Help with nonlinear robost fitting on nlinfit function. fminsearch works fine but not nlinfit.

Question

Mamed 2012-1-15

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/26169-help-with-nonlinear-robost-fitting-on-nlinfit-function-fminsearch-works-fine-but-not-nlinfit

Hi

Im trying to fit some data i have to a function. I have used fminsearch and it has worked fine. Except I wanna introduce robost fitting as there are some point that make the whole fitting wrong.

I searched and found that the stat tool already has a function nlinfit that does everything for you so i thought i would use it.

However when i try to use the function in the program it says that i get NaN or Infinity which i dont get with fminsearch and with a separate m-file the data doesnt go into the other function and it says that the values are unknown.

[ndata] = given below
X = ndata(:,2:8)
dT = ndata(:,1);
par0=[2 0.2 1.5 -0.5 -0.5 -0.5]';

this doesn't send X into funtest.

par = nlinfit(X,dT,funtest,par0);

this one return NaN

%funnfit = @(par,ndata)(par(1).*ndata(:,1).^(0.5*par(2)).*ndata(:,2).^(0.5*par(3)).*ndata(:,3).^par(4).*ndata(:,4).^par(5).*ndata(:,5).^par(6));
%par = nlinfit(ndata(:,2:8),ndata(:,1),funnfit,par0);

funtest

function sse=funtest(params,A)
Pcu  = A(:,1); Ptot = A(:,2); n    = A(:,3); T    = A(:,4); nn   = A(:,5); Tr = A(:,6); L= A(:,7);
k = params(1); x = params(2); y = params(3); z = params(4); a = params(5); b = params(6);
sse=k.*Pcu.^(0.5*x).*Ptot.^(0.5*y).*n.^z.*Tr.^a.*L.^b;
ndata
   1.0e+003 *
0.1085 1.3946 1.8685 0.7500 0.1000 2.6000 0.1800 0.2600 
0.1007 1.3643 2.0014 1.0000 0.1000 2.6000 0.1800 0.2600 
0.0943 1.3446 2.3106 1.5000 0.1000 2.6000 0.1800 0.2600
0.0877 1.3276 2.9261 2.4700 0.1000 2.6000 0.1800 0.2600
0.1080 1.8367 3.6807 3.0000 0.1000 2.6000 0.1800 0.2600 
0.0680 0.7929 1.1946 0.7500 0.0750 2.6000 0.1350 0.2600       
0.0657 0.7889 1.3187 1.0000 0.0750 2.6000 0.1350 0.2600           
0.0624 0.7848 1.6006 1.5000 0.0750 2.6000 0.1350 0.2600
0.0607 0.7925 2.1888 2.4700 0.0750 2.6000 0.1350 0.2600

any help?

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

bym 2012-1-15

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/26169-help-with-nonlinear-robost-fitting-on-nlinfit-function-fminsearch-works-fine-but-not-nlinfit#answer_34189

在 MATLAB Online 中打开

you need to tell nlinfit that funtest is a function by adding a @ in front of it. e.g.

    ndata = 1.0e+003 * ...
    [0.1085 1.3946 1.8685 0.7500 0.1000 2.6000 0.1800 0.2600;
    0.1007 1.3643 2.0014 1.0000 0.1000 2.6000 0.1800 0.2600;
    0.0943 1.3446 2.3106 1.5000 0.1000 2.6000 0.1800 0.2600;
    0.0877 1.3276 2.9261 2.4700 0.1000 2.6000 0.1800 0.2600;
    0.1080 1.8367 3.6807 3.0000 0.1000 2.6000 0.1800 0.2600;
    0.0680 0.7929 1.1946 0.7500 0.0750 2.6000 0.1350 0.2600;       
    0.0657 0.7889 1.3187 1.0000 0.0750 2.6000 0.1350 0.2600;          
    0.0624 0.7848 1.6006 1.5000 0.0750 2.6000 0.1350 0.2600;
    0.0607 0.7925 2.1888 2.4700 0.0750 2.6000 0.1350 0.2600];
X = ndata(:,2:8);
dT = ndata(:,1);
par0=[2 0.2 1.5 -0.5 -0.5 -0.5]';
par = nlinfit(X,dT,@funtest,par0);
par =
      5.5529
      1.0964
      0.7167
     -0.2858
     -0.0487
     -0.2793

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

bym 2012-1-15

generated this warning

Warning: The Jacobian at the solution is ill-conditioned, and some

model parameters may not be estimated well (they are not identifiable).

Use caution in making predictions.

> In nlinfit at 223

请先登录，再进行评论。

Answer 2

Mamed 2012-1-16

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/26169-help-with-nonlinear-robost-fitting-on-nlinfit-function-fminsearch-works-fine-but-not-nlinfit#answer_34200

Ok thanks, that solves that but now i get the same error as the other one, that is that the modelfun returns NaN values. ndata is longer than what i typed up there. But fminsearch doesnt return an error. So i dont understand what the problem is.

Edit: I changed the starting values and it works now, however when plotting the r it says that the max of the residual is 4 but when i calculate the residuals by recalculating the function with the new parameters it gives a residual of 30.

How do i make sure that the value are the optimal ones when it all depends so much on the starting values and i just guess them???