Balanced-14.abk for Arena now available

[Mike S.]

In response to the pleasant release of Arena 2.0, I have created an ABK version of my "Balanced" opening book for engine tests, the

Balanced-14.abk

Free Download: http://remixshare.com/?file=t1uo6hcg1u (~1 MB zip)

It is meant as an update, or replacement of the old Xmas2640.abk. The total number of moves in the Balanced-14 is 44% bigger, and it contains more opening variations (some very rarely played openings are still missing). See the readme text for more details.

Enjoy, and a happy 2009 to all!


P.S.

CTG versions of the Balanced book (and other test books) are available from my homepage:

http://members.aon.at/computerschach/li ... #downloads

In addition to that, I have created some other, more experimental books in Fritz format... maybe still interesting for some:

History107.ctg: Openings from the age of the old masters (1840-1946), incl. ratings.
http://remixshare.com/get.php?file=jn5ihnzpln

PB5moves.ctg: By popular demand available again: (a new version of) my 5moves book.
http://rapidshare.com/files/147461005/P ... G.zip.html

MyDailyBook.ctg: Experimental "balanced" version which avoids sicilian mainlines (for Playchess).
http://rapidshare.com/files/137629786/N ... TG.7z.html

[pichy]

What would be the best settings for Balanced-14.abk

Your current settings came like this:

Balanced-14.abk used in Arena 2.0:

Menu: Book - Manage - Move selection

Priority -------------+5
Number of games-------+9
Win percentage---------+2
Select:"Move with priority 0 nerver played"

[Mike S.]

What would be the best settings for Balanced-14.abk

Yes, these are the settings which I have figured out to work ok, in the sense of big variety (but with a somewhat higher probability for popular openings, still):

Prio 5/Num. 9/W% 2

Note also the settings on the left side:

min. games 2 (as a safety net against book blunders from one game only)
min. won games 0 (black moves can be ok without a win in the database)

min. white percentage 50% (less than that is poor statistics for a white move)
min. black percentage 33% (for example, at least two draws in three games)

Min. 50% for white moves may seem a bit restrictive sometimes, for example 2.f4 for the KG. got only 47% in the database. After e4 e5 Nf3 Nc6, 3.Nc3 got only 46%. So, if a user wants to increase the variety for his engine tests even more, lowering the requirement for white moves to 45% would make the book to choose these moves too, sometimes. The general opening quality (by high standards) would be slightly reduced though.

These are the general settings. - I don't use to edit many priorities of single moves because my books are neutral test books only, intended for the "same book for all" test concept, not top performance(*) books. I have spent the most time not with that, but for compiling the "Balanced" database in a step-by-step process with careful selction of games for many (beginnings of) variations.

*) The MyDailyBook.ctg (NoSici) had a surprisingly good result in Klaus Ritter's book test tournament though: http://chessstars.de/pageID_5104739.html But it has to be seen experimental, especially considering the smaller book depth. Some other books, like Xmas..., also got mediocre to good results in Sedat Canbaz' book tests. I can only assume that it is a result of other phenomenons, not of an absolute quality of my books... Maybe Rybka has a "levelling" effect in self-play with different books?

[pichy]

What would be the best settings for Balanced-14.abk

Yes, these are the settings which I have figured out to work ok, in the sense of big variety (but with a somewhat higher probability for popular openings, still):

Prio 5/Num. 9/W% 2

Note also the settings on the left side:

min. games 2 (as a safety net against book blunders from one game only)
min. won games 0 (black moves can be ok without a win in the database)

min. white percentage 50% (less than that is poor statistics for a white move)
min. black percentage 33% (for example, at least two draws in three games)

Min. 50% for white moves may seem a bit restrictive sometimes, for example 2.f4 for the KG. got only 47% in the database. After e4 e5 Nf3 Nc6, 3.Nc3 got only 46%. So, if a user wants to increase the variety for his engine tests even more, lowering the requirement for white moves to 45% would make the book to choose these moves too, sometimes. The general opening quality (by high standards) would be slightly reduced though.

These are the general settings. - I don't use to edit many priorities of single moves because my books are neutral test books only, intended for the "same book for all" test concept, not top performance(*) books. I have spent the most time not with that, but for compiling the "Balanced" database in a step-by-step process with careful selction of games for many (beginnings of) variations.

