autolearn not working

[Trushin Igor]

Hello.

SpamAssasin in local.cf set:

use_bayes 1
bayes_auto_learn 1
bayes_auto_expire 1
use_bayes_rules 1
bayes_path /var/db/bayes/bayes
bayes_file_mode 0666
bayes_min_ham_num 10
bayes_min_spam_num 20
bayes_auto_learn_threshold_nonspam -10
bayes_auto_learn_threshold_spam 20
bayes_journal_max_size 102400
bayes_expiry_max_db_size 150000

IP of sender is in trusted_networks.

Header X-Spam-Report message:

X-Spam-Report: Content analysis details: (-98.7 points, 5.0 required,0.5 bayes score)
  ____
  pts  rule name              description
  ---- ---------------------- --------------------------------------------------
  -100 ALL_TRUSTED            Passed through trusted hosts only via SMTP
   1.3 MISSING_SUBJECT        Missing Subject: header
  ____


In spamd.log -
spamd: result: . -98 - ALL_TRUSTED,MISSING_SUBJECT scantime=0.5 ,size=969,user=(unknown),uid=58,required_score=5.0,rhost=localhost,raddr=127.0.0.1,rport=60081,mid=<569856404.20090702121054@...>,autolearn=no


We see that with -98.7 points and option
"bayes_auto_learn_threshold_nonspam -10",
but "autolearn=no". Why?

[Benny Pedersen]

On Thu, July 2, 2009 08:43, Trushin Igor wrote:

> We see that with -98.7 points and option
> "bayes_auto_learn_threshold_nonspam -10",
> but "autolearn=no". Why?

why learn anything from trusted ip ?


--
xpoint

[Karsten Bräckelmann-2]

On Thu, 2009-07-02 at 10:43 +0400, Trushin Igor wrote:
> use_bayes 1
> bayes_min_ham_num 10
> bayes_min_spam_num 20

>   -100 ALL_TRUSTED            Passed through trusted hosts only via SMTP
>    1.3 MISSING_SUBJECT        Missing Subject: header

50_scores.cf:  score MISSING_SUBJECT 2.307 1.285 2.476 1.762

SA 3.2.5 stock scores. So you are using score-set 1, network tests
enabled, but Bayes DISABLED.

Forgot to restart your spamd daemon? Well, or you just did not yet learn
sufficient ham and spam beyond your (incredible low) thresholds, so you
are not using score-set 3 with Bayes.


> We see that with -98.7 points and option
> "bayes_auto_learn_threshold_nonspam -10",
> but "autolearn=no". Why?

20_compensate.cf:  tflags ALL_TRUSTED      nice userconf

tflags userconf rules are ignored for auto-learning [1]. Thus, as far as
the AutoLearnThreshold plugin is concerned, the message scored 1.285,
which is not below your auto-learn ham threshold.

Also note, that auto-learning uses a non-Bayes score-set internally for
training decisions, which is either score-set 0 or 1 -- coincidentally
in this case, what has been used anyway.

  guenther


[1] http://spamassassin.apache.org/full/3.2.x/doc/Mail_SpamAssassin_Plugin_AutoLearnThreshold.html

--
char *t="\10pse\0r\0dtu\0.@ghno\x4e\xc8\x79\xf4\xab\x51\x8a\x10\xf4\xf4\xc4";
main(){ char h,m=h=*t++,*x=t+2*h,c,i,l=*x,s=0; for (i=0;i<l;i++){ i%8? c<<=1:
(c=*++x); c&128 && (s+=h); if (!(h>>=1)||!t[s+h]){ putchar(t[s]);h=m;s=0; }}}