Manipulating Graphs with NetworkX

NetworkX is a powerful pure-Python library for creating, manipulating, and analyzing graphs. This guide covers essential tasks using NetworkX.

Inspecting Directed Graphs: In-Edges and Out-Edges

For directed graphs (nx.DiGraph), you may want to know which edges point into or out of a node.

Example:

import networkx as nx

# Create a directed graph
G = nx.DiGraph()

# Add some edges
G.add_edges_from([(1, 2), (2, 1), (2, 3), (3, 1)])

node = 1

# In-edges: edges ending at this node
in_edges = G.in_edges(node)
print("In-edges:", list(in_edges))    # Output: [(2, 1), (3, 1)]

# Out-edges: edges starting from this node
out_edges = G.out_edges(node)
print("Out-edges:", list(out_edges))  # Output: [(1, 2)]

• G.in_edges(node) returns edges directed into node
• G.out_edges(node) returns edges directed out of node

Visualizing Graphs with PyGraphviz and Graphviz

NetworkX can interface directly with PyGraphviz and Graphviz, popular software packages for graph visualization.

Installing Graphviz and PyGraphviz

Before you can use these visualization features, you need to install both PyGraphviz and Graphviz. On most systems, the easiest way is:

conda install -c conda-forge graphviz pygraphviz

Rendering Graphs as Images

Once you have everything installed, you can easily convert a NetworkX graph into a Graphviz object and save it directly as an image (PNG, PDF, etc.):

import networkx as nx

# Create or load a graph
G = nx.DiGraph()
G.add_edges_from([(1, 2), (2, 1), (2, 3), (3, 1)])

# Convert to a pygraphviz AGraph
A = nx.nx_agraph.to_agraph(G)

# Write to DOT format (optional, for inspection/sharing)
A.write("G.dot")

# Render and save as a PNG image
A.draw("G.png", prog="dot")

• nx.nx_agraph.to_agraph(G) converts your NetworkX graph to a PyGraphviz (Graphviz) AGraph object.
• .write("G.dot") saves the graph in DOT format (plain text for Graphviz).
• .draw("G.png", prog="dot") renders the image using the Graphviz “dot” layout engine.

You can open G.png in any image viewer to see the visualization.

Tip: Try changing the prog argument to other Graphviz layout engines like "neato" or "fdp" for different layouts.

Further Customization

You can customize node shapes, colors, and more using AGraph’s methods or by passing attributes on creation; see the pygraphviz documentation for more details.

Serializing and Deserializing Graphs

When working with graphs, you may want to save (serialize) them to a file and load (deserialize) them later. There are two common approaches in NetworkX:

A. Using JSON (Node-Link Data)

Why?

• Human-readable
• Cross-language interoperability

Serialize to JSON:

import networkx as nx
import json

G = nx.path_graph(4)
data = nx.node_link_data(G)

with open("graph.json", "w") as f:
    json.dump(data, f)

Deserialize from JSON:

with open("graph.json", "r") as f:
    data = json.load(f)

G = nx.node_link_graph(data)

Tip: JSON is a popular serialization format and can be easily used with web apps or other languages.

B. Using Pickle

Why?

• Fastest method in Python
• Carries all Python-specific information
• Warning: Not safe for untrusted data, and not interoperable with other languages.

Serialize using Pickle:

import networkx as nx
import pickle

G = nx.cycle_graph(5)

with open("graph.gpickle", "wb") as f:
    pickle.dump(G, f)

Deserialize using Pickle:

with open("graph.gpickle", "rb") as f:
    G = pickle.load(f)