Return DKL(P∥Q) in bits.
DKL(P∥Q)=i∑pilog2qipi
Skip terms where pi=0. It is not symmetric, and it is not a distance:
DKL(P∥Q) generally differs from DKL(Q∥P), which is
exactly why the direction matters when it appears in a loss.
Input
p = [0.5, 0.5]
q = [0.5, 0.5]
Output
0.0