*) The MyDailyBook.ctg (NoSici) had a surprisingly good result in Klaus Ritter's book test tournament though: http://chessstars.de/pageID_5104739.html But it has to be seen experimental, especially considering the smaller book depth. Some other books, like Xmas..., also got mediocre to good results in Sedat Canbaz' book tests. I can only assume that it is a result of other phenomenons, not of an absolute quality of my books... Maybe Rybka has a "levelling" effect in self-play with different books?

Check this Opening created by Rybka 2.2 without opening Book against Thinker 5.4 using your Blalanced-14.abk.

[Event "Arena tournament"]
[Site "JORGE"]
[Date "2008.12.26"]
[Round "6"]
[White "Rybka v2.2n2.mp.w32"]
[Black "SP-x86-Passive-Thinker"]
[Result ""]
[BlackElo "2830"]
[ECO "D02"]
[Opening "Queen's Pawn"]
[Time "12:08:13"]
[Variation "3.g3 c6 4.Bg2 Bg4"]
[WhiteElo "2850"]
[TimeControl "300"]
[Termination "unterminated"]
[PlyCount "73"]
[WhiteType "program"]
[BlackType "program"]

1. Nc3 {(Nb1c3 Ng8f6 Ng1f3 Nb8c6 e2e4 d7d5 e4xd5 Nf6xd5 Bf1c4) +0.11/11 8}
d5 2. Nf3 {(Ng1f3 Ng8f6 e2e3 Nb8c6 Bf1b5 Bc8d7 OO e7e6) +0.08/11 11} Nc6 3.
d4 {(d2d4 Ng8f6 g2g3 Bc8f5 Nf3h4 Qd8d7 Bc1f4 OOO) +0.09/11 19} Nf6 4. g3
{(g2g3 g7g6 Bc1f4 Nf6h5 Bf4e3 Bf8g7 Bf1g2) +0.09/10 6} Bg4 5. h3 {(h2h3
Bg4f5 Nf3h4 Bf5e4 f2f3 Be4g6 Nh4xg6) +0.09/10 7} Bxf3 6. exf3 {(e2xf3 e7e5
Bf1b5 e5xd4 Qd1xd4 Bf8d6 Bc1g5) +0.12/9 2} Qd7 7. Bg2 {(Bf1g2 e7e6 OO Bf8d6
Rf1e1 h7h6 f3f4) +0.09/10 7} O-O-O 8. O-O {(OO e7e6 Rf1e1 Bf8d6 Bc1g5
Kc8b8) +0.14/9 1} h5 9. Bg5 {(Bc1g5 Qd7f5 Qd1d2 e7e6 Rf1e1 Bf8b4) +0.09/9
7} Qf5 10. f4 {(f3f4 Nf6e4 Nc3xe4 d5xe4 Qd1e2 Nc6xd4 Qe2xe4) +0.07/9 6} e6
11. Re1 {(Rf1e1 Bf8e7 Qd1d2 b7b6 Re1e3 Kc8b7) +0.07/9 16} Bb4 12. Qd2
{(Qd1d2 Nf6e4 Bg2xe4 d5xe4 Bg5xd8 Rh8xd8 Qd2e3 Nc6xd4) 0.00/9 18} Rd7 13.
f3 {(f2f3 Bb4a5 Re1e3 Kc8b8 Bg2f1 Nc6b4 Re3e2) -0.07/10 9} Nh7 14. Bh4
{(Bg5h4 Nh7f6 Bh4g5 Bb4a5 Re1e3 Kc8b8 Bg2f1 Nc6b4 Re3e2) 0.00/10 3} Qg6 15.
Kh2 {(Kg1h2 Bb4e7 Bh4xe7 Rd7xe7 h3h4 Nh7f6) -0.03/8 4} Be7 16. Bxe7
{(Bh4xe7 Rd7xe7 h3h4 Nh7f6 Bg2f1 a7a6 Kh2g2) -0.09/9 4} Rxe7 17. h4 {(h3h4
Nh7f6 Ra1d1 a7a6 Qd2d3 Qg6h6 f4f5) -0.09/10 8} Kb8 18. Bf1 {(Bg2f1 Nh7f6
Nc3a4 Rh8d8 Kh2g2 Rd8d6 Na4c5) +0.12/10 3} Nf6 19. Bb5 {(Bf1b5 Nc6b4 Re1c1
c7c6 Bb5e2 Nb4a6 Nc3a4 Na6c7) +0.31/11 4} Nb4 20. Rac1 {(Ra1c1 c7c6 Bb5e2
