kmeans is suddenly behaving differently

Dear Matlab-Community,

just a few days ago I built a skript that required clustering and used the function kmeans(Data,Clusters,Options) and it worked straight away and I got good results with the following settings, analog to a setup from the examples:

   opts = statset('Display','final');
   [idx1, C1] = kmeans(2,datZ,'Distance','sqeuclidean','Replicates',5,'Options',opts);

Today, if I run the same script I get: Error using kmeans Too many input arguments."

The help for kmeans also returns a different function (?!) now:

kmeans Trains a k means cluster model.

   Description
    CENTRES = kmeans(CENTRES, DATA, OPTIONS) uses the batch K-means
   algorithm to set the centres of a cluster model. The matrix DATA
   represents the data which is being clustered, with each row
   corresponding to a vector. The sum of squares error function is used.
   The point at which a local minimum is achieved is returned as
   CENTRES.  The error value at that point is returned in OPTIONS(8).

Did the function get updated and change its functionality? Does anyone know what is going on?

If I then try to adapt to the suggestion from the help I get:

   Error using kmeans (line 46)
   Data dimension does not match dimension of
   centres

/Update: This only occurs in R2017a and not in R2016b! kmeans works like expecter kmeans(Data,Clusters,...) like expected. The Bug/Issue is only found in R2017a.

Answer 1

[Jan]

This means that you have installed the FEX: NETLAB toolbox written by Ian T. Nabney, which contains a kmeans function also. Obviously it was inserted on top of the Matlab path, such that builtin functions are shadowed. You can check this by:

which kmeans -all

[Maximilian Weber]

Thank you for your reply.

Is there a way to decide which of both functions I want to use? Can I block the function of the NETLAB toolbox?

[Stephen23]

Change the path order:

https://www.mathworks.com/help/matlab/matlab_prog/function-precedence-order.html

https://www.mathworks.com/help/matlab/matlab_env/what-is-the-matlab-search-path.html

Use the buttons "Move Up" and "Move Down".

[Jan]

@Maximilian: Matlab chooses the file, which appears at first in the path. So if you re-order the path, you cannot call kmeans of the NETLAB toolbox anymore. This is prefered only, when its parent folder is the current folder, or when it is called from another function in the same folder -- but as far as I remember this behavior has been changed in the past. Therefore I'd never rely on this.

You can convert the toolbox folder to a package by inserting a + in the name, see https://www.mathworks.com/help/matlab/matlab_oop/scoping-classes-with-packages.html. Then you can call the NETLAB kmeans specifically without shadowing the original file.

But the best method remains not to use names of builtin functions. Renaming NETLAB's kmeans to kmeans_netlab might be a mess, because it might require to replace the name in all other files of the toolbox. This is simply an evil problem.

[Maximilian Weber]

Understood! Thank you both for your replies.

The odd thing is: I didn't change anything on the paths and when I started R2017a earlier today everything was fine again. Does Matlab change orders by itself sometimes?

Anyways: I followed your suggestions and moved the correct toolbox up, above the others. So far so good.

[Jan]

@Maximilian: No, Matlab does not modify the path by its own. You could do this temporarily by addpath and as long as the new path is not stored by savepath, teh changes vanish with a restart.

If the pathdef.m file was made writable for users, another user of your machine might have changed the path there. This would be evil and modifying matlabroot\toolbox\local\pathdef.m should be forbidden on multi-user machines.