On 09/22/12 22:38, John Hardin wrote: > On Sat, 22 Sep 2012, James wrote: >> On 09/22/12 20:50, Glenn Sieb wrote: >>> On 9/22/12 8:36 PM, Glenn Sieb wrote: >>>> On 9/22/12 8:32 PM, James wrote: >>>>> It didn't help. :-( >>>>> I got spam with a low score. >>>>> X-Spam-Status: No, score=2.4 required=3.0 tests=BAYES_00,HTML_MESSAGE, >>>>> INVALID_DATE,MIME_HTML_ONLY,RDNS_NONE,URIBL_DBL_SPAM autolearn=no >>>>> version=3.3.2 >> >> I haven't trained my bayes, it seemed complex when I last looked. > > The "BAYES_00" in the rule hits above strongly suggests that mistrained Bayes > is the primary culprit (if this is indeed a FN). > > If you haven't trained bayes, then return your required score to 5 and > disable Bayes and see if it behaves better. > > Then, collect a few hundred ham and spam messages, wipe your bayes database, > train it properly using those messages, and reenable it. > > Training is pretty simple. All you have to do is collect representative ham > and spam messages in a couple of mail folders, and tell sa-learn to process > them. The critical bit that usually causes problems is you must run sa-learn > as the user that the MTA is running spamassassin as, or make sure that your > configuration defines a system-global Bayes database. > > You will probably want to set up a nightly cron job to run sa-learn against > your ham and spam corpus folders. Then you can just add misclassified > messages (spams from your inbox, and hams from your spam quarantine) to those > folders as you encounter them. >
I wrote this little script to update the bayes rules. I can do this on my imap account but my pop3 account gets way more spam and the messages are no longer on the machine with sa once I pop them off. Any comments on my script? #!/bin/bash IFS=$'\n' FOLDERLIST=`find Maildir -name .INBOX\* -type d;` for i in $FOLDERLIST; do echo "Processing ""$i" # `sudo sa-learn"--ham "$i"` done #`sudo sa-learn --spam Maildir/.Junk