Nb4a6 Be2d3 Qg6h6 Bd3xa6 b7xa6 Qd2d3) +0.49/11 5} c6 21. Be2 {(Bb5e2 Nb4a6
Be2d3 Qg6h6 Bd3xa6 b7xa6 Qd2d3 a6a5 Qd3a6 Re7b7) +0.65/12 4} c5 22. dxc5
{(d4xc5 Nb4c6 Nc3b5 Rh8d8 Be2d3 Qg6h6) +0.81/9 0} a6 23. Nb5 {(Nc3b5 Nb4c6
Nb5d6 Rh8d8 c2c4 Re7d7 Be2d3 Qg6h6) +0.98/11 8} Nc6 24. Nd4 {(Nb5d4 Rh8c8
c2c3 Nf6d7 Nd4xc6+ Rc8xc6 b2b4 Qg6f6 Kh2g2) +0.85/10 6} Nxd4 25. Qxd4
{(Qd2xd4 Re7c7 Be2d3 Qg6h6 c2c4 Rh8d8) +1.01/9 1} Rc8 26. Bd3 {(Be2d3 Qg6h6
c2c4 Re7d7 Rc1d1) +1.07/8 1} Qh6 27. c3 {(c2c3 Nf6d7 b2b4 Qh6f6 Kh2g2)
+0.99/8 3} Rec7 28. b4 {(b2b4 Nf6g8 f4f5 Ng8e7 f5xe6) +1.13/8 2} Rc6 29. c4
{(c3c4 Rc6c7 Rc1d1 Rc7d7 Kh2g2 Rc8c7 Qd4e5) +1.25/10 3} R6c7 30. Rcd1
{(Rc1d1 Rc7d7 Kh2g2 Rc8c7 Qd4e5 Rd7d8 c4xd5) +1.17/10 3} Qh8 31. cxd5
{(c4xd5 Nf6xd5 Bd3e4 Nd5f6 Rd1d2 Rc7d7 Qd4e5+ Rc8c7) +1.23/9 3} exd5 32.
Re5 {(Re1e5 a6a5 a2a3 Rc8e8 b4xa5 Qh8f8 Re5xe8+ Qf8xe8 a5a6) +1.89/10 1} g6
33. Kg2 {(Kh2g2 Rc8d8 Rd1e1 Qh8g7 a2a3 Nf6g8) +1.98/9 2} Rd8 34. f5 {(f4f5
a6a5 f5xg6 f7xg6 a2a3 a5xb4) +2.13/8 3} Qg7 35. fxg6 {(f5xg6 f7xg6 Rd1e1
Nf6d7 Re5xd5) +2.21/7 0} fxg6 36. Rde1 {(Rd1e1 Rd8f8 Re5g5 Nf6g8 Qd4xg7)
+2.43/8 2} Nh7 37. Bc2 {(Bd3c2 Rc7f7 a2a4 Rf7c7) +2.27/7 6} *

So far the score is 3.0 to 2 in favour of Rybka without Opening book:

Engine Score Ry SP S-B
1: Rybka v2.2n2.mp.w32 3.0/5 ······ 01011? 6.00
2: SP-x86-Passive-Thinker 2.0/5 10100? ······ 6.00

5 of 30 games played
Name of the tournament: Arena tournament
Site/ Country: JORGE, United States
Level: Tournament Game in 5 Minutes
Hardware: AMD Athlon(tm) Processor 1200 MHz with 512 MB Memory
Operating system: Microsoft Windows XP Professional Service Pack 2 (Build 2600)
PGN-File: C:\Program Files\Arena\Arena.pgn
Website:
E-Mail Address:

[Mike S.]

Check this Opening created by Rybka 2.2 without opening Book against Thinker 5.4 using your Blalanced-14.abk.

The Balanced-14 is not really suitable for this type of book vs. no book tests, because - as explained - it is not a "complete" book. In your example, actually both engines played without book support, because Rybka chose 1.Nc3 but the Balanced-14 does not contain playable continuations for this rare first move at all (that the move appears in the book is only due to games where transpositions into common variations happened, and has been disabled).

