Training Data for Machine Learning

Chapter 2. Getting Up and Running

Introduction

There are many tools available that help us when we work with data: we have databases to smoothly store data and web servers to smoothly serve data. And now, training data tools to smoothly work with training data.

In addition to tools, there are established processes and expectations for how databases integrate with the rest of your application. But what about training data? How do you get up and running with training data? Throughout this chapter, I’ll cover key considerations, including installation, annotation setup, embedding, end user, workflow, and more.

It’s important to note why I referenced smoothly working with training data earlier. I say “smoothly” because I don’t have to use a database. I could write my data to a file and read from that. Why do I need a database, like Postgres, to build my system? Well, because Postgres brings a vast variety of features, such as guarantees that my data won’t easily get corrupted, that data is recoverable, and that data can be queried efficiently. Training data tools have evolved in a similar way.

In this chapter I will cover:

• How to get up and running

• The scope of training data tools

• The benefits you get from using training data tools

• Trade-offs

• The history that got us to where we are today

Most of this is focused on things that will be relevant to you today. I also include some brief sections on history to demonstrate why these tools matter. Additionally, I will also answer ...