[Spambayes] stand alone filter.

[Eric S. Johansson]

skip at pobox.com wrote:
>     Eric> so far the interfaces in hammy.py are proving sufficient.  I'm
>     Eric> using the dbm classifier.  which reminds me.  any protections
>     Eric> against simultaneous access i.e.  using hammy.py in multiple
>     Eric> processes with the same features file?
> 
> There is no protection there, but it should be easy to add something similar
> to the recent changes I made to lock pickle file access using my lockfile
> module.  Those changes may only be available by checking out the source from
> Subversion.  I'm not sure we have a 1.1 alpha release cut yet with them.

okay.  I put my own locks around it.  Shared for score checking and exclusive 
for training.  So far, I am impressed.  My test cases produce charts that looked 
very similar to yours.  It's trained up reasonably well in a few hours with 
relatively small numbers of messages and a few ah-my-hem bugs making training 
fail.  :-)   will see what it looks like in the morning.  Also, it'll be 
interesting see what happens if I overtrain (i.e. train every message that's 
wrong).  Worse comes to worse, I can retrain from scratch using the five-day 
history.

I was serious about needing some volunteer help with Thunderbird extensions for 
my project.  I don't know why they make the environment so bloody miserable to 
work with.  Layers upon layers of complexity.  One would think they're trying to 
make a career out of Thunderbird and Firefox.

Anyway, my extension only needs to grab a header from the current message, 
extract a field and do an Ajax style call back to my server.  oh yes, and move 
the current message to the designated "forget me" box.  Not real hard in any 
other environment but Thunderbird has me running away.  Oh well, someday.


---eric