Author: Mike S.
Date: 18:31:49 11/01/03
Go up one level in this thread
On November 01, 2003 at 15:21:42, Rolf Tueschen wrote:
>On November 01, 2003 at 09:30:02, Mike S. wrote:
>
>>(...)
>Thanks for the response BTW. But I must tell you that I am not willing to give
>you some extra lessons in experimental design and stats.
Thanks, I suffer from an overdose of statistics anyway, when reading the message
boards and computerchess homepages.
>You are in defense for ChessBase.
:-)) If I had told you I bought a cat and wrote a poem about the moon, you'd
probably reply too, that I did this in defense of ChessBase.
>You want to defend the intended practice of
>lending the general.ctg from ChessBase to The King.
I'm not so sure that ChessBase is happy about that practise. - In recent
King/SSDF discussions (also before the test started a few days ago), I've always
written that my idea would be, to generate a simple but solid "neutral" opening
tree for that purpose, from a good game collection. Mixing products would not be
my first idea. But OTOH, this would have meant one more disadvantage for King,
raising protest from the other side so to speak. In such a simple tree,
computerchess specific editing (not necessarily engine specific but more
general) would be missing, in addition to the fact that the King 3.23 tests
start very late after the relase, which was long ago and competitors have
released new versions since then. Therefore, I think it's quite ok too, as a
compromise. It isn't meant as a kind of new rule IMO, but as an exception from
the existing testing practise, also because the common usage of King (among
computerchess fans) is so much different from the usual out of the box/default
philosophy.
>Therefore you rely on the results of Michael42.
I don't especially rely on these results, I took them as an illustration how
books can make a very small difference only, sometimes. But I don't think these
results are typical in this extreme form, book vs. no book and such a small
difference.
>But
>these results don't mean anything and your relying is biased and typical for a
>CSS spin doctor in defense of a ChessBase marketing idea.
This is a strange remark. What marketing idea? The general.ctg is not a product
of it's own, it's an old book sold with Fritz 6 some years ago.
>>He had very good results with a special book he has compiled... I don't see why
>>this should do worse with learning on. I think it would benefit from the
>>learning function just like other books do as well. I really think the book
>>comparison is valid.
>
>No, it's not. He makes clowns out of Sandro, Jeroen and other book doctors.
I don't think so!
But books are overrated in this circus. Good, useful, can decide games
*somtimes*, but in the general opinion much too overrated. The engines rule the
game, not the opening books.
>created his book for whom??? For SHREDDER or both? But if he created his book
>for his SHREDDER, why then could he have the idea to let FRITZ [!!] play with
>it??? No, Mike, if you can't judge such nonsense out of your own you have no
>idea what testing conditions are all about! Excuse me, no insult intended. That
>is simply incredible and Sandro already told you!
But this is explained very well on M42's page. It's a computerchess specific
book, not for a specific engine, but based on computerchess games and practical
experience with several different engines. Maybe this book would even have been
a better (and less controversial) replacement for CM9000.OBK in the King tests.
>(...) Tell me what is going
>on there??? Why must you defend such nonsense experiments???
It is the nature of an experiment, that something unusual is being done (but
carefully). I don't particularly defend that experiment - I wasn't aware that it
needs defense :-))
Maybe this is all very normal from my viewpoint, because my own experiments were
even much more unusual, sometimes (not primarily aiming for rankings, but more
for other explorations like what is the value of the bishop pair in
computerchess, how can engines make use of a large developement advantage,
etc.).
I've even played experimental engine games with odds. Fritz 7 can give odds of a
rook to Doctor? 3.0 and still win:
[Event "Turmvorgabe"]
[Site "P3/700, 10'/40"]
[Date "2003.01.01"]
[Round "2"]
[White "Fritz 7"]
[Black "Doctor? 3.0"]
[Result "1-0"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/1NBQKBNR b Kkq - 0 1"]
[PlyCount "98"]
[EventDate "2003.01.01"]
{W=11.4 ply; 370kN/s B=15.9 ply; 57kN/s} 1... Nf6 {
[%eval -510,19] [%emt 0:00:17]} 2. d4 {[%eval -487,12] [%emt 0:00:21]} e6 {
[%eval -508,19] [%emt 0:00:22]} 3. Nf3 {[%eval -484,11] [%emt 0:00:18]} c5 {
(d5) [%eval -509,17] [%emt 0:00:15]} 4. c4 {(e3) [%eval -478,10] [%emt 0:00:18]
} Qa5+ {(cxd4) [%eval -542,17] [%emt 0:00:21]} 5. Nc3 {
[%eval -462,11] [%emt 0:00:10]} Ne4 {(d5) [%eval -538,17] [%emt 0:00:22]} 6.
