You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@spamassassin.apache.org by Kelly Corbin <kc...@theiqgroup.com> on 2005/04/05 19:04:01 UTC
Spamassassin only autolearning ham, not spam after upgrade to 3.0.2
I have 4 machines configured identically (with the exception of the -m
option due to differences in resources on each machine) with
SpamAssassin and spamass-milter. I recently upgraded to 3.0.2 from 2.64
and everything seems to be working pretty good with the exception of one
machine. After watching the mail log, I noticed that it is not
autolearning any spam, no matter how high it scores. It does autolearn
ham however, and the other 3 machines autolearn spam fine.
I've looked at everything I can think of (configuration files, file
permissions, checked FAQ's, searched list archives, etc.) and can't
figure out why it won't autolearn any spam.
Any ideas?
Thanks!
Kelly
--
--------------------------------------------
-- Kelly Corbin
-- Network Administrator
--
-- http://www.theiqgroup.com
--
-- The IQ Group, Inc.
-- 6740 Antioch Suite 260
-- Merriam, KS 66204
-- (913)722-6700 x105
-- Fax (913)722-7264
--------------------------------------------
Re: Spamassassin only autolearning ham, not spam after upgrade to
3.0.2
Posted by Matt Kettler <mk...@evi-inc.com>.
Kelly Corbin wrote:
>
> What is this 'learned-points'?
That's what score the BAYES_* rules would have given this message based
on existing learning.
This is basically used to prevent SA from automatically learning
anything that noticeably contradicts the existing training.
> Is my database poisoned on the affected
> machine?
Possibly. It's either poisoned, or it's just not trained on a wide
enough variety of spam.
It looks like SA's existing training tells it to regard that message as
BAYES_00. (ie: less than 1% chance of being spam). I'm basing the
BAYES_00 claim on the learned points being -2.599, which matches the
score of the BAYES_00 rule.
Re: Spamassassin only autolearning ham, not spam after upgrade to
3.0.2
Posted by Kelly Corbin <kc...@theiqgroup.com>.
Here's my auto-learn lines from the machine that doesn't work:
debug: auto-learn: currently using scoreset 3, recomputing score based
on scoreset 1.
debug: auto-learn: message score: 23.316, computed score for autolearn:
24.06
debug: auto-learn? ham=0.1, spam=10, body-points=16.82,
head-points=9.84, learned-points=-2.599
debug: auto-learn? no: scored as spam but learner indicated ham (-2.599
< -1)
debug: is spam? score=23.316 required=6
And here's my output from the machine that's learning OK:
debug: auto-learn: currently using scoreset 3, recomputing score based
on scoreset 1.
debug: auto-learn: message score: 25.916, computed score for autolearn:
24.06
debug: auto-learn? ham=0.1, spam=10, body-points=16.82,
head-points=9.84, learned-points=0.001
debug: auto-learn? yes, spam (24.06 > 10)
debug: Learning Spam
What is this 'learned-points'? Is my database poisoned on the affected
machine?
Thanks!
Kelly
Kevin Peuhkurinen wrote:
> Kelly Corbin wrote:
>
>> I have 4 machines configured identically (with the exception of the -m
>> option due to differences in resources on each machine) with
>> SpamAssassin and spamass-milter. I recently upgraded to 3.0.2 from 2.64
>> and everything seems to be working pretty good with the exception of one
>> machine. After watching the mail log, I noticed that it is not
>> autolearning any spam, no matter how high it scores. It does autolearn
>> ham however, and the other 3 machines autolearn spam fine.
>>
>> I've looked at everything I can think of (configuration files, file
>> permissions, checked FAQ's, searched list archives, etc.) and can't
>> figure out why it won't autolearn any spam.
>>
>> Any ideas?
>>
> Take an email with lots of hits and save it as 'spam-email', then run
> 'spamassassin -t -D < spam-email' and see what the debug has to say
> about it. Feel free to post the Bayes-specific parts of the debug here
> if you aren't sure of how to read it.
>
>> Thanks!
>>
>> Kelly
>>
>
--
--------------------------------------------
-- Kelly Corbin
-- Network Administrator
--
-- http://www.theiqgroup.com
--
-- The IQ Group, Inc.
-- 6740 Antioch Suite 260
-- Merriam, KS 66204
-- (913)722-6700 x105
-- Fax (913)722-7264
--------------------------------------------
Re: Spamassassin only autolearning ham, not spam after upgrade to
3.0.2
Posted by Kelly Corbin <kc...@theiqgroup.com>.
That did the trick! I just copied over the databases from one of the
good machines and right away it started doing the autolearn=spam.
Thanks for all your help.
Kelly
Kevin Peuhkurinen wrote:
> Kelly Corbin wrote:
>
>> Here's my auto-learn lines from the machine that doesn't work:
>> debug: auto-learn: currently using scoreset 3, recomputing score based
>> on scoreset 1.
