Re: [exim] Possible Spamassassin related feature request

On Tue, 21 Jun 2011 08:18:24 -0700, Marc Perkel <marc@???> wrote:
> Yes - there are email streams that I have that I know are ham and ones I

> know are spam. My idea was to tap about 10% of these emails (randomly)
> to train the bayes and hopefully improve accuracy.

Then Ted's suggestion of the pipe transport is one answer.
An alternative is to save a random sample into a local mailbox and use
SA's "sa-learn" from cron.

How you implement the randomisation is up to you.

Graeme