pyod

What it is and what it does

PyOD is a comprehensive anomaly detection library that provides 61 detectors spanning tabular, time series, graph, text, image, and audio data under a single API. It supports three usage layers: a classic fit/predict interface for direct detector selection, an ADEngine orchestration layer that automatically selects and compares detectors, and an agentic investigation layer where AI agents drive detection workflows through natural language. The library depends on numpy, scipy, scikit-learn, joblib, matplotlib, and numba for numerical computation, parallel training, and JIT acceleration.

PyOD is built for both traditional machine learning workflows and modern AI-agent integration. It exposes detectors through a standard scikit-learn-compatible API, offers an MCP server for LLM-compatible agents, and provides a Claude Code skill for agentic anomaly investigations. The library has been actively maintained since 2017, supports Python 3.9 through 3.13, and carries no known security vulnerabilities.

Use it for:

• Detect fraudulent transactions or pricing anomalies in financial data using tabular detectors.
• Identify equipment failures or system anomalies in time-series sensor or telemetry data.
• Automate anomaly detection workflows by letting an AI agent choose and tune detectors via natural language.
• Detect out-of-distribution or novel samples in image or text datasets for content moderation.
• Build ensemble anomaly detection systems by comparing multiple detectors on the same dataset.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

PyOD detects anomalies and outliers across tabular, time series, graph, text, image, and audio data using 61 detectors, with optional agentic workflows for AI-driven investigation.

Yes. PyOD is production-stable (BSD-2-Clause, no vulnerabilities, actively maintained), widely adopted (5M+ monthly downloads), and offers genuine breadth—61 detectors across six data modalities with both classical and agentic APIs. Install if you need anomaly detection; the low dependency friction and backward-compatible API make it a safe choice.

Install

pyod on PyPI

pip install pyod

uv add pyod

poetry add pyod

Installing pyod

Before you install

Low friction: pure Python wheel with six well-established scientific dependencies (numpy, scipy, scikit-learn, joblib, matplotlib, numba). Active maintenance—released 12 days ago with 9958 GitHub stars and no known vulnerabilities.

License in practice

BSD-2-Clause permissive license allows commercial and private use with minimal restrictions; attribution required but no copyleft obligations.

Quickstart

pip install pyod

from pyod.models.iforest import IForest
clf = IForest()
clf.fit(X_train)
y_test_scores = clf.decision_function(X_test)

Verify before relying

• Whether the 61 detectors cover all claimed modalities equally or if some modalities have fewer detector options.
• Performance characteristics and scalability limits for large datasets or real-time streaming scenarios.
• Whether the agentic workflow (od-expert skill, MCP server) is production-ready or still experimental.

Package facts

Evidence: pyod-3.6.4-py3-none-any.whl

Keywords: anomaly detection, outlier detection, machine learning, deep learning, unsupervised learning, time series anomaly detection, graph anomaly detection, nlp anomaly detection, image anomaly detection, multimodal, agentic ai, foundation models, fraud detection, novelty detection, out-of-distribution detection, outlier ensembles, pytorch, python

Tags

More Artificial Intelligence packages

Further reading