Re: [qmailtoaster] Spamassassin/bayes not working well?

[Jake Vickers]

Steve Huff wrote:



also, for Jake: have you seen this spam-training script?

http://www.ruwenzori.net/code/teach-sa/

it's a bit more complex than the sa-learn script you have on your 
site; however, it not only processes spam/ham into SA's Bayesian DB 
but also reports them back to Razor2/Pyzor/DCC, if they are configured 
on the system.  i haven't had time to test it out yet, but it looks 
worthwhile; might it perhaps be a candidate for inclusion in a future 
Toaster release?
No, I hadn't seen that one. I'll have to look at it more in depth. The 
one on my site was one that was sent to me. Another user sent one to me 
that would only process spam messages after they were a certain date old 
and would automatically look for "spam-folder" directories (as long as 
they were named the same). That way the users had a couple days to look 
in their spam folder before the messages went away. I'll be putting this 
one up soon.

Thanks for the link!

-
QmailToaster hosted by: VR Hosted 
-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]

Re: [qmailtoaster] Spamassassin/bayes not working well?

[Steve Huff]

On Aug 17, 2006, at 9:44 AM, Steve Huff wrote:



On Aug 14, 2006, at 6:03 PM, Jared Markell wrote:

I've  been training spamassassin's Bayesian for awhile now using a  
crontab
job. It "sa-learn"s usually 200 spams a day and usually some Ham  
we can let

it chew on.

But no matter how hard we try, spamassassin.. just.. well, sucks.  
Some
things are being marked as spam, but a LOT is not. Here are sample  
headers
from one of the emails that was being marked as spam, and it tells  
me that
Bayes is actually HELPING spam get through.. any ideas on how to  
fix this?

Is my Bayes library corrupt? How can I fix it?


another thought: configure one or more of the distributed spam  
checksum services (Razor2, Pyzor, DCC).  i recently set up Razor2  
according to the instructions here () and it seems to be working  
fine, imposing only a minimal load on my creaky server.


oops, these instructions:

http://wiki.apache.org/spamassassin/RazorSiteWide


also, for Jake: have you seen this spam-training script?

http://www.ruwenzori.net/code/teach-sa/

it's a bit more complex than the sa-learn script you have on your  
site; however, it not only processes spam/ham into SA's Bayesian DB  
but also reports them back to Razor2/Pyzor/DCC, if they are  
configured on the system.  i haven't had time to test it out yet,  
but it looks worthwhile; might it perhaps be a candidate for  
inclusion in a future Toaster release?


on second thought, this script would need some modification to fit in  
with Qmail-Toaster.  still worth a look, though.


-steve

--
If this were played upon a stage now, I could condemn it as an  
improbable fiction. - Fabian, Twelfth Night, III,v





-
QmailToaster hosted by: VR Hosted 
-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]

Re: [qmailtoaster] Spamassassin/bayes not working well?

[Steve Huff]

On Aug 14, 2006, at 6:03 PM, Jared Markell wrote:

I've  been training spamassassin's Bayesian for awhile now using a  
crontab
job. It "sa-learn"s usually 200 spams a day and usually some Ham we  
can let

it chew on.

But no matter how hard we try, spamassassin.. just.. well, sucks. Some
things are being marked as spam, but a LOT is not. Here are sample  
headers
from one of the emails that was being marked as spam, and it tells  
me that
Bayes is actually HELPING spam get through.. any ideas on how to  
fix this?

Is my Bayes library corrupt? How can I fix it?


another thought: configure one or more of the distributed spam  
checksum services (Razor2, Pyzor, DCC).  i recently set up Razor2  
according to the instructions here () and it seems to be working  
fine, imposing only a minimal load on my creaky server.


also, for Jake: have you seen this spam-training script?

http://www.ruwenzori.net/code/teach-sa/

it's a bit more complex than the sa-learn script you have on your  
site; however, it not only processes spam/ham into SA's Bayesian DB  
but also reports them back to Razor2/Pyzor/DCC, if they are  
configured on the system.  i haven't had time to test it out yet, but  
it looks worthwhile; might it perhaps be a candidate for inclusion in  
a future Toaster release?


-steve


--
If this were played upon a stage now, I could condemn it as an  
improbable fiction. - Fabian, Twelfth Night, III,v





-
QmailToaster hosted by: VR Hosted 
-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]

Re: [qmailtoaster] Spamassassin/bayes not working well?

[fuzzy]

Jared Markell wrote:
> Alright gents..
>
> I've  been training spamassassin's Bayesian for awhile now using a crontab
> job. It "sa-learn"s usually 200 spams a day and usually some Ham we can
> let
> it chew on.
>
> But no matter how hard we try, spamassassin.. just.. well, sucks.

agreed, bayesian alone is imperfect.  are you running  a 'stock' toaster
install?  no modifications?  how long has it been up and running?  what is
your delete threshold, 12? tried reducing the mark score from 5 to maybe
4.5 or 4, and see how that affects mail traffic?  i still mark at 5 and
delete at 6...no ham casualities.  chkuser line at 10 recipients and 3
wrong recipients in the tcp.smtp file.

my first suggestion would be to google sare ninjas rules-du-jour.  do some
reading there.  they offer more rulesets to aid sa in marking spam.  the
rules can be cron'ed to update nightly.  and you can choose from different
rule subsets, and get as aggressive as you dare.  second suggestion would
be to jump over to the wiki and enable surbl.  third suggestion look at
your blacklist file. again, wiki or search the mailing list for blacklists
or visit jake's site to add more. fourth suggestion, get a book on sa or
search the web, and become a ninja yourself, writing/modifying rulesets,
changing scores as you see fit, etc.  fifth suggestion would be to enable
sa buttons plugin in squirrelmail and get the users to work training sa
with squirrellmail(good luck with that).  sixth suggestion, go to the
source...spamassassin mailing lists or forums for hints/tricks.

sa tweaking ultimately is your (our,the admin's) responsibility and it
seems to be one that cannot be automated to any degree of satisfaction,
the target is very "fluid". i have done four of my six suggestions, and
spam is *greatly* reduced compared to pre-toaster levels, but not
eliminated.  blacklists seem to be the most effecive use of my time, surbl
and rules-du-jour a distant second, judging from my spamd log and the
rejections i see.

realistically, i do not expect to be spam-free.  my limited experience
tells me there is a reward for time invested, and it is linear. i don't
have that many users, and i just don't have that much time.

"fight the good fight every moment..."

fuzz


-
 QmailToaster hosted by: VR Hosted 
-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]