[Spambayes] re-org - making a package &c.
neale at woozle.org
Mon Jan 13 11:42:56 EST 2003
Carel Fellinger <carel.fellinger at chello.nl> writes:
> The problem with initialisation is that there is no data to start
> with, so such a "default adaptor" can't come into play until you've
> gathered some spam and ham. But as part of this extra fine tuning
> step I proposed, it sure seems interesting to try to derive good
> settings for the options at several moments in time and see whether
> they differ widely for all us early adaptors.
With my scheme at least, you are expected to at some point have some ham
and some spam. Maybe not initially, but after a week or two you are
supposed to be collecting examples of both.
In any case, what you propose would work as a tuning tool, to be run
whenever you want to tune your config. I would look at the existing
test programs and try to figure out a way to combine them. I believe
Tim wrote it so that one set of trained data can be used over and over
for multiple types of scoring. That and the existing support modules
should make it little more than chaining scoring methods together.
You still interested in doing this, Carel?
More information about the Spambayes