How to deal with extremely high spam levels

[David Relson]

Bob,

With quantities like you have (very few ham, lots of spam), another
option occurs to me.

Run contrib/randomtrain or contrib/bogominitrain.pl with whatever ham
and spam you have saved.  These scripts score each message and, whenever
there's an error (spam scored as ham or vice versa), will train
bogofilter with the problem message.  They can be run "off-line" and
will find a (roughly) minimal set of messages that need (should) be
added to the wordlist to enable bogofilter to do a good job.

If your many spam fit into only a spam number of categories, the
resultant wordlist will be very small.

David