Autolearn ham?

[Diego Pomatta <in...@abelsonsa.com.ar>]

Hello list!

I was under the impression that autolearn was going to start self 
learning ham at some point, when enough messages had been processed.

Checking the spamd log I see only no-learning or spam-learning. And I 
can't feed anything to sa-learn because we drop spam into oblivion. (The 
only spam I can scavenge are the mails that SA didn't catch, but by that 
time those mails are in the users e-mail client in a host computer, not 
in the server.
I have a spam mailbox that users can forward spam to, but as far as I 
know, one can't feed forwarded mail to sa-learn. I use those mainly to 
learn myself and write rules accordinly, if necessary.


 From spamd log:
----------------------
info: spamd: clean message (0.4/5.0) for simscan:512 in 0.1 seconds, 
5877 bytes.
info: spamd: result: . 0 - AWL,HTML_MESSAGE,J_CHICKENPOX_74,RDNS_NONE 
scantime=0.1,size=5877,user=simscan,uid=512,required_score=5.0,...........,_*autolearn=no*_

...
info: spamd: identified spam (11.2/5.0) for simscan:512 in 0.5 seconds, 
8576 bytes.
info: spamd: result: Y 11 - 
AWL,FUZZY_CREDIT,HTML_IMAGE_RATIO_02,HTML_MESSAGE,MIME_BASE64_TEXT,MPART_ALT_DIFF_COUNT,RDNS_NONE,SUBJECT_NEEDS_ENCODING,SUBJ_ILLEGAL_CHARS 
scantime=0.5,size=8576,user=simscan,uid=512,required_score=5.0,........,*_autolearn=spam
_----------------------*

My local.cf shows:
use_bayes 1
use_bayes_rules 1
bayes_auto_learn 1

So, is there any other option that needs to be set somewhere? Because I 
don't think that bayes is even being used at all.
Maybe an autolearn ham score threshold or something like that?

Thanks
/Diego

Solved: Autolearn ham?

[Diego Pomatta <in...@abelsonsa.com.ar>]

Matthias Leisi escribió:
>> Is there any way to tweak the score used to trigger autolearning for HAM ?
>>
>> Diego
>> Ah, SA version is 3.2.1
>>     
>
> http://spamassassin.apache.org/full/3.2.x/doc/Mail_SpamAssassin_Plugin_AutoLearnThreshold.html
>
> --- cut ---
> The following configuration settings are used to control auto-learning:
>
> bayes_auto_learn_threshold_nonspam n.nn (default: 0.1)
>     The score threshold below which a mail has to score, to be fed into
> SpamAssassin's learning systems automatically as a non-spam message.
> --- cut ---
>
> -- Matthias
>
>
>
>
>   
Thanks, that was what I wanted. I'd searched around but hadn't found 
that page.
I included de plugin in init.pre, added the 
"bayes_auto_learn_threshold_nonspam" setting in local.cf, upped the 
threshold a little bit;
and bayes has already autolearned a couple of hams.

Regards
Diego

Re: Autolearn ham?

[Matthias Leisi <ma...@leisi.net>]

> Is there any way to tweak the score used to trigger autolearning for HAM ?
>
> Diego
> Ah, SA version is 3.2.1

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

--- cut ---
The following configuration settings are used to control auto-learning:

bayes_auto_learn_threshold_nonspam n.nn (default: 0.1)
    The score threshold below which a mail has to score, to be fed into
SpamAssassin's learning systems automatically as a non-spam message.
--- cut ---

-- Matthias

Re: Autolearn ham?

[Diego Pomatta <in...@abelsonsa.com.ar>]

Lars Ippich escribió:
> Diego Pomatta schrieb:
>   
>> Ok, but then why is "auto-learn" learning spam, but not ham?
>>
>> Diego
>>     
>
> Have a look at this, maybe it explains what you are wondering about:
> http://wiki.apache.org/spamassassin/AutolearningNotWorking
>
> Lars

Yup. I'd read that already.
It explains maybe why mail may not be learnt as spam when it says that 
"SpamAssassin <http://wiki.apache.org/spamassassin/SpamAssassin> 
requires at least 3 points from the header and 3 points from the body, 
to auto-learn as spam. If either section contributes fewer points, the 
message will not be auto-learned."

Cool, but in my case it is auto-learning spam fine ;)
Is there any way to tweak the score used to trigger autolearning for HAM ?

Diego
Ah, SA version is 3.2.1

Re: Autolearn ham?

[Lars Ippich <la...@speicherplatz4you.de>]

Diego Pomatta schrieb:
> Ok, but then why is "auto-learn" learning spam, but not ham?
> 
> Diego

Have a look at this, maybe it explains what you are wondering about:
http://wiki.apache.org/spamassassin/AutolearningNotWorking

Lars

Re: Autolearn ham?

[Diego Pomatta <in...@abelsonsa.com.ar>]

Lars Ippich escribió:
> Diego Pomatta schrieb:
>   
>> [...] And I
>> can't feed anything to sa-learn because we drop spam into oblivion. (The
>> only spam I can scavenge are the mails that SA didn't catch, but by that
>> time those mails are in the users e-mail client in a host computer, not
>> in the server.
>> [...]
>> So, is there any other option that needs to be set somewhere? Because I
>> don't think that bayes is even being used at all.
>> Maybe an autolearn ham score threshold or something like that?
>>     
>
> There is a threshold of 200 hams and 200 spams, which have to been fed
> to the bayes filter by sa-learn before it actually starts classifying
> mails as spam or ham.
>
> Lars
>
>
>   
Ok, but then why is "auto-learn" learning spam, but not ham?

Diego

Re: Autolearn ham?

[Lars Ippich <la...@speicherplatz4you.de>]

Diego Pomatta schrieb:
> [...] And I
> can't feed anything to sa-learn because we drop spam into oblivion. (The
> only spam I can scavenge are the mails that SA didn't catch, but by that
> time those mails are in the users e-mail client in a host computer, not
> in the server.
> [...]
> So, is there any other option that needs to be set somewhere? Because I
> don't think that bayes is even being used at all.
> Maybe an autolearn ham score threshold or something like that?

There is a threshold of 200 hams and 200 spams, which have to been fed
to the bayes filter by sa-learn before it actually starts classifying
mails as spam or ham.

Lars