How to properly enable autolearn for spam ?

[bougatoyta]

Hi,

I do not clearly understand what I need to do to enable autolearn on PMG.

I have a zimbra mail server that is "protected" by PMG but quite a bit of obvious spam are passing by, even if we refuse anything that as a spamscore > 6.

I read somewhere that PMG need 250 mails with a score > 12 to learn is it still true ?

How can I efficiently train PMG ?

Regards

 

[fgams]

A score of 6 sounds high to me, we use 3.

 

[bougatoyta]

A score of 6 sounds high to me, we use 3.

I will consider this possibility, but it might raise a lot more false positive

 

[Stoiko Ivanov]

How can I efficiently train PMG ?

I would first check out the getting started article on our wiki:
https://pmg.proxmox.com/wiki/index.php/Getting_started_with_Proxmox_Mail_Gateway

with those settings you usually get quite good results already.

I will consider this possibility, but it might raise a lot more false positive

You can always put mails between 3 and 6 into quarantine - in my experience with a score of >4 very few false positives happen (with 3 it's also quite usable)

Else regarding the Bayes training- checkout the links from:
https://forum.proxmox.com/threads/auto-whitelist.66435/post-298470

In my experience Bayes usually does not help too much - so my general recommendation is to disable it.

If this does not help - post some logs of problematic mails (PMG logs which spamassassin rules fired - so this helps in finding potential for optimization)


I hope this helps!