I need some help with fitting a curve that is non-linear. This is for an embedded application so what I want to do is precompute all the values (they're discrete) and simply do a look up do get the value. I hate a set of data points that generally give the shape of the curve (think time-shifted exponential decay) but obviously don't encompass all possibilities or I wouldn't need to compute anything.

For example, I have 1024 possible values. My data set might look something like this:

1024 -> 8
950 -> 10
900 -> 13
850 -> 18
ect....


I want to be able to this: valueof999 = precomputed_vals[999];

Any ideas?

3/19/2008 11:59:33 AM

http://en.wikipedia.org/wiki/Interpolation

3/19/2008 12:38:32 PM

Go look up nonlinear regression.

3/19/2008 1:02:17 PM

Right, I guess I should have mentioned that I know about non-linear regression and interpolation. I should have asked if anyone knew of any software or pre-existing packages that do this. I don't want to sitdown and write something myself at this point, but I will if I have to.

3/19/2008 1:05:52 PM

if you only have one set of data points, why not just use the spline function in matlab

3/19/2008 1:14:36 PM

^Great, thats exactly what I was trying to find.

3/19/2008 1:31:53 PM