TopolearnReal graphs come from tables and files, and you often need standard test graphs to try ideas on.
Most graphs start as a table of edges. Ways to load them:
- From a list:
nx.Graph([(u, v), ...])ornx.DiGraph(...), with attributes as a third dict element,(u, v, {"weight": 3}). - From pandas:
nx.from_pandas_edgelist(df, source="src", target="dst", edge_attr=["weight"], create_using=nx.DiGraph), and back withnx.to_pandas_edgelist(G). - From a dict of lists or dict of dicts:
nx.Graph({"a": ["b", "c"]}).
File formats in nx.readwrite:
- Edge lists:
nx.read_edgelist(path, nodetype=int, data=[("weight", float)])andnx.write_edgelist. Simple, but node attributes are lost and node IDs come back as strings unless you passnodetype. - GraphML (
read_graphml,write_graphml): XML that keeps node and edge attributes and is readable by Gephi and Cytoscape. - GML, GEXF, Pajek, adjacency lists.
- JSON for web visualization:
nx.node_link_data(G)andnx.node_link_graph(data).
Generators create standard graphs: nx.complete_graph(5), nx.path_graph(4), nx.grid_2d_graph(3, 3), random models like nx.erdos_renyi_graph(n, p, seed=1) and nx.barabasi_albert_graph(n, m, seed=1), and classic datasets like nx.karate_club_graph(). Pass seed= for reproducible random graphs.
Pickle is not recommended for long-term storage; prefer a standard format.
Example
import networkx as nx
import pandas as pd
edges = pd.DataFrame({"src": ["a", "b", "c"], "dst": ["b", "c", "a"],
"weight": [1.5, 2.0, 0.5]})
G = nx.from_pandas_edgelist(edges, "src", "dst", edge_attr="weight",
create_using=nx.DiGraph)
print(G, G["c"]["a"]["weight"]) # DiGraph with 3 nodes and 3 edges 0.5
nx.write_edgelist(G, "g.edgelist", data=["weight"])
H = nx.read_edgelist("g.edgelist", create_using=nx.DiGraph,
data=[("weight", float)])
print(sorted(H.edges(data="weight"))) # [('a', 'b', 1.5), ('b', 'c', 2.0), ('c', 'a', 0.5)]
nx.write_graphml(G, "g.graphml")
print(nx.read_graphml("g.graphml").number_of_edges()) # 3
K = nx.karate_club_graph()
print(K.number_of_nodes(), K.number_of_edges()) # 34 78Remember
from_pandas_edgelistturns edge tables into graphs- GraphML and GEXF keep attributes; edge lists are minimal
- Generators give standard and random test graphs; pass
seed=
Watch out for
- Reading an edge list of integers without
nodetype=int, so nodes '1' and 1 differ.