1. New Library-- Fuzzy Logic (long post)
- Posted by Hawke <mdeland at NWINFO.NET>
Oct 01, 1998
-
Last edited Oct 02, 1998
I'm in the very last stages of my new library.
It works, I'm just finalizing the example programs now.
I've finished the documentation this evening, and
I'll include that here tonight, so that you can
read it, and decide if you want it. An effort to
determine interest, and help prevent listserv clutter.
Without further ado, here is fuzzy.doc, unattached
and not zipped, since many on the list cannot handle
those types of files. Those that want the library
will be able to get it zipped as I'll be asking Rob
to put it on the main EU page.
Hopefully, this library will interest a few.
Take care, and care of. --Hawke'
----------------begin fuzzy.doc
Fuzzy Logic Library
9-28-98
Hawke' <mdeland at nwinfo.net> <trumpyt at nwinfo.net>
Version History
===============
=current, public release=
version 1.4
several functions were further speed optimized by
prevention of recursion for the innermost
dimension of sequences (which generally holds
a long sequence of atoms). It duplicates a
lot of code this way, but it saves many many
clockticks and helps prevent cache thrashing.
version 1.3
seperated:FuzzyNOT,AND & OR, and FindTruth, into
three functions apiece for each of the above
types of calculations. There is a generic
version for each of the above, which provides
extensive (but perhaps, not complete) error
checking, as well as a version for truth
tables that are {atom,atom} and a version
for truth tables that are {seq,seq}. Neither
of the last two versions mentioned perform
any error checking to speak of. They are
provided for the instances where you know
ahead of time that all of your data and all
of your truth tables are properly formatted,
matched, and equal in nestings and length.
This means extra speed for when you have
known good data. Conversly, you can opt for a
small speed penalty for data that is unknown
to you (user entered?) and/or debugging needs.
version 1.2
added: use of {} to bypass automatic truth table
calculation in FuzzyNOT,AND, & OR
version 1.1
1>many speed optimizations tried, tested, implemented.
2>discovered: minimum and maximum using
"return a*(a>=b)+b*(b>a)" & viceversa
bogs down in certain cases, *badly*
version 1.05
added: logging to file upon error, including full
logging of your *entire* dataset(s).
version 1.0
first stable, properly calculating version
(my head hurts now :)
What is this???
===============
This is a library that will (blisteringly fast) allow you
to implement 'fuzzy logic' into your software.
"Great!", you exclaim, "but what does that do for me?"
Simply put, you are no longer bound by 1's and 0's.
You can ask questions like "Is it a little brown?" or
"Is that guy skinny, slim, normal, chubby, fat or obese?"
or even go crazy and compare images for use with visual
recognition systems, robotic vision, and security, or
compare waveforms for speech recognition perhaps.
An example, easily implemented, would be:
suppose you had a database of mug shots, and someone
came in with a suspect photo. Using this library, you
could sift through (rather quickly indeed) your entire
database and determine which pictures in the database
are potential matches. You cannot do this with simple
"if pic_suspect=pic_database[current] then Beep()"
type comparisons. Why? Because images are often slightly
different in lighting and such. If one single pixel
has a slightly different color/greyscale value, you
won't be alerted to that possible match.
Functions/Procedures/Usage
==========================
You have FindTruth, FuzzyNOT, FuzzyAND, and FuzzyOR at your
disposal. FuzzyNOT,AND, & OR all call FindTruth automatically
so that you do not need to worry with storing intermediate
results.
This can be overridden by using {} for the truth table values.
Why would you want to do that? For combinational logic
values like a NAND. First you would find any NOT's upon
your data sets and truth tables. You would then pipe
the results through FuzzyAND using {} as a truth table
value, that way you don't take the 0..1 ranged data
and attempt to truth table it. Fuzzy.ex has more examples
of this, but here is a quick one:
Old={18,65} --under 18=not old, over 65=old
--0..1 for 18..65
Tall={4,7} --under 4ft=not tall, over 7ft=tall
--0..1 for 4ft..7ft
--Who is Tall and Young?
result=FuzzyNOT(AgeData,Old}
result=FuzzyAND(HeightData,Tall,result,{})
Now, result has the 0..1 ranged values that answer the question.
Let's look at each function more specifically now.
FindTruth()
