Fig. 7
From: Stochastic block models with multiple continuous attributes
Collaborative filtering in the protein interaction network. For each of the 82 nodes, we fit a model to the remaining 81 node network and given the node’s closest neighbors (based on network connectivity) sought to predict its 6-dimensional attribute vector. The reported error is the root mean squared error (denoted by \(\mathcal {E}\)) between the difference between the true attribute vector (xi) and its predicted attribute vector \((\hat {\textbf {x}}_{i})\). The mean error in xi using the attributed SBM is 0.74, as opposed to the neighborhood average, weighted neighborhood average, and SBM, resulting in errors of 0.74, 0.62, 0.62, and 0.94, respectively