Kohonen Map

Command:

Math -> Neural Network -> Kohonen...

The Kohonen network (or "self-organizing map", or SOM, for short) has been developed by Teuvo Kohonen. The basic idea behind the Kohonen network is to setup a structure of interconnected processing units ("neurons") which compete for the signal. While the structure of the map may be quite arbitrary, this package supports only rectangular and linear maps.

The chess pattern in the center shows the structure of the Kohonen network, each rectangle symbolizing a single neuron. Within each neuron, the observations of the data are indicated. The type of symbol used for the data can be set by clicking the setup button (). The position of a single observation has no meaning; each observation is shifted just enough to make the observations attached to a particular neuron visible (if the check box "Spread Objects" is checked).

Observations can be marked by selecting the intended marking mode (button ) and drawing either a lasso or a rectangle around the corresponding observations.

On the right part of the window the parameters of the Kohonen network may be set up:

X-size	The number of neurons in horizontal direction. Note that a linear Kohonen network may be created by setting either X-Size or Y-Size to 1.
Y-size	The number of neurons in vertical direction. Note that a linear Kohonen network may be created by setting either X-Size or Y-Size to 1.
No. of Neighbors	The size (radius) of the initial neighborhood. The number of neighbors decreases during the training linearly towards 1. Note that the number of neighbors should not be greater than half the greatest size parameter (X-Size and Y-Size). Otherwise the learning process will be slowed down.
Alpha	The initial learning rate. The learning rate will decrease linearly to a zero value during the training.
No. of Iterations	The number of training steps. The number of training steps may be varied between 10 and 10000.
Use Class Colors	If this option is active the data points are shown using the class colors.
Cyclic Net	Checking this check box creates a cyclic network (either a torus or a circle).
Show Training	The training progress is displayed during the training if this check box is checked. Note that this option considerably slows down the training.
Spread Objects	The observations attached to a single neuron are spread over the neuron rectangle. This option does not change the functionality of the network; it is just a simple way to visualize the number of observations per neuron.
Connect Objects	This option shows the observations with a connecting line between them. The line is drawn in the order of the data of the data matrix. Connecting the objects may be especially useful when time series are fed into the network.