I had a problem learning a neural network with sparse input data to solve a problem with controlled regression. When I do the average normalization (subtract the average and then divide the standard deviation) by the input, I get many NaN values. I am wondering if anyone has experience solving this problem. What is the correct way to scale sparse input?
thanks joe
It looks like your data is so sparse that the standard deviation is sometimes zero.
Test this, and if so, do not divide it by input (stdev normalization is not needed in this case).