Foreword

Machine learning (ML) has become big business in the last few years: companies are using it to make money, applied research has exploded in both industrial and academic settings, and curious developers everywhere are looking to level up their ML skills. But this newfound demand has largely outrun the supply of good methods for learning how these techniques are used in the wild. This book fills a pressing need.

Applied machine learning comprises equal parts mathematical principles and tricks pulled from a bag—it is, in other words, a true craft. Concentrating too much on either aspect at the expense of the other is a failure mode. Balance is essential.

For a long time, the best—and the only—way to learn machine learning was to pursue an advanced degree in one of the fields that (largely separately) developed statistical learning and optimization techniques. The focus in these programs was on the core algorithms, including their theoretical properties and bounds, as well as the characteristic domain problems of the field. In parallel, though, an equally valuable lore was accumulated and passed down through unofficial channels: conference hallways, the tribal wisdom of research labs, and the data processing scripts passed between colleagues. This lore was what actually allowed the work to get done, establishing which algorithms were most appropriate in each situation, how the data needed to be massaged at each step, and how to wire up the different parts of the pipeline.