Summary

This chapter was intended to expose you to several new types of challenging data that, although infrequently found in simple teaching examples, are regularly encountered in practice. Despite popular adages that tell us that “one can’t have too much of a good thing” or that “more is always better,” this is not always the case for machine learning algorithms, which may be distracted by irrelevant data or have trouble finding the needle in the haystack if overwhelmed by less important details. One of the seeming paradoxes of the so-called big data era is the fact that more data is simultaneously what makes machine learning possible and what makes it challenging; indeed, too much data can even lead to a so-called “curse of dimensionality.”

As disappointing as it is to throw away some of the treasure of big data, this is sometimes necessary to help the learning algorithm perform as desired. Perhaps it is better to think of this...