I personally prefer to develop my own machine learning solutions. One may, of course, chalk this up to ego. However, in the first chapter, I introduced the notion that there are different types of problems. Some of these problems may be solved by machine learning algorithms. Some problems may only require general machine learning algorithms, while some require specialized algorithms derived from the general algorithms. In the majority of this book, I've shown the general algorithms, and readers are free to adapt these to their own specific problems.

I, too, recognize the value of having general machine learning algorithms as being part of the solution. Imagine that you are developing a program to reorganize your personal photos on your computer. There is no need to spend a protracted amount of time getting a convolutional neural network trained upon a corpus of faces. The main task is to organize the photos, not facial recognition! Instead, one may just use a model that is already trained. These sorts of ready-made solutions are suitable for problems in which the ready-made solution is a small part. Increasingly, there is a demand for such solutions.

As such, many machine learning algorithms are provided now as a service. Amazon Web Services has its own offering, as do Google Cloud and Microsoft Azure. Why did I not choose to introduce those in this chapter? Here's another thing you should know about me: I like to work offline. I find being connected to the internet while working only serves as a distraction—Slack messages, emails, and various other sites compete for my scarce attention. No, I prefer to work and think while offline.

The cloud companies do offer machine learning as a service, and they all require internet access. MachineBox, to its credit, provides a Docker image. A Docker pull is all that is required. A once-off internet connection is required to download the files. But once that's done, the entire workflow may be developed offline—or as is the case for all the code in this chapter, on a plane.

This is MachineBox's main benefit: you are not beholden to a corporate entity that requires an always-on connection to their cloud services. But of course, that's not all. MachineBox is famous for its developer friendliness. That I am able to write the majority of this chapter's code in-flight is testament to their developer friendliness. To be fair, even as a seasoned machine learning library author, facial recognition is still pretty awesome.