There are two interfaces:
VectorNormalizer. Merger is used to combine gradient tensors
produced across different examples (i.e., to combine the gradients within a minibatch). VectorNormalizer is
used to normalize an array, usually into a probability distribution.
ClassDescriptionNormalizes the exponential values of the input array.NoopNormalizer returns a copy in
NoopNormalizer.normalize(double)and is a no-op in place.Normalizes, but first subtracts the minimum value (to ensure positivity).Normalizes the input by applying a logistic sigmoid to each element.A functional interface that generates a normalized version of a double array.