classification calculation method -- time to make Fisher the default?
David Relson
relson at osagesoftware.com
Mon Jan 13 18:15:04 CET 2003
Greg,
I've used all three algorithms.
During the pre-production evaluation period and for the first 3 weeks of
production usage, I used the Graham algorithm. Of course that far back, it
was the only choice.
The switch to Robinson was made at the end of October for its improved
ham/spam discrimination.
The switch to Robinson-Fisher was made at the beginning of December to take
advantage of its ham/spam/unsure tristate results. I'm very pleased with
it. A few spam that have avoided my spam and unsure folders since
then. Due to some database problems and a lexer defect (since corrected),
those messages didn't have X-Bogosity header lines. Excepting those
hiccups, bogofilter and the Robinson-Fisher algorithm are doing a
superlative job of classifying spam and keeping it away from my users.
I think it's time to change the default algorithm from Robinson to
Fisher. I have no reservations about supporting the change.
David
More information about the Bogofilter
mailing list