Networkx

You'll reach for this when you need to analyze relationships between entities, whether that's social networks, citation graphs, biological pathways, or transportation systems. It handles the full workflow: creating graph structures with attributes, running algorithms like PageRank and shortest paths, detecting communities, and visualizing networks with matplotlib. The documentation covers four graph types (directed, undirected, multi-edge variants), synthetic network generators for testing (Barabási-Albert, Watts-Strogatz), and I/O for various formats including GraphML and pandas DataFrames. The real strength is the breadth of algorithms available out of the box, from centrality measures to flow problems, which saves you from implementing graph theory fundamentals yourself.