In this section we introduce operations needed for a TS-SOM analysis. First, we give an overview how different data structures are used with TS-SOM. The first operation is training. It builds and organizes a TS-SOM structure, which is stored in the name space with a given name. Suppose that its name is som. In addition, the training creates a weight matrix, which is stored in a data frame called the name of the TS-SOM with the extension _W, for instance, som_W. In that data frame, each data record contains the weights of one neuron, and it is connected to the corresponding neuron through its index. The TS-SOM structure maintains all the structural information about the neural network (see the figure below).
The second important operation is SOM classification. This operation divides the data records of a given data frame into subgroups according to located BMU (neuron). The result is a classified data which includes as many classes as there are neurons in the TS-SOM. Like a weight matrix, one class corresponds to one neuron and includes the indexes of the data records classified to that neuron (see the figure). If the neuron does not get any data record in the classification, then an empty class is created corresponding to that neuron.
