Notes:
We are using three main C functions: fast_dists_by_unit_arma2_par
if we are using openmp (parallel="yes"
), fast_dists_and_trans
for small n, and fast_dists_and_trans_by_unit_arma
for N>20. (Where 20 is chosen without a lot of profiling. Basic idea is not to hold large matrices in memory. Matrices are faster when N is not to large, but bog down when we have large number of observations.