Sjeng 2.7 released

chessfurby · Post by **chessfurby** » Mon Sep 03, 2007 9:29 pm

To quote GCP (whose ccc account is not working atm):

Hi all,

Deep Sjeng 2.7 was just released. For people who have bought Deep Sjeng 2.5 this is a free update. Just install Deep Sjeng 2.7 and your code for Deep Sjeng 2.5 will still work.

For the silly people who have not bought it yet we offer a free 30-day trial just like the previous time

http://www.sjeng.org/deepsjeng2

Noteworthy new features:
* Engine +/- 50 ELO stronger than Deep Sjeng 2.5
* *Vastly* improved passed pawn and kingsafety understanding
* SMP improvements
* Adjustable board graphics, GUI usability tweaks
* GUI supports pondering
* Improved analysis options in GUI
* Opening book now a lot bigger

The new engine is actually an improved version of "Stoofvlees" who played in the last Leiden tournament. "Stoofvlees", err, I mean Deep Sjeng 2.7, has its evaluation function entirely automatically learned from watching Grandmaster games. Aside from making my job a lot easier (no more tuning to do at all!), the engine has noticeably improved in strength, particularly in the areas where it was less optimal before. Passed pawn handling is simply on another level compared to Deep Sjeng 2.5. There is also a strong effect on the playing style which will be easily noticeable: the engine plays very humanlike, which isn't surprising, given that it learned chess from watching GM's!

Overall I'm very happy with the result of this experiment, and it will be quite interesting to see how much further it can be taken...

www.sjeng.org

Cheers

Graham Banks · Post by **Graham Banks** » Mon Sep 03, 2007 9:59 pm

Thanks Jen.

I'll take a look at Gian-Carlo's CCC account to make sure all is okay.

Regards, Graham.

gerold · Post by **gerold** » Mon Sep 03, 2007 11:05 pm

Very nice Thanks.

Good luck,

Gerold

GenoM · Post by **GenoM** » Tue Sep 04, 2007 1:01 am

How can an engine be stronger than other one by +/-50 ELO points?

by -50 ELO points stronger?

I think correct is to say that engine is stronger by ~50 ELO points.

Dr.Ex · Post by **Dr.Ex** » Tue Sep 04, 2007 1:50 am

chessfurby wrote:
The new engine is actually an improved version of "Stoofvlees" who played in the last Leiden tournament. "Stoofvlees", err, I mean Deep Sjeng 2.7, has its evaluation function entirely automatically learned from watching Grandmaster games. Aside from making my job a lot easier (no more tuning to do at all!), the engine has noticeably improved in strength, particularly in the areas where it was less optimal before. Passed pawn handling is simply on another level compared to Deep Sjeng 2.5. There is also a strong effect on the playing style which will be easily noticeable: the engine plays very humanlike, which isn't surprising, given that it learned chess from watching GM's!

Overall I'm very happy with the result of this experiment, and it will be quite interesting to see how much further it can be taken...

www.sjeng.org

Cheers

Played two quick Blitzgames on my A64 X2. Not exactly above Super Grandmaster level and not very humanlike either...

In general it's quite amazing what stupid openings strong engines can afford to play against slightly weaker engines and still win the majority of games.
This would be definitely not possible in human chess.

[Event "Blitz:3'+2""]
[Site ""]
[Date "2007.09.04"]
[Round "1"]
[White "Deep Sjeng 2.7"]
[Black "Rybka 1.0 Beta 32-bit"]
[Result "0-1"]
[ECO "A40"]
[Annotator "0.25;0.11"]
[PlyCount "80"]
[TimeControl "180+2"]

