[jfim]

If both your training and testing datasets are representative of actual data, wouldn't the normalization function be nearly equivalent in both datasets?