Bd2 {(Dc2) [%eval -472,12] [%emt 0:00:19]} Nxc3 {[%eval -570,18] [%emt 0:00:15]
} 7. Bxc3 {[%eval -475,12] [%emt 0:00:08]} Qxa2 {[%eval -572,18] [%emt 0:00:14]
} 8. e3 {(e4) [%eval -469,12] [%emt 0:00:24]} Na6 {
[%eval -573,17] [%emt 0:00:16]} 9. Bd3 {(d5) [%eval -481,11] [%emt 0:00:17]}
cxd4 {[%eval -584,17] [%emt 0:00:16]} 10. exd4 {[%eval -478,11] [%emt 0:00:13]}
Nb4 {[%eval -584,17] [%emt 0:00:13]} 11. Be2 {[%eval -500,12] [%emt 0:00:23]}
d5 {(b5) [%eval -589,17] [%emt 0:00:13]} 12. c5 {[%eval -472,12] [%emt 0:00:27]
} Nc6 {[%eval -589,18] [%emt 0:00:12]} 13. O-O {[%eval -475,11] [%emt 0:00:17]}
g6 {(b6) [%eval -588,18] [%emt 0:00:14]} 14. b4 {
(Lb5) [%eval -419,11] [%emt 0:00:11]} Bh6 {[%eval -559,16] [%emt 0:00:16]} 15.
Qd3 {[%eval -394,11] [%emt 0:00:17]} Ne7 {(e5) [%eval -532,16] [%emt 0:00:14]}
16. Ne5 {(Ta1) [%eval -256,11] [%emt 0:00:05]} f6 {
[%eval -410,17] [%emt 0:00:14]} 17. Ra1 {[%eval -272,11] [%emt 0:00:06]} Qxa1+
{[%eval -415,19] [%emt 0:00:18]} 18. Bxa1 {[%eval -291,12] [%emt 0:00:04]} fxe5
{[%eval -418,19] [%emt 0:00:21]} 19. dxe5 {[%eval -297,12] [%emt 0:00:13]} a5 {
(0-0) [%eval -423,19] [%emt 0:00:19]} 20. b5 {
(bxa5) [%eval -303,13] [%emt 0:00:32]} Bg7 {
(0-0) [%eval -431,17] [%emt 0:00:14]} 21. Qa3 {
(Lg4) [%eval -306,13] [%emt 0:00:24]} Bd7 {(a4) [%eval -426,19] [%emt 0:00:19]}
22. c6 {(Lb2) [%eval -294,12] [%emt 0:00:14]} bxc6 {
[%eval -506,18] [%emt 0:00:12]} 23. b6 {[%eval -287,13] [%emt 0:00:12]} Bc8 {
(Tb8) [%eval -489,17] [%emt 0:00:12]} 24. Qd6 {[%eval -259,12] [%emt 0:00:08]}
Kf7 {(a4) [%eval -365,17] [%emt 0:00:12]} 25. Bb2 {
(Dc7) [%eval -244,13] [%emt 0:00:18]} a4 {(Te8) [%eval -403,16] [%emt 0:00:14]}
26. Ba3 {(Lg4) [%eval -234,12] [%emt 0:00:07]} Re8 {
[%eval -328,17] [%emt 0:00:10]} 27. Qc7 {[%eval -241,11] [%emt 0:00:06]} Bf8 {
[%eval -305,16] [%emt 0:00:12]} 28. Bd3 {(Lf1) [%eval -228,12] [%emt 0:00:21]}
Kg8 {[%eval -306,16] [%emt 0:00:18]} 29. g4 {
(b7) [%eval -222,11] [%emt 0:00:10]} Kg7 {(Kf7) [%eval -313,16] [%emt 0:00:12]}
30. Kf1 {(Lf1) [%eval -206,11] [%emt 0:00:10]} Kf7 {
(c5) [%eval -316,17] [%emt 0:00:10]} 31. Ke1 {
(Kg1) [%eval -181,11] [%emt 0:00:15]} Bh6 {[%eval -325,17] [%emt 0:00:14]} 32.
