Matt Kettler said:

> As for disabling the network checks for auto-learning, that makes sense to 
> me as well, since the bayes code learns from text tokens, not IPs.

Actually, not quite right, if you're scanning with network tests, it'll
do the auto-learn score test with network tests as well.

But regarding the use of Bayes in auto-learn determination causing
feedback, that's the big danger.

BTW, one possible way to avoid FP/FNs getting into the auto-learn data
further, is to modify the learn() sub to add to the existing verification
steps:

  - recomputed hits must be < bayes_auto_learn_threshold_nonspam or
    > bayes_auto_learn_threshold_spam

  - for spam, must have 3 head hits and 3 body hits

add this one:

  - previous hits must be < bayes_auto_learn_threshold_nonspam or
    > bayes_auto_learn_threshold_spam

that would mean both the existing main score and the recomputed score
must agree that the mail is spam or ham.

comments?

--j.


-------------------------------------------------------
This SF.Net email is sponsored by: INetU
Attention Web Developers & Consultants: Become An INetU Hosting Partner.
Refer Dedicated Servers. We Manage Them. You Get 10% Monthly Commission!
INetU Dedicated Managed Hosting http://www.inetu.net/partner/index.php
_______________________________________________
Spamassassin-talk mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/spamassassin-talk

Reply via email to