Since any modification of a Self-Organizing Map in the 2D display like the creation, deletion or movement of units or weights may destroy the relative position of the units in the map we strongly recommend to generate these networks only with the available BIGNET (Kohonen) tool. See also chapter for detailed information on how to create networks. Outside xgui you can also use the tool convert2snns. Information on this program can be found in the respective README file in the directory SNNSv3.3/tools/doc. Note: Any modification of the units after the creation of the network may result in undesired behavior!
To train a new feature map with SNNS, set the appropriate standard functions: select init function KOHONEN_Weights, update function Kohonen_Order and learning function Kohonen. Remember: There is no special activation function for Kohonen learning, since setting an activation function for the units doesn't affect the learning procedure. To visualize the results of the training, however, one of the two activation functions Act_Euclid and Act_Componnent has to be selected. For their semantics see section .
After providing patterns (ideally normalized) and assigning reasonable values to the learning function, the learning process can be started. To get a proper appearance of SOMs in the 2D-display set the grid width to 16 and turn off the unit labeling and link display in the display panel.
When a learning run is completed the adaption height and adaption radius parameters are automatically updated in the remote panel to reflect the actual values in the kernel.