Using scikit-learn with other scientific Python tools

Suppose you want to apply machine learning to a problem in Python. You need algorithms that learn patterns from data, and you need to understand how each algorithm is meant to be used. Writing every algorithm yourself is slow and error-prone, so practitioners usually rely on a library: a collection of tested, reusable tools that someone else has already built. scikit-learn is one such library.

This article answers three questions before you go further into the chapter: what scikit-learn is, what it provides, and why it is worth knowing about. It also shows you how to use the learning resources that come with it, and how the library fits into a larger working setup.

Before looking at what the library contains, it helps to see the order in which you should engage with it. The source gives a concrete instruction: while you read the chapter, you should also browse the scikit-learn user guide and API documentation for more detail and for the many options available for each algorithm. Learning what the library offers comes first. Getting it onto your own system is a separate step that makes it usable.

The phrase open source tells you something about access, not only about cost. Because the source code is available to anyone, you are not limited to trusting a black box. If you want to know what a particular step is doing behind the scenes, the code that performs that step can, in principle, be read. This transparency is one of the features that makes the library valuable to people who want to understand machine learning rather than just call functions that produce results.

Open source projects of this kind are also not frozen at one moment in time. The source describes scikit-learn as constantly being developed and improved, with a very active user community. Improvements arrive continually, and the people who use the library also take part in shaping it.

scikit-learn offers two main things. The first is a set of machine learning algorithms. The source describes them as state-of-the-art, and it does not list every specific algorithm the library contains. The second is documentation: each algorithm is accompanied by comprehensive documentation describing that algorithm.

The two parts belong together. The algorithms give you working tools, and the documentation explains what each tool is for and what options it offers. If you want a complete list of the algorithms, or the detailed options for one of them, the documentation is the right place to look, which is why the source recommends it so strongly.

The source notes that scikit-learn works well with a number of other scientific Python tools, which the chapter discusses later. Keep this in mind as you learn: scikit-learn is usually one piece of a larger working setup rather than a complete environment by itself. The source does not describe the details of these connections here, so this article does not claim more than that. The chapter is where those details are covered.

Other scientific Python tools that a learner may meet alongside scikit-learn include NumPy, pandas, SciPy, and matplotlib. Read the diagram below as a map of relationships: scikit-learn sits among these tools, and it works with them rather than replacing them.

The learning resources that accompany scikit-learn are the user guide and the API documentation. The source describes the online documentation as very thorough. It also explains that this chapter supplies the machine learning prerequisites needed to understand that documentation in depth. The two work together: the chapter gives you the background, and the documentation gives you the detail for each algorithm.

The source describes scikit-learn as a very popular tool and as the most prominent Python library for machine learning. It is widely used in both industry and academia. Those two settings have different goals. Companies usually want models that solve business problems, while universities and research groups often want to study and compare methods.

A library that serves both audiences has to be accurate, well documented, and easy to learn from. Several features described in the source line up with those needs: the algorithms are described as state-of-the-art, each one comes with comprehensive documentation, the code is open for anyone to examine, and the project keeps improving with an active community. Taken together, these explain why it is worth knowing about before you go further.