So, with SA 2.6x and having an autolearn bayes db per user setup, is the journal method recommended, or is it not needed in such a setup? Just trying to figure out the best configuration for per user bayes autolearning.
To be honest with you, I can't help you on this part.. I'm low volume.
Also, is it recommended to keep the default autolearn threshholds of 12 and 0.1, or are there any better values that others have found for an autolearn setup?
Personally, I dropped my autolearn threshold to be sub-zero. However, I'm in a situation where I have a few custom comp-rules that can actually make an email slightly negative in the first place.
Quite frankly, I think that having a threshold at or above 0 is outright dangerous to the integrity of your bayes database. Doing so means that any carefully crafted spam message that doesn't hit any non-bayes rules will get autolearned as ham. And heck, every once in a while a spam misses the entire ruleset just due to general spam-filter avoidance techniques, and not spamassassin specific tuning (ie: text mangling spams).
I'm just of the opinion that autolearning should be something that an email earns by rule, not by default if no rules prohibit it. However, with SA 2.60's default ruleset having only a few negative scoring rules (habeas SWE and bonded sender), it's impractical to use a negative score as a learning threshold.
Comp rules were removed in 2.60 as a result of abuse by spammers, but perhaps it might be worthwhile to bring some of them back with small negative score limits (ie: no scores assigned by the GA to less than -0.1). Admittedly a spammer could still abuse them for autolearning, but at least this way they'd have to intentionally try and won't as easily stumble into autolearning by accident.
The tightly controlled score limits would prevent the abuse-for-whitelisting we saw in the past where spammers were tacking on -16 points worth of comp rules to a spam because so many rules had scores in the -5 to -6 ballpark.
------------------------------------------------------- This SF.net email is sponsored by: SF.net Giveback Program. SourceForge.net hosts over 70,000 Open Source Projects. See the people who have HELPED US provide better services: Click here: http://sourceforge.net/supporters.php _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
