5 Visualizations That Make Machine Learning Make Sense
Machine learning is an expanding field that can feel like magic for outsiders. For practitioners, that magic is math. For everyone else, there are better ways to get started. If you want to become comfortable with machine learning, visualizations are the way to start.
Visualizations make it possible to see what is going on in a machine learning algorithm. The best ones allow some interaction that will help you remember what you learn. Below, I have collected a list that can help you understand machine learning without taking a whole course.
Decision Trees
For your first introduction to machine learning, let’s begin with decision trees, a basic algorithm of machine learning. Based on what they already know about an item, decision trees focus on finding the best questions to ask when categorizing that item, just like a person would. Watch as a tree is trained:
Random Forest
Once you understand decision trees, you can move on to random forest, a very commonly used machine learning algorithm. Random forest combines many decision trees together to make better predictions. When data scientists want to start analyzing new data, random forest is often the first algorithm they use because it works reasonably well for most problems. Full disclosure, I’m one of the authors for this visualization:
Neural Networks
At this point, you are ready to look at the neural networks that underly most of the recent excitement about artificial intelligence. Neural networks are inspired by the human brain and use many connected “neurons” to make decisions. Start with the TensorFlow visualization of neural networks. Unlike the previous two demos, this one runs machine learning in your browser:
Recurrent Neural Networks
The neural networks shown above are great for solving problems where each data point has a list of unordered descriptions, but they do not work as well when working with ordered data like sentences. Go play with a neural net that tries to predict what you will write next:
Everything Else
Unfortunately, not every common machine learning algorithm has a visualization. Until people make one for each technique, we have to work with static visualizations that compare algorithms. To see how a variety of machine learning algorithms classify the same data, take a look at this comparison:
Next Steps
At this point, you may know more about machine learning than you will ever need, but if this list has inspired you to delver deeper into machine learning, you can move on to online courses, books, or tutorials. Also, explore distill.pub for more fun, useful visualizations!
