You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@spamassassin.apache.org by Matthew Yette <my...@mapolce.com> on 2005/08/19 20:24:14 UTC
Ham not auto-learning?
Running the sa-stats.pl version 0.9 that produces a chart with stats on
what rules are hit for spam and ham most frequently, I notice that of
all 13,411 autolearns performed, every one of them was for spam. Ham has
0 messages autolearned. Wouldn't, for example, a message that comes in
and has been whitelisted (and therefore scoring ~ -100) be autolearned?
My bayes thresholds are set for 12.1 (spam) and -12.0(ham).
--
Matthew Yette
Senior Engineer - NOC/Operations
MA Polce Consulting, Inc.
myette@mapolce.com
315-838-1644 (w)
315-356-0597 (f)
AIM/Yahoo: MAPolceNOC
MSN: noc@mapolce.com
Re: Ham not auto-learning?
Posted by Craig McLean <cr...@craig.dnsalias.com>.
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Matthew Yette wrote:
| Running the sa-stats.pl version 0.9 that produces a chart with stats on
| what rules are hit for spam and ham most frequently, I notice that of
| all 13,411 autolearns performed, every one of them was for spam. Ham has
| 0 messages autolearned. Wouldn't, for example, a message that comes in
| and has been whitelisted (and therefore scoring ~ -100) be autolearned?
| My bayes thresholds are set for 12.1 (spam) and -12.0(ham).
Matthew,
If I recall correctly, bayes learning thresholds are compared against a
message score *before* whitelist adjustments are made, so unless a
message scores -12 using just the standard rules (unlikely) it will
never be learned as ham. Just set the ham threshold to 0 and you'll see
any message hitting no positive scoring tests being learned as ham.
Regards,
Craig.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)
iD8DBQFDBiVFMDDagS2VwJ4RAkBVAJ9IHh/KpJ3uZRG+pZYQ7Mo77cPiaQCgvEOw
F4d9wRpAt5ZHl2jHGfSE7RQ=
=cXb8
-----END PGP SIGNATURE-----
Re: Ham not auto-learning?
Posted by Steve Martin <st...@planomartins.com>.
I'm going to guess that whitelist isn't taken into consideration.
-12 for autolearning of ham is pretty extreme, I'm not surprised you
aren't seeing any autolearning. The default is .1
On Aug 19, 2005, at 1:24 PM, Matthew Yette wrote:
> Running the sa-stats.pl version 0.9 that produces a chart with
> stats on
> what rules are hit for spam and ham most frequently, I notice that of
> all 13,411 autolearns performed, every one of them was for spam.
> Ham has
> 0 messages autolearned. Wouldn't, for example, a message that comes in
> and has been whitelisted (and therefore scoring ~ -100) be
> autolearned?
> My bayes thresholds are set for 12.1 (spam) and -12.0(ham).
>
> --
> Matthew Yette
> Senior Engineer - NOC/Operations
> MA Polce Consulting, Inc.
> myette@mapolce.com
> 315-838-1644 (w)
> 315-356-0597 (f)
> AIM/Yahoo: MAPolceNOC
> MSN: noc@mapolce.com
>
--
Steve Martin http://www.cheezmo.com/
Smart Calibration, LLC http://www.smartcalibration.com/
The Widescreen Movie Center http://www.widemovies.com/
Letterboxed Movie TV Schedule http://www.widemovies.com/lbx.html