Skip to main content
. 2017 May 3;3(5):e1602548. doi: 10.1126/sciadv.1602548

Fig. 3. The neoSBM on synthetic data.

Fig. 3

(A) The stochastic blockmodel likelihood surface shows four distinct peaks corresponding to a sequence of locally optimal partitions. (B) Block density diagrams depict community structure for locally optimal partitions, where darker color indicates higher probability of interaction. (C) The neoSBM, with partition 1 as the metadata partition, interpolates between partition 1 and the globally optimal stochastic blockmodel partition 4. The number of free nodes q and stochastic blockmodel log likelihood as a function of θ show three discontinuous jumps as the neoSBM traverses each of the locally optimal partitions (1 to 4).