Explore different random selection methods in graphical Voronoi #5
Description
Currently, the graphical_voronoi feature type chooses a simple random sub-sample of points from which to define the regions, and thereby the categorical graph for that particular node. However, there may be value in choosing points which are "more spaced out" -- for example, using similar techniques as those used to initialize K-means algorithms. This should be implemented and explored to see if it improves performance.