networkx

Use nx.from_pandas_edgelist with your DataFrame, specifying the source and target columns and optionally edge attributes like weight. This builds a graph directly from tabular edge data without manual add_edge loops.

Call nx.shortest_path with source and target nodes, optionally passing a weight attribute for weighted graphs. For all pairs, use nx.all_pairs_shortest_path, and for negative weights use the Bellman-Ford variant.

NetworkX reads and writes edge lists, adjacency lists, GraphML, GML, GEXF, JSON node-link data, and Pajek formats. It also converts to and from pandas DataFrames, NumPy arrays, and SciPy sparse matrices.

NetworkX loads entire graphs into memory, so very large networks can be slow. Use approximate algorithms with the k parameter for centrality, sparse matrix representations, and fast generators like fast_gnp_random_graph for large sparse graphs.

Import the community module from networkx.algorithms and call greedy_modularity_communities for modularity-based detection or label_propagation_communities for a faster approach. The result is a list of node sets you can color in visualizations.

Force-directed layouts like spring_layout use random initialization, so positions change between runs. Pass a fixed seed parameter, such as nx.spring_layout(G, seed=42), to get reproducible layouts.