Smoothing data on a(n uncomplete) cartesian grid
INPUT:
-ijk : list of the positions, which is assumed to be an (n,3) int array
it is typically returned by transpose(numpy.where())
- data : an array of data sampled from the grid size(ijk.shape[0],d)
where d is the datas dimension
-sigma : the kernel parameter
OUTPUT
- data, which is the smoothed data