Request for New Glaurung analysis

[Eraserheads]

I heard from Tord that there is a new and exciting Glaurung (not yet available) which tends to make exchange sacs often. Can someone post some analysis with it of some of the sacs of the game below? Thanks!

The post below was copied from the Thinkers' forum.


=-==========================================
Every once in a while, I get to play the way I like to play.

Here is a game where I got to sacrifice not just one, two, or three pieces against a computer, and I drew the game!

The game was eventually drawn.

[Event "Blitz:3'"]
[Site "?"]
[Date "2008.05.02"]
[Round "?"]
[White "Reyes, Michael"]
[Black "Genius, Chess"]
[Result "1/2-1/2"]
[ECO "B19"]
[Annotator "Reyes,Michael"]
[PlyCount "76"]


{24MB, Fritz11.ctg, WINDOWS-MIX6LP5} 1. e4 c6 2. d4 d5 3. Nc3 dxe4 4. Nxe4 Bf5
5. Ng3 Bg6 6. h4 h6 7. Nf3 Nd7 8. h5 Bh7 9. Bd3 Bxd3 10. Qxd3 e6 11. Bf4 Ngf6
12. O-O-O Be7 13. Kb1 Nd5 14. Bd2 O-O 15. c4 N5f6 16. Bc3 Qc7 17. Rhe1 Bd6 18.
Ne4 Nxh5 19. d5 cxd5 20. Neg5 hxg5 21. Nxg5 Ndf6 22. cxd5 Nf4 23. Qf3 N6xd5 24.
Bxg7 Kxg7 25. Qg4 Ng6 26. Rxe6 Nf6 27. Rxf6 Kxf6 28. Rxd6+ Qxd6 29. Ne4+ Ke5
30. Nxd6 Kxd6 31. Qg3+ Ne5 32. f4 Nc4 33. Qd3+ Kc5 34. Qf5+ Kb6 35. Qf6+ Kb5
36. Qf5+ Kb4 37. a3+ Nxa3+ 38. bxa3+ Kxa3 1/2-1/2 Drawn by perpetual.
===============================================

[Eelco de Groot]

Hi Michael,

That looks like a fine game against Genius. I did not do any deep analysis but here is some output from Glaurung Mjolnir, Glaurung in the 080422 version but with altered code.

The output is from the Shredder GUI that made an analysis starting from the end of the game and commenting if Glaurung's choices were different with more than half a pawn. At the end though the time-settings were maybe too low because I had not instructed something like a fixed or average time per move but a time for the whole game. The analysis for the early opening did not go very well I think, the engine seemed to skip from move thirteen to move three, maybe that was as those were still in book but so was move three. If necessary I could try again. Glaurung seems to find some alternative moves for the computer but that is not surprising when your copy of Genius only had three minutes for the whole game. With this quick analysis your moves look mostly fine, but the sacrifices are it seems not understood and Glaurung overestimated the computer's chances to win!

This version of Glaurung Mjolnir, it is Build No. 47 03-05-2008 6:21 hr, Mjolnir 006b settings, overevaluates passed pawns sometimes and gives really really big scores for them, Tord's version is probably different

Best, Eelco

[Event "Blitz:3'"]
[Site "?"]
[Date "2008.05.02"]
[Round "?"]
[White "Reyes, Michael"]
[Black "Genius, Chess"]
[ECO "B19 - Caro Kann/Classical Variation"]
[Result "1/2-1/2"]

