[om-list] Degree of Belief vs. Fuzzy Set Membership
Mark Butler
butlerm at middle.net
Fri Aug 29 12:06:50 EDT 2003
Hello everybody,
I was talking with Thomas about epistemology, anf thought I would repeat
an idea here.
My preferred way of tracking degree of belief is in log odds form, using
the following formula I found in an expert systems text:
log odds = log10( p / (1 - p) )
where p is the probability between 0 and 1 and p / (1 - p) is the
odds "o".
The log odds comes out as a number between minus infinity and plus
infinity.
The following approximations apply:
p odds log odds
------ ------ ------
zero zero -infinity
0.001 1:1000 -3
0.01 1:100 -2
0.1 1:10 -1
0.5 1:1 0
0.9 10:1 1
0.99 100:1 2
0.999 1000:1 3
...
one infinite +infinity
I like this formula because it is nice and symmetrical, "0" is the value
for no evidence one way or the other, is easy to understand and explain,
and emphasizes the level of evidence required to move from highly
probable to absolutely certain and from highly unlikely to absolutely false.
Now of course, degree of belief is a different concept than degree of
set membership. For example, there is a big difference between saying
"I am certain the weather is mild outside" and "I don't know what the
weather is like outside". In the first case the degree of belief is
infinite and the degree of membership in the "weather is warm" set is
~1/2, whereas is in the second case the degree of belief is zero, and
the degree of membership is unknown.
Degree of belief is only a necessary expedient where you have incomplete
information, whereas fuzzy set membership is a valid concept even when
you know everything.
Comments?
- Mark Butler
More information about the om-list
mailing list