Docker containers allow developers to isolate and package certain code and some parts of the surrounding operating system as a binary file. This file can then be run on any other machine with a similar OS, with no changes needed. Because Docker images do not include the whole OS, they are relatively small (a few gigabytes in size) and can be pulled over the internet. At the same time, they are fairly isolated and can be run with little exposure. Multiple containers can run on the same machine at once. Using Docker software, an image can be compiled into a set of layers, similar to how classes inherit from each other. For this compilation, you should use a short text file, commonly called a Dockerfile, that's convenient to store in a Git repository. 

Compared to, say, Python packages, Docker containers may be a better way to deliver your code to a customer as they won't need to install anything, except the image. In fact, many CI/CD and web services use Docker internally.

For more information on Docker and its application for data-driven applications, please refer to this video by Andy Terrel: https://www.youtube.com/watch?v=i8vrWFZW2xk.

In the repository for this book, you will find a Dockerfile that was used to generate a corresponding image—one that has Python 3.7, all the Python packages that we used throughout this book, and the code from the repository, installed. Alternatively, you can download this image instead of Python.