-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Tom Meunier wrote:
> Hi Dave, hope I can help here beyond parroting what I've read &
> played with...
Urghs. Seems to me that you used a figure of speech that I do not
understand (I'm german).

> I are but a simple sysadmin, not a programmer.

I'm both, but I haven't programmed in perl yet :-)

> Why would you want
> the Bayes score wrapped into the decision about whether to learn it?

I want the final score... see below.

> If I get something *REALLY* spammy, high enough to be learned, but
> then it scores BAYES_00, it may be dropped below my autolearn
> threshold.  That is the stuff I want going through bayes, not the
> BAYES_99 stuff.

See below... I have a different goal.

[score in the "safety zone" is auto-learned]
> I didn't see the complete debug output as it was processed by
> SpamAssassin, so I can't really comment, except keep in mind that
> Bayes autolearning uses either scoreset 0 or 1.  This may be scoreset
>  3 or 4, which may be different.  I don't know.

Ah. That may explain it.

>> I'm curious about the "safety zone"... why did the programmer(s)
>> decide to put it in? Even if they saw sense in it, most open source
>>  programs do let you make senseless configurations. Or, to say it
>> in other words: They not only let you shoot yourself in the foot,
>> they give you an assortment of guns already loaded and pointed
>> downwards[2] :-)
>
> I imagine they define "shooting yourself in the foot" as exactly what
>  you're attempting to do.  Unattended autolearning in the sweet spots
>  for false positives and false negatives.  I'd never autolearn that
> close to my threshold.  I'd hand-feed.  But I guess it's good that
> they allow you to do it now.

What I'm trying to do is described en detail on
http://maligree.yi.org/imapspamfilter.php (working draft, not done yet).

I want to sort my mails into three IMAP folders: NotSpam, DontKnow and
Spam. What goes to NotSpam should be auto-learned as ham; what goes to
Spam should be auto-learned as spam; messages to DontKnow should not be
auto-learned. I patched an IMAP server to automatically feed mails
manually moved into one of the folders NotSpam or Spam into sa-learn, so
that the folders NotSpam and Spam only contain mails already learned.
The mechanism I use to determine in which folder the mail will be stored
is simple: maildrop (a tool similar to procmail) looks for the
X-Spam-Level header and sorts the mail according to that header into the
IMAP folders. So the final score counts. If auto-learning depends on
another score than that one the sorting mechanism uses, some mails would
not be fed into the bayes filter.
Maybe I can configure maildrop to feed mails into sa-learn instead of
configuring spamassassin to auto-learn, but I'm not sure about that...

I hope I made my point clear; I'm not that good in english that I really
know whether I'm making myself clear or not.

cu
 Dave KLiczbor


NB: You don't have to answer to both the list and my private address.
The list would suffice, if that's not too much trouble for you :-)

- --
Q:      What do little WASPs want to be when they grow up?
A:      The very best person they can possibly be.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.2 (GNU/Linux)
Comment: Using GnuPG with Debian - http://enigmail.mozdev.org

iD8DBQE/Vmuu9gy0Ccu0VlMRAisFAJ9Od0NkHMQZyBs6qQ8CoDhSKDYwhgCfcpOF
pTLRJJwVD+uvLLH2smjiyZs=
=JA7Y
-----END PGP SIGNATURE-----



-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
Spamassassin-talk mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/spamassassin-talk

Reply via email to