I want to calculate the k-nearest neighbors using either sklearn, scipy, or numpy but from a rectangular distance matrix that is output from scipy.spatial.distance.cdist.