On Wed, 14 Dec 2005, Eben King wrote:
> On Wed, 14 Dec 2005, steve szmidt wrote:
>
> > On Wednesday 14 December 2005 12:01, Eben King wrote:
> > > I put Spamassassin on my machine to deal with the spam issue
>
> > > the two-week average of "% spams caught" has
> > > gone from 38% two weeks ago to 44% now.
>
> > > What am I doing wrong? Shouldn't it be improving?
> >
> > You also need to tell it how good email (ham) looks like. It should be a 50 -
> > 50 split when you train it.
>
> [eben@pc eben]$ sa-learn --ham --mbox mbox
> Learned from 37 message(s) (63 message(s) examined).
> [eben@pc eben]$ sa-learn --ham --mbox mail/linux
> Learned from 537 message(s) (695 message(s) examined).
> [eben@pc eben]$ sa-learn --ham --mbox mail/{bugtraq,sent-mail,sent-mail-nov-2005}
> Learned from 421 message(s) (423 message(s) examined).
>
> Cool, I'll see how it acts now.
OK, did that over 4 weeks ago (the middle of December). It's not better; in
fact it's worse -- 4-week rolling average at 28% of spam caught, 1-week at
18%. Still no false positives, which I guess is a Good Thing. How do I see
what SA thinks is going on? It doesn't appear to be learning (and I've been
running every spam through "sa-learn --spam" and every non-spam through
"sa-learn --ham"), as I plotted cumulative non-spam received (after manually
removing the spam and rubbing SA's nose in them), and its R^2 on a linear
best-fit line is 0.9864.
-- I firmly believed we should not march into Baghdad ...To occupy Iraq would instantly shatter our coalition, turning the whole Arab world against us and make a broken tyrant, into a latter-day Arab hero assigning young soldiers to a fruitless hunt for a securely entrenched dictator [.] - George Bush Sr.----------------------------------------------------------------------- This list is provided as an unmoderated internet service by Networked Knowledge Systems (NKS). Views and opinions expressed in messages posted are those of the author and do not necessarily reflect the official policy or position of NKS or any of its employees.
This archive was generated by hypermail 2.1.3 : Fri Aug 01 2014 - 16:28:56 EDT