repetitive training

[Tom Anderson]

On Tue, 2004-03-09 at 06:59, Greg Louis wrote:
> Great!  A procedure that is easy to reproduce "in the lab" for testing.
> I may very likely run such an experiment.

I'd like to see the results of that.

> Is this per user?  If so, the results aren't as good as I get ("no" fn
> or fp and an average of about 5 unsures/day), but pretty close, and
> with what seems to be significantly less effort.  (By "no" I mean
> unmeasurably few, like one in 120,000 or so; for me that translates to
> one in about six weeks.)

Well, the FN/unsure numbers are mine alone.  Milage varies from user to
user, but overall I've seen a decline in FN and unsures.  However, the
main reason they are still relatively high is because I'm very resistant
to tuning.  I only use small increments at a time and no automatic
tuning.  Since I never get any hams in my unsures, I can lower both the
spam_cutoff and ham_cutoff a bit more to reduce both FN and unsures.  I
go at least a week or two between changes in these numbers, usually
longer, and only change them 0.1-0.5.  Of the FN and unsures I receive,
they are all of two types: really long messages with many paragraphs of
plain english (as I've described here before), and virii with no body
and only an attachment.  If I killed the virii in procmail, my FN and
unsure numbers would decrease significantly.  I'm still holding out to
see if I can get bogofilter to filter these though.  I'm also in the
process of writing a prefilter that will apply some heuristics to the
headers to help polarize them, as too many hammy tokens are contained in
the x-headers, dates, etc., of spam.

Tom

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL: <http://www.bogofilter.org/pipermail/bogofilter/attachments/20040309/824746bb/attachment.sig>