spamassassin-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sam <liste-spamassas...@ingescom.com>
Subject Re: sa-learn spam and Bayes_50
Date Tue, 27 Oct 2009 23:54:10 GMT
John Hardin a écrit :
> On Tue, 27 Oct 2009, Sam wrote:
>
>> John Hardin a écrit :
>>>  On Tue, 27 Oct 2009, Sam wrote:
>>>
>>> >  And after learning with sa-learn, it is still saying bayes_50 >  
>>> whereas sa-learn told it has learned it.
>>>
>>>  Okay, basic Bayes troubleshooting questions:
>>>
>>>  (1) Are you running sa-learn as the same user that SA itself is
>>>  running as, so that you're training the Bayes database that SA is
>>>  actually using to score messages?
>>>
>>>  (2) Please run sa-learn --dump magic and send us the results.
>>
>> 1) For all users there is only one database in /var/bayes. I've done
>>    some tests with su Debian-exim and it is same result.
>>
>> 2) lenny:/home/samuel# sa-learn --dump magic
>> 0.000          0          3          0  non-token data: bayes db version
>> 0.000          0     112532          0  non-token data: nspam
>> 0.000          0        844          0  non-token data: nham
>> 0.000          0    1935545          0  non-token data: ntokens
>
> Okay, good. About the only comment I can make based on this is, you 
> might want to learn a bunch of ham. You want the database to kinda 
> reflect your actual raw spam/ham ratio, but yours is a little strongly 
> skewed towards spammy tokens...
>
Thanks to everybody for yours comments.
If I understand well, the few french spam I give to sa-learn are too 
little front of the tons of english spam feed to sa-learn.

It could be interesting (but not existing I think) to have one bayes for 
each langage if I understand that this the problem in my case.

Thanks a lot.
Sam.



Mime
View raw message