PySAD
What it is
PySAD is a free, open-source Python framework for detecting anomalies in streaming data. Its online models update as each new instance arrives, supporting sequential detection for univariate and multivariate data. The repository describes 16 online detectors, including xStream, LODA, RS-Hash, Half-Space Trees, and Robust Random Cut Forest. PySAD also includes stream simulators, evaluators, preprocessors, statistic trackers, postprocessors, and probability calibrators. Integrations allow batch anomaly detectors from PyOD to run in a streaming setting. Experiments can use supervised, semi-supervised, or unsupervised methods. The documentation notes that streaming approaches may retain one instance or a small recent window to limit memory and processing needs. Installation is available through pip or the GitHub source; the current README lists Python 3.10 or later on Linux, macOS, and Windows. PySAD is self-hosted and distributed under the BSD 3-Clause license. The README says 17 labeled benchmark datasets download on first use. The maintainer welcomes issues and pull requests, while general questions and project demonstrations belong in GitHub Discussions.
For and against
For
- Models update as each data instance arrives
- Includes 16 online detectors
- Connects PyOD batch detectors to streaming use
- Supports univariate and multivariate data
- BSD 3-Clause licensed
Against
- Requires Python 3.10 or newer
- Self-hosted deployment
- 17 benchmark datasets download on first use
The verdict
PySAD focuses on sequential anomaly detection and includes tools for preparing and evaluating streaming workflows. Its Python requirement and self-hosted deployment are worth considering before adoption.
Who it suits
PySAD suits Python users building online anomaly-detection workflows for streaming univariate or multivariate data. It supports experiments across supervised, semi-supervised, and unsupervised settings.
Plans and prices
PySAD
Open-source Python framework
Read 2 Oct 2026 · github.com
On the record
- Benchmarks
- The README says 17 labelled benchmark datasets are downloaded on first use.github.com · read 2 Oct 2026
- Data and learning settings
- The framework supports models for univariate and multivariate data and experiments in supervised, semi-supervised, and unsupervised settings.pysad.readthedocs.io · read 2 Oct 2026
- Detectors
- The repository describes 16 online detectors, including xStream, LODA, RS-Hash, Half-Space Trees, and Robust Random Cut Forest.github.com · read 2 Oct 2026
- Evaluation tools
- PySAD includes stream simulators, evaluators, preprocessors, statistic trackers, postprocessors, and probability calibrators.pysad.readthedocs.io · read 2 Oct 2026
- Installation
- The project can be installed with pip or from its GitHub source, and its current README lists Python 3.10 or newer on Linux, macOS, and Windows.github.com · read 2 Oct 2026
- License
- The project is distributed under a BSD 3-Clause license.github.com · read 2 Oct 2026
- Maintainer
- The GitHub profile identifies Selim Firat Yilmaz as an AI researcher based in London, UK.github.com · read 2 Oct 2026
- Online detection
- Its models update as each new data instance arrives for online or sequential anomaly detection.pysad.readthedocs.io · read 2 Oct 2026
- Purpose
- PySAD is an open-source Python framework for anomaly detection on streaming multivariate data.pysad.readthedocs.io · read 2 Oct 2026
- PyOD integration
- PySAD provides integrations that let batch anomaly detectors from PyOD run in a streaming setting.pysad.readthedocs.io · read 2 Oct 2026
- Resource use
- The documentation says streaming methods may store only an instance or a small window of recent instances to meet memory and processing constraints.pysad.readthedocs.io · read 2 Oct 2026
- Security
- The repository links to a security policy, but the policy page could not be opened during this research.github.com · read 2 Oct 2026
- Support
- The maintainer says issues and pull requests are welcome and aims to reply within a few days; questions and show-and-tell go in GitHub Discussions.github.com · read 2 Oct 2026
