Figure 14
From: Bayesian inference of transition matrices from incomplete graph data with a topological prior
Results of the application of model selection on BaCon and the noninformative Bayesian method with model selection on the powergrid and ecosystem datasets for the tasks of transition matrix inference (first row), node ranking (second row) and clustering (third row). The fourth row illustrates the prior concentration parameters α resulting from the model selection in the powergrid and ecosystem datasets for BaCon . The value for BaCon stabilizes around 0.1 for the ecosystem and around 10 for the powergrid, because ecosystem has biased edge weights, and powergrid has equiprobable edge weights. Hyperparameter for noninformative Bayesian approach prefers lower values of the hyperparameter – and favors biased distributions –, because some edges never appear