How to fix the coefficients values using the fitnlm() function

Hi all,

I am trying to use non linear regression fit using the fitnlm function. Is there a way to fix a specific

coefficient to a known value? in this specific case I would like to fix b(2) to 50.

Thank you in advance for your help.

I report here my code

x = data (:,1);                      
y = data (:,2);                     
plot (x, y, 'o')
hold on

%----- Model function for 1:1 binding -------

b0 = [0.06; 50; 0.5];

modelfun = @(b,x)b(1)*(b(2)+x+b(3)-sqrt((b(2)+x+b(3)).^2-4*b(2)*x))./(2*b(2)); 

%options for fitnlm
options = statset('Display','iter','TolFun',1e-10)

modelFit=fitnlm(x,y,modelfun, b0,'Options',options)

results=table2array(modelFit.Coefficients)

plot(x, modelfun(results(:,1), x))
hold off

Answer 1

[Star Strider]

There are likely several approaches you could trake to this.

This one seems to work:

b0 = [0.06; 50; 0.5];
modelfun = @(b,x)b(1)*(b(2)+x+b(3)-sqrt((b(2)+x+b(3)).^2-4*b(2)*x))./(2*b(2)); 
modelfun = @(b,x) modelfun([b(1),50,b(3)],x)

%options for fitnlm
options = statset('Display','iter','TolFun',1e-10)
modelFit=fitnlm(x,y,modelfun, b0,'Options',options)
results=table2array(modelFit.Coefficients)
plot(x, modelfun(results(:,1), x))
hold off

It avoids your having to re-write your ‘modelfun’ function to ‘freeze’ ‘b(2)’ at 50.

[favret]

thank you very !much it was really what I was looking for!

[Star Strider]

As always, my pleasure!

[Ankit Labh]

Is it possible to give an upper and lower limit of parameters in fitnml?

Thanks

[the cyclist]

This is quite old, and I am only just seeing it, so sorry if this information is not useful anymore. But the answer is yes, you can get confidence intervals for the fitted parameters. Take a look at this documentation section, regarding the coefficients in the model object.

[Sebastien Abadi]

If I understand Ankit's question correctly, they are asking about whether you can constrain parameters to only be within a certain range before performing the fit. I think doing so is possible in other toolboxes like the curve fitting toolbox, but I haven't seen a clear way of doing so with fitnlm.