Am 21.09.2016 um 17:23 schrieb Thomas Barth:
Am 21.09.2016 um 16:13 schrieb li...@rhsoft.net:
#bayes
use_bayes 1
use_bayes_rules 1
bayes_auto_learn 1
so your setup either don't use that config (amavais or something like
that part of the game then you don't have just spamassassin)
or you have not trained enough spam *and* ham - or you train the wrong
bayes-database likely by calling "sa-learn" with the wrong user
https://wiki.apache.org/spamassassin/SiteWideBayesSetup
I cant do that because I dont have spam mails. I dont make
store&forward. I didnt thought that I need the spam uncompressed in a
folder for autolearning, I thought it works when sa is analyzing the
mail
how do you imagine autolearning from start with nothing trained?
just rely on rules and the train on false postives and negatives, in
other words every rejected message as spam and every passed as ham won't
work and when you think about it 10 seconds it should be obvious
anyways, you can't tell me that there are no mails which didn't make it
trugh the filters which where spam to find 200 of them and 200 ham
should be even more easy as long as you don#t delete your mail after read
My mailsystem checks mails in real time and blocks mail during
connection. If there is a false positive the sender gets an error and I
get a call of the sender to check it (last call was over a year ago :-).
But I have a compressed copy in the quarantine folder so that I can
check the reason anyway.
don't change the fact that you need the stuff which was wrong classified
and tell SA if it's good or bad to make the filter better
