The way I was looking at it is, there are cases where if certain tests fail
together you would want to give it more or less (probably never for less,
but strange how people find uses for odd things) weight than would normally
be assigned in the weighting.  Kind of like in routing or network filters
where you start with general rules then make some specific ones as
exceptions.

For example under normal circumstances Spamcop might be 5 and Sniffer 5,
combined with other tests, they might not be a real strong indicator, but if
both fail you might want a weight of 15 instead of 10, so you'd have (from
Madscientist's example):

SNS  BOOL  (SPAMCOP * SNIFFER) 5

Of course, Madscientist's thoughts make sense too, and if I let them sink in
more, might be related to what I'm saying (need a little more time to
process than I have right now).

I like the idea.

Chuck Frolick
ArgoNet, Inc.

-----Original Message-----
From: [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED]]On Behalf Of Madscientist
Sent: Friday, April 26, 2002 3:39 PM
To: [EMAIL PROTECTED]
Subject: RE: [Declude.JunkMail] What do you think of this?


| >S_N_R  BOOL  (SPAMCOP * REVDNS) 90000000
| >
| >...
| >
| >What about that?
|
| It would certainly work.  The question in my mind, though, is
| how useful
| would this be?  Is this something that would commonly be
| used?  I think
| that with the weighting system, the value of Boolean test
| logic is less
| than it would have been before.
|                                  -Scott

Perhaps... However, many folks might tend to think in logical (binary)
rather than numerical (weighting) terms. I suspect that in many cases,
users think to themselves first in terms of "If this test and that test
are true then it's spam... " and then massage their thinking to use the
weighting system.

Declude is so comprehensive that there are a great many tests available.
Combining them is a challenge.

Mathematically speaking, the addition of Boolean algebra to the system
would provide the last necessary mechanism to practically allow for
discontinuous functions when determining the solution set of the
IsThisSpamFn().

Ok, that was too much I guess...

Think of it this way... When there were just a few test, a simple
weighting system could be tweaked easily to simulate the thresholds that
people use when evaluating how individual tests combine to give a high
spam confidence.

So, somebody can have a few tests that are "almost" good enough by
themselves, weight them so that if they all fired it passes the
threshold and you're sure you've got spam.

Just as easy to add to the mix one or two that if they happen you're
sure with that test alone...

If you look at that scenario, it's really a weighted binary equation,
not a linear one. (See structured criticality below)

Now, add all of the tests that Declude now has available and it's just
not practical to tweak those groups into something that will work -
because they all interfere with each other in the one final result - the
sum.

To handle the extra complexity you need a way to keep rules from
interfering with that sum until they pass a particular threshold of
their own - and a good clean way of doing that is to combine Boolean and
linear arithmetic so that the Boolean part can do the "gating" that's
required.

In complex systems (natural ones too - like sand piles & thunder storms)
this is analogous to "structured criticality". Structured criticality
cannot be reasonably expressed with purely linear functions - nonlinear
functions are required.

That's the scientific argument anyhow.

My $1.32

:-)
_M


---
[This E-mail was scanned for viruses by Declude Virus
(http://www.declude.com)]

---

This E-mail came from the Declude.JunkMail mailing list.  To
unsubscribe, just send an E-mail to [EMAIL PROTECTED], and
type "unsubscribe Declude.JunkMail".  You can E-mail
[EMAIL PROTECTED] for assistance.  You can visit our web
site at http://www.declude.com .

---
[This E-mail was scanned for viruses by Declude Virus (http://www.declude.com)]

---

This E-mail came from the Declude.JunkMail mailing list.  To
unsubscribe, just send an E-mail to [EMAIL PROTECTED], and
type "unsubscribe Declude.JunkMail".  You can E-mail
[EMAIL PROTECTED] for assistance.  You can visit our web
site at http://www.declude.com .

Reply via email to