So, whenever the "no book engine" chooses very rare moves by calculation, the book support is virtually removed from the opponent too if the book doesn't contain continuations. That may happen with the Balanced-14 a little bit more often (1.Nc3, 1.f4, 1.e4 Nf6...), but will most probably also happen with other books sometimes, if the opponent engine chooses "off theory" moves without book.

Again: It is intended as a test book for "same book for all" engine matches.

It will be suitable for matches between different books (but not very good, IOW. not in a "performance" sense), because then opponent's moves like 1.Nc3 should have a very low probability. Maybe the Balanced-14 covers ~90% of what's normally being played up to the 14th move, but not 99.9%...

An interesting comparison would be:

A. Rybka 3 with an official Rybka 3 book against various opponents (with their books), and then
B. Rybka 3 with Balanced-14.abk (or with Balanced-16.ctg if you use Fritz) against the same opponents,
rest of conditions identical each.

Basically, since my books are certainly shorter than most others, it only means that the battle between the engines begins earlier in the "B" games except if Rybka manages to find many book moves by calculation. But I am almost sure that the Elo performance difference would not be dramatic. Maybe -30 Elo?

[pichy]

Check this Opening created by Rybka 2.2 without opening Book against Thinker 5.4 using your Blalanced-14.abk.

The Balanced-14 is not really suitable for this type of book vs. no book tests, because - as explained - it is not a "complete" book. In your example, actually both engines played without book support, because Rybka chose 1.Nc3 but the Balanced-14 does not contain playable continuations for this rare first move at all (that the move appears in the book is only due to games where transpositions into common variations happened, and has been disabled).

So, whenever the "no book engine" chooses very rare moves by calculation, the book support is virtually removed from the opponent too if the book doesn't contain continuations. That may happen with the Balanced-14 a little bit more often (1.Nc3, 1.f4, 1.e4 Nf6...), but will most probably also happen with other books sometimes, if the opponent engine chooses "off theory" moves without book.

Again: It is intended as a test book for "same book for all" engine matches.

It will be suitable for matches between different books (but not very good, IOW. not in a "performance" sense), because then opponent's moves like 1.Nc3 should have a very low probability. Maybe the Balanced-14 covers ~90% of what's normally being played up to the 14th move, but not 99.9%...

An interesting comparison would be:

A. Rybka 3 with an official Rybka 3 book against various opponents (with their books), and then
B. Rybka 3 with Balanced-14.abk (or with Balanced-16.ctg if you use Fritz) against the same opponents,
rest of conditions identical each.

Basically, since my books are certainly shorter than most others, it only means that the battle between the engines begins earlier in the "B" games except if Rybka manages to find many book moves by calculation. But I am almost sure that the Elo performance difference would not be dramatic. Maybe -30 Elo?

What I'm doing is Rybka v2.2 with rybka 2.0 Opening book against Thinker 5.4 with different Arena openings and the two openings that get the better performance against Rybka v2.2 will play a match against each other with Rybka and Thinker reversing engine and openings.

PS: it seems that the free version of rybka v2.2 is too strong, that even without an opening book can manage to hold even one of the strongest free engine available after Rybka v2.2 with different books.

Congratulations Thinker 5.4 with your Balanced-14.abk held Rybka v2.2 without an opening book even after 30 games.

Now Rybka v2.2 will use HS-Book.abk versus Thinker 5.4 using your Balance-14.abk and later another match with both engines and openings reversed. Or Rybka v2.2 with Balanced-14.abk Vs Thinker 5.4 with HS-Book.abk.

Engine Score SP Ry S-B
1: SP-x86-Passive-Thinker 15.0/30 ······························ 101001100110100100=11=01101001 225.00
1: Rybka v2.2n2.mp.w32 15.0/30 010110011001011011=00=10010110 ······························ 225.00

30 games played / Tournament is finished
Name of the tournament: Arena tournament
Site/ Country: JORGE, United States
Level: Tournament Game in 5 Minutes
Hardware: AMD Athlon(tm) Processor 1200 MHz with 512 MB Memory
Operating system: Microsoft Windows XP Professional Service Pack 2 (Build 2600)
PGN-File: C:\Program Files\Arena\Arena.pgn
Website:
E-Mail Address: