Fitting data from file

Hello,

I am trying to fit data from a file. For that I have created data with simply a sine wave function. Here is my code:

load test_fit

Error using load
Unable to find file or directory 'test_fit.txt'.

x = test_fit(:,1);
y = test_fit(:,2);
fo=fit(y,x,'a*sin(b*x+c)+d');
plot(fo,y,x)

But the results I got are clearly not good:

I'm just wondering if I should specify somwhere the number of itterations or the fitting constrains?

Thank you in advance for any help or comments!

Answer 1

[Rik]

You can specify those parameters, but what is much more effective is providing good initial guesses (and not mixing up x and y):

test_fit=load(websave('test_fit.txt','https://www.mathworks.com/matlabcentral/answers/uploaded_files/1137860/test_fit.txt'));
x = test_fit(:,1);
y = test_fit(:,2);
plot(x,y,'*')
fo=fit(x,y,'a*sin(b*x+c)+d','StartPoint',[1 4 0 2])

fo = 
     General model:
     fo(x) = a*sin(b*x+c)+d
     Coefficients (with 95% confidence bounds):
       a =           1  (1, 1)
       b =           4  (4, 4)
       c =      0.3927  (0.3927, 0.3927)
       d =           2  (2, 2)

hold on,plot(fo)

[Piotr]

Hi Rik,

Thank you that is really helpful. I have one more question I would like to combine the fitting with the reading data from a an image, but getting an error. Maybe you can help me to solve it?

That is my code:

I=imread('testf5.png');
x = [size(I,2)/2 size(I,2)/2];
y = [0 size(I,1)];
c = improfile(I,x,y);
figure
subplot(2,1,1)
imshow(I)
hold on
plot(x,y,'r')
subplot(2,1,2)
plot(c(:,1,1),'r')
fo=fit(x,y,'m*sin(n*x+o)+p','StartPoint',[125 100 0 125])

but it doesnt like the y column format (I got this error mesage: Y must be a column vector)

[Rik]

If the fit function requires column vectors, why don't you transpose x and y to be column vectors?

But there is a fundamental problem here: you have 2 points, to which you are fitting an equation with 4 unknowns. That is not going to be a reliable fit. It seems like you want to use the profile as your data, but you don't.

[Piotr]

Okay I transformed to column vector with:

fo=fit(x(:),y(:),'m*sin(n*x+o)+p','StartPoint',[125 100 0 125])

and also got the error about insuficient point number which I don't understand becaouse there are many points:

[Rik]

Where exactly are you telling the fit function that you want to use all those points? (Hint: you don't)

[Piotr]

This I don't fully understand, do I need to then read the data from the graph? I started to use Matlab only recently

[Rik]

You provide x and y as input arguments to the fit function, but that only contains the coordinates of the line you're using for improfile, not the data you're plotting. You probably want to use fit(x_adapted,c,___). That first variable should contain the x position for each value in c. Currently when you're plotting your data, you are implicitly using the index as the x-value. That's fine for getting a first impression of your data, but not for actual analysis.

You probably want this:

x_adapted = linspace(y(1),y(2),numel(c));

I would also suggest doing a basic Matlab tutorial. The Onramp tutorial is provided for free by Mathworks.

[Piotr]

Okay, thank you now that is clear but still got an error with fit weights. Even if I define wieghts I still get it

Error using fit>iFit

X, Y and WEIGHTS cannot have NaN values.

Error in fit (line 116)

[fitobj, goodness, output, convmsg] = iFit( xdatain, ydatain, fittypeobj, ...

Error in testfitimage (line 20)

fo=fit(transpose(x_adapted),c(:,1,1),ft)

websave('testf5.png','https://www.mathworks.com/matlabcentral/answers/uploaded_files/1139785/testf5.png');
I=imread('testf5.png');
x = [size(I,2)/2 size(I,2)/2];
y = [0 size(I,1)];
c = improfile(I,x,y);
figure
subplot(2,1,1)
imshow(I)
hold on
plot(x,y,'-o')
subplot(2,1,2)
plot(c(:,1,1),'-o')
x_adapted = linspace(y(1),y(2),numel(c(:,1,1)));

ft = fittype('m*sin(n*x+o)+p');
options=fitoptions(ft);
options.StartPoint = [125 100 0 125];
options.Weights = [1,1];

plot(x_adapted(:),c(:,1,1))

fo=fit(transpose(x_adapted),c(:,1,1),ft)

Error using fit>iFit
X, Y and WEIGHTS cannot have NaN values.

Error in fit (line 116)
[fitobj, goodness, output, convmsg] = iFit( xdatain, ydatain, fittypeobj, ...

[Walter Roberson]

x = [size(I,2)/2 size(I,2)/2];
y = [0 size(I,1)];
c = improfile(I,x,y);

y = 0 is outside the image coordinates, so the profile gives NaN at that location.

[Rik]

If you look at the code below, you see that c starts with a NaN. So we should remove it.

I = imread(websave('testf5.png','https://www.mathworks.com/matlabcentral/answers/uploaded_files/1139785/testf5.png'));
x = [size(I,2)/2 size(I,2)/2];
y = [0 size(I,1)];
c = improfile(I,x,y);
c = c(:,1,1); % select red chanel
x_adapted = linspace(y(1),y(2),numel(c));
x_adapted = x_adapted.'; % transpose 1x1491 to 1491x1

subplot(2,1,1)
imshow(I)
hold on
plot(x,y,'-o')
hold off

subplot(2,1,2)
plot(x_adapted,c)

ft = fittype('m*sin(n*x+o)+p');
options=fitoptions(ft);
options.StartPoint = [125 100 0 125];
options.Weights = [1,1];

find(isnan(x_adapted)),find(isnan(c))

ans =

  0×1 empty double column vector

ans = 1

Here is the problem, so we'll have to remove that point.

L = isnan(x_adapted) | isnan(c); % the first part is just good habit
x_adapted(L) = [];
c(L) = [];

% Now we can safely to the fit:
fo=fit(x_adapted,c,ft,StartPoint=[125 100 0 125])

fo = 
     General model:
     fo(x) = m*sin(n*x+o)+p
     Coefficients (with 95% confidence bounds):
       m =        0.15  (-6.376, 6.676)
       n =         100  (99.9, 100.1)
       o =   -0.001634  (-87.15, 87.15)
       p =       129.3  (124.6, 133.9)

figure,plot(fo)

So the fit is not great, you will have to adjust the start point.

[Rik]

If my answer helped you, please consider marking it as accepted answer.