Papers

[matanox]

What would be the best way to systematically learn the science behind this package? It might be a less obvious topic to find good reads about than e.g. the latest neural network architecture.

[MaxHalford]

Great question! As you say, online machine learning is not as popular as other subfields of machine learning, so there's much less papers and comprehensive reading.

Each River model has a References section where we list the relevant papers. That's the best we can do for now.

[matanox]

Thanks again for developing and maintaining this library and the two which are being merged/integrated into it! For some reason I can't click the "Mark as answer" new widget that Github place here, but this question can be closed obviously!

[matanox]

Just as one case in point though, is there any special paper or technique deeply involved in the incremental version of the logistic regression implementation?

[MaxHalford]

For logistic regression, I would refer to this part of scikit-learn's documentation. River's implementation produces the same results :)

[jacobmontiel]

Some resources that should provide some foundations:

• The “Machine Learning for Data Streams with Practical Examples in MOA” textbook is a resource intended to help students and practitioners enter the field of machine learning and data mining for data streams.

• "A survey on concept drift adaptation". Although a little outdated, it is a great introduction to concept drift.

• "Machine learning for streaming data: state of the art, challenges, and opportunities"

This list is by no means complete, the field of data streams is continuously evolving (pun intended) and we expect it to keep growing.