Training Amavis

@lbutlr kremels at kreme.com
Mon Feb 1 08:32:30 CET 2016


> On Jan 31, 2016, at 9:49 PM, listsb-amavis at bitrate.net wrote:
> 
> 
>> On Jan 31, 2016, at 23.07, @lbutlr <kremels at kreme.com> wrote:
>> 
>> I get daily mails from wordpress verifying backups and these are all tagged as spam (at a very high score in the 7-13 range).
>> 
>> How do I train amavis? Do i just run  normal sa-learn as root? As the user? as the scan user?
> 
> you don't train amavis.  you train spamassassin.  they are two different pieces of software, which work well together.  while training spamassassin is good to do regardless of if you are having a problem or not, blindly training it to solve a specific problem is not a sensible approach.

I ma not blindling trainmen it. i wam training false positives as ham.

What I need to know is what user to train them as so that amavis will use the bases database that I am training to.

They all hit BAYES_99 and BAYES_999, some hit other rules as well.

X-Spam-Status: Yes, score=10.2 required=5.0 tests=BAYES_99,BAYES_999,
	HEADER_FROM_DIFFERENT_DOMAINS,NO_RELAYS,TVD_SPACE_RATIO,TVD_SPACE_RATIO_MINFP
	autolearn=no autolearn_force=no version=3.4.1


> instead, look at the *actual* scoring the message was given [X-Spam-Status header], and see which rule[s] are the ones which significantly contributed to the score.

Yes, that’s what I’ve done.

> then you can determine the right way to solve the problem.

Training falsely classified mail is *always* a good idea.

The question still remains, do I train SA as root, as the user (which is a problem for most of the users since they are virtual users in a database) or as the vscan user?

That is to say:

sa-learn -u *WHAT* --ham /path/to/ham

 

-- 
Stone circles were common enough everywhere in the mountains. Druids
built them as weather computers, and since it was always cheaper to
build a new 33-Megalith circle than to upgrade an old slow one, there
were generally plenty of ancient ones around --Lords and Ladies



More information about the amavis-users mailing list