gpartinorm¶
Normalises each Feature in each Batch in a special way: After subtracting the mean of each vector x, it subtracts the mean(abs(x)) from it, just to divide it by (mean(abs(x))+max(abs(x)))/2.
Arguments
- gs: The Number of Nodes of the Graph (Graph Size)
- param: The Number of Features in each Feature vector
- alpha=0.01: A numeric Constant to remove divergences from dividing (instead of 1/a it uses 1/(abs(a)+alpha))