>> Benny Pedersen wrote: >>> >>> >>> On Sun, July 26, 2009 15:29, snowweb wrote: >>> >>>> 0.000 0 258 0 non-token data: >>>> nspam >>>> 0.000 0 160 0 non-token data: >>>> nham >>> >>> try to have them more or less equal to have good bayes >>> db >>> >>> so if less then 1000 in diff is fine >>> >>> if more then 1000 adjust learning scores >>> >>>> I see from that, that I've not trained as many HAM as I >>>> thought! OK, I'm off >>>> in search of some more HAM! Thanks guys. >>> >>> super you found the problem finaly >>> >>> -- >>> xpoint >>> >> >> Sorry, it hasn't solved it :( >> >> As you can see below, I now have more than 200 of both >> SPAM & HAM trained: >> >> [r...@s1 Maildir]# sa-learn --dump magic >> [1644] warn: FuzzyOcr: Cannot find executable for >> tesseract >> 0.000 0 3 0 non-token data: >> bayes db version >> 0.000 0 278 0 non-token data: >> nspam >> 0.000 0 221 0 non-token data: >> nham >> 0.000 0 34120 0 non-token data: >> ntokens >> 0.000 0 1245088823 0 non-token data: >> oldest atime >> 0.000 0 1248616170 0 non-token data: >> newest atime >> 0.000 0 1248620830 0 non-token data: >> last journal sync atime >> 0.000 0 0 0 non-token data: >> last expiry atime >> 0.000 0 0 0 non-token data: >> last expire atime delta >> 0.000 0 0 0 non-token data: >> last expire reduction count >> >> but here are the spam headers of a message, which show >> that bayes is not being used: >> >> X-Spam-Flag: NO >> X-Spam-Checker-Version: SpamAssassin 3.2.4 (2008-01-01) >> on s1.snowweb.info X-Spam-Level: ****** >> X-Spam-Status: No, score=3.0 required=4.7 >> tests=RELAYCOUNTRY_US autolearn=no version=3.2.4 >> X-Spam-Report: >> * 3.0 RELAYCOUNTRY_US Relayed through United States >> of America >> * 0.0 HTML_MESSAGE BODY: HTML included in message >> X-Spam-Relay-Country: US US US US US US US US US >> >> This is not a random event. Bayes is not being used for >> any message. > > Which user you run the spam check with? sa-learn shows > root's magic. If use for example use user spam to check > the spam, you should call sa-learn as > > sa-learn -u spam --dump magic
One can also use bayes_sql_override_username spam in local.cf, if the bayes is in a SQL database. I have that, and there seems to be no difference if I use -u or not.