New new script to train bogofilter
Przemyslaw Brojewski
przemek at my.tenbit.pl
Fri Jul 4 15:38:34 CEST 2003
Hello,
I am using bogofilter 0.13.7 for two weeks, and I think, the way I use
it just might give some more light on this subject.
I had no previous spams to train with, so I decided to use bogofilter
untrained. I configured it to use three state evaluation. Used example
values for ham cutoff of 0.1 and spam cutoff of 0.95.
I check e-mail roughly 4 times a day. On each round, every messages
that got "Unsure" rating got fed to bogofilter -s or -n. Any message
that bogofilter is sure about is not going to training, unless
the program got it the wrong way.
I receive about 250 messages a day, of which about 200 are spams.
There happen one or two false negatives now and then (sorry, I don't
intend to mesure it, I want a solution that works unattended, and
bogofilter is doing fine so far). I haven't noticed any false positives,
not even at the very beginning, when I was carefully watching for them.
finally, my database sizes:
goodlist: .MSG_COUNT 143
spamlist: .MSG_COUNT 232
-rw-r--r-- 1 przemek users 1196032 Jul 4 15:20 goodlist.db
-rw-r--r-- 1 przemek users 1499136 Jul 4 15:20 spamlist.db
Regards,
Przemyslaw Brojewski
More information about the Bogofilter
mailing list