From: "Alexander Piavka" <[EMAIL PROTECTED]>
On Sat, 15 Jul 2006, Magnus Holmgren wrote:
On Tuesday 11 July 2006 23:16, Alexander Piavka took the opportunity to write:
> Hi , i'd like to know if its possbile and how, to ignore specific rule
> scores (like ALL_TRUSTED) then calculating the autolearn threshold for
> spam and ham?
"Like" ALL_TRUSTED, eh? If you have a problem with ALL_TRUSTED you likely have
a bad trusted_networks setting. Adding a host to trusted_networks means that
you trust it not to forge headers and not to originate spam, meaning that if
ALL_TRUSTED fires then the message *should* definitely be ham, otherwise your
assumption that the host can be trusted is wrong.
No i've no problem with ALL_TRUSTED , it's just i thoght it's not a good
idea to learn every mail from trusted networks as ham, i wanted to make a
bayes autolearn independent of the sending source and thus ignore ALL_TRUSTED
and some more tests. Since this way bayes would learn from much more ham
messages than spam messages,esspecialy since most spam messages we get are
the same. Thus i thougth since the bayes databese size is limited it
should have learn from at least as much spam mail as ham, to have more
spam mails detected by bayes.
But probably i'm wrong or not?
One might say two things. The first is a startled "Well duh!" The second
is, "if you have ALL_TRUSTED" appear as a rule hit on every message then
you're being silly.
ALL_TRUSTED does not mean a damn thing with respect to whether a message
is ham or spam. It just says that the received headers are likely to be
accurate in as much as "you" or a "trusted agent" oversees the header
generation.
{o.o}