{W=13.1 ply; 1.010kN/s B=11.8 ply; 109kN/s} 1. d4 {0} g6 {0.11/13 6} 2. e4 {0}
Nf6 {0.16/14 8} 3. Nc3 {0} d5 {0.07/14 24} 4. e5 {0} Ne4 {0.12/13 2} 5. Nxe4 {
(Lb5+) 0.25/14 13} dxe4 {0.00/10 0} 6. c3 {(Le3) 0.35/14 12} Bg7 {
(Sc6) -0.02/12 7} 7. f3 {(Sh3) 0.42/13 11} Bf5 {(Dd5) 0.17/12 8} 8. g4 {
(fxe4) 0.82/12 4} Bd7 {-0.11/12 6} 9. fxe4 {0.69/14 6} c5 {-0.01/11 3} 10. h3 {
(Le3) 0.65/14 10} Nc6 {-0.03/12 5} 11. Nf3 {0.65/14 5} cxd4 {(Db6) -0.03/11 6}
12. cxd4 {0.61/14 7} Qb6 {-0.03/12 6} 13. a3 {(Lf4) 0.69/12 3} O-O {
(h5) -0.19/10 4} 14. b4 {0.72/13 7} Rad8 {(Tfd8) -0.13/11 4} 15. Ra2 {
(b5) 0.39/12 5} Be6 {-0.02/11 13} 16. Rd2 {0.18/12 6} f5 {-0.01/11 11} 17. exf6
{(gxf5) -0.09/12 4} Rxf6 {-0.65/11 4} 18. Rd3 {(Lg2) -0.27/12 12} Rxf3 {
(Lc4) -0.76/11 4} 19. Qxf3 {-0.88/12 6} Nxd4 {-0.94/11 2} 20. Qg2 {-1.30/12 14}
Qc6 {-1.09/11 3} 21. Be3 {-1.54/12 6} Bb3 {-1.23/11 2} 22. b5 {(g5) -1.54/11 5}
Nc2+ {-1.16/12 3} 23. Qxc2 {-1.46/11 3} Qxc2 {-1.28/14 2} 24. Rxd8+ {-1.83/11 3
} Kf7 {-1.19/13 0} 25. Be2 {-1.68/12 2} Qb1+ {(Dxe4) -1.34/12 2} 26. Bd1 {
-1.66/12 3} Bc4 {-1.36/12 2} 27. Kf2 {-1.73/13 4} Qxb5 {(Lxb5) -1.52/12 14} 28.
a4 {-1.59/13 3} Qc6 {(Db1) -1.48/12 1} 29. Kf3 {-1.59/13 3} Bc3 {
(b6) -1.48/11 8} 30. g5 {(Lh6) -1.61/13 3} Qc7 {(La5) -1.56/11 1} 31. Ra8 {
-1.75/13 3} b6 {(Dd6) -1.52/11 1} 32. h4 {(Lc2) -1.54/14 5} Qd7 {
(Le5) -1.59/12 3} 33. Bc2 {-1.85/14 4} Be6 {(Ld2) -1.98/11 2} 34. Bf4 {
(Kg2) -1.71/12 2} Bd4 {(Ld2) -2.13/11 1} 35. Rb8 {(Le3) -2.75/12 3} Bc5 {
-5.58/12 10} 36. Be3 {-4.23/12 3} Bg4+ {-5.48/9 1} 37. Kg2 {-6.10/12 3} Bxe3 {
-5.48/9 1} 38. Bb3+ {-6.51/12 3} Be6 {-5.88/9 2} 39. Rd1 {-6.51/12 4} Bd2 {
-6.35/7 2} 40. Rd8 {(Lc2) -6.73/12 2} Qxd8 {-6.90/9 0} 0-1

[Event "Blitz:3'+2""]
[Site ""]
[Date "2007.09.04"]
[Round "1"]
[White "Deep Sjeng 2.7"]
[Black "Rybka 1.0 Beta 32-bit"]
[Result "0-1"]
[ECO "B01"]
[Annotator "0.24;0.13"]
[PlyCount "58"]
[TimeControl "180+2"]

