Embedding Bias Direction (PCA)

This method extracts a bias direction from group prototypes using PCA and measures the projection gap across groups. Large gaps indicate systematic embedding bias.

Usage

from shortcut_detect.geometric import BiasDirectionPCADetector

detector = BiasDirectionPCADetector()
detector.fit(embeddings, group_labels)
print(detector.report_)

Outputs

• projection_gap: max–min projection across groups
• explained_variance: variance explained by the bias direction
• group_projections: per-group projection means and support

Reference

Bolukbasi et al. 2016: https://arxiv.org/abs/1607.06520