On 2024-01-30 at 09:59:52 UTC-0500 (Tue, 30 Jan 2024 09:59:52 -0500)
joe a <joea-li...@j4computers.com>
is rumored to have said:
Advisable to "prune" Bayes data based on age?
Yes. That is why it has an expiration model. Expiration may be de facto
blocked on some busy systems so you may need to explicitly force it
occasionally. The command "sa-learn --dump magic" will show you
expiration and other Bayes metadata.
While cleaning up recent Ham/Spam, found my "saved SPAM" goes back to
2013.
Why that's over . . . wait, I need to take off my socks . . .
I've still got some almost 3x as old. BUT: I do not use it for training
SA today.
So, how old is "too old". For saved SPAM?
I would suggest a year as the outer edge of Bayes usefulness.
I find it helpful to keep my decades of garbage because I use them (and
my ham archive) in developing prospective rules. There are non-obvious
fingerprints in some spam that imply decades-long spamming operations.
--
Bill Cole
b...@scconsult.com or billc...@apache.org
(AKA @grumpybozo and many *@billmail.scconsult.com addresses)
Not Currently Available For Hire