[Spambayes] Spammie on Outlook 2000 - questions

[Mark Hammond]

> <snip>
> > > I found one message in Spam Maybe that ended up being good. I
> > > did "Spam Recover" and got a rule error from OL about unable to
> > > move message to (as it happens) python-list. Closed message box
> > > and hit Spam recover again - no errors and message was moved
> > > to inbox.
> >
> > Odd.  I've tried to reproduce that error and can't.  Even if I
> > delete the original folder it finds it's way back to the inbox
> > without an error.  Make sure you let the list know if this is a
> > regular occurance.
> >
> Consider this notice: frequent rule errors upon mail fetch.

Can you clarify exactly what error?  This is the Outlook rule engine
displaying the error in a message box, not spambayes?

> These are of the
> form "couldn't move message to <folder>" and there are a
> couple of folders
> involved.

I have never seen this, but *do* see a number of messages fail to filter.
In general, this happens as Outlook is shutting down or starting up, but
occasionally it seems to happen for no good reason.

A "problem" is that I recently reconfigured my Outlook mail for "corporate"
mode (rather than "Internet Only").  It is possible to do this even if you
only use internet mail.  Corporate mode changes outlook *radically*.  One
large change is that the mail spooler now runs in its own process
(mapisp32.exe), and Outlook spools from that.  If you kill Outlook but leave
mapisp32 running, you often end up with messages in your inbox that the
Outlook rule engine missed (as did spambayes etc - but we catch what we
missed next startup).  In "corporate" mode, "Send and Receive" works
completely differerently (no way to see the total number of messages you are
pulling down) and all the "Preferences" dialogs change.

If anyone wishes to see this for themselves, simply hit the "reconfigure
mail support" button in your "Accounts" setup.  You don't lose your accounts
or mail, and you can re-configure later - but stuff changes all over the
place.

So, I see a little rule strangeness too, but I am not sure how much of this
is related to spambayes - I know not all of it is.

> I am also getting _many_ python-list messages (31
> just now) into
> Spam Maybe and some (17) just stayed in the inbox - generally most
> python-list messages go to the python-list folder.

The occasion I have seen this (Outlook failing to filter *any* message), I
had to manually kill Outlook and this new mapisp32.exe.  Then it came good.

Oh, I forgot to mention that a client insists I run the Outlook PGP plugin,
and this often causes that error too - it appears to have a memory leak, and
Outlook displays an "Out of resources" dialog, before all filters stop
working (and you can no longer even read mails, etc)

My point is that Outlook is complex, and extensible, so your problem could
possibly lie in a few places.  From what I can tell, most people have
generally successful Outlook rules working alongside spambayes.

> Looking over my Spam corpus (copied from deleted messages
> folder), I noticed
> that there are a lot of otherwise legit messages that just
> got deleted.

I'm not sure what you mean here.  You created the spam corpus by hand from
your "deleted items" folder, so I am not sure what you mean by "I noticed
that there are a lot of otherwise legit messages that just got deleted."
Deleted by whom?  Certainly if you trained spambayes incorrectly, it will
give incorrect results.

> Currently, I am getting few spam and many spam maybe - is this
> characteristic of a young spammie?

Not really, but possibly characteristic of poor training.  This really is
spambayes' biggest drawback IMO - not knowing when it is trained well enough
to do a good job.  Well, to be fair, it is too much to expect spambayes to
know this, when the people behind it don't even know the answer to that yet
<wink>

Anecdotally, it seems people get good results with a fairly balanced, well
filtered corpa.  A huge imbalance in spam:ham ratios may behave strangely,
even though there is some special weighting in that case.

And don't forget about "Filter Now" - you can pick random folders, and have
spambayes score them *without* doing a filter.  This can help you see how
well you existing mail would have scored with your new training data.  There
is no reason you can not "filter now" your "deleted items" to see how it
would go.  Particularly, do refilter your "ham" and "spam" folders, checking
that you haven't accidently misclassified any messages.

Hope this helps.  Let us know if it doesn't.

Mark.