online-ml/river

Implements single-pass learning algorithms optimized for concept drift and streaming data, supporting linear models, decision trees, anomaly detection, and time series forecasting without requiring historical data access. Features composable pipelines with integrated feature preprocessing, drift detection, and progressive validation for production-like event-driven workflows. Designed for compatibility with Python's broader ML ecosystem while prioritizing per-sample efficiency over batch processing throughput.

5,746 stars and 118,463 monthly downloads. Used by 5 other packages. Actively maintained with 53 commits in the last 30 days. Available on PyPI.