Prev: face extraction from an image
Next: for loop using class property as variable
From: Jackson Fung on 14 Mar 2010 11:16
i got a problem to get my RBF to work. I trained the data with positive values but somehow the predicted data got me negative values. wondering what is the cause of this?pls help..
From: John D'Errico on 14 Mar 2010 11:35
"Jackson Fung" <jacksonfhs(a)yahoo.com> wrote in message <hniujl$dnv$1(a)fred.mathworks.com>...
> i got a problem to get my RBF to work. I trained the data with positive values but somehow the predicted data got me negative values. wondering what is the cause of this?pls help..

There is NO requirement that such an interpolant
cannot yield numbers that lie outside the range of
your data, especially if you use it to extrapolate.
This is the nature of extrapolation.

Almost as bad, if there are large enough internal
holes in your data, then an interpolant can still
give you ringing behavior. This is especially true
if the relationship is highly nonlinear.

If your model is an approximation, and not a true
interpolant, then simple lack of fit (or poor data)
can result in negative numbers where none
existed before.

Simply throwing a neural net at some random
data will give you randomly garbage results.

John
From: Greg Heath on 15 Mar 2010 07:28
On Mar 14, 11:16 am, "Jackson Fung" <jackson...(a)yahoo.com> wrote:
> i got a problem to get my RBF to work. I trained the data with positive values but somehow the predicted data got me negative values. wondering what is the cause of this?pls help..

Thanks for the overwhelming amount of detail.

Hope this helps.

Greg
 | 
Pages: 1
Prev: face extraction from an image
Next: for loop using class property as variable