Quote:
Originally posted by Jeremy Howard
Kind of - there's actually two kinds of reports "explicit" and "implicit". The 'empty junk' creates an "implicit report". We will deal this these differently, and probably add an option to allow you to not train from implicit reports at all.
No. The system is not currently doing any learning from these reports - they are not yet being used at all, except for our own internal statistics, so we can get some idea of how we might be able to use them in the future. Before we use this information, we will provide appropriate docs and options, and make an announcement. We will clear the reports at that time and start afresh.
No.
|
Any further developments on these items above? Any further thoughts on Bayesian filters?
I'm referring specifically to posts
http://www.emaildiscussions.com/...765#post257765 and
http://www.emaildiscussions.com/...06#post259406.
Thanks