This book is not a book about data science. It is a book about how to use Scala, a programming language, for data science. So, where does programming come in when processing data?

Computers are involved at every step of the data science pipeline, but not necessarily in the same manner. The style of programs that we build will be drastically different if we are just writing throwaway scripts to explore data or trying to build a scalable application that pushes data through a well-understood pipeline to continuously deliver business intelligence.

Let's imagine that we work for a company making games for mobile phones in which you can purchase in-game benefits. The majority of users never buy anything, but a small fraction is likely to spend a lot of money. We want to build a model that recognizes big spenders based on their play patterns.

The first step is to explore data, find the right features, and build a model based on a subset of the data. In this exploration phase, we have a clear goal in mind but little idea of how to get there. We want a light, flexible language with strong libraries to get us a working model as soon as possible.

Once we have a working model, we need to deploy it on our gaming platform to analyze the usage patterns of all the current users. This is a very different problem: we have a relatively clear understanding of the goals of the program and of how to get there. The challenge comes in designing software that will scale out to handle all the users and be robust to future changes in usage patterns.

In practice, the type of software that we write typically lies on a spectrum ranging from a single throwaway script to production-level code that must be proof against future expansion and load increases. Before writing any code, the data scientist must understand where their software lies on this spectrum. Let's call this the permanence spectrum.