{W=12.4 ply; 857kN/s B=11.8 ply; 100kN/s} 1. e4 {0} d5 {0} 2. exd5 {0} Qxd5 {0
} 3. Nc3 {0} Qe6+ {0} 4. Qe2 {0.24/13 7} Nf6 {(Sc6) 0.13/13 6} 5. Nb5 {
(Sf3) 0.07/14 9} Qd7 {(Db6) -0.11/14 12} 6. Qe5 {(Dc4) 0.00/13 3} Nc6 {
(Sa6) -0.07/13 7} 7. Qxc7 {0.37/13 7} Qe6+ {-0.23/14 14} 8. Kd1 {0.23/13 8} Qf5
{(a6) -0.07/14 16} 9. Nf3 {(f3) 0.08/11 5} Ne4 {(Sg4) -0.21/13 2} 10. Ke1 {
(Sbd4) -0.08/12 4} a6 {-0.25/13 11} 11. d3 {0.02/13 11} axb5 {-0.51/13 6} 12.
dxe4 {0.00/13 9} Qxe4+ {-0.48/11 0} 13. Be2 {0.00/12 10} g6 {(Dxc2) -0.56/11 1}
14. c3 {-0.25/12 5} Bg7 {-0.69/11 1} 15. Kf1 {(Df4) -0.32/12 5} b4 {-0.91/11 2}
16. cxb4 {-0.35/13 5} Qxb4 {-0.94/12 1} 17. a3 {-0.68/13 7} Qb3 {-0.98/11 2}
18. Nd2 {(h4) -1.38/12 12} Qc2 {-1.45/12 4} 19. Qb6 {-1.38/12 8} Nd4 {
(Txa3) -1.62/12 2} 20. f3 {(Lb5+) -1.51/10 6} O-O {-2.03/10 1} 21. Kf2 {
(h4) -1.86/12 4} Nxe2 {-2.33/11 1} 22. Kxe2 {-2.27/12 4} Be6 {-3.01/11 4} 23.
Qxb7 {(Db4) -2.55/12 5} Ra5 {(Tad8) -3.51/11 2} 24. Qb4 {-3.62/10 5} Rd5 {
-4.39/12 2} 25. Ke1 {-3.40/10 3} Rfd8 {-4.67/12 8} 26. Rf1 {-5.96/10 3} Bd4 {
-8.78/13 6} 27. Rf2 {(f4) -9.29/11 5} Bxf2+ {-8.82/5 1} 28. Kxf2 {-9.34/10 1}
Rxd2+ {-9.13/5 1} 29. Bxd2 {-9.43/11 3} Rxd2+ {-9.36/6 1} 0-1

fern · Post by **fern** » Tue Sep 04, 2007 3:00 am

He means "more or less THAN 50 points, of course. It could be 40 or 60.

Come on, Eugen, pick up better jokes...

gerold · Post by **gerold** » Tue Sep 04, 2007 3:58 am

fern wrote:He means "more or less THAN 50 points, of course. It could be 40 or 60.

Come on, Eugen, pick up better jokes...

Right Fernando. 50 elo stronger + / - 10 points.

Ryan Benitez · Post by **Ryan Benitez** » Tue Sep 04, 2007 4:09 am

I think it could be argued that losing to Rybka 1.0 is very human like.

I think Sjeng is an interesting engine and am happy to see a new release.

Ryan

Dr.Wael Deeb · Post by **Dr.Wael Deeb** » Tue Sep 04, 2007 8:31 am

Dr.Ex wrote:
chessfurby wrote:
The new engine is actually an improved version of "Stoofvlees" who played in the last Leiden tournament. "Stoofvlees", err, I mean Deep Sjeng 2.7, has its evaluation function entirely automatically learned from watching Grandmaster games. Aside from making my job a lot easier (no more tuning to do at all!), the engine has noticeably improved in strength, particularly in the areas where it was less optimal before. Passed pawn handling is simply on another level compared to Deep Sjeng 2.5. There is also a strong effect on the playing style which will be easily noticeable: the engine plays very humanlike, which isn't surprising, given that it learned chess from watching GM's!

Overall I'm very happy with the result of this experiment, and it will be quite interesting to see how much further it can be taken...

www.sjeng.org

Cheers
Played two quick Blitzgames on my A64 X2. Not exactly above Super Grandmaster level and not very humanlike either...

In general it's quite amazing what stupid openings strong engines can afford to play against slightly weaker engines and still win the majority of games.
This would be definitely not possible in human chess.

[Event "Blitz:3'+2""]
[Site ""]
[Date "2007.09.04"]
[Round "1"]
[White "Deep Sjeng 2.7"]
[Black "Rybka 1.0 Beta 32-bit"]
[Result "0-1"]
[ECO "A40"]
[Annotator "0.25;0.11"]
[PlyCount "80"]
[TimeControl "180+2"]

{W=13.1 ply; 1.010kN/s B=11.8 ply; 109kN/s} 1. d4 {0} g6 {0.11/13 6} 2. e4 {0}
Nf6 {0.16/14 8} 3. Nc3 {0} d5 {0.07/14 24} 4. e5 {0} Ne4 {0.12/13 2} 5. Nxe4 {
(Lb5+) 0.25/14 13} dxe4 {0.00/10 0} 6. c3 {(Le3) 0.35/14 12} Bg7 {
(Sc6) -0.02/12 7} 7. f3 {(Sh3) 0.42/13 11} Bf5 {(Dd5) 0.17/12 8} 8. g4 {
(fxe4) 0.82/12 4} Bd7 {-0.11/12 6} 9. fxe4 {0.69/14 6} c5 {-0.01/11 3} 10. h3 {
(Le3) 0.65/14 10} Nc6 {-0.03/12 5} 11. Nf3 {0.65/14 5} cxd4 {(Db6) -0.03/11 6}
12. cxd4 {0.61/14 7} Qb6 {-0.03/12 6} 13. a3 {(Lf4) 0.69/12 3} O-O {
(h5) -0.19/10 4} 14. b4 {0.72/13 7} Rad8 {(Tfd8) -0.13/11 4} 15. Ra2 {
(b5) 0.39/12 5} Be6 {-0.02/11 13} 16. Rd2 {0.18/12 6} f5 {-0.01/11 11} 17. exf6
{(gxf5) -0.09/12 4} Rxf6 {-0.65/11 4} 18. Rd3 {(Lg2) -0.27/12 12} Rxf3 {
(Lc4) -0.76/11 4} 19. Qxf3 {-0.88/12 6} Nxd4 {-0.94/11 2} 20. Qg2 {-1.30/12 14}
Qc6 {-1.09/11 3} 21. Be3 {-1.54/12 6} Bb3 {-1.23/11 2} 22. b5 {(g5) -1.54/11 5}
Nc2+ {-1.16/12 3} 23. Qxc2 {-1.46/11 3} Qxc2 {-1.28/14 2} 24. Rxd8+ {-1.83/11 3
} Kf7 {-1.19/13 0} 25. Be2 {-1.68/12 2} Qb1+ {(Dxe4) -1.34/12 2} 26. Bd1 {
-1.66/12 3} Bc4 {-1.36/12 2} 27. Kf2 {-1.73/13 4} Qxb5 {(Lxb5) -1.52/12 14} 28.
a4 {-1.59/13 3} Qc6 {(Db1) -1.48/12 1} 29. Kf3 {-1.59/13 3} Bc3 {
(b6) -1.48/11 8} 30. g5 {(Lh6) -1.61/13 3} Qc7 {(La5) -1.56/11 1} 31. Ra8 {
-1.75/13 3} b6 {(Dd6) -1.52/11 1} 32. h4 {(Lc2) -1.54/14 5} Qd7 {
(Le5) -1.59/12 3} 33. Bc2 {-1.85/14 4} Be6 {(Ld2) -1.98/11 2} 34. Bf4 {
(Kg2) -1.71/12 2} Bd4 {(Ld2) -2.13/11 1} 35. Rb8 {(Le3) -2.75/12 3} Bc5 {
-5.58/12 10} 36. Be3 {-4.23/12 3} Bg4+ {-5.48/9 1} 37. Kg2 {-6.10/12 3} Bxe3 {
-5.48/9 1} 38. Bb3+ {-6.51/12 3} Be6 {-5.88/9 2} 39. Rd1 {-6.51/12 4} Bd2 {
-6.35/7 2} 40. Rd8 {(Lc2) -6.73/12 2} Qxd8 {-6.90/9 0} 0-1

