On Saturday 10 January 2004 16:23, Richard Urwin wrote:
> On Saturday 10 Jan 2004 3:16 pm, Anne Wilson wrote:
> > On Saturday 10 January 2004 14:44, Richard Urwin wrote:
> > > If I can get SpamAssassin in it is more
> > > powerful.
> >
> > I have to question that.  I know it's better known, but I get far
> > better rates under POPFile than I did when I tried SA.  POPFile
> > functions typically at 99.87% accuracy.  I don't think you can
> > better that.
>
> That's a very impressive number, but I don't see what POPFile does
> that SA doesn't. You do need to train SA, and it doesn't start
> using Bayesian analysis until it's database has 200 messages in. I
> don't have any experience of how easy, hard or accurate this bit of
> SA is. I run it at work, but the setup precludes training the
> Bayesian bit.

I suspect that's the crucial bit.  SA must be very capable to keep its 
good name, but it does take a lot of training.  POPFilter, OTOH, 
seems to train very quickly.  I had to re-install a couple of weeks 
ago, so right now my statistics are

Emails classified:      3,457
Classification errors:  15
Accuracy:       99.56%

Traffic over the holidays has been predictably light.  I would expect 
this to be back to the 99.87% that I had before Christmas within the 
next week or so.  As you can see, since I reinstalled on 22 Dec it 
has only made 15 classification errors (most of them in the first 48 
hours) - I always correct them immediately, to speed up the learning.

Anne
-- 
Registered Linux User No.293302
Have you visited http://twiki.mdklinuxfaq.org yet?


Want to buy your Pack or Services from MandrakeSoft? 
Go to http://www.mandrakestore.com

Reply via email to