Congratulations to the winners (Hitman's performance was fantastic) and a
big thanks to you Sedat for all the hard work you put into running this
tournament.
There are some tough book makers out there and I'm
glad that my entry at least beat Nelsons low baller! Speaking of Mr.
Hernandez, here is the output from Victor's Aquarium script:
[You must be registered and logged in to see this link.]Looking at the "longer and better" columns you'll see that Hitman topped the best book list as both white *and* black.
Also, here is a simple take on the "better" book exit column:
Top 13 as white, in order:
Hitman H15a
GuardianAngel H15a
NoFear H15a
Sequoia Rybka 4.1
Attacker H15a
Atlantis H15a
Istanbul H15a
Awesome H15a
Defender H15a
Don H15a
Muntaqim H15a
Alfil H15a
Bulbul8X H15a
Top 13 as black, in order:
Hitman H15a
Awesome H15a
Rompecoco H15a
Prodigy H15a
Defender H15a
Don H15a
Neapus H15a
Muntaqim H15a
SicilianFinale H15a
Kerch H15a
ASLevels H15a
TheWar H15a
Sequoia Rybka 4.1
Of course there are lots of different ways to analyze the data.
And well done to Audy with his brave Critter!
[You must be registered and logged in to see this link.]By
[You must be registered and logged in to see this link.] Very interesting stuff, which offers very good clues regarding how the winners did it and how the losers went astray.
First
off, relative length of time in book versus opponents, both colors.
Hitman was in book 9.5 moves longer than his opponents with both colors,
beating out former champion Defender. Hitman just did the same thing
only better:
Hitman H15a 9.50
Defender H15a 7.30
JapanTopOne H15a 4.00
Don H15a 3.60
S.W.Y.D H15a 3.30
SicilianFinale H15a 2.35
Attacker H15a 2.30
Sequoia Rybka 4.1 2.00
Bullseye H15a 1.90
Rompecoco H15a 1.75
...
LastPlaceFun H15a (3.30)
Jakal H15a (3.40)
BookNewSD H15a (3.70)
Kuscu H15a (3.80)
Morphius H15a (4.35) spotted opponents over four book moves! Way to go! What courage!
Secondly,
relative book exit. Here I assumed the opening position was +0.15 and
any deviation from that (or -0.15 if black) was a positive or negative.
So here, Nick's book Sequoia achieved the best book-exits, averaging 13
centipawns of evaluation improvement per game from both side of the
board.
Sequoia Rybka 4.1 13.00
NoFear H15a 12.30
GuardianAngel H15a 10.95
MadJackal H15a 8.40
Istanbul H15a 8.30
SicilianFinale H15a 7.00
TheWar H15a 6.50
SeekWanda H15a 6.30
AlvinsAngels H15a 6.15
Don H15a 6.05
...
SweetDreams H15a (6.80)
Terminator H15a (7.45)
Sirena H15a (11.55)
FICSTop H15a (15.85) I thought nobody could possibly do worse than this!
Arilovsu H15a (17.05) I was wrong! Congratulations to Arilovsu!
Hitman
and Defender really stand out as similar in strategic style. In other
words, regardless of what openings they chose, they had the same grand
strategy for winning this tournament and they both carried out their
plans with considerable craftsmanship and skill. What is interesting to
observe is not only their 1-2 finish in time in book, but their telling
average book-exits: Hitman only 2.80, Defender 0.30. So clearly they
did NOT follow an IDeA-type book-exit evaluation optimization path, but
instead followed lines that exited in deceptively equal positions yet
greatly favored them. In other words, their much-deeper analysis of
book positions, from far deeper outposts on the tree, told them that
seemingly equal positions were actually winners. This is bookmaking at
its finest and is the result of very long, hard work! (I have yet to
analyze their openings in detail to see if I can figure the source of
their inspiration in my database.)
[You must be registered and logged in to see this link.][You must be registered and logged in to see this image.][You must be registered and logged in to see this link.] PS - One possible source of their mediocre book-exits and great outcomes
is that they compared Houdini 1.5 (which they knew everyone would use)
against some other top engine's output and spotted major deviations in
their evaluations. That would explain why at book-exit HOUDINI thinks
the position is equal, but another engine might see a significant edge
and be correct.
[You must be registered and logged in to see this image.]