Using fittype for three dimensions

Hello!

I am attempting to use the fittype function to fit a softmax surface to a sigmoid surface using only one coefficient, beta. I couldn't understand the literature on how to do this, so I attempted to do it in a for loop where I would only fit the two dimensional curve for one dependent axis, y, while iterating over the dependent surface, theta. The output would then be an array of the coefficients I need. My efforts resulted in this code:

 x = linspace(0,1);
 z = zeros(1,10);

 for n = 1:10
    theta = n/10; 
    y = x.^2./(x.^2 + theta.^2);
    myfittype = fittype('1/(1+exp(beta*(theta-x)))', 'dependent',{'y'},'independent',{'x'}, 'coefficients',{'beta'});
    myfit = fit(x',y',myfittype);
    z(n) = myfit.beta;
 end

The problem is that fittype doesn't recognize theta as an input variable, so its undefined right now. Does anyone know how to use fittype to directly fit surfaces, or how I can fix this particular for loop?

Answer 1

[Prajit T R]

Hi Colin

The variable 'theta' is not being recognized by the fittype function- hence the error. Try this modified code instead:

x = linspace(0,1);
z = zeros(1,10);
for n = 1:10
  theta = n/10; 
  y = x.^2./(x.^2 + theta.^2);
  myfittype = fittype('1/(1+exp(beta*(theta-x)))', 'dependent',{'y'},'independent',{'x'}, 'coefficients',{'beta','theta'});
  myfit = fit(x',y',myfittype);
  z(n) = myfit.beta;
end

The only change is that 'theta' has been defined as a co-efficient.

Cheers