I've made this a text post since it seems submitting to your own blog is a no no. Its been a little sparse in here, so I figured I would submit some of my own investigation into Mini Batch k-Means and Elkan's k-Means. Turns out the speed improvements are not quite as good as advertised when compared with more advanced exact k-Means implementations.
Any feed back on other things to investigate in the future is also welcome.
[link][3 comments]