On Thu, 2003-01-16 at 20:07, Marco Tabini wrote:
> On Thu, 2003-01-16 at 13:46, Adrian 'Dagurashibanipal' von Bidder wrote:
> > [no cc:s please, I read the list]
> 
> Fair enough. Sorry about that.

No problem.

> > > > Huh!? Ok, forget it.
> > > 
> > > Thanks for your summary dismissal. 
> > 
> > I'm interested in fighting spam. But I'm also somewhat strict in the use
> > of opensource software - I won't use binary only software.
> 
> That's your prerogative, and I respect your point of view, although I do
> not necessarily agree with you. I've explained my reasons for the binary
> only release, and mentioned that I'm open to providing the source code.

Where? I must have overlooked it.

> I simply ask that you also respect my point of view and avoid comments
> like "forget it", which really do not add to anything other than my
> blood pressure :-)

I apologize, I should probably have packaged that statement in a bit
more wrapping. It would have conveyed the same meaning, though, I fear.

> > > And exactly what happens if *you* happen to disagree with spamassassin?
> > 
> > I tweak the rules. It's occasionally necessary, but by now bogofilter
> > works very well so I give it more and more weight. I'd say there's <1%
> > false negatives and <.1% false positives (and those are mostly from
> > discussions about spam, sometimes containing parts of actual spam
> > messages - so I can understand that the system gets confused).
> 
> My point is, wouldn't it be better if the system were able to learn to
> tweak its own rules, particularly if you could train it to do so
> directly from within Evo, rather than by having to change the rules
> manually? I'm not interested in creating "the Spamassassin killer"
> (pardon the pun)--I'd rather learn from it--but perhaps there is a more
> convenient solution.

I'd not want my spamfilter in evo - I read mail from various machines,
sometimes with different MUAs. My spamfilter should always work.

But I agree with you that a self-tuning spamfilter is the ultimate goal,
and that MUA functions to easily train it are very convenient. The
bogofilter part of my setup already does that, I think that spamassassin
will go in the same direction - in the end, a combination with a
bogofilter-type approach and a spamfilter (rules based) approach, but
with self-tuning weights on the rules will be the best approach.

cheers
-- vbi

-- 
OpenPGP encrypted mail welcme - my key: http://fortytwo.ch/gpg/92082481

Attachment: signature.asc
Description: This is a digitally signed message part

Reply via email to