Overview
PyOD is a Python library designed for detecting anomalies or outliers in multivariate data. Established in 2017, it offers over 50 detection algorithms ranging from classical methods like LOF to cutting-edge techniques like ECOD and DIF. The library is engineered for high performance, utilizing `numba` and `joblib` for JIT compilation and parallel processing, enabling fast training and prediction through the SUOD framework. Version 2 incorporates a PyTorch-based framework for deep learning models, expanding its capabilities. PyOD also leverages LLM-based model selection to automate tuning. It integrates with ADBench for comprehensive benchmarking and is compatible with distributed systems like Databricks. The library supports various probabilistic, linear, and proximity-based models, providing a unified interface for outlier detection tasks.