Kohonen maps

Hello everyone,
I'm currently trying to figure out how to use the Kohonen maps. I'm confused with the determination of the number of input and output neurons. For example, if we have dataset with grades of students who have performed a few tests and I want to use that set to train a network. How to determine a number of input and output neurons in that case. I'm also interested in the process of reduction of data dimensionality.
Is there any reference to more concrete example done step by step. I understand example from the book where we have character recognition, and number of characters determined the number of output neurons.
Thans in advance