Do what Excel does in curve fitting in Matlab

Hello all, thanks in advance for your help!

I know that Matlab is loaded with all sorts of curve fitting software that can utilize a whole bunch of different functions in order to give the best fitting possible.

However, I just have some simple data that I wish to fit a linear standard line to- make the line go through 0 and show the r squared value, and show the equation. I also wish to show this for my plot that has more than one "function" plotted (this is because I noticed that using the tools method in the plot interface only allows one curve to be fit, and doesn't show the rsquared value).

If anybody can list out what I can add to my code to implement these simple aspects, that will be great.

Thanks again. sam

Answer 1

[Teja Muppirala]

Wayne gives a good way to do it, but in the latest version of the Statistics Toolbox, there is some great functionality that makes this stuff even easier::

x = 1:100;
y = 2*x+ 4*randn(size(x));

F = LinearModel.fit(x,y,'linear','intercept',false)
plot(F)

F.Rsquared

See "help LinearModel" for more information

[Samuel]

Thanks for the help. I actually have the statistics toolbox when i check the versions, but strangely when I adapt and run the code as you wrote, it gives me an error saying

Undefined variable "LinearModel" or class "LinearModel.fit".

and typing help LinearModel gives an error

LinearModel not found.

i have matlab r2011b for mac and windows. what might be the problem?

thanks!

[Teja Muppirala]

Hello Samuel,

The LinearModel class was just introduced recently in the R2012a release.

You may have to use the other methods mentioned instead.

Answer 2

[Wayne King]

Are you really sure you want to force the line to go through zero?

Do you have the Statistics Toolbox? If so, use regress()

    x = 1:100;
    y = 2*x+ 4*randn(size(x));
    X = ones(length(y),2);
    X(:,2) = x';
    [beta,bint,r,rint,stats] = regress(y',X);

Fitted line is yhat = beta(1)+beta(2)*x. Rsquared is

stats(1)

   plot(x,y,'k*'); hold on;
   yhat = beta(1)+beta(2)*x;
   plot(x,yhat,'r','linewidth',2);

[Samuel]

Hello Wayne, thanks for the help!

I actually tried implementing your method, but when I execute it, I get an error that states matrix dimensions must agree. It seems that the X matrix should be only 1 column in order for the regress function to work? Any clarification will be helpful! Thanks.

Answer 3

[Daniel Shub]

While it doesn't do everything you want I would suggest looking at the source for lsline

type lsline

It takes care of plotting multiple lines. It makes use of polyfit

doc polyfit

which provides the r^2 value you want

Adding the equation for a line that goes through the intercept seems a bit odd to me ...