[Spambayes] Training

[Michael K. Schummers]

I had a strange quirk that I don't know how to reproduce last week. On
Wednesday, 11/19 toward the end of the day, I was training on some
additional spam to help "balance" the training database. It appeared that it
worked at the time from the statistics on Spambayes Manager "General" tab. I
did this training with the "Rebuild Entire Database" radio button unchecked
but I had left the Rescore radio button checked. I was out of the office
until this morning and found that the General Tab now said I had 1617 Spam
and 0 Ham trained when I came in this morning. Needless to say, everything
was moved to the Spam folder. I would have concluded that I had
inadvertantly left the "Rebuild" button checked except that when I went into
Spambayes Manager to redo the Training to correct things, the radio button
was still unchecked. Any thoughts of what might have caused this kind of
reset in the database statistics? I do recall checking the training
statistics after I was done last Wednesday to see how close to "even" I had
gotten the spam vs. ham training ratio so my guess is that something occured
when/while I exited the Spambayes Manager but I can't think of anyway to
"reconstruct" the problem to see what might have happened. Do any of the log
files record anything that might give me a clue?? It wouldn't have been that
big a problem but my workstation stays on 24x7 and 4 + days of ham had to be
manualled extracted from the Spam folder. I have since retrained with
"Rebuild" checked and things seem pretty much back to normal.
 
Thanks for any thoughts. I am running the Outlook Plugin v 0.81 and had been
running fine for quite a while except (per an earlier post) for the fact
that "normal" operation seems to build up the number of ham scored so that
the ham/spam ration seems to quickly (weeks) become skewed to 3 - 4 x the
number of ham trained vs. spam. I am assuming this imbalance is occuring
because of the number of Unsure that get retrained into Ham but I have kept
detailed enough statistics to tell.
 
Michael K. Schummers
General Manager

 
 
 
 
 
(859) 455-7317
HYPERLINK "mailto:mikes at members-insurance.com"mikes at members-insurance.com

---
Outgoing mail is certified Virus Free.
Checked by AVG anti-virus system (http://www.grisoft.com).
Version: 6.0.543 / Virus Database: 337 - Release Date: 11/21/2003


-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: image/gif
Size: 6113 bytes
Desc: not available
Url : http://mail.python.org/pipermail/spambayes/attachments/20031124/bb66d69d/attachment-0001.gif
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: image/jpeg
Size: 7559 bytes
Desc: not available
Url : http://mail.python.org/pipermail/spambayes/attachments/20031124/bb66d69d/attachment-0001.jpe