Re: SA going downhill

From: Justin Guerin (jguerin_at_cso.atmel.com)
Date: 06/18/04

  • Next message: Richard Weil: "sarge, gnome icons gone" 
    Date: Fri, 18 Jun 2004 11:32:31 -0600
To: debian-user <debian-user@lists.debian.org>

    On Wednesday 16 June 2004 04:45, Antony wrote:
    > Hi all,
    >[ big snip]
    >
    > Is there anything that I can do about this? I know about training it,
    > but doesn't that need thousands of messages to work?
    >
    > A

    Yes, training spamassassin requires thousands of messages to work, both ham
    messages and spam messages. When I first installed spamassassin, it caught
    nothing. I didn't realize that thousands meant thousands. Once I got a
    thousand spam messages to train it on, it started working better and
    better. My hit rate now is probably 75% or greater, with no false
    positives (I've only ever had one false positive). I know 75% isn't
    awesome, but I'm being conservative, because I don't really track it that
    closely, and I don't want to mislead you into thinking it'll be perfect.
    But it's enough that I get most of my spams properly identified.

    I don't have spamassassin set to autolearn, and I've noticed that right
    after I train it with the newest messages, it works great. Then, new style
    spam comes out, and it does poorly for a bit. I then train it on the
    newest spams, and it works great again. That's what autolearn is for, I
    guess, but I'd rather not have to deal with false positives. I could train
    it once a day via a cron job, but I haven't done that yet.

    As always, your mileage may vary, but hopefully not that much! :-)

    Justin Guerin 

    -- 
To UNSUBSCRIBE, email to debian-user-REQUEST@lists.debian.org 
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
  • Next message: Richard Weil: "sarge, gnome icons gone"

    Relevant Pages

    • Re: SA going downhill
      ... accuracy of Spamprobe over Spam Assassin. ... > Do you trust the Bayesian scoring more than the rest of the framework? ... > setup SpamAssassin is quite liberal in what it lets through because no one ... false positives to a minimum. ...
      (Debian-User)
    • Accuracy in SpamAssassin just took a huge leap up.
      ... SpamAssassin, I was having a large number of false positives (spam not ... SpamAssassin, or if I just hadn't trained it with enough e-mails. ...
      (alt.os.linux.suse)
    • Re: Really headache on antispam!
      ... >>> I have seen false positives with bl.spamcop.net, thus, I find that ... You have to have the cpu cycles to run SpamAssassin, ... >> I agree with the assessment of spamcop. ... SPAMCOP is pretty good at blocking spam. ...
      (comp.os.linux.security)
    • Re: [opensuse] Spamassassin - Working - How do I make it smarter??
      ... all of the spam still seems to be getting by. ... How do I train it ... or set "autolearn" to tell it what is spam? ... SpamAssassin version 3.1.7 ...
      (SuSE)
    • Re: [opensuse] Spamassassin - Working - How do I make it smarter??
      ... Spamassassin - Working - How do I make it smarter?? ... all of the spam still seems to be getting by. ... How do I train it ... sa-learn - train SpamAssassin's Bayesian classifier ...
      (SuSE)