4-e6 b2-b4 Bc5-a7 Bd2-e3 Ba7xe3 Nd5xe3 Nc6xb4 Ne2-g3 Rh5-h4 Ng3-f5
27/41 01:30 344,461,922 3,825,696 +1.38 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Qd8-d7 Bc1-g5 Ng4-h6 Kg1-h2 f7-f5 e4xf5 Nh6xf5 b2-b4 Bc5xb4 Nd5xb4 Nc6xb4 Bg2xb7 Ra8-b8 Bb7-g2 O-O c2-c3 Nb4-d5 d3-d4 e5xd4 c3xd4 Rb8-b2 Bg5-c1 Rb2-b6 Ne2-f4 Nd5xf4
28/41+ 01:34 361,956,084 3,826,093 +1.46 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Qd8-d7 Bc1-g5 Ng4-h6 Kg1-h2 f7-f5 e4xf5 Nh6xf5
28/41 01:42 394,024,540 3,829,980 +1.45 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Qd8-d7 Bc1-g5 Ng4-h6 Kg1-h2 f7-f5 b2-b4 Bc5xb4 Bg5xh6 Rh8xh6 Qd1-b1 Bb4-a5 e4xf5 Be6xf5 Qb1xb7 Ra8-b8 Qb7-a6 Ba5-b6 Qa6-c4 Nc6-e7 Ne2-c3 Bf5-e6 Ra1-b1 c7-c6
29/41- 01:54 438,077,798 3,838,242 +1.37 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Qd8-d7 Bc1-g5 Ng4-h6 Kg1-h2 f7-f5 b2-b4 Bc5xb4 Ra1-b1 Nh6-f7 Bg5-c1 a7-a5 e4xf5 Be6xd5 Bg2xd5 Nf7-d8 a2-a3 Bb4-c5 Ne2-c3 Qd7xf5 Nc3-e4
29/41+ 02:07 488,667,046 3,846,984 +1.54 Nf6xg4 Nb1-c3 Bc8-e6
29/41- 02:26 565,353,992 3,860,810 +1.29 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Qd8-d7 h3xg4 Be6xg4 Kg1-h2 Qd7-d8 Bg2-h1 Nc6-d4 f2-f3 Nd4xe2 f3xg4 Ne2-g3 Rf1-f3
29/41- 03:11 744,426,891 3,896,890 +1.11 Nf6xg4 Nb1-c3
29/48 05:34 1,299,718,988 3,883,548 +1.23 Nf6xg4 Nb1-c3 Qd8-d7 Nc3-d5 Ng4-f6 Kg1-h1 Nf6xd5 e4xd5 Nc6-e7 d3-d4 e5xd4 Ne2xd4 O-O c2-c4 Ne7-f5 Nd4-b3 Bc5-b6 a2-a4 a7-a5 Bc1-d2 Qd7-d8 Bd2-c3 Qd8-g5 Bc3-d2 Qg5-g6
30/48- 06:14 1,455,304,054 3,889,502 +1.15 Nf6xg4 Nb1-c3 Qd8-d7 Nc3-d5 Ng4-f6 Kg1-h1 Nf6xd5 e4xd5 Nc6-e7 d3-d4 e5xd4
30/48- 07:29 1,747,937,906 3,892,263 +1.07 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Ng4-f6 c2-c3 Bc5-b6 d3-d4 Qd8-d7 Qd1-d3 e5xd4 Nd5xb6 a7xb6 Ne2xd4 Nc6xd4 c3xd4 d6-d5 e4-e5 Nf6-e4 Rf1-e1 Be6-f5 Qd3-f3 c7-c5 Bc1-e3 O-O
30/48 10:12 2,400,851,013 3,922,305 +1.07 Nf6xg4 Nb1-c3 Bc8-e6 Nc3-d5 Ng4-f6 c2-c3 Bc5-b6 d3-d4 Qd8-d7 Qd1-d3 e5xd4 Nd5xb6 a7xb6 Ne2xd4 Nc6xd4 c3xd4 d6-d5 e4-e5 Nf6-e4 Rf1-e1 Be6-f5 Qd3-f3 c7-c5 Bc1-e3 O-O

Thanks Joel!

So both Houdini and Stockfish think black is much better, however they do not even consider accepting the sacrifice. I, for my part, eat that knight with great gusto and feel quite comfortable with the ensuing position.

[Lyudmil Tsvetkov]

All thanks go to Ralph, in his place I would not have known what to do, but he knew.

I understand you perfectly since I am in the same boat as you with respect to programming. But you still have credit to report the interesting positions and your thoughts on them.

We will see if Stockfish got better when I download the next official release and play some games.

Hmm why would you wait until the next "official" release when the dev releases are also available from Stockfish's website without the need to compile?

An advantage of 162cps for white, I thought I was better, but so much better...

Me too! Using my brain only (around average chess player), I would have thought that the evaluation should return a number close to 0. With my very limited chess knowledge, I could not see any advantage for either white (despite being 1 pawn up) nor black.
So I had an idea: modify a value in the Ralph's patch. He gave a bonus to the king of 200 (cp I guess) or something like that. Hard for me to understand really what's going on when I don't know programming. The thing is, I reduced this bonus to 35 and let stockfish analyze the position. It gave an advantage for white, too big to my taste. So I lowered the value again and again, until finally I reached a bonus of "1". With this seemingly small value, SF would still favor white (by around 35 cp which looked more reasonable to me). So I ran a test in the fishtest with that value. It failed at 15+0.05 time control, although it didn't seem to be harmful overall; it just didn't bring any strength gain.

The interesting thing for me is if latest Stockfish would still sacrifice its knight on g4 in the position from the second game I posted.

I cannot say with 100% accuracy because Marco did not commit yet Ralph's patch and I do not know exactly how the next SF dev release will be, but using Ralph's version of SF I get the following:

Code: Select all

info depth 29 seldepth 44 score cp -34 nodes 755144879 nps 1354269 time 557603 multipv 1 pv f6g4 h3g4 c8g4 g1h2 d8f6 d1e1 h4h3 g2h1 g4f3 c1g5 f6g5 h1f3 c6d4 e2d4 g5f4 h2h1 c5d4 b1d2 d4b2 a1d1 h8h4 e1e3 e8e7 f1g1 a8h8 e3f4

So yes, SF would still sacrifice its knight on g4. On the other hand, the evaluation is totally different from the "old SF", if you look at Joel Martus analysis... 107 cp (with respect to black's viewpoint?), while Ralph's version gives an advantage to white of 34 cp.

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?