[ale] Procmail... ?

[Jason Day]

On Thu, Sep 25, 2003 at 02:43:46PM -0400, Christopher Fowler wrote:
> Here is part of the headers from one that got through
> 
> ----- Cut Here -----
> X-Spam-Status: No, hits=4.4 required=5.0
>         tests=BAYES_60,HTML_60_70,HTML_IMAGE_RATIO_06,HTML_WEB_BUGS,
>               MAILTO_TO_REMOVE,MAILTO_WITH_SUBJ,MAILTO_WITH_SUBJ_REMOVE,
>               REMOVE_PAGE,SUBJ_REMOVE
>         version=2.50
> X-Spam-Level: ****
> X-Spam-Checker-Version: SpamAssassin 2.50 (1.173-2003-02-20-exp)
> ----- Cut Here -----

First of all, you need to upgrade your version of spamassassin.  I see
that 2.60 is the latest version, but I'm still using 2.55.  When I
upgraded from 2.50 to 2.55 it caught a *lot* more spam.

Second, are you training your bayes classifier?  Any time you get a spam
that the bayes classifier thinks is ham (or vice versa) you need to
correct it with sa-learn.  IIRC, you are using mutt, so put these in
your ~/.muttrc:

---
macro index X "|spamassassin -D -r" "report spam to Vipul's Razor"
macro index H "|sa-learn --no-rebuild --single --ham" "learn message as HAM"
macro index S "|sa-learn --no-rebuild --single --spam" "learn message as SPAM"
---

Only add the top line if you also use razor, otherwise ignore it.  If
you do use razor though, reporting a spam to razor will also cause the
bayes classifier to relearn the message as spam (if it didn't already).

Whenever you get a message that was mis-classified, just return to the
index view, hilight the message (or tag it with `t'), and press H or S
as appropriate (or `;' then H or S if you've tagged multiple messages).

If you want to verify that the bayes classifier re-learned the spam,
pipe the message back through spamassassin in test mode: return to index
view, hilight the message, press the `|' key, and you should get a
prompt saying "Pipe to command: " type this:
spamassassin -d | spamassassin -t

The spamassassin -d will remove any previous SA markup which might
confuse SA or the bayes classifier, and the spamassassin -t will report
the results.  The bayes score should be 90 or greater.

HTH,
Jason
-- 
Jason Day                                       jasonday at
http://jasonday.home.att.net                    worldnet dot att dot net
 
"Of course I'm paranoid, everyone is trying to kill me."
    -- Weyoun-6, Star Trek: Deep Space 9