I found the detailed results page tough to read. Here's a quick key on
how to read it. Maybe it's overkill.
For those trying to read the results... Taking the last line from test 1,
"explore_rate = 1 - ego_base wins 62.7 + 3.8 % games (158 ames)"
The first part "explore_rate = 1 - ego_base" is a description of the
white player.
"explore_rate = 1" is a description of the variable changing in the
experiment
"-" is a separator, not part of an equation
"ego_base" is the program used to play white (sometimes omitted)
The second part "wins 62.7 + 3.8 % games" is a win (for white) of 62.7%
of the time (with a confidence of +/- 3.8%)
The last part is obvious "(158 ames)". The number of games played to
come to that conclusion (158)
I don't know what confidence level is used (e.g. 95%), but I bet some
simple match could figure it out.
Łukasz Lew wrote:
My student - Filip Gruszczynski, made extensive testing of alternative
formulas for UCT.
Over 30 000 games; ~10 different algorithms with various constants
(including BAST).
http://students.mimuw.edu.pl/~fg219435/Go/
From the article:
"It seems, that the more simple approach we take, the better. EGO_SQRT
seems to be the best when compared with other ego programs, though
EGO_BAST seems to be a bit more effective against GnuGo - possibly
with better constants it could win even more."
--
Taking the opportunity, I also would like to announce that development
of libego will be stopped at least for 4 months (I'm on 3 month
internship + vacations :] ).
Best Regards,
Lukasz Lew
_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/
_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/
