Generally, you will not want to use this randomizer as a pure neural network
randomizer. More on this later in the description.
Generate random numbers that fall within a Gaussian curve. The mean
represents the center of the curve, and the standard deviation helps
determine the length of the curve on each side.
This randomizer is used mainly for special cases where I want to generate
random numbers in a Gaussian range. For a pure neural network initializer, it
leaves much to be desired. However, it can make for a decent randomizer.
Usually, the Nguyen Widrow randomizer performs better.
Uses the "Box Muller" method.
Ported from C++ version provided by Everett F. Carter Jr., 1994