Fig. 3
From: Manifold learning and maximum likelihood estimation for hyperbolic network embedding
Benchmarking on artificial networks (γ=2.5). Artificial networks with N=500 nodes, average node degree 2m=10, scaling exponent γ=2.5 and temperature T={0,0.3,0.6,0.9} were embedded to hyperbolic space with LaBNE, HyperMap and LaBNE+HM. a Real vs inferred radial coordinates in the four networks. Only LaBNE’s coordinates are shown, because all the methods follow the same strategy to infer them. b Connection probabilities as a function of hyperbolic distances measured with the coordinates inferred by each method. c Greedy routing efficiency when the inferred hyperbolic coordinates are used as addresses to send packets between 1000 randomly selected source-target pairs. d Hop stretch of successful packet deliveries for the considered source-target pairs. Red diamonds indicate the average hop stretch. e Time needed by each method to embed the networks to hyperbolic space