On Fri, 30 Oct 2015, j...@lexoncom.com wrote:
On Fri, 30 Oct 2015, j...@lexoncom.com wrote:
I already cleaned the db to make sure I dont have it broken.
Would it be better to turn off the autolearn.
Teach sa ham and spam from over 200 messages and then turn back the
autolearn?
How big is your userbase and ham email volume?
If both are fairly small, I'd leave autolearn turned off and do purely
manual classification and training. That's what I do and I have good
results, but I'm only supporting 5 users.
similar to yours
i have been running sa for few years so i do have like
80000-100000 entries in auto-whitelist per user
i cleared it and i will start over
with no auto-whitelist enabled for now
auto-whitelist (AWL) has nothing to do with bayes or autolearn. Its name
is misleading, it is actually more of a score averaging facility to allow
for an occasionally spammy-looking email from someone with a hammy
history.
Turn off autolearn to start while you're evaluating the performance of
your initial corpora. Train any FPs and FNs (keeping them as part of your
reference training corpora), and get your DNS issues resolved.
not sure where is the problem with dns
as i have the caching server setup
Are you sure that your DNS server is actually the one being used? Can you
check the DNS server's logs to see queries coming in from your network and
beign recursively resolved?
Perhaps post your DNS server's config file?
Once things are stable and working smoothly for a while, then you can turn
autolearn back on if you feel your mail volume justifies it.
--
John Hardin KA7OHZ http://www.impsec.org/~jhardin/
jhar...@impsec.org FALaholic #11174 pgpk -a jhar...@impsec.org
key: 0xB8732E79 -- 2D8C 34F4 6411 F507 136C AF76 D822 E6E6 B873 2E79
-----------------------------------------------------------------------
...the Fates notice those who buy chainsaws...
-- www.darwinawards.com
-----------------------------------------------------------------------
Tomorrow: Halloween
