>>> On 3/31/2012 at 6:27 PM, RW <rwmailli...@googlemail.com> wrote: > On Sat, 31 Mar 2012 12:17:52 -0400 > joea wrote: > > >> Beyond that, where can I find the difference, in a SPAM learning >> sense, between "sa-learn --spam filename" and "spamassassin -r < >> filename"? >> >> If I do the sa-learn on the same file, after doing spamassassin, it >> tells me 0 tokens. If I then do "sa-learn --forget filename", then >> "sa-learn --spam filename" it tells me 1 token learned. > > Are you sure that's what it says and not tokens from 1 message? The > reason I ask is that it's practically impossible for Bayes to find only > one token, so it would be a sign that something is wrong if some part > of spamassassin is telling you that.
Sorry, yes, "tokens from 1 message" is what it reports. I should know better than to shorten messages. > > >> I infer from this they perform similar or the same function, from a >> Bayes sense. > > spamassassin -r is mainly for reporting spam to SpamCop, Pyzor, etc, > training Bayes is just a side-effect. But there's no grounds for > thinking the training is going to be any different to running sa-learn > or auto-training. Thanks.