1. e4 c6 2. d4 d5 3. Nc3 {+0.54/13 2s} (3. e5 {+1.07/13
15s} Bf5 4. Nf3 e6 5. Bd3 Bxd3 6. Qxd3 c5 7. O-O Nc6
8. dxc5 Bxc5 9. Nc3 Nge7 10. Be3 Bxe3 11. Qxe3) 3... dxe4
4. Nxe4 Bf5 5. Ng3 Bg6 6. h4 h6 7. Nf3 Nd7 8. h5 Bh7 9. Bd3
Bxd3 10. Qxd3 e6 11. Bf4 Ngf6 12. O-O-O Be7 13. Kb1 Nd5
14. Bd2 O-O 15. c4 N5f6 16. Bc3 Qc7 17. Rhe1 Bd6 18. Ne4
Nxh5 19. d5 cxd5 20. Neg5 {-4.50/15 4s} (20. cxd5 {+0.17/15
22s} Nf4 21. Qd4 e5 22. Qd2 Rfe8 23. Nxd6 Qxd6 24. Nxe5
Nxg2 25. Bb4 Nc5 26. Re2 Rxe5 27. Rxe5 Qxe5 28. Bxc5)
20... hxg5 21. Nxg5 Ndf6 22. cxd5 {-5.92/13 0s} (22. Bxf6
{-5.15/13 12s} Nxf6 23. cxd5 exd5 24. Rh1 g6 25. Ka1 Rad8
26. Qh3 Nh5 27. Qb3 Bc5 28. Rxd5 Rxd5 29. Qxd5 Bxf2)
22... Nf4 23. Qf3 N6xd5 24. Bxg7 {-6.68/13 2s} (24. Ne4
{-6.05/13 16s} Rfc8 25. g3 Ng6 26. Rc1 Qb6 27. Bd4 Rxc1+
28. Rxc1 Qxd4 29. Nxd6 Ne5 30. Qe4 Qxe4+ 31. Nxe4 Rd8
32. f4) 24... Kxg7 25. Qg4 Ng6 {+5.86/12 1s} (25... Kf6
{+7.05/12 24s} 26. Rxd5 Rac8 27. Ne4+ Ke7 28. Qg5+ f6
29. Qg7+ Rf7 30. Qxf7+ Kxf7 31. Nxd6+ Kg7 32. Nxc8 Nxd5
33. Nxa7 Qb6 34. f3 Qxa7 35. Rxe6) 26. Rxe6 Nf6 {+2.96/12
0s} (26... Nc3+ {+5.80/12 17s} 27. bxc3 Qb6+ 28. Kc2 fxe6
29. Nxe6+ Kf6 30. Nxf8 Rxf8 31. Qd4+ Qxd4 32. cxd4 Rc8+
33. Kb3 Nf4 34. g3 Nd5 35. Re1 Rc3+ 36. Kb2) 27. Rxf6 Kxf6
28. Rxd6+ Qxd6 {+0.86/13 1s} (28... Ke7 {+2.86/13 24s}
29. Rd1 Rfd8 30. Re1+ Kf6 31. Ne6 Qa5 32. b4 Qd5 33. Nxd8
Rxd8 34. Rc1 Ne5 35. Qf4+ Kg7) 29. Ne4+ Ke5 30. Nxd6 Kxd6
31. Qg3+ {-2.25/15 32s} (31. Qb4+ {0.00/15 18s} Ke6
32. Qb3+ Kf6 33. Qc3+ Kf5 34. Qc2+ Kg5 35. Qc1+ Kh5
36. Qh1+ Kg5 37. Qc1+) 31... Ne5 32. f4 {-3.39/14 9s}
(32. Qa3+ {-2.45/14 42s} Kd7 33. Qb3 Kc7 34. Qc3+ Nc6
35. Qg3+ Kc8 36. Qf4 a5 37. b3 Kd7 38. Qd2+ Kc7 39. Qf4+
Kb6 40. Qe3+ Ka6 41. a4 Rfd8) 32... Nc4 {+0.58/14 3s}
(32... Nc6 {+3.35/14 30s} 33. f5+ Kd7 34. Qd3+ Ke7 35. a3
Kf6 36. g4 Rfe8 37. Qd6+ Kg5 38. Qd7 Ne5 39. Qxb7 Kxg4
40. Qe4+ Kg3) 33. Qd3+ Kc5 34. Qf5+ Kb6 {0.00/15 1s}
(34... Kd4 {+0.84/15 31s} 35. Qf6+ Ke4 36. Qe7+ Kd3
37. Qd7+ Ke3 38. Qh3+ Kf2 39. Qf3+ Kg1 40. Qd1+ Kxg2
41. Qd5+ Kg3 42. Qxc4 f5 43. Qd4 Rfe8 44. Kc2 Rac8+
45. Kb3) 35. Qf6+ Kb5 36. Qf5+ Kb4 37. a3+ Nxa3+ 38. bxa3+
Kxa3 1/2-1/2

[Eraserheads]

Thanks for the reply elco! Interesting to see the huge eval jump between moves 24 and 25. I am looking for a UCI engine which can serve as a Junior-engine-style alternative. Glaurung tops my choice list!

[ozziejoe]

I also think Glaurung beta is a bit aggressive. You can see here that it has lower draw percentage than glaurung epsilong/5 (last percentage in line, e.g., 28 versus 32%)


3 Glaurung 080420 : 2839 43 44 180 40.0 % 2910 28.9 %

5 Glaurung 2-epsilon/5 : 2737 43 44 180 26.9 % 2910 32.8 %



if anything, I think Glaurung is too aggressive. For example, it thinks white has a full point advantage in this test opening position. It is true that white has the more aggresive, attacking position, but black is fine I think.

[d]r1b2k1r/ppqn1pp1/3bp2p/2p5/3P2Q1/3B1N2/PPP2PPP/R1B2RK1 w - - 0 13[d]


at depth 17, glaurung beta gives white a full point advantage (.94) in this position. It loves the attacking side. Its evaluation comes down with deeper depth (to about .54), but its initial impression is "attack is good."


In contrast, depth 17, glaurung epsilon 5 gives the position .66 for white..

rybka , at depth 15 (probably true depth of 17), scores the position much lower than glaurung (.24)

I suspect glaurung could be improved by making her more conservative, but that would also be less interesting to many people.


