[Spambayes] Training good messages has no effect

[Mark Hammond]

> I am using the outlook plugin of SpamBayes and since I
> updated to 0.6 I have
> a problem with some mailing lists. I keep training the emails
> as good, but
> it classifies them as spam anyway.

I think SpamBayes is training OK.

> Loaded message database from 'C:\Documents and
> Settings\hscheidler\Application
> Data\SpamBayes\default_message_database.db'
> Bayes database initialized with 150 spam and 13414 good messages

This is a huge imbalance.  Most of our testing has been done with roughly
equal spam and ham.  Anecdotally, 4:1 ratios can work OK - but you have
nearly 100:1

I suggest you train with far fewer ham.  SpamBayes doens't help you do this
very well, so the simplest is to create a temporary folder, copy a couple
hundred ham into it, do a complete re-train and see how things go.

> Current version is 0.6, latest is 0.6.

Current is actually 0.7, but we keep managing to screw up the website where
this is stored.  "Check Latest Version" should start reporting 0.7 soon.

Mark.