Surface Extrapolation
in [Matlab]
|
From: Panagiotis Mouzakis on 19 May 2010 16:29 Hi, The only I would like to know is how we can make 2d extrapolation (surface extrapolation) as with griddata the only we can do is interpolation.
From: Richard Willey on 20 May 2010 09:46 Hi Panagiotis If you need to perform extrapolation and predict values "outside" your existing data set you should probably start by specifying some kind of parametric model that defines the relationship between your variables. If your model is linear, you can solve it using MATLAB's backslash command. If your model is nonlinear, your going to need to use optimization techniques to solve the nonlinear regression problem. Three different MathWorks toolboxes contain functions that can be used to solve this problem 1. Curve Fitting Toolbox can solve surface fitting problems: model z = f(x,y) 2. Statistics Toolbox has a function called "nlinfit" to solve the same sorts of problems 3. Optimization Toolbox has a lot of great solvers which can certainly be used to solve this type of problem From my perspective: Curve Fitting Toolbox is the easiest way to solve this problem. The toolbox has a nice Graphical User Interface. CFT provides code to automatically suggest good starting conditions for the optimization solver. The toolbox also provides "fit objects" that make it very easy to work with yur fit. The main reason to consider using Statistics Toolbox is range of features supported in the toolbox. If you are at all concerned that your independent variables might be correlated with one another (or alternatively autocorrelated) it would probably be better to use Statistics Toolbox. Optimization Toolbox provides the most flexible solution. If you need to constrain the problem in any way, Optimization Toolbox is the way to go. "Panagiotis Mouzakis" <panos.mouzakis(a)gmail.com> wrote in message news:ht1hn1$hk8$1(a)fred.mathworks.com... > Hi, > The only I would like to know is how we can make 2d extrapolation (surface > extrapolation) as with griddata the only we can do is interpolation. >
|
Pages: 1 Prev: 5 Minute timing on a while loop Next: Plotting raw spectrogram |