anomalybench

anomalybench bundles anomaly detection algorithms, loaders for benchmark datasets and a command-line runner so that detectors can be compared on the same data under the same protocol. The base install covers classical detectors, ARIMA forecasting, graph detectors and the benchmark workflows; optional extras add deep-learning detectors (PyTorch and TensorFlow), streaming detectors (River) and Prophet-based forecasting. The bundled datasets span tabular, image, time-series and graph data.

Install

1
pip install anomalybench

The project targets Python 3.12 only for now; 3.13 is blocked until the runtime and dependency stack are validated there. Extras: deep, streaming, forecasting and all-detectors.

Package Metadata

  • Current release: 0.6.1
  • Requires Python: >=3.12,<3.13
  • Status: alpha

Where It Fits

Use it when the question is "which detector, on which kind of data, under which budget" rather than "how do I run one detector". It pairs with the anomaly-detection articles on this site, which lean on the same distinction between point, contextual and collective anomalies, and with cfad for the distribution-shape end of the problem.

Loading mathematical content