On Sat, 14 Aug 2010 14:46:50 +0200
Julien Valroff <jul...@kirya.net> wrote:
> Hi Stevan,
>
Hello Julien,
[...]
> My bad!! I have cleared my db after upgrading to 3.9.1 rc1 to switch to
> the osb tokenizer. The source tracking is only effective after the
> training period:
> src/dspam.c:
> if (CTX->totals.innocent_learned + CTX->totals.innocent_classified > 2500) {
> if (CTX->result == DSR_ISSPAM &&
> strcmp(CTX->class, LANG_CLASS_VIRUS) != 0 &&
> _ds_match_attribute(agent_config, "TrackSources", "spam")) {
> ...
>
> I didn't remember this limitation... all my apologies for the noise!
>
No problem.
> btw, why is it so? Do we consider it is not worth logging when training?
> I would personally prefer inaccurate statistics than no statistics at
> all.
>
The reason for that is RABL. RABL was/is supposed to use the data from
TrackSources to automatically make a DNSBL and tracking sources while in
training could influence the accuracy of such a automatically made DNSBL. IHMO
it would be better to disable TrackSources only if RABL queue is active. So
people like you that use TrackSources for logging/graphing would get what they
need while admins using TrackSources for RABL would still get accurate data for
their RABL.
> Cheers,
> Julien
>
--
Kind Regards from Switzerland,
Stevan Bajić
> --
> Julien Valroff <jul...@kirya.net>
> http://www.kirya.net
> GPG key: 4096R/290D20C5
> 092F 4CB5 5F19 E006 1CFD B489 D32B 8D66 290D 20C5
------------------------------------------------------------------------------
This SF.net email is sponsored by
Make an app they can't live without
Enter the BlackBerry Developer Challenge
http://p.sf.net/sfu/RIM-dev2dev
_______________________________________________
Dspam-user mailing list
Dspam-user@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/dspam-user
