Starting a couple of months ago, I noticed a marked increase in spam, from one or two a week to dozens a day. I’ve set my configuration according to the documentation. I manually move or copy every email to a spam or archived folder and train the Bayesian filter on each using cron jobs. It appears to be learning. I run sa_dump weekly and see:
0.000 0 3 0 non-token data: bayes db version
0.000 0 281 0 non-token data: nspam
0.000 0 17247 0 non-token data: nham
0.000 0 1027047 0 non-token data: ntokens
I know it’s active because it does catch some spam. Here are the numbers of rule hits when I search against a training corpus of 4994 spam.
BAYES_00=4663
BAYES_05=12
BAYES_20=24
BAYES_40=29
BAYES_60=0
BAYES_80=0
BAYES_95=0
BAYES_99=4
(I’m not sure why they don’t quite add up).
It seems that Bayes hardly ever gets a high score. Is that because my ham to spam ratio is out of whack?
The network tests appear to be doing most of the work of catching spam. And the percentage caught is dropping, from 58% during the first week of July to 47% from July 22-29.
And again, I didn’t have to mess with any of this until several months ago. What changed? Did Site5 stop running sa-update or something?
]
MENU
Answer #1
Hello,
In this situation, I would recommend that you open a ticket with our support department. You can do this by submitting a ticket though backstage, or by emailing [email protected]