[Spambayes] Can I upgrade without losing my training?

Lanny Trager ltrag at optonline.net
Fri Sep 5 14:15:51 EDT 2003

First, I must apologize to my fellow list members for the paucity of details 
regarding my specific setup.  

I humbly and profusely apologize.

I am running SpamBayes on RedHat 9 with the latest RedHat kernel 
(2.4.20-20.9), using Kmail and of course pop3proxy.

I was until this morning running "spambayes-1.0a4", I have upgraded to 
spambayes-1.0a5 and so far all of the problems that I was having seem to have 
been resolved by the upgrade.

In spambayes-1.0a4 I first noticed that not all of my incoming e-mail was 
showing up on the review page.  Some would show up after I went to the config 
page and hit the save button.  I had trained 42 ham and 42 spam by pasting 
the source including full headers into the Train on a message box.  SpamBayes 
was not doing very well in catching spam though.  This just didn't jive with 
what I had read about SapmBayes so I started searching the mailing list 
archives for a clue.  It was begining to look like my database was corrupt. 

Having found some messages that suggested that my system might be using 
Dumbdbm and that that was not the preferred database, I downloaded 
which_database.py and ran it.  It  told me that I had Pickle, Dumbdbm, 
Dbhash, and Bsddb available.  But claimed that my storage hammie.db was none.

After a bit of fiddling, including going back a version to 1.0a3 I managed to 
corrupt the database enough so that I no longer got the X-spambayes: ham - 
spam or unsure.  I was now getting X-Spambayes-Exception: 
exceptions.AssertionError() in probability() at
spambayes/classifier.py line 307: assert hamcount <=nham

This was actually helpfull to me, the mystery was over, the database was 
definitively currupted.  By then I had subscribed to the list and recieved 
notification that 1.0a5 was available.

This morning, I started from scratch with 1.0a5 and all seems well so far.  


\ /
 X   ASCII Ribbon Campaign
/ \  Against HTML Email

More information about the Spambayes mailing list