lsqcurvefit error: Function value and YDATA sizes are not equal.

Hello, I'm trying to get the following code to work:

[ud,pd] = assignp3data

x0=[1; 1] ;
%
p2fit = lsqcurvefit(@p32,x0,ud,pd) ;
K2 = p2fit(1); C2 = p2fit(2) ;
%
[xmin,Jmin]=fminsearch(@p32,x0) ;
%
p2_predict = ((K2./ud(n)).*(ud(n)/C2).^K2.*exp(-(ud(n)/C2).^K2)) ;
%

with the function

function J = p32(K2,C2)
[ud,pd] = assignp3data(1807387)
N = numel(ud,pd) ;
for n = 1:N
    J = sum((pd(n) - (K2./ud(n)).*(ud(n)/C2).^K2.*exp(-(ud(n)/C2).^K2)).^2) ;
end
end

however I keep getting the error in the title.

If it helps, the question I am trying to anwer is given here:

Thanks in advance for any help that can be provided

Answer 1

[Star Strider]

One possibility is that the output of ‘p32’ is a single value, that being the last value of ‘J’ since it was not subscripted in the loop:

J = sum((pd(n) - (K2./ud(n)).*(ud(n)/C2).^K2.*exp(-(ud(n)/C2).^K2)).^2) ;

Another possibility could be that the y-data are a column vector and the output of ‘p32’ is a row vector, because this:

J(n) = sum((pd(n) - (K2./ud(n)).*(ud(n)/C2).^K2.*exp(-(ud(n)/C2).^K2)).^2) ;

will produce a row vector. To get it to produce a column vector, add a second ‘column’ subscript:

J(n,:) = sum((pd(n) - (K2./ud(n)).*(ud(n)/C2).^K2.*exp(-(ud(n)/C2).^K2)).^2) ;

[cxctvz]

Thanks for the quick reply. Do you know how I could make the loop output all the values (31 of them) that I need?

I tried to add in the

J(n)

and

J(n,:)

as well, however the second gives the same error and the first gives some other different errors.

[Star Strider]

I cannot improve on what I already wrote without your data.

[cxctvz]

If it helps you to see where I have gone wrong, I have attached my data set here

ud =

    0.5000
    1.0000
    1.5000
    2.1000
    2.6000
    3.1000
    3.6000
    4.1000
    4.6000
    5.1000
    5.7000
    6.2000
    6.7000
    7.2000
    7.7000
    7.8000
    8.2000
    8.7000
    9.3000
    9.8000
   10.3000
   10.8000
   11.3000
   11.8000
   12.3000
   12.9000
   13.4000
   13.9000
   14.4000
   15.4000
   17.0000


pd =

   -0.0107
    0.0177
    0.1369
    0.1075
    0.1973
    0.1746
    0.1812
    0.1320
    0.2040
    0.1093
    0.1226
    0.0931
    0.1089
    0.0695
    0.0538
    0.0803
    0.0955
    0.0793
    0.0052
    0.0332
    0.0098
    0.0114
    0.0064
    0.0028
    0.0038
    0.0015
    0.0014
    0.0010
    0.0006
    0.0001
    0.0001

thanks

[Star Strider]

You wrote your objective function incorrectly. The parameters must be one vector and the first argument, and the independent variable must be the second argument. (Note that you can write it ass an anonymous function, as I did here.)

Try this:

% % % MAPPING: b(1) = K,  b(2) = C
p32 = @(b,u) (b(1)./u).*(u/b(2)).^b(1).*exp(-(u/b(2)).^b(1)) ;

That worked with lsqcurvefit when I ran it:

p2fit = lsqcurvefit(p32,x0,ud,pd) 

I realise this is homework, and that normally we do not provide anything closely resembling complete code for homework. Please explore the relevant lsqcurvefit documentation on the corrrect syntax for objective functions.

A slightly different apprroach woule be required to use ‘p32’ with fminsearch or some of the other optimisation functions. Those would require calculating the norm of the difference between the dependent variable and the objective function, and optimising with respect to the paarameter vector.

[cxctvz]

Thank you for all of your help. I think I understand this a bit better now, and I will try to explore what I can do with fminsearch as I believe this is what we were supposed to use.

[Star Strider]

My pleasure.

If my Answer helped you solve your problem, please Accept it!