On Tue, 11 May 2004, Mark wrote:
Lucky Leavell wrote:
OS: 9.0 SA Version: 2.55 (from SuSE archive)
I have ample ham and spam counts (6,000 spam, 38,000 ham) but SA is still not using Bayesian analysis. I did manually set use_bayes to 1 in /etc/mail/spamassassin/local.cf though it is supposed to be the default but still no luck (based on the X-Spam-Status tests= line(s) in the message headers).
Aside: I managed to get bogofilter working and use it via procmail but that is after SA processing.
Thank you, Lucky
Have you set a path to where you'd like the system-wide bayes database?
Just as an example: bayes_path /home/mybayes
This is probably it. I have only run sa-learn for my user account and only intend to run it for accounts with sufficient ham/spam history.
You'll soon see whether the database is created and used.
My database is in ~/.spamassassin/
Second, you need to tell spamassassin what kind of maildir you have. I use a single file rather than a folder with separate messages, so I have to say:
sa-learn --ham --mbox /path/to/ham sa-learn --spam --mbox /path/to/spam
I finally figured this one out after sa-learn reported it processed one message instead of the 6000 or so in the file!
After that I run sa-learn rebuild.
Did that...
Admittedly, I've had to install a lot of third-party rulesets to keep the old spam detection rates high and upgrade to 2.63, but spamassassin has been working on tiptop form for a long time now and my only problem is the occasional false positive, usually from this list.
Still using 2.55 downloaded from SuSE with default rulesets. I am far more concerned about false positives than an occasional missed spam message. Fortunately, most of these have been noncritical. Thank you, Lucky