This Counterpropagation initialization function initializes all weight vectors of the Kohonen layer with random input patterns from the training set. This guarantees that the Kohonen layer has no dead neurons.
The weights of the Grossberg layer are all initialized to 1.