[Spambayes] ageing out database entries

[Tim Peters]

[Anthony Baxter]
> Large training sets were shown to lead to a slight decrease in
> accuracy, but we're talking tens of thousands of messages in either
> direction. It'll be in the spambayes list archives, back in the dark
> ages.

Was that result unique to you, Anthony?  I ran tens of thousands of
messages, and didn't see that.  Guido ran tens of thousands but didn't keep
testing long enough to report anything.  Who else ran against test sets that
large?  And you always got somewhat different results because you mixed ham
from multiple users, right?