Graph Entropy

An information entropy computed from graph structure, treating samples as nodes and similarities as edge weights. It considers both local neighborhood structure and global distribution, and is used to measure the uniformity and pattern richness of sample distribution in embedding space.