>> debug: auto-learn: message score: 23.316, computed score for
>> autolearn: 24.06
>> debug: auto-learn? ham=0.1, spam=10, body-points=16.82,
>> head-points=9.84, learned-points=-2.599
>> debug: auto-learn? no: scored as spam but learner indicated ham
>> (-2.599 < -1)
>> debug: is spam? score=23.316 required=6
>>
>>
>> And here's my output from the machine that's learning OK:
>> debug: auto-learn: currently using scoreset 3, recomputing score based
>> on scoreset 1.
>> debug: auto-learn: message score: 25.916, computed score for
>> autolearn: 24.06
>> debug: auto-learn? ham=0.1, spam=10, body-points=16.82,
>> head-points=9.84, learned-points=0.001
>> debug: auto-learn? yes, spam (24.06 > 10)
>> debug: Learning Spam
>>
>> What is this 'learned-points'? Is my database poisoned on the
>> affected machine?
>>
> I'm guessing here that the email is hitting BAYES_00 (which has a score
> of -2.599 by default, and which is the "learned points"). SA now has
> some code to ensure that emails that hit low BAYES scores will not be
> autolearned as spam and emails that hit high BAYES scores will not be
> autolearned as ham, no matter what they score otherwise. I'm assuming,
> then, that all or most of your emails are hitting BAYES_00 to BAYES_40
> only. This means that indeed your Bayes database is pooched.
> The easiest solution is likely to just delete the database from this
> machine and copy over the database from one of your other systems,
> provided that they are handling similar types of emails.
>
>
--
--------------------------------------------
-- Kelly Corbin
-- Network Administrator
--
-- http://www.theiqgroup.com
--
-- The IQ Group, Inc.
-- 6740 Antioch Suite 260
-- Merriam, KS 66204
-- (913)722-6700 x105
-- Fax (913)722-7264
--------------------------------------------
Re: Spamassassin only autolearning ham, not spam after upgrade to
3.0.2
Posted by Kevin Peuhkurinen <ke...@meridiancu.ca>.
Kelly Corbin wrote:
> Here's my auto-learn lines from the machine that doesn't work:
> debug: auto-learn: currently using scoreset 3, recomputing score based
> on scoreset 1.
> debug: auto-learn: message score: 23.316, computed score for
> autolearn: 24.06
> debug: auto-learn? ham=0.1, spam=10, body-points=16.82,
> head-points=9.84, learned-points=-2.599
> debug: auto-learn? no: scored as spam but learner indicated ham
> (-2.599 < -1)
> debug: is spam? score=23.316 required=6
>
>
> And here's my output from the machine that's learning OK:
> debug: auto-learn: currently using scoreset 3, recomputing score based
> on scoreset 1.
> debug: auto-learn: message score: 25.916, computed score for
> autolearn: 24.06
> debug: auto-learn? ham=0.1, spam=10, body-points=16.82,
> head-points=9.84, learned-points=0.001
> debug: auto-learn? yes, spam (24.06 > 10)
> debug: Learning Spam
>
> What is this 'learned-points'? Is my database poisoned on the
> affected machine?
>
I'm guessing here that the email is hitting BAYES_00 (which has a score
of -2.599 by default, and which is the "learned points"). SA now has
some code to ensure that emails that hit low BAYES scores will not be
autolearned as spam and emails that hit high BAYES scores will not be
autolearned as ham, no matter what they score otherwise. I'm assuming,
then, that all or most of your emails are hitting BAYES_00 to BAYES_40
only. This means that indeed your Bayes database is pooched.
The easiest solution is likely to just delete the database from this
machine and copy over the database from one of your other systems,
provided that they are handling similar types of emails.
Re: Spamassassin only autolearning ham, not spam after upgrade to
3.0.2
Posted by Kevin Peuhkurinen <ke...@meridiancu.ca>.
Kelly Corbin wrote:
> I have 4 machines configured identically (with the exception of the -m
> option due to differences in resources on each machine) with
> SpamAssassin and spamass-milter. I recently upgraded to 3.0.2 from 2.64
> and everything seems to be working pretty good with the exception of one
> machine. After watching the mail log, I noticed that it is not
> autolearning any spam, no matter how high it scores. It does autolearn
> ham however, and the other 3 machines autolearn spam fine.
>
> I've looked at everything I can think of (configuration files, file
> permissions, checked FAQ's, searched list archives, etc.) and can't
> figure out why it won't autolearn any spam.
>
> Any ideas?
>
Take an email with lots of hits and save it as 'spam-email', then run
'spamassassin -t -D < spam-email' and see what the debug has to say
about it. Feel free to post the Bayes-specific parts of the debug here
if you aren't sure of how to read it.
> Thanks!
>
> Kelly
>