Cerebro v3.03b, WB/UCI released!

Antonio Senatore have released Cerebro v3.0.3b, it is available from the download pages of WBEC-Ridderkerk

A big thanks to Antonio

Best wishes,
Leo

Create book:
book {source file} {target file} [maxply] [mn]

where:
{source file} is the name of a .pgn file
{target file} can be:
Blancas.obk (book in .obk format for White)
Negras.obk (book in .obk format for Black)
[maxply] the depth of the book lines (default = 60)
[mn] minimal number of games to include a move (default = 2)


a)
if use maxply 26 and low, time cca 4s (base cca 50000 games)
if use maxply > 28 / sleep (little RAM or some bug)
b) Some bug.
Fruit 2.3.1 - Cerebro 3.03b (ph-book_try1)
1.d4 d5 2.c4 e6 3.Nf3 Nf6 4.Nc3 c6 5.Bg5 h6 6.Bxf6 Qxf6 7.e3 Nd7 8.Bd3 dxc4 9.Bxc4 Nc5??
game 8, previously solid.
Database without move 9.-Nc5 ...
ph-book_try1.rar
maxply=26, mn=4, pack 256KB/unpack cca 1MB.
ph-book_try2.rar
maxply=26, mn=2, pack 0.5MB/unpack cca 2MB.
c) if use maxply=27, then create blancas.obk, but sleep for negras.obk (probably size file limit, cca 1.1MB?)
Best wishes Pavel Háse

Hi Pavel:

Yes, you are right; Cerebro 3.03b has a bug in its routine to select moves from its opening books. The error is not frequent but it can happen mainly when you are using the tournament mode book and Cerebro can't find a good move in it. For that reason the engine played 9...Nc5 in its game against Fruit (are you testing Cerebro against Fruit? Cerebro is not rival for that engine ). I have fixed that bug and right now I'm sending Cerebro 3.03d to Leo.

Regarding your troubles by building books with more than 26 plies, that is new to me. With a database of 50000 games, if it took you about 4s to build a book with 26 plies, then to build a book with 60 plies (the maximum) and with mn = 1 (so all moves in the database will be included in the book) it should take you about 30s (or less because when you use the command

book games.pgn blancas.obk 60 1

the book only includes all the moves (until ply 60) in games won by White or moves from games ended in a draw). So there must be something wrong, maybe in the engine, maybe in the database you used. Is that database very commented? That can be a source of troubles; I always work with a "clean" database (although Cerebro can manage commented database) because the time to build a book is lower (anyway, with a very commented database you shouldn't have troubles to build a book of 28 plies). For your words it seems that Cerebro is entering in an unending buckle. In the Oliver Deville site you can find big databases; if you can, download any of those and try with:

book blancas.obk 60 1

and

book negras.obk 60 1

If you don't receive any error and you can finish the building in a reasonable time (let's say 30s, 60s as maximum) then maybe the problem is in your database (a corrupted game, or a comment that doesn't start with "{" or that doesn't end with "}" or a variant that doesn't start with "[" or ends without "]"). Sorry, but I can't tell you more about this problem if I can't work with your database. I tested the engine with a clean database of 83660 games and I couldn't reproduce your error by building books of 60 plies. Anyway thank you very much again for your notification; I will continue trying to reproduce the problem you had.

Regards
Antonio

PS: There is no limit in the size of the opening books.

.

Hi Antonio.
Yes, database includes some comments (source MegaBase2008). The same i use for first and second layer my ph-book2008.bin (Polyglot) -> errors (type: --;x`+y";[];FEN= etc.) cleaned.
I will try whitout comments.
Test match Cerebro 3.03b (+my book 4 plies) - Smarthink 1.10 (original book), 15m+10s, so far 2.0:16.0, 2 games to go.
Fair won:
[Event "test-Smarthink_1.10, Blic:15m+10s"]
[Site "RUBINUS"]
[Date "2008.06.22"]
[Round "12"]
[White "Cerebro 3.03b"]
[Black "SmarThink 1.10"]
[Result "1-0"]
[ECO "B12"]
[Annotator "0.79;0.67"]
[PlyCount "117"]
[TimeControl "900+10"]

