Figure 1
A Network State Space was made from the 3000 simulated networks (1500 directed and undirected each, across the five mechanisms, each with 20 nodes) by pairwise comparing 18 properties of each network in a stacking ensemble method. A two-dimensional NMDS projection (Nonmetric Multidimensional Scaling ordination, calculated using the metaMDS function in the R package vegan25) of this state space (with identical axis scaling) shows that networks cluster by mechanism. In this plot points are each a network, with their radius proportional to the governing parameter. Circles are undirected networks and triangles are directed networks. Classification ROC curves and corresponding AUC values (inset numbers) for the ability to classify these 3000 networks. When AUC = 1 the classifier can identify all true positives without including any false positives. Random classifiers produce AUC = 0.5. Each network was classified with an independently simulated state space. Empirical Classifications of 1191 empirical networks spanning 18 kinds of systems. Each circle represents a mechanism. Inset numbers are the number of networks classified as that mechanism. If a network was classified as more than one mechanism the resulting pie graph was made with equal splits for each mechanism because we cannot yet confidently assign probabilities to mixture networks as shown in Fig. 4.
