Fischer or Robinson when using -u -e -p with maildrop
David Relson
relson at osagesoftware.com
Wed Nov 5 16:15:30 CET 2003
On Wed, 5 Nov 2003 23:01:34 +0800
Yusuf Goolamabbas <yusufg at outblaze.com> wrote:
> Hi, I am presently using the Robinson mechanism in bogofilter where I
> invoke it via maildrop with -r -u -e -p. Should I switch to the
> Fischer method, would I gain/lose anything ? I believe with -u -e -p,
> I will not take into account the "unsure" state which Fischer offers
> me
>
> Regards, Yusuf
Hello Yusuf,
Long ago, Gary Robinson suggested using the Fisher modification to his
algorithm. At the end of the calculation a chi-square test is done.
This results in low scores being squeezed towards 0 and high scores
towards 1, thus reducing the number of mid-range scores. I think of
this as a "certainty" test -- the chi-square test looks at the spam/ham
score and the number of tokens involved and produces a result that
indicates bogofilters "certainty" that the message is ham or spam.
By all means, drop the "-r" option and let bogofilter use its default
algorithm (Fisher). Whether you set a ham_cutoff value to take
advantage of tri-state results is up to you. Some people, like me,
swear by it and other people won't touch it with a ten foot pole.
David
More information about the Bogofilter
mailing list