It says on the Github readme that ball tree works with any metric; I was using it with a custom metric over discrete vectors (like hamming distance) and noticed that it may be generating centroids that are not discrete; thus, such centroid may be equidistant to all points in hamming space.
It says on the Github readme that ball tree works with any metric; I was using it with a custom metric over discrete vectors (like hamming distance) and noticed that it may be generating centroids that are not discrete; thus, such centroid may be equidistant to all points in hamming space.