Sure, but bogofilter has no pre-made rules, so it makes a bad first impression, and it’s tricky (as in, I wouldn’t know offhand how to do it) to implement site-wide. Rspamd wins on both these points (in fact, its pre-made rules are subjectively better than those of SpamAssassin), and it still has bayesian learning.
True. I use bogofilter locally just for myself, and I had a set of ~100k ham and spam e-mails for initial training.
For single user mail folder/server and local use, it's very simple to set up, compared to anything server side. You basically use 3 commands after initial training and don't need to worry about anything else.
I don't want to keep an e-mail archive server side (I use pop3), and that would include a spam filter's database, there are not many other filtering options with that requirements.
Anyway, I agree it would not make a good first impression without training data, at all. Though if you train it with what you actually receive, it's quite awesome. I get > 99% accuracy for both false positives/negatives.
