numpy - Python: Efficiently sample an n-dimensional distribution from density array -

i have n-dimensional distribution have estimated gauissian kernel density have stored n-dimensional array. need perform 2d kohonen map fitting underlying distribution.

the simplest way of doing in python found far use newc module neurolab.

however, module requires cloud or points , need sample n-dimensional array recover points correspond original density distribution.

what efficient way perform such sampling? alternatively, there kohonen map modules work directly density arrays?