yzhao062 / pyod / 24023212910

94%

master: 95%

push

github

feat: add MultiModalEncoder and MultiModalOD for multi-modal anomaly detection

MultiModalEncoder (early fusion): encodes each modality with its own
encoder, concatenates embeddings. Supports per-sample missing data via
mean imputation from training embeddings, passthrough for numeric
features, and per-modality weights.

MultiModalOD (score fusion): runs a separate detector per modality,
standardizes scores with training-time scalers, combines via
average/maximization/median. Missing modalities at test time are
imputed with the training mean score (0 after standardization).

Both classes handle missing data gracefully and integrate with
EmbeddingOD for text/image modalities. 81 tests, 5 Codex review rounds.

303 of 306 new or added lines in 4 files covered. (99.02%)

11733 of 12414 relevant lines covered (94.51%)

10.4 hits per line