Configuring a Self-Organizing Map

Choosing the self-organizing map technique

You can choose between two techniques for creating a self-organizing map. The difference between the two techniques is the type of error they try to minimize during the construction of the map. If you are using self-organizing maps to identify clusters, you should select the Preserve topology of sample space technique (default). If you are more concerned with the accuracy of the self-organizing maps, you should select the Maximize resolution of fit technique. In most cases, you will be interested in identifying clusters.

Preserve topology of sample space

The Preserve topology of sample space technique tries to minimize the topology error—are similar cells clustered close to each other? Results Analytics calculates the topology error as the sum of the error values for each bin divided by the number of bins. For each cell, if the second best matching cell is adjacent to it, the error value is 0; otherwise, the error value is 1.

Maximize resolution of fit

The Maximize resolution of fit technique tries to determine the quality of the fit. For each data point, Results Analytics calculates the distance between the data point and the cell points that best match the data point's value. The average value of all of these distances is the error value. If all the cell values match the data point values, the error is low. (In the ideal case, each data point would have its own cell, with the cell's value equal to the data point value. This would result in a perfect fit with a zero error.)

Setting the number of iterations

You can enter the number of iterations that Results Analytics performs (the outer loop described in The Self-Organizing Maps Algorithm). If you have a large number of data points and/or parameters, you can accelerate the generation of the self-organizing maps by decreasing the number of iterations, although the accuracy of the maps will suffer. Conversely, to increase the accuracy of the self-organizing maps, you can increase the number of times Results Analytics repeats this process. However, you will probably not experience much improvement in the accuracy of the maps beyond the default 10 iterations.

The self-organized map displayed by Results Analytics for each parameter is the best map chosen from amongst the maps generated by all of the iterations.