So I'm building a 3-layer backprop neural network to help me transform a feature vector input of roughly 50 elements into a score from 0 - 100. EDIT - should mention I plan on having at least 1000 training vectors (depending on how many I can find lol).
I've found some good rules of thumb about choosing the number of neurons in the hidden layer: http://www.heatonresearch.com/node/707
But how should I go about choosing the number of input neurons, and the number of output neurons? It seems like such an obvious question but I haven't seen it addressed in my research thus far.
Thanks for any help.
[link][3 comments]