Neuron weights are visible in Neuro-Lab's XML files. Based on those weights I want to discard inputs that the network has decided are not useful. Is a non-useful input one with a large negative weight or on with with a value close to zero, regardless of sign?
Size:
Color:
QUOTE:
Based on those weights I want to discard inputs
It's more complex than basing it on the weights. I'd suggest Googling "ANN input significance". If inputs are of different scale, (one with a scale of 0 to 1 and another with a scale of -50 to plus 50, for instance) some weight magnitude exists only to adjust that. Actually more the difference in standard deviation, but needs to be considered.
Size:
Color: