Analyze and Visualize Network Data

This skill guides graph creation, manipulation, analysis, and visualization with NetworkX in Python. It covers relationships between entities in social, biological, transportation, citation, and knowledge networks. The workflow starts with creating or loading a graph, examines its structure, computes relevant measures, draws the network, and exports the graph or analysis results.

Graph selection distinguishes undirected Graph structures from directed DiGraph structures. MultiGraph and MultiDiGraph support multiple edges between nodes. Nodes can be hashable Python objects, including numbers, strings, tuples, and custom objects. Node and edge attributes can hold weights, types, or relationships. The guidance includes examining neighbors, checking whether nodes and edges exist, modifying structures, and working with subgraphs. Removing a node also removes its incident edges.

Analysis spans shortest paths, degree and betweenness centrality, closeness centrality, PageRank, clustering, connectivity, connected components, and community detection. Its algorithm references also cover flows, matching, trees, coloring, isomorphism, and traversal. These operations address different structural questions: routes through the network, relative node importance, connected groups, and local clustering.

Synthetic graph generation includes complete and cycle graphs, known examples such as the karate club graph, grids, trees, and random network models. Examples include Erdős-Rényi, Barabási-Albert, and Watts-Strogatz networks for testing, simulation, or modeling.

Use the skill when a task involves graph data structures, network comparison, graph metrics, or drawing relationships. Random generators suit experiments requiring synthetic networks, while graph I/O guidance helps choose formats for existing data. Select the graph type to match directionality and multiple-edge requirements before applying analysis.

Large networks require attention to memory because graph data is loaded into memory each time a script runs. The guidance recommends sparse matrices for large sparse graphs, loading necessary subgraphs, efficient or compressed formats, and approximate algorithms for very large networks. Floating-point results are inherently approximate, which can affect minimum and maximum computations. Set random seeds for reproducible graph generation and force-directed layouts. Treat results as subject to environment-specific validation and expert review.

Inputs can be manually added nodes and edges, edge lists, GraphML, GML, JSON, CSV, adjacency matrices, or tabular relationship data. Outputs include graph structures, paths, centrality measures, clustering values, communities, saved graph files, and metric tables. GraphML preserves attributes; metric results can be exported as CSV.

Create an attributed graph:

import networkx as nx

# Create empty graph
G = nx.Graph()

# Add nodes (can be any hashable type)
G.add_node(1)
G.add_nodes_from([2, 3, 4])
G.add_node("protein_A", type='enzyme', weight=1.5)

# Add edges
G.add_edge(1, 2)
G.add_edges_from([(1, 3), (2, 4)])
G.add_edge(1, 4, weight=0.8, relation='interacts')

Pandas supports conversion between DataFrames and graph edge lists. NumPy arrays and SciPy sparse matrices provide matrix representations. Matplotlib drawing examples customize node colors, centrality-based sizes, edge widths, labels, and spring, circular, Kamada-Kawai, or spectral layouts. Publication-oriented examples export PNG and PDF figures. The visualization reference also covers interactive Plotly and PyVis views, three-dimensional networks, and bipartite layouts. Dedicated references organize graph basics, algorithms, generators, I/O, and visualization guidance.

NetworkX uses the 3-clause BSD license.