REBEL 15.1
1. New network architecture, version 10.
2. Increased data set from 2.75 billion positions to 3.6 billion positions.
3. Search improvements.
4. Estimated elo gain based on GRL testing ~56 elo.
5. Rebel 15.1 development diary.
http://rebel13.nl/windows/rebel-15.html
REBEL 15.1
Moderator: Ras
-
Rebel
- Posts: 7434
- Joined: Thu Aug 18, 2011 12:04 pm
- Full name: Ed Schröder
REBEL 15.1
90% of coding is debugging, the other 10% is writing bugs.
-
Steve Maughan
- Posts: 1308
- Joined: Wed Mar 08, 2006 8:28 pm
- Location: Florida, USA
Re: REBEL 15.1
Brilliant — thanks Ed!
http://www.chessprogramming.net - Juggernaut & Maverick Chess Engine
-
Wolfgang
- Posts: 989
- Joined: Sat May 13, 2006 1:08 am
-
Werewolf
- Posts: 2058
- Joined: Thu Sep 18, 2008 10:24 pm
Re: REBEL 15.1
Thanks Ed.Rebel wrote: ↑Wed Jul 13, 2022 7:15 pm REBEL 15.1
1. New network architecture, version 10.
2. Increased data set from 2.75 billion positions to 3.6 billion positions.
3. Search improvements.
4. Estimated elo gain based on GRL testing ~56 elo.
5. Rebel 15.1 development diary.
http://rebel13.nl/windows/rebel-15.html
Chris was working on some private CS Tal I seem to remember, is it still private?
-
ernst
- Posts: 354
- Joined: Thu Mar 09, 2006 6:00 pm
Re: REBEL 15.1
Thank you Ed!
Would you be willing to add Large Pages support? That should give about 10% speed-up for free.
Would you be willing to add Large Pages support? That should give about 10% speed-up for free.
This post may either be cause or result of misunderstandings.
-
Damir
- Posts: 2905
- Joined: Mon Feb 11, 2008 3:53 pm
- Location: Denmark
- Full name: Damir Desevac
Re: REBEL 15.1
and syzygy tbs support as well. Rebel tend to make a blunder against Stockfish in endgames, from roughly equal sometimes and than suddenly to outright losing..
-
Wolfgang
- Posts: 989
- Joined: Sat May 13, 2006 1:08 am
Re: REBEL 15.1
1300 games played. +78 to v15x2Wolfgang wrote: ↑Wed Jul 13, 2022 10:24 pm Thanks Ed
CEGT 40/4: https://cegt.forumieren.com/t1771-testing-rebel-15-1nn
Starting soon...
and +105 to v15.0 https://cegt.forumieren.com/t1771-testi ... 5-1nn#3359
-
James Constance
- Posts: 358
- Joined: Wed Mar 08, 2006 8:36 pm
- Location: UK
Re: REBEL 15.1
[fen]r3k1r1/pnp1bp1p/1p2p3/4Np1N/3q4/5QP1/PPP2P1P/R3K2R w KQq - 0 15[/fen]
Qc6+ is the only good move here. However, in multipv, Rebel is giving +9 to losing moves.
Qc6+ is the only good move here. However, in multipv, Rebel is giving +9 to losing moves.
Code: Select all
1. +- (9.54 --): 15.Qc6+ Kf8 16.Nd7+ Qxd7 17.Qxd7 Nd6 18.0-0-0 Rg6 19.Qxc7 e5 20.Kb1 Rh6 21.Rhe1 Rxh5 22.Rxe5 Nc8 23.Qc6
2. +- (9.37 --): 15.Nc6 Qe4+ 16.Qxe4 fxe4 17.0-0-0 Nd6 18.h4 f5 19.Rd2
3. +- (8.96 --): 15.0-0 Qxe5 16.Qxf5 -
Rebel
- Posts: 7434
- Joined: Thu Aug 18, 2011 12:04 pm
- Full name: Ed Schröder
-
Rubinus
- Posts: 1213
- Joined: Thu Jan 18, 2007 4:05 pm
- Location: Prague
- Full name: Pavel Háse
Re: REBEL 15.1
AMD Ryzen 5 4500U with Radeon Graphics
W=23.3 plies; 3 255kN/s; C30-39-LC0.ctg
B=29.4 plies; 6 663kN/s; 3 793 TBAs; C30-39-LC0.ctg
15+10, 6 cpu, Rebel 15x2.
[pgn][Event "Rebel15-Ethereal1375_C30-39"]
[Site "?"]
[Date "2022.07.21"]
[Round "7"]
[White "Rebel 15"]
[Black "Ethereal 13.75"]
[Result "0-1"]
[ECO "C30"]
[Annotator "-0.08;-1.13"]
[PlyCount "52"]
[EventDate "2022.??.??"]
1. e4 e5 2. f4 Bc5 3. Nf3 d6 4. c3 Bb6 5. Na3 Nf6 {preferably f5} 6. Nc4 $6 {O evaluation can only be sought after fe5} Nxe4 {Both outside the library} 7. Qc2 $2 {First move
by your own power and immediately a losing mistake. Just the fact that instead of a move in the game
expected Nf6 - the stopper further on - shows that cutting off the variations in the
Rebel has some flaws. The only right thing to do was to change to b6 with equilibrium.} Bf2+ {
(Nf6)} 8. Ke2 ({It's supposed to be a lot worse} 8. Kd1 d5 9. Ncxe5 Bb6 10. d4 O-O
11. Be3 c5 {I don't think it would be easy for me to play, if the advantage lies in
the ability to find that move. I think Rebel's a little lost in the shuffle.} 12.
