On Mon, 12 Apr 2021 09:40:47 -0400 Greg Troxel wrote:
> 3) sa-learn does not document that it is no longer for BAYES, but a > general interface to mechanisms that learn. It always was in theory. > 4) There is a bonus of txrep_learn_penalty for learning spam, > default 20. If the user says it is spam by calling learn, then I > don't understand why it isn't just treated as score 20. Probably it's to keep retraining straightforward if you make a mistake. > (I've added -L to my script that calls sa-learn.) That sounds like a bad idea because you will be feeding it bogus data. It's probably better to just turn-of TxRep when training from an historic corpus and only run your periodic training on new mail.