Cluster Models/Methods

Current supported models

The current set of supported cluster models can always be listed by running the function listmodels() at the matlab prompt. For example, running this command reveals that K-means is a supported cluster model in this toolbox. You can access the matlab help file for the supported cluster models by following the CCToolbox\mixtures link from the main CCToolbox matlab help screen. You can also access the matlab help file by typing help <method> at the matlab prompt.

Naming convention

There is a specific naming convention that is followed in this toolbox for the set of regression mixture models. The naming convention is at first based on the standard linear and spline regression mixture models. These two models are listed by listmodels() as follows.

       lrm    - Linear Regression Mixtures (LRM) with y = XB + e
       srm    - Spline Regression Mixture (SRM) with y = XB + e
    

This listing shows that the method name 'lrm' should be used to run lrm() and the method name 'srm' should be used for srm(). All of the other regression mixture models are named by adding a suffix containing the transformation variables that are learned for that specific model (see the background section for a discussion of the transformation variables and of curve-aligned clustering in general). For example, the following curve-aligned LRM models are listed by listmodels().

       lrm_d      - LRM with transformation  [x]B + d
       lrm_b      - LRM with transformation  [x+b]B
       lrm_d_b    - LRM with transformation  [x+b]B + d
       lrm_cd     - LRM with transformation c[x]B + d
       lrm_ab     - LRM with transformation  [ax+b]B
       lrm_d_ab   - LRM with transformation  [ax+b]B + d
       lrm_cd_b   - LRM with transformation c[x+b]B + d
       lrm_cd_ab  - LRM with transformation c[ax+b]B + d
    

Notice how the measurement space transformation variables are always listed before the time transformation variables for each method name (on the left). A similar but shorter list is returned for the curve-aligned SRM models.

       srm_d      - SRM with transformation y=[x]B+d
       srm_cd     - SRM with transformation y=c[x]B+d
       srm_b      - SRM with transformation y=[x+b]B
       srm_ab     - SRM with transformation y=[ax+b]B
       srm_d_b    - SRM with transformation y=[x+b]B+d
    

Gaussian mixtures ('gmix') and K-means ('kmeans') are also provided in the base set.

Calling a method

You should use curve_clust() to call all of the clustering methods. This procedure is demonstrated in the tutorial section. In short, ops.method should contain the string of the method that is to be used, and then this options structure is passed-in to curve_clust(). For example, these two lines will run spline regression mixtures with space alignment using the curves in trajs and the options in ops.

>> ops.method = 'srm_cd';
>> model = curve_clust(trajs,ops);