Let's go through process of how SOMs learn:
- Each node's weights are initialized by small standardized random values. These act like coordinates for different output nodes.
- The first row's input (taking the first row from all of the variables) is fed into the first node.
- Now, we have got two vectors. If V is the current input vector and W is the node's weight vector, then we calculate the Euclidean distance, like so:
- The node that has a weight vector closest to the input vector is tagged as the best-matching unit (BMU).
- A similar operation is carried out for all the rows of input and weight vectors. BMUs are found for all.
- Once the ...