Re: BlameThorstenAndJenny: doesn't appear to classify some mails correctly after learning

Martin Steigerwald <[email protected]> Wed, 4 Nov 2009 22:10:17 +0100
Newsgroups gmane.mail.spam.crm114
Message-ID <[email protected]>
Am Mittwoch 04 November 2009 schrieb Eric S. Johansson:
> On 11/3/2009 4:44 PM, Martin Steigerwald wrote:
> > Maybe its a treshold issue? Those after learning wronly classified
> > mails are all above 5 but below 10.
> >
> > martin@shambhala:~/Mail/.trash.directory/Unsicher/cur>  grep
> > CRM114-Status *
> > 1257260447.22316.TzOey:2,S:X-CRM114-Status: UNSURE (   8.75  )
> > 1257260448.22316.oyQo1:2,S:X-CRM114-Status: UNSURE (   9.35  )
> > 1257261491.22316.huxyz:2,S:X-CRM114-Status: UNSURE (   8.84  )
> > 1257261491.22316.huxyz:2,S:X-CRM114-Status: UNSURE (   8.14  )
> > 1257280958.22316.OnA0L:2,S:X-CRM114-Status: UNSURE (   8.98  )
> > 1257280993.22316.Tapxy:2,S:X-CRM114-Status: UNSURE (   7.99  )
> > 1257280994.22316.czbwZ:2,S:X-CRM114-Status: UNSURE (   9.18  )
> > 1257281001.22316.Xp889:2,S:X-CRM114-Status: UNSURE (   8.63  )
> > 1257283921.22316.6uatN:2,S:X-CRM114-Status: UNSURE (   7.12  )
> >
> > ("Unsicher" is the german word for "unsure")
> 
> I have been experiencing a similar or common even the same bug for a
>  couple years now.  I bet if you look at the statuses and plotted them
>  you'd find that there is a pileup of scores near the threshold on the
>  red and a smaller hump on the green.  this kind of problem has
>  rendered twopenny blue almost useless because you spend your day
>  cleaning out hundreds of messages that should have been classified as
>  spam and are not training properly.
> 
> I'm more than willing to run experiments on different classifiers but I
>  need a simpler interface than even mail Reaver.  I want no cache and
>  an output that makes it easier to determine what's going on without
>  schlepping the message in both directions.

Actually I just like to work as before the update. ;)

It worked perfectly for me. Now I have at least 30 mails in my unsure 
folder that are wrongly classified. Something is really not working as 
smooth as before here. Fortunately I have policyd-weight running on my 
mail server, otherwise I it might have been 500 or 1000 mails already.

I am considering setting ham treshold to +5 instead of +10. But I have a 
mail that has even been classified below +5 after learning. At least it 
should mitigate the problem somewhat.

Just want things work as quickly as possible, cause I had a bit lot of 
issues to debug. Okay, I run Debian testing/unstable and I compile my own 
quite recent. But at the moment I long at least a bit for Debian Lenny 
with bog standard stuff ;).

Ciao,
-- 
Martin 'Helios' Steigerwald - http://www.Lichtvoll.de
GPG: 03B0 0D6C 0040 0710 4AFA  B82F 991B EAAC A599 84C7

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day 
trial. Simplify your report design, integration and deployment - and focus on 
what you do best, core application coding. Discover what's new with
Crystal Reports now.  http://p.sf.net/sfu/bobj-july

_______________________________________________
Crm114-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/crm114-general
signature.asc (application/pgp-signature, 198 B)
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (GNU/Linux)

iEYEABECAAYFAkrx7bkACgkQmRvqrKWZhMeLNwCeL4eQUJS1fTjUg7oyD7Z+ziZH
Ot4An3qQuS0zQQOithtQzSBB3BQqU4sh
=gU/L
-----END PGP SIGNATURE-----