Convenience function for near neighbor distances about verde

[leouieda]

Using k=[2] doesn't always work because sometimes the nearest neighbors are too close (like in flight line data). It might be best to do a mean/median of the k nearest neighbors instead:

def neighbor_distances(coordinates, projection=None, k_nearest=2, reduction="median"):
    points = np.transpose([coordinates])
    # Get the second closest (the closest are points themselves)
    tree = cKDTree(points)
    k_distances, labels = tree.query(points, k=k_nearest)
    distances = getattr(np, reduction)(k_distances, axis=1)
    return distances

spacing = neighbor_distances(coordinates).mean()

This way, if k_nearest=2 then it would be the original nearest neighbor distance. But if k_nearest=10, we'll take the median of the 10 nearest neighbors to each point.

[leouieda]

It's probably not worth the trouble to make this too complicated in terms of custom reduction functions. The median distances to k nearest neighbors is a good robust metric and we should stick to that. It's easy to explain as well.

def median_distance(coordinates, k_nearest=1, projection=None):
    ...
    return distances