File Exchange

image thumbnail

error_ellipse

version 1.0.0.0 (111 KB) by AJ Johnson
Plot an error ellipse depicting confidence interval given a covariance matrix.

58 Downloads

Updated 23 Jul 2015

View License

% ERROR_ELLIPSE - plot an error ellipse, or ellipsoid, defining
% confidence region
% ERROR_ELLIPSE(C22) - Given a 2x2 covariance matrix, plot the
% associated error ellipse, at the origin. It returns a graphics handle
% of the ellipse that was drawn.
%
% ERROR_ELLIPSE(C33) - Given a 3x3 covariance matrix, plot the
% associated error ellipsoid, at the origin, as well as its projections
% onto the three axes. Returns a vector of 4 graphics handles, for the
% three ellipses (in the X-Y, Y-Z, and Z-X planes, respectively) and for
% the ellipsoid.
%
% ERROR_ELLIPSE(C,MU) - Plot the ellipse, or ellipsoid, centered at % MU, a vector whose length should match that of C (which is 2x2
% or 3x3).
%
% ERROR_ELLIPSE(...,'Property1',Value1,'Name2',Value2,...) sets
% the values of specified properties, including:
% 'C' - Alternate method of specifying the covariance matrix
% 'mu' - Alternate method of specifying the ellipse (-oid) center
% 'conf' - A value betwen 0 and 1 specifying the confidence interval.
% the default is 0.5 which is the 50% error ellipse.
% 'scale' - Allow the plot the be scaled to difference units.
% 'style' - A plotting style used to format ellipses.
% 'clip' - specifies a clipping radius. Portions of the ellipse, -oid,
% outside the radius will not be shown.
%
% NOTES: C must be positive definite for this function to work
% properly.

Cite As

AJ Johnson (2019). error_ellipse (https://www.mathworks.com/matlabcentral/fileexchange/4705-error_ellipse), MATLAB Central File Exchange. Retrieved .

Comments and Ratings (77)

Larry

@Matthias Y The confidence intervals are different depending on the number of dimensions. For two-dimensional data, try changing 0.68 to 0.3934. (A useful table is provided on page 3 of the documentation.)

In your example, error_ellipse(covm,'conf',0.3934) should produce a circle (or rather, near-circle) with a radius equal to the standard deviation.

Thanks a Lot you saved me time, Appreciate Sir.AJ Johnson

donald tong

LIGANG SUN

Thank you so much!!

Matthias Y

Is the way the confidence interval is plotted correct?
Take the 2D example. Isn't the std-dev the distance from the origin to the outermost points of the ellipse for each axis. If I do this:

samples = randn(200,2);
covm = cov(samples);
error_ellipse(covm,'conf',0.68);
std_devx = sqrt(covm(1));

My std-dev for the x axis (std_devx) is somehow not matching the error_ellipse for 0.68% confidence.

Farah Jabin

works, easy

Qi An

Perfect. Saved me hours of reading and code writing.

works fine. thank you!

M.B

Nice piece of code.

Aurican

Ke Jing

Thank you, good for learning

Alceu Costa

Ahmed Fasih

Thanks AJ! Delicious BSD license :)

Ahmed Fasih

What's the license on this submission, anyone know?

Should the second argument to qchisq not be the degrees of freedom, rather than the number of parameters?

This was excellent for learning about Kalman filters.

Serena

Thank you,it was useful!
I have a question about the k value since I'm not familiar with statistics: why do you get it from a k square distribution and not from a normal distribution?

Christian

Burcu

Really very helpful. But did not understand how to calculate k value from conf "k = sqrt(qchisq(conf,r))" Does anybody know how this relation holds? Thanks..

Thanks. Saved a lot of my time.

Takuma

Anthony

Very helpful and easy to use.

gunes bay1r

thanks for conf. interval comment phil fox

Claus

Ooops, there's an obvious mistake in my area/volume calculation from above. I forgot to factor in the quantile for the desired percentile, i.e., you have to multiply the area/volume with the quantile of the desired confidence interval (conf). You can use the embedded functions from error_ellipse to calculate Pconf (for the default of conf=0.5, Pconf(2D)=1.1774/Pconf(3D)=1.5382). The formular for area (volume accordingly) should read:
area=prod(sqrt(eig(C)))*pi*Pconf

Claus

@Jenn Risk & @me: Finally got a solution for area/volume. It's all based on the eigenvalues of the CovMatrix (the prerequisite for error_ellipse).
Area (2D): area=prod(sqrt(eig(C)))*pi
Volume (3D): volume=prod(sqrt(eig(C)))*4/3*pi

Hope that helps. Cheers!

