river

Overview

River implements online machine learning algorithms in Python, where models update one sample at a time through learn_one and predict_one. It covers linear models, decision trees, nearest neighbors, anomaly and drift detection, clustering, bandits and time series forecasting, plus metrics and progressive validation. It resulted from merging creme and scikit-multiflow.

Who it is for: Python developers doing ML on streaming data or with concept drift

What it needs

• no GPU
• Needs Python 3.11+, pandas (optional, for mini-batch)
• Models: linear models, decision trees, random forests, nearest neighbors, clustering