Document worth reading: “Metrics for Graph Comparison: A Practitioner’s Guide”

Comparison of graph development is a ubiquitous exercise in data analysis and machine learning, with numerous functions in fields equal to neuroscience, cyber security, social neighborhood analysis, and bioinformatics, amongst others. Discovery and comparability of buildings equal to modular communities, rich golf tools, hubs, and timber in data in these fields yields notion into the generative mechanisms and purposeful properties of the graph. Often, two graphs are in distinction by the use of a pairwise distance measure, with a small distance indicating structural similarity and vice versa. Common choices embody spectral distances (usually often known as $lambda$ distances) and distances based totally on node affinities. However, there has of however been no comparative study of the efficacy of these distance measures in discerning between frequent graph topologies and completely completely different structural scales. In this work, we study typically used graph metrics and distance measures, and present their means to discern between frequent topological choices current in every random graph fashions and empirical datasets. We put forward a multi-scale picture of graph development, whereby the influence of worldwide and native development upon the area measures is taken under consideration. We make ideas on the applicability of varied distance measures to empirical graph data draw back based totally on this multi-scale view. Finally, we introduce the Python library NetComp which implements the graph distances used on this work. Metrics for Graph Comparison: A Practitioner’s Guide