On Sat, 20 Sep 2003 05:20:36 +1000, Poe Lim <[log in to unmask]> wrote:

>> >Wots a "Bayesian filter"?

When I asked, I thought it was joke you made up and wondered HOW you knew
about
Bayesian Statistics, a branch of statistical inference few know about, and
my
doctoral dissertation advisor, the late L J Savage was the undisputed
Father
of Bayesian Statistics in the USA, just as Sir Ronald Fisher (UK) and
Jersey
Neyman were considered the Farthers of Classical Statistics, the kind most
people know about, the frequentist approach.

Now it's my turn to point to some google web links:
http://www-gap.dcs.st-and.ac.uk/~history/Mathematicians/Savage.html

about the man himself (or use keyword "Leonard Jimmie Savage").  His book,
"The Foundations of Statistics" made his Father of Bayesian Statistics
fame.

http://www.ktl.fi/setti/projects/jkhthesis/thesisb5/node64.html
read the 2nd paragraph ONLY -- which I suspect is the only part
understandableby a layman.  :-)


Bayesian statistics incorporates PERSONAL (subjective) probabilities of the
USER into the inference part given the DATA, as opposed to Classical
statistics where data alone counts.  Thus, given the SAME data, all
classical statisticians come to the same conclusion, whereas every
Bayesian statistician would come to a different conclusion according to
their PERSONAL opinion prior to observing the data.



>It works by learning the difference between normal email and spam, but the
>difference you will find in the frequency and choice of words for each
>category. If you do a search on Google, you'll find quite a few links.

Thanks to your pointer, I didn't realize that there is such an animal as
SPAM Conference, and there is such a thing as a "Bayesian filter" though
it is a COMPLETE MISNOMER!!!

The Bayesian filter is a FREQUESTIST filter.  The user has NO INPUT into
the probabilities!   It not only is frequentistic and NON-Bayesian, it is
a very dumb (unintelligent) approach to filtering.  :-))

For example, it says the word FREE in the subject has a correct probability
of O.9999 of the SPAMS.

While it does make SOME sense, it's the other 0.0001 that I worry about
filtering the WRONG stuff!  What if *I* want to tell some friends about
some freebies on the internet that's worth looking up??


That reminded me that MY SPAM filter (which I paid $29, but trashed, in
favor
of a freebie which does a much better job) was a "Bayesian filter" in the
sense discussed in the SPAM COnferences.

I trashed that stupid filter when I noticed it filtered my OWN EMAIL (a
copy
of it) to Strike, suich had the word "Bastard" in the subject!  It was just
one Old Bastard greeting another.  :-)   When I deleted the word Bastard
in the Subject line, then the email would go through.


Not only a VERY DUMB filter, it gives "Bayesian" a bad name!  :-))

As most gadgets go, it's the filter between the ears that is ultimately
the ONLY trustworthy one.   ;-)


ElPezNeuvo.