In this chapter, we looked at clustering, which is an unsupervised learning approach. We use unsupervised learning to explore data, rather than for classification and prediction purposes. In the experiment here, we didn't have topics for the news items we found on reddit, so we were unable to perform classification. We used k-means clustering to group together these news stories to find common topics and trends in the data.

In pulling data from reddit, we had to extract data from arbitrary websites. This was performed by looking for large text segments, rather than a full-blown machine learning approach. There are some interesting approaches to machine learning for this task that may improve upon these results. In the Appendix of this book, I've listed, for each chapter, avenues for going beyond the scope of the chapter and improving upon the results. This includes references to other sources of information and more difficult applications of the approaches in each chapter.

We also looked at a straightforward ensemble algorithm, ECA. An ensemble is often a good way to deal with variance in the results, especially if you don't know how to choose good parameters (which is especially difficult with clustering).

Finally, we introduced online learning. This is a gateway to larger learning exercises, including Big data, which will be discussed in the final two chapters of this book. These final experiments are quite large and require management of data as well as learning a model from them.

In the next chapter, we step away from unsupervised learning and go back to classification. We will look at deep learning, which is a classification method built on complex neural networks.