PyOD is a comprehensive Python library for outlier and anomaly detection, offering a unified interface to over 50 classical and deep learning algorithms. It caters to researchers and practitioners needing to identify unusual patterns in multivariate data, providing optimized performance and ease of use.

How It Works

PyOD integrates a wide array of detection algorithms, including probabilistic, linear, proximity-based, ensemble, and neural network models. It leverages Numba and Joblib for JIT compilation and parallel processing, enhancing performance. The library also supports advanced features like the SUOD framework for accelerated training on large datasets and LLM-based model selection for automated tuning.

Quick Start & Requirements

• Install via pip: pip install pyod or conda: conda install -c conda-forge pyod.
• Requires Python 3.8+, joblib, matplotlib, numpy>=1.19, numba>=0.51, scipy>=1.5.1, scikit_learn>=0.22.0.
• Optional dependencies include PyTorch for deep learning models.
• Official documentation: https://pyod.readthedocs.io/en/latest/

Highlighted Details

• PyOD 2.0 includes 12 new PyTorch-based neural models, bringing the total to 45 algorithms.
• Features LLM-based automated model selection.
• Supports distributed systems via Databricks integration.
• Comprehensive benchmark paper and ADBench dataset available for performance comparison.

Maintenance & Community

• Actively maintained with frequent updates and releases.
• Extensive documentation and numerous tutorials available.
• Related projects include ADBench, TODS, and PyGOD.
• Citations are encouraged for the PyOD 2.0 paper and the original JMLR publication.

Licensing & Compatibility

• Released under the MIT License.
• Compatible with commercial and closed-source applications.

Limitations & Caveats

• Saving neural network models may present challenges, with workarounds available in GitHub issues.