Choosing a Machine Learning Model from a Library

Before you start a machine learning book, it helps to know what kind of help it is trying to give you. Some books walk through the equations behind every algorithm. Others teach you to build each algorithm yourself from first principles. Others focus on getting useful results by applying tested tools. This article explains which of these approaches the book chooses, which it deliberately leaves out, and how you can decide whether its emphasis matches what you need.

The main idea is one of purpose. Read in this way, the book's focus is a question of what you should be able to do after reading it. Its answer is that you should be able to apply machine learning algorithms to problems, choosing and using models that already exist, rather than prove why those algorithms work or code them yourself.

Machine learning books generally fall into one of three camps. The first walks through the equations behind every algorithm. The second teaches you to build each algorithm yourself from first principles. The third concentrates on getting useful results by applying tested, ready-made tools. Each camp serves a different reader, so the first step is to identify which camp a given book belongs to.

The authors say they made a deliberate choice to keep mathematics in the background and concentrate on the practical side of using machine learning algorithms. The goal is to help you apply these algorithms to problems, not to prove why they work in full mathematical depth. Your reading goal is therefore practical: understand which algorithm suits a problem and how to apply it.

The diagram below compares what the book covers with what it leaves out. Use it as a map of the book's scope before you read the chapters.

The book deliberately sets aside two things: detailed mathematical analysis and writing algorithms from scratch. It does not take the analysis of each algorithm down to great detail, and it does not teach you to implement learning methods yourself.

The authors point readers who want the mathematics to a separate resource. They recommend The Elements of Statistical Learning by Trevor Hastie, Robert Tibshirani, and Jerome Friedman, which the source says is available for free at the authors' website. The book hands off mathematical depth to a dedicated text rather than trying to cover everything itself.

The second part of the book's focus is how it expects you to work with algorithms. Rather than teaching you to write each learning method yourself, the book concentrates on using the large collection of models already implemented in scikit-learn and other libraries. The practical skill being developed is choosing and applying an available model to a task.

The example below illustrates the working style the book assumes. It does not implement a learning algorithm. It obtains a model that scikit-learn already provides, and that is the kind of object this book works with throughout. This example was written for illustration and is not taken from the source.