{Mobile AMD Sempron(tm) Processor 3000+ 1799 MHz W=13.2 ply; 1 335kN/s; 4 125
TBAs B=15.4 ply; 530kN/s; 6 911 TBAs} 1. d4 {Oba mimo knihovnu 0.01/0 1} c6 {0
.01/0 0} 2. e4 {0.01/0 0} d5 {0.01/0 0} 3. e5 {0.01/0 0} Bf5 {0.01/0 0} 4. Nf3
{0.01/0 0} e6 {0.01/0 0} 5. Be2 {0.01/0 0} Ne7 {0.01/0 0} 6. O-O {0.01/0 0} Bg4
{0.01/0 0} 7. Nbd2 {0.79/12 34} Bh5 {(Jd7) 0.67/15 145} 8. Nb3 {(h3) 0.90/13 43
} Nf5 {(Jg6) 0.51/14 33} 9. Bf4 {(h3) 0.48/13 82} Nd7 {(Se7) 0.35/14 27} 10.
Qd2 {(h3) 0.41/12 38} h6 {(Se7) 0.29/13 28} 11. h3 {0.70/12 48} Nh4 {(Sg6) 0.
22/14 26} 12. Nxh4 {(Dd1) 0.61/13 35} Bxe2 {0.19/15 25} 13. Qxe2 {0.58/15 34}
Qxh4 {0.21/15 25} 14. Bd2 {0.59/13 33} Nb6 {(0-0-0) 0.22/14 25} 15. f4 {0.75/
13 29} Nc4 {0.22/13 40} 16. Be1 {(a3) 0.71/14 31} Qd8 {0.09/14 24} 17. f5 {
(Sc3) 0.69/1330} Qg5 {(exf5) 0.15/13 23} 18. fxe6 {0.92/13 29} fxe6 {0.13/13 8}
19. Rf3 {(Sc3) 1.00/12 23} O-O-O {(Se7) 0.18/14 23} 20. Na5 {1.11/13 22} Nxa5 {
0.45/16 57} 21. Bxa5 {1.06/10 1} Rd7 {(Ve8) 0.43/16 25} 22. Qd3 {(Vaf1) 1.33/
13 26} Qh5 {(Se7) 0.48/16 33} 23. Raf1 {1.21/14 53} Qe8 {0.48/16 21} 24. Qb3 {
(Dc3) 1.12/1225} h5 {(Se7) 0.49/15 24} 25. Bd2 {(Sb4) 1.19/12 24} Be7 {
(Kb8) 0.36/1522} 26. Qa4 {(Sb4) 1.03/13 47} Kb8 {0.36/15 20} 27. Bb4 {0.95/13
22} g5 {(Dd8) 0.37/15 20} 28. Bxe7 {1.56/12 20} Rxe7 {(Dxe7) 0.37/15 19} 29.
Rf6 {1.52/14 21} Ka8 {(c5) 0.56/14 32} 30. Qb4 {1.91/14 20} a6 {(g4) 0.58/16 26
} 31. Qd6 {2.06/12 18} Ka7 {0.65/15 30} 32. Qc5+ {(b3) 2.01/12 20} Kb8 {0.65/
15 18} 33. c3 {(b3) 1.99/12 19} g4 {(Dd7) 0.62/15 18} 34. Qd6+ {(h4) 1.95/13 19
} Kc8 {0.80/16 17} 35. h4 {1.97/13 18} Qd8 {(Dd7) 1.23/16 43} 36. Qc5 {
(Vxe6) 1.79/1336} Kb8 {(Dc7) 0.75/13 16} 37. g3 {(Kf2) 1.91/12 13} Qe8 {
(Kc7) 0.77/1516} 38. Qd6+ {1.87/14 33} Ka7 {(Kc8) 0.80/17 16} 39. a4 {
(b3) 2.06/1416} Ka8 {(Dd8) 1.00/15 16} 40. Kh2 {(b4) 2.43/14 15} Ka7 {
(Dd8) 1.01/1617} 41. b4 {(Kg2) 2.39/13 15} Rg8 {(Ka8) 1.58/16 39} 42. Qc5+ {
(Vh6) 2.45/1315} Ka8 {1.80/16 37} 43. b5 {(Kg2) 2.18/14 29} cxb5 {1.77/14 14}
44. axb5 {2.40/10 1} Qd7 {1.77/14 14} 45. bxa6 {(Dd6) 2.40/13 28} bxa6 {1.80/
13 5} 46. Ra1 {(Da3) 2.43/13 25} Qb7 {2.76/17 99} 47. Rf2 {2.56/12 13} Rc8 {2.
77/16 20} 48. Rxa6+ {2.90/13 13} Qxa6 {2.78/16 4} 49. Qxe7 {2.76/10 1} Rb8 {
(Vh8) 2.80/1515} 50. Qf6 {3.12/13 13} Rc8 {3.03/15 30} 51. Qh6 {3.31/14 13}
Rxc3 {3.23/13 11} 52. Qxh5 {3.24/10 3} Kb7 {(Db6) 3.60/13 28} 53. Qxg4 {3.52/
12 13} Rc7 {(Db6) 3.77/12 11} 54. Rb2+ {(h5) 3.97/13 13} Ka8 {(Ka7) 4.30/16 27}
55. Qf3 {(h5) 4.44/13 13} Qc8 {4.35/15 12} 56. Qa3+ {4.92/12 13} Ra7 {4.80/15
26} 57. Qd6 {4.47/13 25} Rd7 {(Va1) 8.73/13 56} 58. Qxe6 {(Va2+) 16.48/12 12}
Qb7 {26.47/13 34} 59. Ra2+ {19.50/11 12} 1-0

