Nlinfit error with func2str

Hello,

I am trying to fit a curve using this function: t is the time vector, x the force measurement, d0 and nu are constants. I want to extract Er, tau1 and tau2.

function [Er,tau1,tau2] = viscoelasticcone(t,x,d0,nu)

          fitfunctionconical=sprintf('Er*2*tan(35*pi/180)/(pi*(1-%d^2))*%d^2*(1+(tau1-tau2)/tau2*exp(-t/tau2))',nu,d0);
          ftconical=fittype( fitfunctionconical, 'independent', 't', 'dependent', 'F');
          [fitresult,R,~,CovB,MSE] = nlinfit(t,x, ftconical,[1e3 0.001 0.001]);
          fit=ftconical(fitresult,t);
          [Ypred,delta] = nlpredci(ftconical,t,fitresult,R,'Covar',CovB,'MSE',MSE,'SimOpt','on');
          confidence{k}=delta;
          lower = Ypred - delta;
          upper = Ypred + delta;
  end

If I use this function however, i get the following error:

Undefined function 'func2str' for input arguments of type 'fittype'.

Error in nlinfit (line 204)
        m = message('stats:nlinfit:ModelFunctionError',func2str(model));

Any help ?

Michael

Answer 1

[Walter Roberson]

You need to call fit() on fit objects returned from fittype(), not nlinfit()

[Michael Lherbette]

Hello Walter,

What about the nlinfit ?

I use fit with option equal to NonLinearLeastSquares but I have a real trouble with the fitting.

          s = fitoptions('Method','NonlinearLeastSquares');
          fitfunctionconical=sprintf('Er*2*tan(35*pi/180)/(pi*(1-%d))*%d^2*(1+(tau1-tau2)/tau2*exp(-t/tau2))',nu,d0);
          ftconical=fittype( fitfunctionconical, 'independent', 't', 'dependent', 'F','coefficients',{'Er','tau1','tau2'},'option',s);
          fitresult = fit(t',x, ftconical);
          coeffvals = coeffvalues(fitresult);

I get wrong values for the coefficients and the curve is not at all fitted.

[Michael Lherbette]

All right,

Here my final code:

          ftVisco= @(b, t) b(1)*2*tan(35*pi/180)/(pi*(1-nu))*d0^2*(1+(b(2)-b(3))/b(3).*exp(-t./b(3)));
          [fitresultnlinfit, R,~,CovB,MSE,Errorinfo] = nlinfit(t',x, ftVisco, [1e3 0.001 0.001]);
          fitresultnlinfit
          [Ypred,delta] = nlpredci(ftVisco,t',fitresultnlinfit,R,'Covar',CovB,'MSE',MSE);
          confidence=delta;
          lower = Ypred - delta;
          upper = Ypred + delta;

It seems to work well.

Thank you Walter for pointing me out that we can't use fittype for nlinfit.

Cheers