bayes_auto_learn_threshold_nonspam 1.1 bayes_auto_learn_threshold_spam 9.0 bayes_auto_learn 1
I am getting "autolearn=ham" for various scores. Some get learned as ham with a negative score, but others with a negative score are not learned. Shouldn't it learn all <1.1 as ham?? Maybe I'm missing something about the autolearn threshhold? Thanks - John
The score used isn't the same as the final message score, and there's lots of other conditions to be aware of.
1) the score, computed as if bayes were disabled (including a change to scoreset 0 or 2) must be in the range
2) whitelists, blacklists, AWL and gtube are ignored in the learning score computation
3) the autolearn must not contradict existing bayes learning (don't autolearn as ham anything that would have gotten BAYES_99)
4) There must be no other opportunistic processes that have the bayes DB locked for writing (other autolearns, expiry, sync, etc).
