[Spambayes] Guidance re pickles versus DB for Outlook

[T. Alexander Popiel]

In message:  <15843.43038.341350.515691@slothrop.zope.com>
             Jeremy Hylton <jeremy@alum.mit.edu> writes:

>I just did a cvs update in spambayes and tried to restart my pspam
>code, but nothing is working anymore :-(.  I'm sorry I haven't had
>time to read every message on the proposed changes, but there's been a
>flurry of activity and I've got a day job.

Yeah, there's been a bunch of changes, mostly revolving around the
removal of update_probabilities.

>Anyway, here's a traceback.  Can anyone suggest quickly how I would
>fix this?  The pspam code calls learn() for a bunch of messages and
>then calls update_probabilities() at the end.  Is that the default
>now?  Or is that a discontinued feature?  Are the APIs documented anywhere?

You need to remove the _last_ argument passed to learn(), dealing
with whether or not to run update_probabilities().  Judging by later
mail, it sounds like you removed the second-to-last argument (ham vs.
spam), and are now calling everything ham.

You should also remove the call to update_probabilities(), since it's
toast, but it sounds like you already did that.

The API is unfortunately not documented yet; until someone goes back
through and updates it, the code itself will be the best documentation.

Addressing a concern from a later mail, the MetaInfo class exists to
make it easier to derive subclasses of the classifier and its parts,
without having to touch everything.  For the two counts that it currently
holds, it is certainly overkill, but it provides a nice way to expand
that set without having to go mucking around in as many places.  You
are correct that the revision number it holds is useless; it is cruft
left over from one of the interim schemes for doing away with
update_probabilities(), and as such the revision number should be
removed.  Perhaps I'll patch that.

- Alex