Building Machine Learning Powered Applications

Part I. Find the Correct ML Approach

Most individuals or companies have a good grasp of which problems they are interested in solving—for example, predicting which customers will leave an online platform or building a drone that will follow a user as they ski down a mountain. Similarly, most people can quickly learn how to train a model to classify customers or detect objects to reasonable accuracy given a dataset.

What is much rarer, however, is the ability to take a problem, estimate how best to solve it, build a plan to tackle it with ML, and confidently execute on said plan. This is often a skill that has to be learned through experience, after multiple overly ambitious projects and missed deadlines.

For a given product, there are many potential ML solutions. In Figure I-1, you can see a mock-up of a potential writing assistant tool on the left, which includes a suggestion and an opportunity for the user to provide feedback. On the right of the image is a diagram of a potential ML approach to provide such recommendations.

Figure I-1. From product to ML

This section starts by covering these different potential approaches, as well as methods to choose one over the others. It then dives into methods to reconcile a model’s performance metrics with product requirements.

To do this, we will tackle two successive topics:

Chapter 1

By the end of this chapter, you will be able ...