Other games later (test Smarthink 1.10 from 13 opponents, 20xSpike, Delfi and 2xCerebro to go).
Best wishes Pavel Háse

Database cca 53 000 games (source cca 48k Mega2008, 3k CEGT-extreme,quad, 2k playchess, a few corr., whitout shorter 15 moves and remis 20 moves)
ph-book_try3-68x2.rar
maxply=68, mn=2, pack 0.9MB/unpack cca 4.1MB.
ph-book_try3-68x3.rar
maxply=68, mn=3, pack 320KB/unpack cca 1.2MB. My advice, database past select (First and second layer for Polyglot maxply=10 and 20, other layers use database 600 000 games).
ph-book_try3-68x4.rar
maxply=68, mn=4, pack 210KB/unpack cca 0.85MB. Probably over short.
Best wishes Pavel Háse

Hi Pavel:

I see that you could build the books without problems this time. So I guess that your previous errors were due to the comments of your database. It's good to know that the building routine works fine

Now two things:

1) The maximum ply is 60 (30 moves for each side). So if you use as maxply 68, Cerebro will use 60 anyway.

2) Very good books can be built of the following way:

a- Use a database with very good games (from players with 2600 elo or more).
b- Split the database in two so one of them only has games won by White or ended in a draw and the other only has games won by Black or ended in a draw.
c- Use the commands:

book WhiteGames.pgn blancas.obk 60 1
book BlackGames.pgn negras.obk 60 1

As the database has very good games, there is no problem to use mn=1. Of course, with a big database (25000 games each one) the opening books also can be quite big (about 6MB uncompressed/1MB compressed).

Regards
Antonio

Hi Antonio.
Solid book I try later.
- 2600+? (2700+?) No. Mistakes is here. Well Rybka rating CEGT is cca 3000. Whithout computer analyze not recommended. For plies=1 new correspondence games is better - night analyze good engines, good player.
Regard
Pavel Háse

[Event "new-08-2008"]
[Site "?"]
[Date "2008.08.17"]
[Round "2.2"]
[White "Cerebro 3.03d"]
[Black "Rybka 3 Dynamic 1-cpu"]
[Result "0-1"]
[ECO "C63"]
[Annotator "0.44;0.24"]
[PlyCount "80"]
[EventDate "2008.08.07"]
[EventType "tourn"]
[TimeControl "5400+30"]