h4 {(Kf1) [%eval -166,11] [%emt 0:00:09]} Bg7 {
(Lf8) [%eval -340,16] [%emt 0:00:09]} 33. Kd1 {
(Lf1) [%eval -156,11] [%emt 0:00:10]} Bh6 {(Lf8) [%eval -351,16] [%emt 0:00:11]
} 34. Ke1 {[%eval 0,13] [%emt 0:00:04]} Bf4 {[%eval -332,17] [%emt 0:00:09]}
35. Kd1 {(Kf1) [%eval -125,12] [%emt 0:00:13]} Bh2 {
[%eval -332,17] [%emt 0:00:13]} 36. Ke1 {[%eval -112,11] [%emt 0:00:08]} d4 {
[%eval -261,16] [%emt 0:00:11]} 37. Kd1 {(Le4) [%eval -81,11] [%emt 0:00:11]}
g5 {(Lf4) [%eval -216,15] [%emt 0:00:10]} 38. Bxh7 {
(hxg5) [%eval 78,11] [%emt 0:00:07]} Ba6 {[%eval -33,15] [%emt 0:00:13]} 39. h5
{[%eval 191,12] [%emt 0:00:13]} c5 {[%eval 11,14] [%emt 0:00:09]} 40. Bg6+ {
[%eval 222,11] [%emt 0:00:06]} Kf8 {(Kg7) [%eval 172,16] [%emt 0:00:08]} 41.
Bxc5 {(Lxe8) [%eval 350,11] [%emt 0:00:06]} d3 {
(a3) [%eval 425,15] [%emt 0:00:17]} 42. Qd7 {
(Dc6) [%eval 1069,11] [%emt 0:00:07]} Bc4 {[%eval 1225,18] [%emt 0:00:19]} 43.
b7 {[%eval 1372,11] [%emt 0:00:08]} Bb3+ {[%eval 32756,17] [%emt 0:00:15]} 44.
Kd2 {[%eval 1422,11] [%emt 0:00:05]} Bf4+ {[%eval 32757,9] [%emt 0:00:02]} 45.
Kxd3 {[%eval 2350,11] [%emt 0:00:10]} Ba2 {
(Tab8) [%eval 32758,9] [%emt 0:00:02]} 46. Bxe7+ {
[%eval 32762,6] [%emt 0:00:01]} Kg8 {[%eval 32761,0] [%emt 0:00:00]} 47. Bh7+ {
[%eval 32763,5] [%emt 0:00:02]} Kxh7 {[%eval 32764,1] [%emt 0:00:00]} 48. Bf8+
{[%eval 32764,3] [%emt 0:00:01]} Re7 {[%eval 32765,0] [%emt 0:00:01]} 49. Qxe7+
{[%eval 32765,2] [%emt 0:00:00]} Kg8 {(Kh8) [%eval 32766,0] [%emt 0:00:00]} 50.
Qg7# {[%eval 32766,2] [%emt 0:00:01]} 1-0
But from a total of 8 such games against Fritz 7 and Tiger 14, Doctor 3.0 won 5
and lost 3, which means the rook more than compensates for the strength
difference, in average.
(Piece values have been expressed in Elo, and I wanted to see how the method to
evaluate these by match results, works. But I didn't continue this.)
>I would be interested if SHREDDER has an extra implementation in its engine?
>That is what I would ask myself from a science view. But nobody seems to be
>interested in such questions. The results show into that direction. But then I'm
>a layman in CC [computerchess].
In the developement advantage experiment mentioned, Shredder (5.32 at that time)
did very well, too. For example a win with Black (!) against Tiger 14 from this
position:
[D]rnbqkbnr/pppppppp/8/8/3PP3/2NB1N2/PPPBQPPP/R3K2R b KQkq - 0 7
(But Tiger 14 also won his game with Black against Shredder 5.32, in return!)
Most probably, general priniciples of the early opening have been "implanted"
very good, in Shredder. OTOH, in the FRC rating list where games are played
without books too, because these start from Fischerrandom positions, Shredder
7.04 is sligthly behind Hiarcs 9 (!) and Fritz 8. This could indicate that it
uses patterns which fit to the classic starting position better. But that are
just assumptions; the 3 engines mentioned are within a small performance
bandwidth of 0.6% only, on that rating list.
http://www.beepworld.de/members53/frc-list/
mfg.
M.Scheidl
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.