Bd3 cxd4 13. Bxd4 Nc6 14. Bxb6 Qxb6 15. Re1 Bf5 16. Re2 Rae8 17. a4 a5 18. Ke1
Nxe5 19. fxe5 Rxe5 $1 20. Kf1 {That Qg1 is easy to see through...} Re6 21. Rae1 Rh6
22. Re3 Bg4 23. c4 Bxf3 24. gxf3 Qf6 25. cxd5 Nd6 {Believe it or not, -4. I would
...I don't think I could handle the loss after 8.Kd1 against Black.} 26. b3 g6 27. Qd2
Rh4 28. Bc2 Rc8) 8... d5 9. d3 dxc4 10. dxe4 Bc5 11. Nxe5 O-O 12. Nxc4 Re8 {Behind the dog is a white king trapped in the center. The compensation is luxurious. Yet it is
a position that Stockfish usually holds against other programs, while
...opponents lose to it.} 13. Ne3 Qh4 {(Nd7) Even here the expected move in
the curtain suggests that the Rebel has some problem in appreciating the activity of the figures and
of the king's specs - it can be overridden, but it could lead to other
other mistakes.} 14. g3 Qh5+ 15. Kf2 Nd7 16. Be2 {(b4)} Qh3 17. e5 $2 {(b4) Tim is
It's decided. After the expected b4, perhaps white could still survive.} Nxe5 $1 {
There's nothing in the curtain, Rebel saw the move, he just misjudged it somehow.
Definitely the knight bind, which he could have gotten rid of with move b4..} 18. fxe5
Bf5 {We improve with the pace of the piece.} 19. Bf1 Qh5 20. Qa4 Rxe5 21. Bg2 Bh3 22.
Bf3 Qg5 23. Qf4 {(Re1)} Qe7 24. Re1 g5 {(Re8)} 25. Qc4 Qf6 {(Rf5)} 26. Qe2 Rae8
0-1[/pgn]
So far 0:14=24, 31,58%, +- 134 (ever 1.e4 e5 2.f4)
W=23.3 plies; 3 255kN/s; C30-39-LC0.ctg
B=29.4 plies; 6 663kN/s; 3 793 TBAs; C30-39-LC0.ctg
15+10, 6 cpu, Rebel 15x2.
[pgn][Event "Rebel15-Ethereal1375_C30-39"]
[Site "?"]
[Date "2022.07.21"]
[Round "7"]
[White "Rebel 15"]
[Black "Ethereal 13.75"]
[Result "0-1"]
[ECO "C30"]
[Annotator "-0.08;-1.13"]
[PlyCount "52"]
[EventDate "2022.??.??"]
1. e4 e5 2. f4 Bc5 3. Nf3 d6 4. c3 Bb6 5. Na3 Nf6 {preferably f5} 6. Nc4 $6 {O evaluation can only be sought after fe5} Nxe4 {Both outside the library} 7. Qc2 $2 {First move
by your own power and immediately a losing mistake. Just the fact that instead of a move in the game
expected Nf6 - the stopper further on - shows that cutting off the variations in the
Rebel has some flaws. The only right thing to do was to change to b6 with equilibrium.} Bf2+ {
(Nf6)} 8. Ke2 ({It's supposed to be a lot worse} 8. Kd1 d5 9. Ncxe5 Bb6 10. d4 O-O
11. Be3 c5 {I don't think it would be easy for me to play, if the advantage lies in
the ability to find that move. I think Rebel's a little lost in the shuffle.} 12.
Bd3 cxd4 13. Bxd4 Nc6 14. Bxb6 Qxb6 15. Re1 Bf5 16. Re2 Rae8 17. a4 a5 18. Ke1
Nxe5 19. fxe5 Rxe5 $1 20. Kf1 {That Qg1 is easy to see through...} Re6 21. Rae1 Rh6
22. Re3 Bg4 23. c4 Bxf3 24. gxf3 Qf6 25. cxd5 Nd6 {Believe it or not, -4. I would
...I don't think I could handle the loss after 8.Kd1 against Black.} 26. b3 g6 27. Qd2
Rh4 28. Bc2 Rc8) 8... d5 9. d3 dxc4 10. dxe4 Bc5 11. Nxe5 O-O 12. Nxc4 Re8 {Behind the dog is a white king trapped in the center. The compensation is luxurious. Yet it is
a position that Stockfish usually holds against other programs, while
...opponents lose to it.} 13. Ne3 Qh4 {(Nd7) Even here the expected move in
the curtain suggests that the Rebel has some problem in appreciating the activity of the figures and
of the king's specs - it can be overridden, but it could lead to other
other mistakes.} 14. g3 Qh5+ 15. Kf2 Nd7 16. Be2 {(b4)} Qh3 17. e5 $2 {(b4) Tim is
It's decided. After the expected b4, perhaps white could still survive.} Nxe5 $1 {
There's nothing in the curtain, Rebel saw the move, he just misjudged it somehow.
Definitely the knight bind, which he could have gotten rid of with move b4..} 18. fxe5
Bf5 {We improve with the pace of the piece.} 19. Bf1 Qh5 20. Qa4 Rxe5 21. Bg2 Bh3 22.
Bf3 Qg5 23. Qf4 {(Re1)} Qe7 24. Re1 g5 {(Re8)} 25. Qc4 Qf6 {(Rf5)} 26. Qe2 Rae8
0-1[/pgn]
So far 0:14=24, 31,58%, +- 134 (ever 1.e4 e5 2.f4)