Claus

Worked just great for me and saved a LOT of time writing my own code. Thanks!

I'm currently working on a area/volume calculation. @Ahmed Fasih: det(covMatrix) may definately be used as a measure, however, it's not the actual area/volume.

Ahmed Fasih

Jenn Risk: just use det(CovMatrix) for area/volume.

M ChC, Volkan Ozturk:
figure; handle(1)=error_ellipse(A); hold on; handle(2)=error_ellipse(B); handle(3)=error_ellipse(C); set(handle(2), 'color','r', 'linestyle',':'); set(handle(3), 'color','k', 'linestyle','--');

Marcel

Perfect piece of code! I am using it alot. Had to change the API though to make it a bit more flexible in terms of plot properties (e.g. passing it a parent axes handle).

Ipsita

??? function h = error_ellipse(varargin)
|
Error: Function definitions are not permitted at the prompt or in scripts.

I get the above error, can anyone help ? I am using MATLAB 7.6.0(2008a). Thanks

Jenn Risk

Very helpful optically - but can it calculate the area of the error ellipse? I couldn't easily get it to adapt to this. Any help would be great!

David

Sorry, but my previous comment contained some errors:
- C = [1 -2; -2; 4]. This is not a valid covariance matrix.
- "positive semi-definite symmetric (that is, they may include negative values off the diagonal)." This is not true.

What is true, is that C must be a positive semi-definite matrix. But that is just a minor comment, because the implementation allows for that.

Conclusion:
This Matlab function works. Thanks!

David

% NOTES: C must be positive definite for this function to
% work properly.

This is quite a strange restriction, because covariance matrices are positive semi-definite symmetric (that is, they may include negative values off the diagonal). For example, it is not possible to visualize the following very simple, and valid, covariance matrix: C = [1 -2; -2; 4].

Does anybody know of a Matlab function which is capable of visualizing all kinds of covariance matrices (both 2D and 3D)? Thanks!

Pearl

Nora

John

function works great and is easy to use. thank you

Dan Stowell

Delightful tool, easy to use and worked straight away

R Lewis

Did the job flawlessly.

Alex Yakubovich

re: change color, multiple ellipses on same plot

To have multiple ellipses, just type 'hold on' in the command window before calling error_ellipse. You can change the color once you've plotted the ellipse in the property editor (open pe in view>property editor)

Adam Attarian

precisely what i needed! hooray.

M ChC

how do I change the color of the ellipse?

Dagmar Schoch

Trevor Wood

excellent, thanks.

Siladitya Dey

Marino Pagan

Thank you very much!!! Excellent!

Matthew Nelson

very useful. thank you

phil fox

Be aware that the default confidence interval is 0.5 rather than more standard values such as 0.683, 0.9 etc.. Use the 'conf' parameter to change.

Volkan Ozturk

How can i see two plots on the same graphics. I have to do some comparison.

Regards

kevin chen

handy

Feng Rice

Excellent.

mohan palani

Sujai Kumar

John Magirk

perfect! works great AND easy to use.

Chris Willy

Great code, very useful. Saved me lots of time.

Matt Irwin

Great code, thanks.

Damien Dusha

You just saved me a day and most of my hair :)

Valentin Kuklin

Amazing. Good work! Thanks a lot!

M Fellows

Very Useful!

ryan jessup

worked perfectly! Thanks!

oxana govokhina

Excellent

Laurent Nguyen

Works perfectly

Juan Pablo Nieto

It works excellent

Paul Thompson

There is NOT an error at line 117.

Kamini Singha

works fabulously.

Niklaus Hugi

Gordon Dryden

There is an error at line 117. The line reads: XYZ = [X(:),Y(:),Z(:)]*sqrt(eigval)*eigvec'; but should be: XYZ = [X(:),Y(:),Z(:)]*eigvec*sqrt(eigval)*eigvec';

Mark Brown

Doesn't run. I get an error...
??? Undefined command/function 'getopt'.
Error in ==> error_ellipse at 61
prop = getopt(default_properties, varargin{:});

Una Kusan

Now it's working fine.
Thanks again!
U.

Una Kusan

It doesn't work on my matlab platform (6.1).

Updates

1.0.0.0

Updated license.
Added white paper describing the math behind the confidence region computations.

1.0.0.0

Fixed major blunder: missing function.

MATLAB Release Compatibility
Created with R12.1
Compatible with any release
Platform Compatibility
Windows macOS Linux