Spamassassin autolearn

[felix_84]

Hi. Can anyone tell what should be the correct settings for bayes autolearn?

Where shoud i set thresholds and other

 

[felix_84]

Anyone?

 

[x3ppy]

I am also interested in this question.
For starters, I did these actions:
cp /var/lib/pmg/templates/local.cf.in /etc/pmg/templates/local.cf.in
vim /etc/pmg/templates/local.cf.in
# ...
bayes_auto_learn 1

Log before:
SA score=1/5 time=3.518 bayes=undefined autolearn=no autolearn_force=no

Log after:
SA score=0/5 time=4.529 bayes=undefined autolearn=ham autolearn_force=no

 

[felix_84]

I did the same and added this lines in /etc/pmg/templates/local.cf.in

use_bayes_rules 1
bayes_auto_learn 1
bayes_auto_learn_threshold_nonspam -0.1
bayes_auto_learn_threshold_spam 10

Nonspam messages learned but spam is not.

sa-learn --dump-magic shows following:

0.000 0 46 0 non-token data: nspam
0.000 0 522 0 non-token data: nham

But this 46 tokens was learned a long time ago, and now this counter does not changing even though i have many spam messages with score 10 and more

 

[x3ppy]

I will observe my training for some time and write an answer.

 

[x3ppy]

2019.11.08
0.000 0 4 0 non-token data: nspam
0.000 0 473 0 non-token data: nham
2019.11.11
0.000 0 9 0 non-token data: nspam
0.000 0 499 0 non-token data: nham

 

[heutger]

Train spam by yourself. Ham is working well, but spam need to be done manually. And It's better not to tweak as original setting is very conservative not to mis-train maybe-spam, which really may be no spam.