[spambayes-dev] RE: [Spambayes-checkins] spambayes/Outlook2000 msgstore.py, 1.61, 1.62

[Tim Peters]

[Mark Hammond]
> Modified Files:
> 	msgstore.py
> Log Message:
> Fix [ 782709 ] not match between actual score and what's shown in
> outlook We can't trust potentially large properties in the data used
> to create the msg object.  Thanks Tim, Tony, everyone.

Excellent, Mark!  I confirm that all the (subtle, unless you're looking for
them) symptoms went away for me.  This had a major-league good effect on my
score distributions too:  I've been mildly puzzled for a long time that the
scores-after-training in my ham and spam Outlook data had much higher
variance than in standalone non-Outlook tests.  Now I suspect my modest
1,000-msg training database is much bigger than I really need <0.9 wink>.

BTW, the ham msg I posted about before, scoring 0.18 or 0.03 (depending on
where you looked), now scores a much more satisfying 0.000972835.

In effect, the Outlook addin has been acting much like a body-only
classifier?  Wow.  No wonder I had to keep training Laura Creighton's
two-liners as ham <wink>.