On 07/12/2014 12:41 PM, Axb wrote:
per default, no "single* SA rule should tag a msg as spam.
I'm trying to "trust the defaults". But what would be the hazards of
leaving BAYES_99 at 3.5 and upping BAYES_999 to 1.5? It seems that I
should be able to trust Bayes to declare a message spam on its own.
if you can't imagine the "hazards" you better not do it ;-)
That's clear enough advice. And I think I'll take it. But briefly, what
are the hazards? Thunderbird's spam filter does well, and I've always
figured it was bayes. And it doesn't seem to wait for 200 spam/hams
before it kicks in.
Also, after considering advice from this list, I've turned autolearn
back on. But that advice also implied that a system-wide bayes database
works well. I'm still using per-user databases. This is all complex
enough that I'm very willing to listen to people with more real-world
experience.
I'm not getting complaints from my users. But I don't feel that I am
doing as well as our ISP's admins did in sorting the spam/ham.
And thank you for your help.
-Steve Bergman
