Re: EuOS, 3D distance alg's, and fuzzy logic
Michael Sabal wrote:
>[3D Distance Algorithms]
>What is there is fully functional rotation/revolution code
will try to check that out, sounds intriguing...
>[Fuzzy logic]
>Thanks for the long and informative post about fuzzy logic
>operators ;).
ya, i'm sorry it was long, but i couldn't think of any other
way to explain what i felt i needed to, in such a way that
everyone could understand, without simply tagging the
documentation, as it did just that... I had started to
make a post and found myself basically repeating the doc...
heh...
>While I expect the ideal is great, reality just
>doesn't hold up. First of all, the laws of probability
this is what's interesting... fuzzy logic and statistical
probability are simultaneously the same and not the same.
you can apply fuzzy logic to statistics and probability.
but you don't have to. you can interpret your resultants
as a statistical probability, but you don't have to.
when you don't do that, it really isn't a percentage
anymore, persay. it's a *strength* of matching, not
the *probability of* a match.
subtle, very very subtle.
some say my last sentence is gobble-de-gook...
if you apply fuzzy logic to any data set, you can
always say your resultant truth is a percentage
likelihood. but you can also say that your resultant
is a bias or a strength that was resolved and
interpreted against an *ideal*.
fuzzy logic is not the best formulae to use to determine
statistical probability, nor can I envision fuzzy logic
operands being used to answer questions like:
"if i flip a coin, how many times will it come up heads?"
this question should be answered using a function plot
of the mean, median and mode upon a bell curve whilst
throwing in discussion of the standard deviation to
obfuscate and confuse anyone listening.
fuzzy logic is more akin to inferential statistics
that are applied to a determinate data set, such as the
one generated above, that attempts to determine the
*strength* (read that validity) of your testing by
analyzing the resultant of several of said function
plots against one another.
this is what allows you to have more than a simple
yes or no answer to the question:
"is my testing of the flipping of a coin accurate?"
and is what gives a computer the ability to answer
that type of a question. if you only have 0 and 1
as an answer to the last question, then due to the
fact that no 2 tests (in theory) will ever plot the
same bell curve, a computer will always answer that
your testing procedures are *not* valid, as any single,
minute, trivial and unimportant difference will always
result in failing to return an exact match of those two
curves that symbolize the results of your tests.
this is why i say that the difference is so very subtle.
the plotting of the difference/alikeness of two or
more bell curves that measure the same event,
begins to resemble a bell curve itself, but that isn't
really a measure of *probability*, now is it?
no, of course not, since the plotting of the sameness
of bell curves that measure the same event will not
be a bell curve, but instead, a *heavily skewed*
bell curve where almost all your plots are in the
upper 5%. (if your testing was indeed a properly
executed testing.)
(and no, i don't know if i could explain that in another way :)
>Also, have you used a search engine like Excite recently,
>which employs fuzzy logic?
well, this is not a fair question IMHO.
why? for starters most search engines use word frequency
count algorithms and dorks on the net generally use
*huge* meta-tags to make sure their site goes to
the top of the list, even if their site really doesn't
have anything to do with your search.
>The sites marked as 86% are almost never what you want,
>while the sites marked 10% are usually right.
>What gives there?
also, a lot of the engines out there now, are *paid*
to have certain sites _always_ appear at the top
of a search return list, no matter the search.
and yes, they pay *big* to do that...
>Nice idea, but fundamentally flawed.
with all statistics, resultants are flawed.
statistics is the measurement of *strength* and the
measurement of *probability* (seperately, depends on
the branch) and as such, getting a 1 or 0 as an answer
is *not* within the realm of viability, which means
that all resultants *must* be flawed (as in: not(1or0) )
in order for those resultants to actually be _not_ flawed :)
enjoyed the talk michael :)
--Hawke'
> Enuf ov mai opinyunz,
ditto
|
Not Categorized, Please Help
|
|
