Question about MLP neural network


I have a question about using this MLP neural network. I trained some network and export the
result to c++ code. Then I realized in the exported code, all the input variables are nomalized first
by the mean and rms of the variable distrubutions in the train sample. I want to ask what I can do if I want to use this neural network output code to real sample which has different input
variable distrubutions? Just replace the mean and rms with the one in the real sample? Thanks in advance.



No, you don’t have to change the normalization.
Inputs are normalized for training since a normal range is most suited to sigmoids. That way, training is more efficient, especially at the beginning.

These normalization factors have to be seen as parameters of the NN, fixed during the learning.


I see. Thanks a lot!