On Mon, 10 May 2004, Roland Hill wrote:
> >> 0.000 0 2 0 non-token data: bayes db > >> 0.000 0 81 0 non-token data: nspam > >> 0.000 0 355 0 non-token data: nham > > > > spamd/spamassassin starts using bayes if you have 200 ham and 200 spam, > > this isnt yet fullfilled. so collect some more spam first. > > Thanks for your reply. > > My concern is more that I fed 20 odd spams to bayes via sa-learn, yet the > nspam value did not increase by this at all (81 before and after). Should > it have increased as I have assumed? yes, should have been happen. try sa-learn with option -D when learning ham/spam. maybe there is something faulty. > Also, as my spam volume is low (ISP uses SA, plus it is my home network) > am I right in assuming that sa-learn should feed the bayes db (say 20-30 > at a time), and then once 200 is hit, bayes will then be used? yes, 200 is the minimum for each spam / ham, spamd needs for using bayes. below that number spamd wont use it. > Or put another way, do I need to feed the bayes db a minimum of 200 in one > hit from the get go (as opposed to being drip fed)? hmm, my englich sux a bit, so i might get you wrong. but, it doesnt count, when or how many spam/ham bayes learns at a time. the fact is, you need 200 of each, otherwise (below 200) spamd skips these test. but spamd learns ham / spam meanwhile, if bayes_auto_learn is set to 1 (done per default), even if bayes itself is not used for tests yet. regards, Matthias
