When I generate a Matlab function of a neural network, it comes with the normalizat​ion/re-nor​malization constants. Will these constants affect the neural network performance on another data set? Why?

The normalization/re-normalization (for example, mapping the data to [-1,1] or [0,1]) is built for the training data. What does make the normalization constants work to another data set? And if the new data have large values than the training data?

Answer 1

[Greg Heath]

Not if both datasets have similar summary statistics (e.g., mean/std for mapstd or min/max for mapminmax).

That is why sometimes it is better to explicitly normalize before training and explicitly denormalize after training.

Hope this helps.

Thank you for formally accepting my answer

Gregg