Doing this right?
JoeHill
joehill at sympatico.ca
Thu May 5 18:06:34 CEST 2005
On Wed, 4 May 2005 11:34:04 -0400
Greg Louis disseminated the following:
> > bogofilter -n < falsepos
>
> Right.
>
> > spamicity=0.494604
> > after running the retraining command on the same message, I get this:
> > joehill at node3:~/mail$ bogofilter -vv < falsepos
> > X-Bogosity: Unsure, tests=bogofilter, spamicity=0.264029, version=0.94.4
> >
> > Am I correct in assuming that the lower score is because of the retraining
> > command I ran on the message?
>
> Exactly correct. Training increased the database's nonspam counts by 1
> for every token in that message, and the effect was to lower the spam
> "probability" value for each token; that, in turn, lowered the overall
> spam score as you saw.
...and yet, after 'retraining', a subsequent mail from the same person (replying
to my reply) again received a relatively high score:
X-Bogosity: Unsure, tests=bogofilter, spamicity=0.498903, version=0.94.4
Luckily, I had upped my spam threshold to .50, but I'm still curious as to why
mail from this particular person is getting scored so high after retraining with
a misclassified message from them.
--
JoeHill / RLU #282046 / www.freeyourmachine.org
12:00:42 up 73 days, 13:10, 7 users, load average: 0.04, 0.01, 0.00
+++++++++++++++++++++++++++
Rule $19.99 (Brad `Squid' Shapcott): The Internet *isn't* *free*. It just has an
economy that makes no sense to capitalism.
More information about the Bogofilter
mailing list