yzhao062 / pyod / 23997194235

94%

master: 94%

push

github

feat: add EmbeddingOD for multi-modal anomaly detection via foundation model embeddings

Add EmbeddingOD wrapper that chains embedding encoders (sentence-transformers,
OpenAI, HuggingFace, or any callable) with any of PyOD's 50+ detectors,
enabling text and image anomaly detection through the standard PyOD API.

Implements the two-step approach validated by NLP-ADBench (EMNLP 2025):
pretrained embeddings + classical OD outperforms end-to-end methods.

Includes benchmark-informed presets (for_text, for_image), encoder registry
with string resolution, optional PCA dimensionality reduction, and
setup.py extras for optional dependencies.

464 of 566 new or added lines in 7 files covered. (81.98%)

11432 of 12110 relevant lines covered (94.4%)

10.38 hits per line