Faster Seeding than K-means++ several orders of magnitude using binary tree. The drop in quantization error is insignificant.

Tree K-means implementation of seeding before Lloyd's algorithm.

Tree K-means implementation runs in O(log(n)kd) time in comparison to K-means++ which runs in O(nkd) while at the same time not requiring significant memory overhead. This is a huge computational advancement as seeding for large datasets is extremely computationally expensive.

Written in Cython using Numpy dependency and compiled using distutils. You can change the dataset in k_tree_means.pyx file and modify compiler directives to Cython in setup.py.

Then one can compile it in command line python setup.py build_ext --inplace within the directory and your package is built.

I also built the package and commited it, you can see it in build folder, just for the sake of an example.