[Spambayes] Training with cache on the POP proxy

[Jim McAtee]

Regarding training, I'm wondering what is the recommended method of using
cached messages.

I've run about 300 spam and 1100 ham messages through the interface, mostly
by having the server process existing Outlook Express dbx folder files, but
also using the web interface.  Now when I go to the review page, probably 95%
or more of the messages are shown as either 'ham' or 'spam', with the
remaining 'unsure'.  Is it recommended to train on either the ham or spam
messages for which SpamBayes has already made a correct decision?  Will this
reinforce the decision making, make it less accurate, or have no effect?