Cluster Visualization

Interpreting patterns of gene expression with self-organizing maps: methods and application to hematopoietic differentiation.
-> [pdf] SOMs take a fundamentally different approach. They attempt
to provide an ‘‘executive summary’’ of a massive data set
by extracting the n most prominent patterns (where n is the
number of nodes in the geometry) and arranging them so that
similar patterns occur as neighbors in the SOM. As with all
exploratory data analysis tools, the use of SOMs involves
inspection of the data to extract insights.
SOMs are widely used in data mining because they have
many desirable mathematical properties, including scaling well
to large data sets. In our own hands, we have indeed found
them valuable in analyses involving hundreds of experiments.

A SOM calculated on the twoday-survey data, with a 4x4 grid, and pre-ordered columns: Such a chart wont scale well with a high number of variables.

-> Plotting Eight Direction Arranged Maps or Self-Organizing Maps
-> SOM package in R
-> EDAM in R

Same plot, but SOM with a finer grid: (Obviously there are not enough data sets to do that)

EDAM-Plot of Iris: Raabe, N. (2003). Vergleich von Kohonen Self-Organizing-Maps mit einem nichtsimultanen Klassifikations- und Visualisierungsverfahren. Diploma Thesis, Department of Statistics, University of Dortmund.
-> Raabe + Diplomarbeit

Diplomarbeit vergleicht SOM + EDAM hinsichtlich Visualisierungs- (=Topologie-Erhaltungs-)-Güte und Klassifizierungsgüte. Conclusio: SOM erhält räumliche Distanzen (=Toplogie) besser, EDAM klassifiziert besser (Unterschied sei aber bei hochdimensionalen Daten möglicherweise weniger stark).

Computation of 200 steps of the evolutionary algorithm for our survey data (500 cases x 26 variables) took more than 5 minutes in R.