reclus

Rectangle cluster plot - any clustering method - any dimensionality of the data. This can be used to plot the results of any clustering algorithm (k-means, agglomerative, model-based clustering), where the input is the cluster labels.

Input Arguments (see below for their usage):

• CLUSLABS is a vector of n cluster labels.
• TRULABS is a vector of n true class labels.
• STRENGTH is a measure of 'goodness' or certainty in the clustering. See below for more information.
• THRESH is a cutoff value. Observations with a STRENGTH value above this are displayed in bold, black font.

Output Arguments:

• None

Synopsis

reclus(cluslabs,trulabs,str,thresh)

RECLUS(CLABS) plots the rectangles, where the area of each rectangle represents the proportion of points falling into that cluster. The data are plotted using their observation number.

RECLUS(CLABS,TRULABS) plots the rectangles as before, but the position of the symbols as case labels matches the same position when the true class labels are used as the plotting symbol - see the options below for 3 or 4 arguments. This allows the user to see which cases correspond to specific symbols plotted with the true class label.

RECLUS(CLABS,TRULABS,STRENGTH) plots the rectangles as above, where each symbol color represents a measure of the STRENGTH of cluster membership. This could be the P(X|cluster) (obtained from MIXCLASS: 1 - ERR) or it could be the silhouette value from SILHOUETTE. A colorbar is included to indicate the color scale.

RECLUS(CLABS,TRULABS,STRENGTH,THRESH) plots the points as above. The value THRESH is used to indicate which observations have a classification certainty greater than THRESH; these values are plotted in bold. Thus, the color indicates the probability that it belongs to the cluster on a continuous scale, and the bold indicates a binary value - above or below THRESH - in the case of finite mixture clustering.

References:

Martinez and Martinez, Exploratory Data Analysis with MATLAB, CRC Press, 2004.

JSM paper.