MATLAB and Simulink resources for Arduino, LEGO, and Raspberry Pi

# Nonlinear least square regression

Asked by ameen on 23 May 2013

Hi all i have 17 observation (x and y) the relation between them as follows

y = 0.392 * (1 - (x / J)) ^ i

i want to use nonlinear least square regression to know J and i

## Products

No products are associated with this question.

Answer by Eli Duenisch on 23 May 2013
Edited by Eli Duenisch on 23 May 2013

Do you have the statistics toolbox installed? It supports nonlinear regression - look for NonLinearModel.fit() in the docs.

Tom Lane on 24 May 2013

For modelfun: either write a function or use 'y ~ .392*(1-x/b1)^b2'. The function will be happier if you use names b1/b2 in place of J/I.

For beta0: give a two-element vector with your best guess at b1 and b2. This depends on your data. A good guess gives the function a better shot at solving the problem. If you have real data, you might consider something like b1=2*max(x) to avoid problems with complex numbers.

ameen on 24 May 2013

thank you for your kind reply but i put my x values, then my y values, and put beta0=[0.65,1.1] then i write

mdl=nonlinearmodel.fit(x ,y ,'y=0.392*(1-x/b1)^b2',beta0)

and i received error message ' Undefined variable "nonlinearmodel" or class "nonlinearmodel.fit".'

Tom Lane on 24 May 2013

Try NonLinearModel.fit capitalized just that way. This should work in the most recent releases of MATLAB. Otherwise try nlinfit.