Summary

In this chapter, we've discussed online learning. We've talked about some of the advantages of online learning methods:

• They are efficient and can handle high-speed throughput
• They can work on very large datasets
• And they can adjust to changes in data distributions

Concept drift is a change in the relationship between data and the target to learn. We've talked about the importance of drift, which is that the performance of a machine learning model can be strongly affected by changes to the dataset to the point that a model will become obsolete (stale).

Drift detectors don't monitor the data itself, but they are used to monitor model performance. Drift detectors can make stream learning methods robust against concept drift, and in River, many adaptive models use a drift detector for partial resets or for changing learning parameters. Adaptive models are algorithms that combine drift detection methods to avoid the degradation...