| |
BGonline.org Forums
The Michael Depreli Bot Comparisons
Posted By: Frank Berger In Response To: The Michael Depreli Bot Comparisons (Art Grater)
Date: Tuesday, 15 February 2011, at 5:31 p.m.
Hi Art,
given that the differences we are talking about are very small, a sample of 500 games means close to nothing. For comparing candidate neural nets for BGBlitz I run 8 million games and even this is not enough to answer all questions.
The benchmark works in comparing the deviations of the bots during analyzes, makes rollouts of the differing moves and add up the error.
IMHO an excellent approach to get a quick evaluation of an bots strength.
ciao Frank
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.