MATLAB and Simulink resources for Arduino, LEGO, and Raspberry Pi

Learn moreOpportunities for recent engineering grads.

Apply Today
Asked by Clifford Shelton on 30 Apr 2012

I have a dataset and I want to best fit a sinewave to the plotted data set. This process I think is called a regression...but all the info I come across is about linear regressions only.

Any help would be most appreciated!

*No products are associated with this question.*

Answer by Wayne King on 30 Apr 2012

Accepted answer

You need to know what periods you want to fit. You had another post where you talked about fitting city population for a period of 50 years. You did not say how often the data are sampled, I'll assume yearly. Just substitute your data for y (as a column vector)

t = (1:50)'; X = ones(50,3); X(:,2) = cos((2*pi)/50*t); X(:,3) = sin((2*pi)/50*t); y = 2*cos((2*pi)/50*t-pi/4)+randn(size(t)); y = y(:); beta = X\y; yhat = beta(1)+beta(2)*cos((2*pi)/50*t)+beta(3)*sin((2*pi)/50*t); plot(t,y,'b'); hold on plot(t,yhat,'r','linewidth',2);

If you have the Statistics Toolbox, you can do the same thing with regress()

If you don't know the periods, it is best to use Fourier analysis.

Clifford Shelton on 1 May 2012

Your answer to my question yesterday has been EXTREMELY helpful! Thanks so much. Some issues though:

Let's say I had a data set of city population statistics and it was yearly data for 501 years. And I wanted to see if there was a 25.7 year cycle to that data.

How would I go about fitting a sin wave with a 25.7 year period to my 501 year data set?

Furthermore, when I try to substitute y with my data and create a column vector as suggested in your posted code:

>> X = ones(501,3);

>> X(:,2) = cos((2*pi)/501*t);

>> X(:,3) = sin((2*pi)/501*t);

>> y = Score(:);

>> y = 2*cos((2*pi)/50*t-pi/4)+randn(size(t));

>> y = y(:);

>> beta = X/y;

??? Error using ==> mldivide

Matrix dimensions must agree.

I get the above error. What am I doing wrong in constructing my column vector with the data I call "Score"?

Thank you SO much!

Answer by Richard Willey on 1 May 2012

Here's some simple code that illustrates how to perform nonlinear regression using the 12a release of Statistics Toolbox.

Note: NonLinearModel.fit requires that you provide starting conditions for the various parameters. (Providing good starting conditions helps to ensure that the optimization solvers converge on a global solution rather than a local solution)

%% Generate some data

X = 2* pi*rand(100,1); X = sortrows(X); Y = 9 + 7*sin(2*X + 4*pi) + randn(100,1);

scatter(X,Y)

%% Generate a fit

% Note that we need to pass three sets of input arguments to NonLinearModel % # The X and Y data % # A string describing our model % # Starting conditions for the optimization solvers

% Generate some good starting conditions for the solvers

scatter(X, Y) hold on

B0 = mean(Y); % Vertical shift B1 = (max(Y) - min(Y))/2; % Amplitude B2 = 2; % Phase (Number of peaks) B3 = 0; % Phase shift (eyeball the Curve)

myFit = NonLinearModel.fit(X,Y, 'y ~ b0 + b1*sin(b2*x1 + b3)', [B0, B1, B2, B3])

% Note that all the coefficient estimates are very good except for b3 where % any even integer is equally valid

%% look at the complete set of methods

methods(myFit)

%% Generate a plot

hold on plot(X, myFit.Fitted) hold off

%% Generate a fit using an alternative syntax

myFit2 = NonLinearModel.fit(X,Y, @(b,x)(b(1) + b(2)*sin(b(3)*x + b(4))), [B0, B1, B2, B3])

## 0 Comments