Assuming the changes suggested in pull request #10 are acceptable, we can remove the dependency on SciPy entirely just by reimplementing the get_nearest_neighbor_distances function. This will facilitate goftests being runnable on PyPy, which does not yet have SciPy support.

Here is a short, pure Python nearest-neighbor function backed by a k-d tree: http://code.activestate.com/recipes/577497-kd-tree-for-nearest-neighbor-search-in-a-k-dimensi/. It is published under an MIT license. Since we don't need any other behavior from the k-d tree (we just need to find the distance of each point to its nearest neighbor), I think incorporating that code is the best approach.