Figure 8
From: Bayesian inference of transition matrices from incomplete graph data with a topological prior
Effects of partial constraints on the downstream task of clustering. The left panel shows results on Erdős Rényi graphs, the middle panel shows results on geometric Euclidean random graphs, and the right panel shows results on geometric hyperbolic random graphs. We measure performance with the Adjusted Mutual Information (AMI) between the clusters obtained from the ground truth and the inferred transition matrices. Error bars represent the 95% confidence interval. The positive impact of constraints on network inference translates to the clustering task