If hist, (x, y, z) = numpy.histogramdd() gives you the histogram values at positions (x, y, z) corresponding to the edges of the bins for a three-dimensional function, how one can calculate (interpolate) the histogram values at midpoints namely (x+d/2, y+d/2, z+d/2) where d is the fixed size of bins in all three directions?
Actually, np.histogramdd gives you the bin boundaries in x, y, z, but the counts are "at the midpoints", not the boundaries (strictly speaking they are neither, they are over d x d x d cubes centered at the midpoints).
If---for some reason---you have values at boundaries and want to interpolate and if you are ok with linear interpolation:
np.lib.stride_tricks.as_strided(hist, (2, 2, 2, *map((-1).__add__, hist.shape)), 2 * hist.strides).mean(axis=(0, 1, 2))
This simply takes the average of the 8 nearest neighbors.