Chapter 9: Drift and Drift Detection
Throughout the previous chapters, you have discovered plenty of ways to build machine learning (ML) models that work in an online manner. They are able to update their learned decision rules from one single observation rather than having to retrain completely as is common in most ML models.
One reason that this is great is streaming, as these models will allow you to work and learn continuously. However, we could argue that a traditional ML model can also predict on a single observation. Even batch learning and offline models can predict a single new observation at a time. To get more insight into the added value of online ML, this chapter will go in depth into drift and drift detection.
To get to an improved understanding of those concepts, the chapter will start with an in-depth description of what drift is. You will then see different types of drift, including concept drift, data drift, and retraining strategy issues.
After that, you...
