S.D.A. wrote: > I agree. I switched from SA several months ago, and am quite happy with the > speed, accuracy of Spamprobe over Spam Assassin.
I think this thread has shown that many people have a gross misconception
on how SpamAssassin works and how it is fundimentally different than the
alternatives listed. In every case the alternatives listed have been a pure
Bayesian system. SpamAssassin is *NOT* a Bayesian system. It is a framework
in which a Bayesian system is also included.
Do you trust the Bayesian scoring more than the rest of the framework?
Want to increase its accuracy? The *adjust the scoring*! In the default
setup SpamAssassin is quite liberal in what it lets through because no one
test can set a piece of mail to spam. This means it is highly resilient to
false positives. With proper tuning people can achieve excpetionally high
results but you have to be invormed and understand what's going on. That, in
the long run, is a good thing. So you turst the Bayes scoring? Find, 5 lines
for ya:
# Reweighted high end bayesian classifier scores
score BAYES_60 0 0 1.250 1.250
score BAYES_70 0 0 2.000 2.000
score BAYES_80 0 0 4.250 4.250
score BAYES_90 0 0 5.500 5.500
score BAYES_99 0 0 7.750 7.750
I have SA set up to be run through exiscan-acl (so far not something I've
seen others be able to do). At 4.5 it is marked as spam. At 7.5 it is
rejected at SMTP time. Most spam is simply rejected ouright. The joy of it,
though, is that compared to other systems I haven't had to train SpamAssassin.
Since it is a framework it can use the other tests to decide if mail is spam
or not. Any grossly high, or low, scoring messages are automatically trained.
auto_learn_threshold_spam 10
auto_learn_threshold_ham -1.0
Did I do initial training? Yeah, because I was impatient. Have I done
any training since then? Yeah, but it's down to about maybe 10 messages a
week total when and if I feel like it. The rest is autotrained. As my
reading habits change the Bayes db changes with it. As the spam that hits my
machine changes so does the bayes db. All without intervention from me based
on sound rules in that framework.
So, fine, swtich, glad that the alternatives work for ya. But at least be
fair to that which you switched from when discussing it with others.
--
Steve C. Lamb | I'm your priest, I'm your shrink, I'm your
PGP Key: 8B6E99C5 | main connection to the switchboard of souls.
-------------------------------+---------------------------------------------
signature.asc
Description: OpenPGP digital signature