Detailed analysis by rybka is below



Analysis by Rybka 2.3.2a mp 32-bit :

13.h3 cxd4
² (0.37) Depth: 5 00:00:00
13.Bd2 Nf6
² (0.45) Depth: 5 00:00:00
13.Qh4 c4
² (0.53) Depth: 5 00:00:00
13.Qh4 c4 14.Be2
² (0.45) Depth: 6 00:00:00 17kN
13.Qh4 c4 14.Be2 Kg8
² (0.44) Depth: 7 00:00:00 30kN
13.Qh4 Kg8 14.c3 cxd4 15.Nxd4 Nc5
² (0.42) Depth: 8 00:00:00 105kN
13.Qh4 Kg8 14.c3 cxd4 15.Nxd4 Nc5 16.Bc2
² (0.43) Depth: 9 00:00:01 157kN
13.Qh4 Kg8 14.c3 cxd4 15.Nxd4 g5 16.Qh3 Nc5
= (0.25) Depth: 10 00:00:01 213kN
13.c3 b6 14.Qh3 Bb7 15.Re1 cxd4 16.cxd4 Bb4
² (0.34) Depth: 10 00:00:04 549kN
13.Re1 c4 14.Be2 b6 15.Bd2 Bb7 16.Qh5 Nf6
² (0.34) Depth: 10 00:00:04 590kN
13.Re1 c4 14.Be2 b6 15.Bd2 Bb7 16.Qh5 Nf6
² (0.32) Depth: 11 00:00:04 611kN
13.Re1 c4 14.Be2 b6 15.a4 Bb7 16.Qh4 Kg8 17.a5
² (0.26) Depth: 12 00:00:06 868kN
13.c3 b6 14.Qh3 Bb7 15.Re1 Kg8 16.Bd2 Rd8 17.b4
² (0.32) Depth: 12 00:00:07 1053kN
13.c3 b6 14.Qh3 Bb7 15.Re1 Kg8 16.Bd2 Rd8 17.Rad1 Bf4
² (0.26) Depth: 13 00:00:12 1759kN
13.c3 b6 14.Qh3 Bb7 15.Re1 Kg8 16.Bd2 Rd8 17.Rad1 Bd5 18.a4
= (0.24) Depth: 14 00:00:22 3098kN
13.c3 b6 14.Qh3 Bb7 15.Re1 Kg8 16.Bd2 Rd8 17.Rad1 Bd5 18.a4 c4
= (0.20) Depth: 15 00:00:53 7249kN
13.Qh4 Kg8 14.c3 b6 15.Be3 g5 16.Qh3 Bb7 17.Nd2 Kg7 18.Ne4 cxd4
= (0.22) Depth: 15 00:01:41 12995kN
13.Qh4 Kg8 14.c3 c4 15.Bc2 b6 16.Qe4 Bb7 17.Qe2 Rd8 18.Re1 g6 19.h3
= (0.19) Depth: 16 00:02:07 16373kN
13.c3 b6 14.Qh3 Bb7 15.Re1 Kg8 16.Bd2 Rd8 17.Rac1 Bd5 18.b4 c4 19.Bc2
= (0.24) Depth: 16 00:02:36 19575kN
13.c3 b6 14.Qh3 Bb7 15.Re1 Kg8 16.Bd2 Rd8 17.b3 Bxf3 18.Qxf3 Bxh2+ 19.Kh1 Bd6
= (0.22) Depth: 17 00:05:56 42932kN
13.c3 b6 14.Re1 Bb7 15.h4 Rd8 16.Qh3 Bd5 17.Re2 cxd4 18.cxd4 Qb7 19.Bd2 Bxf3
= (0.16) Depth: 18 00:15:31 110mN

[Tord Romstad]

if anything, I think Glaurung is too aggressive. For example, it thinks white has a full point advantage in this test opening position. It is true that white has the more aggresive, attacking position, but black is fine I think.

[d]r1b2k1r/ppqn1pp1/3bp2p/2p5/3P2Q1/3B1N2/PPP2PPP/R1B2RK1 w - - 0 13[d]


at depth 17, glaurung beta gives white a full point advantage (.94) in this position. It loves the attacking side. Its evaluation comes down with deeper depth (to about .54), but its initial impression is "attack is good."

A large part of the reason why Glaurung likes this position so much for white is the black rook on h8. My program has a huge penalty for rooks which are trapped inside a king which has lost the right to castle, with no open files between the king and the edge of the board.

I suspect glaurung could be improved by making her more conservative, but that would also be less interesting to many people.

Yes. The current version is, as you point out, very speculative. The positional scores are huge. Another example of this is the overevaluation of passed pawns, which Eelco mentioned in another post.

I like this very speculative style of play, but of course the Elo fetishists, who are always the most vocal computer chess fans, will hate it. But as there are already so many programs to cater to the Elo fetishists needs, I think I'll aim for interesting and adventurous play over maximum strength in my next version.

Tord