Why is data visualization useful for machine learning algorithms?

The discipline of data visualization gives us practically infinite ways to show off what's happening with machine learning algorithms. It's worth thinking about exactly why data visualization is so important, and why it frees up so much creative power for so many people who are getting involved in machine learning processes.

To understand the value of data visualization for machine learning, just take a look at any of the algorithms that are used to create these groundbreaking and innovative programs.

One of the simplest is the decision tree. Without getting into activation functions or hidden layers or anything like that, the decision tree is simply sets of binary nodes. But even the simple decision tree is very difficult for people to describe or write about. It's much easier when it's visualized on a screen or on a page. When you see each node and its connections to other nodes, the whole thing becomes readily apparent.

Now let's take one of the most byzantine and elaborate machine learning algorithm types – the neural network.

In some ways, neural networks are really collections of machine learning algorithms. The basic setup consists of an input layer, hidden layers and an output layer. The activation functions help the individual digital neurons to process weighted inputs.

All of these items and all of these processes are much more easily explained through data visualization than they are through verbal or written descriptions. You can say that a neural network has weighted inputs flowing into an input layer, and that they coalesce in some hidden layer and consolidate into a given output, but when you use a visual figure to show how this works, the human eye and the human brain latch onto that in a much more direct and useful way.

In a sense, you can see the power of data visualization even without taking machine learning into account. Back in the days of linear programming, compilers and computer language studios would give programmers the choice to set a step-by-step test program where they could inspect the values of variables in small visual boxes. Again, this helped show what happens in an execution much better than just reading through a code base.

Machine learning is hyper-intensive programming – it's probabilistic programming and that's why data visualization really helps us to get our heads around what's happening with any given algorithm or process.