PSOM assists in building and training the parameterless version of a Self-Organizing Map. Unlike the ordinary SOM as implemented in the SOMnia program the updating of the prototypes (nodes) in the SOM grid does not follow a gradually decreasing learning rate parameter. Instead, the amount of update depends on the current value of the quantization error. PSOM combines two analytical steps. It lets the analyst train a SOM grid to approximate a manifold in input data space according to Berglund and Sitte (2006). Then it performs a stepwise merging of class centers based on statistical neighborhood and the Simple Structure Index of Mazanec (2001). PSOM is written in Matlab and runs under the Matlab Runtime Machine V7.8 (see the download instructions in the read_me_1st.txt file below).

PSOM is based on Berglund, E. and J. Sitte (2006), 'The Parameterless Self-Organizing Map Algorithm', IEEE Transactions on Neural Networks, 17(2), 305-316, and Mazanec, J. A. (2001), 'Neural Market Structure Analysis: Novel Topology-Sensitive Methodology', European Journal of Marketing 35(7-8): 894-916.

Note: Please read and accept our download conditions before!