... and possibly deleting rules where the GA seems to be saying they're
irrelevant -- that would speed up processing too if there are fewer
regexps to match against.

C

On Thu, 2002-02-14 at 15:42, Gene Ruebsamen wrote:
> I tend to agree..
> 
> I would trust the GA on this one, irrelavent rules will tend to get
> low/irrelavent scores over time.  Because the GA caught the spam using other
> rules, the "HUNZA BREAD" rule probably has become irrelavent.
> 
> Instead of manually fiddling with the scores of those rules, I would say
> adding more relavent rules would be the better route to take.
> 
> Gene
> 
> > -----Original Message-----
> > From: [EMAIL PROTECTED]
> > [mailto:[EMAIL PROTECTED]]On Behalf Of
> > Daniel Pittman
> > Sent: Thursday, February 14, 2002 3:20 PM
> > To: [EMAIL PROTECTED]
> > Subject: [SAtalk] Re: SA 2.01 low scores
> >
> >
> > On 14 Feb 2002, Craig Hughes wrote:
> > > I'll investigate -- it's weird, because both GAs seem to assign low
> > > scores to "HUNZA_DIET_BREAD" even though it obviously only appears in
> > > the spam corpus, and not in non-spam.
> >
> > Sure, but how much of the spam corpus do they show up in?
> >
> > > In fact my GA which allows -ve scores give it a -ve score! Now having
> > > said that, all the HUNZA_DIET_BREAD messages in the corpus end up
> > > being identified as spam anyway because of the other rules they
> > > trigger, but it's still very strange behavior by the GA.
> >
> > To me, it sounds very much like the HUNZA_DIET_BREAD rule isn't really
> > very useful for identifying spam, overall, because spam that contains it
> > is already marked by many other, more general, characteristics.
> >
> > So, the score for that fluctuates semi-randomly over time but, because
> > it's never selected for by the GA, it drifts gently toward irrelevance.
> >
> >
> > At least, that's my assumption based on what knowledge I have of GA
> > evolution...
> >
> >         Daniel
> >
> > --
> > We are always getting ready to live, but never living.
> >         -- Ralph Waldo Emerson, _Journals_, 1834
> >
> > _______________________________________________
> > Spamassassin-talk mailing list
> > [EMAIL PROTECTED]
> > https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
> >
> 
> 
> _______________________________________________
> Spamassassin-talk mailing list
> [EMAIL PROTECTED]
> https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
> 
> 


_______________________________________________
Spamassassin-talk mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/spamassassin-talk

Reply via email to