{AMD Athlon(tm) 64 Processor 3200+ 2010 MHz W=17.0 ply; 1 482kN/s; 674 895
TBAs B=17.0 ply; 66kN/s; 1 202 TBAs; KingGambit.ctg} 1. e4 {[%eval 1,0] [%emt
0:00:00]} e5 {[%eval 0,0] [%emt 0:00:00]} 2. Nf3 {[%eval 1,0] [%emt 0:00:00]}
Nc6 {[%eval 0,0] [%emt 0:00:00]} 3. Bb5 {[%eval 1,0] [%emt 0:00:00]} f5 {
[%eval 0,0] [%emt 0:00:00]} 4. d3 {[%eval 1,0] [%emt 0:00:00]} fxe4 {[%eval 0,
0] [%emt 0:00:00]} 5. dxe4 {[%eval 1,0] [%emt 0:00:00]} Nf6 {[%eval 0,0] [%emt
0:00:00]} 6. O-O {[%eval 1,0] [%emt 0:00:00]} Bc5 {[%eval 24,15] [%emt 0:02:09]
} 7. Qd3 {(Sc4) [%eval 1,0] [%emt 0:00:00]} d6 {[%eval 0,0] [%emt 0:00:00]} 8.
Qc4 {[%eval 1,0] [%emt 0:00:00]} Bd7 {[%eval 36,16] [%emt 0:05:40]} 9. Nc3 {
[%eval 44,13] [%emt 0:03:47]} Bb6 {[%eval 0,0] [%emt 0:00:00]} 10. Ng5 {
[%eval 82,14] [%emt 0:04:06]} Qe7 {[%eval 0,0] [%emt 0:00:00]} 11. Nd5 {
[%eval 73,15] [%emt 0:03:57]} Nxd5 {[%eval 0,0] [%emt 0:00:00]} 12. exd5 {
[%eval 73,13] [%emt 0:00:05]} Nd4 {[%eval 75,16] [%emt 0:02:37]} 13. Bxd7+ {
[%eval 71,15] [%emt 0:03:48]} Qxd7 {(Kxd7) [%eval 50,17] [%emt 0:11:22]} 14. c3
{[%eval 108,16] [%emt 0:03:39]} h6 {[%eval 58,17] [%emt 0:09:48]} 15. cxd4 {
[%eval 104,16] [%emt 0:03:31]} hxg5 {[%eval 58,16] [%emt 0:00:00]} 16. dxe5 {
[%eval 108,14] [%emt 0:00:14]} Qf5 {[%eval 58,17] [%emt 0:01:06]} 17. Qa4+ {
(e6) [%eval 97,14] [%emt 0:06:47]} Kf7 {[%eval 49,15] [%emt 0:01:45]} 18. e6+ {
[%eval 120,14] [%emt 0:02:57]} Kg8 {[%eval 48,16] [%emt 0:01:33]} 19. h3 {
[%eval 108,15] [%emt 0:06:00]} Rh4 {(Dxd5) [%eval 0,15] [%emt 0:02:22]} 20. Qd7
{(Db3) [%eval 120,14] [%emt 0:02:45]} Rf8 {(g4) [%eval -25,13] [%emt 0:00:00]}
21. Be3 {[%eval 40,13] [%emt 0:05:18]} g4 {[%eval -25,13] [%emt 0:00:00]} 22.
Qe7 {[%eval 29,14] [%emt 0:04:52]} Qh5 {(Dh7) [%eval -25,12] [%emt 0:00:00]}
23. Qg5 {[%eval -26,13] [%emt 0:04:28]} Bxe3 {[%eval 0,17] [%emt 0:01:33]} 24.
Qxh5 {[%eval -28,12] [%emt 0:00:08]} Rxh5 {[%eval -10,18] [%emt 0:02:21]} 25.
fxe3 {[%eval -34,12] [%emt 0:00:07]} Rxd5 {[%eval -8,20] [%emt 0:02:15]} 26.
Rxf8+ {(Vac1) [%eval -30,14] [%emt 0:02:06]} Kxf8 {[%eval -8,18] [%emt 0:00:00]
} 27. Rf1+ {(Vc1) [%eval -32,16] [%emt 0:02:01]} Ke7 {[%eval -25,19] [%emt 0:
00:49]} 28. Rf7+ {[%eval -30,16] [%emt 0:01:57]} Kxe6 {[%eval -25,18] [%emt 0:
00:00]} 29. Rxc7 {[%eval -26,15] [%emt 0:00:19]} g3 {[%eval -8,20] [%emt 0:04:
58]} 30. Kf1 {[%eval -51,17] [%emt 0:03:47]} Rf5+ {[%eval -4,20] [%emt 0:01:31]
} 31. Kg1 {(Ke1) [%eval -51,17] [%emt 0:01:45]} Rc5 {[%eval -792,19] [%emt 0:
08:37]} 32. Rxc5 {[%eval -56,21] [%emt 0:01:41]} dxc5 {[%eval -505,17] [%emt 0:
00:00]} 33. a4 {(Kf1) [%eval -132,22] [%emt 0:03:16]} Ke5 {[%eval -1126,18]
[%emt 0:02:41]} 34. Kf1 {[%eval -119,21] [%emt 0:01:13]} Ke4 {[%eval -510,16]
[%emt 0:00:00]} 35. Ke2 {[%eval -121,21] [%emt 0:01:29]} c4 {[%eval -505,14]
[%emt 0:00:00]} 36. h4 {[%eval -146,21] [%emt 0:02:52]} g6 {[%eval -510,14]
[%emt 0:00:00]} 37. Kd2 {(a5) [%eval -234,21] [%emt 0:02:39]} b6 {[%eval -1590,
15] [%emt 0:01:35]} 38. Ke2 {[%eval -234,20] [%emt 0:01:14]} a6 {[%eval -510,
14] [%emt 0:00:00]} 39. Kd2 {[%eval -769,21] [%emt 0:02:24]} b5 {[%eval -510,
13] [%emt 0:00:00]} 40. a5 {(axb5) [%eval -984,20] [%emt 0:02:14]} b4 {
[%eval -2809,15] [%emt 0:05:39]} 0-1

Interest?
After 31.Kg1 - evalution both engines (outlying passed pawn theme).

Hi Pavel:

Yes, 31.Kg1 was a very bad move; with 31.Ke1 maybe Cerebro could have sustained the ending (31.Ke1 Rf2 32.Rxg7 Rxb2 33.Rxg3 Rxa2 34.h4=). The move 31.Kg1 allows 31...Rc5 forcing a pawns endings won by Black. I'm working to improve the way as Cerebro plays endings. So far, the engine has practically no knowledge about this stage of the game and it's not rare to see it losing even positions or worse yet, to lose favorable positions :-) Thus, the best strategy to defeat my engine -so far- is to change queens :-) After that, its tactical strength will be low (quite low) and the engine will become very vulnerable.

Regards
Antonio