Samson -
First, thanks for your response that sent me to another white paper I
had not seen. However, the discussion in Girratano and Riley's book
is not quite the same as the topic of the author of the paper, even
though similar and both do discuss MYCIN as the basis of
conversation. And even though Ernest does not encourage theoretical
discussions on this board I'll answer your comment and then go
quietly back into the woodwork. :-)
Something else to take up precious time - but I couldn't resist, so I
read the paper. Interesting... (Nice way to say I don't necessarily
agree with it.) The author failed to note that a rulebased system is
non-monotonic, not monotonic as were the systems with which he was
working, and the monotonicity was the foundation of his argument
against using confidence intervals. Even then, I couldn't quite
agree that he was considering the far richer implementations of CE
than that of the classic CE. And the authors of MYCIN also struggled
with this problem but they seem to have come up with a fairly good
solution though not perfect.
The other point was his misconception of probabilities, which I would
not have expected from a noted author. In the article he was
factoring in the probability of drawing a white ball from a jar after
having drawn a white ball would be reduced the second time if the two
balls in the jar were one white and one black. Not quite. That's
the same theory of asking if you flip a fair coin five times and all
five times are heads, what are the odds that the next flip will be
heads? The same as the first time: 50/50. Prior probabilities are
not the same as asking what are the odds of flipping a coin and
getting a heads all six times. In that event, a prior probability,
it is .5*.5*.5*.5*.5*.5 - not very good. But stopping at each point
and asking what is the probability of the next one being heads is
still the same; 50/50. And on his scale of -1 to +1 that should have
been zero each time. That gambling example was a really hard reality
for me to swallow in my sophomore statistics course but it has stood
me well over the years.
Anyway, back to the initial thought; this probably is not the place
nor the space to debate probabilities, combinations, permutations,
set theory, pattern matching and confidence intervals so I'll quit on
that final note. I did enjoy the paper though because it disagreed
with me and made me have to think - something that I should do more
of these days but don't due to other pressures. Maybe one day I'll
have a position where I can just shut the door and cogitate on the
unknown things of the universe. :-)
SDG
jco
James C. Owen
Senior Consultant
[EMAIL PROTECTED]
http://www.kbsc.com
"To be or not to be. That is the question. Whether 'tis nobler in
the mind to suffer the slings and arrows of outrageous fortune or by
taking up arms to defeat them." (So saith Shakespeare in Hamlet,
from my own cluttered memory so it might not be quite right.)
On Oct 17, 2006, at 12:28 PM, Samson Tu wrote:
There are inherent problems in building management of uncertainties
into rule-based systems. See
D. E. Heckerman, E. Horvitz. On the Expressiveness of Rule-Based
Systems for Reasoning with Uncertainty. AAAI - 87, Sixth National
Conference on Artificial Intelligence, Seattle, WA, 1987.
http://smi-web.stanford.edu/auslese/smi-web/reports/SMI-87-0194.pdf
James C. Owen wrote:
Actually, certainty and/or confidence factors are discussed in
detail in the Giarratano and Riley's book, complete with an
introduction to probability theory as well as many other concepts
not normally used in business software. This approach was used
with great success in diagnosis when designing MYCIN, an early
rulebased system dealing with medical diagnosis, as well as many
of some other types of prognostication software. It isn't
"easily" implemented in straight Jess nor in any other forward
chaining package, BUT it can be done with a bit of thought up
front concerning tracking tables, maximum probabilities, etc.
After all, you wouldn't want to have a 250% chance of something
happening. But it does deal with the problem of a 70% chance that
something will happen and a 10% chance that something will not
happen. The other 20% is the problem child. Is it doubt or
confidence? This kind of logic is also used extensively in the
financial fraud detection and CRM software. _Great_ stuff but
requires 70% thought and 30% coding. Something to which we, as
programmers, are not usually accustomed. :-)
And you are right - a fuzzy logic extension relieves a lot of the
manual programming. But, on the other hand, it also adds some
restrictions to what you can and can not do. I've often wondered
why vendors such as ILOG, FIC, Pega, MindBox, Haley, etc., etc.
don't include something like that. But, then, they've never
really done anything constructive with full opportunistic backward
chaining, have they? Well, time to get off the soap box and get
back to work. However, I, for one, and probably several others on
the list, would be most interested in what you find out using
Jess. SDG
jco
James C. Owen
Senior Consultant
[EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>
http://www.kbsc.com
"Never give in--never, never, never, never, in nothing great or
small, large or petty, never give in except to convictions of
honour and good sense. Never yield to force; never yield to the
apparently overwhelming might of the enemy.'' The speech was made
29 October 1941 to the boys at Churchill's old public [private]
school, Harrow--not Oxford nor Cambridge.
On Oct 17, 2006, at 5:49 AM, Orchard, Bob wrote:
There are no certainty or confidence factors implemented in Jess.
There is an extension
that allows fuzzy reasoning to handle uncertainty, called
FuzzyJess. If this might
suit your needs see:
http://www.iit.nrc.ca/IR_public/fuzzy/fuzzyJToolkit2.html
It is mentioned in Jess in Action. At some point I was going to
add certainty/confidence
factors but I never did ... many ways to implement and I'm still
not convinced of
its usefulness.
Bob.
Bob Orchard
National Research Council Canada Conseil national de
recherches Canada
Institute for Information Technology Institut de technologie de
l'information
1200 Montreal Road, Building M-50 M50, 1200 chemin Montréal
Ottawa, ON, Canada K1A 0R6 Ottawa (Ontario) Canada K1A
0R6
(613) 993-8557
(613) 952-0215 Fax / télécopieur
[EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>
Government of Canada | Gouvernement du Canada
-----Original Message-----
*From:* [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED] *On Behalf Of *m u
*Sent:* Tuesday, October 17, 2006 12:36 AM
*To:* [email protected] <mailto:[email protected]>
*Subject:* JESS: certainty/confidence factor
Hi all,
I've tried to search in JESS in Action book but couldn't find
anything about certainty or confidence factor. Is this
feature not
yet implemented in JESS?
thanks,
irejai
--
Samson Tu email: [EMAIL PROTECTED]
Senior Research Scientist web: www.stanford.edu/~swt/
Stanford Medical Informatics phone: 1-650-725-3391
Stanford University fax: 1-650-725-7944
--------------------------------------------------------------------
To unsubscribe, send the words 'unsubscribe jess-users
[EMAIL PROTECTED]'
in the BODY of a message to [EMAIL PROTECTED], NOT to the list
(use your own address!) List problems? Notify owner-jess-
[EMAIL PROTECTED]
--------------------------------------------------------------------
--------------------------------------------------------------------
To unsubscribe, send the words 'unsubscribe jess-users [EMAIL PROTECTED]'
in the BODY of a message to [EMAIL PROTECTED], NOT to the list
(use your own address!) List problems? Notify [EMAIL PROTECTED]
--------------------------------------------------------------------