[Event "Blitz:3'+2""]
[Site ""]
[Date "2007.09.04"]
[Round "1"]
[White "Deep Sjeng 2.7"]
[Black "Rybka 1.0 Beta 32-bit"]
[Result "0-1"]
[ECO "B01"]
[Annotator "0.24;0.13"]
[PlyCount "58"]
[TimeControl "180+2"]

{W=12.4 ply; 857kN/s B=11.8 ply; 100kN/s} 1. e4 {0} d5 {0} 2. exd5 {0} Qxd5 {0
} 3. Nc3 {0} Qe6+ {0} 4. Qe2 {0.24/13 7} Nf6 {(Sc6) 0.13/13 6} 5. Nb5 {
(Sf3) 0.07/14 9} Qd7 {(Db6) -0.11/14 12} 6. Qe5 {(Dc4) 0.00/13 3} Nc6 {
(Sa6) -0.07/13 7} 7. Qxc7 {0.37/13 7} Qe6+ {-0.23/14 14} 8. Kd1 {0.23/13 8} Qf5
{(a6) -0.07/14 16} 9. Nf3 {(f3) 0.08/11 5} Ne4 {(Sg4) -0.21/13 2} 10. Ke1 {
(Sbd4) -0.08/12 4} a6 {-0.25/13 11} 11. d3 {0.02/13 11} axb5 {-0.51/13 6} 12.
dxe4 {0.00/13 9} Qxe4+ {-0.48/11 0} 13. Be2 {0.00/12 10} g6 {(Dxc2) -0.56/11 1}
14. c3 {-0.25/12 5} Bg7 {-0.69/11 1} 15. Kf1 {(Df4) -0.32/12 5} b4 {-0.91/11 2}
16. cxb4 {-0.35/13 5} Qxb4 {-0.94/12 1} 17. a3 {-0.68/13 7} Qb3 {-0.98/11 2}
18. Nd2 {(h4) -1.38/12 12} Qc2 {-1.45/12 4} 19. Qb6 {-1.38/12 8} Nd4 {
(Txa3) -1.62/12 2} 20. f3 {(Lb5+) -1.51/10 6} O-O {-2.03/10 1} 21. Kf2 {
(h4) -1.86/12 4} Nxe2 {-2.33/11 1} 22. Kxe2 {-2.27/12 4} Be6 {-3.01/11 4} 23.
Qxb7 {(Db4) -2.55/12 5} Ra5 {(Tad8) -3.51/11 2} 24. Qb4 {-3.62/10 5} Rd5 {
-4.39/12 2} 25. Ke1 {-3.40/10 3} Rfd8 {-4.67/12 8} 26. Rf1 {-5.96/10 3} Bd4 {
-8.78/13 6} 27. Rf2 {(f4) -9.29/11 5} Bxf2+ {-8.82/5 1} 28. Kxf2 {-9.34/10 1}
Rxd2+ {-9.13/5 1} 29. Bxd2 {-9.43/11 3} Rxd2+ {-9.36/6 1} 0-1

Hi,
What opening book did Rybka use in these games

Tony Thomas · Post by **Tony Thomas** » Tue Sep 04, 2007 9:59 am

I wonder how people come to the conclusion that it isnt grandmaster strength by playing two games against the strongest ever free engine.

Sjeng 2.7 released

Sjeng 2.7 released

Re: Sjeng 2.7 released

Re: Sjeng 2.7 released

Re: Sjeng 2.7 released

Re: Sjeng 2.7 released

Re: Sjeng 2.7 released

Re: Sjeng 2.7 released

Re: Sjeng 2.7 released

Re: Sjeng 2.7 released

Re: Sjeng 2.7 released