A python/rust library for embedding graphs in 2D space, using force-directed layouts.
pip install graph_force
The first parameter defines the number of nodes in graph. The second parameter is an iterable of edges, where each edge is a tuple of two integers representing the nodes it connects. Node ids start at 0.
import graph_force
edges = [(0, 1), (1, 2), (2, 3), (3, 0)]
pos = graph_force.layout_from_edge_list(4, edges)
This library does not have a function to consume a networkx graph directly, but it is easy to convert it to an edge list.
import networkx as nx
import graph_force
G = nx.grid_2d_graph(10, 10)
# we have to map the names to integers
# as graph_force only supports integers as node ids at the moment
edges = []
mapping = {n: i for i, n in enumerate(G.nodes)}
i = 0
for edge in G.edges:
edges.append((mapping[edge[0]], mapping[edge[1]]))
pos = graph_force.layout_from_edge_list(len(G.nodes), edges, iter=1000)
nx.draw(G, {n: pos[i] for n, i in mapping.items()}, node_size=2, width=0.1)
This methods can be used with large graphs, where the edge list does not fit into memory.
Format of the file:
- Little endian
- 4 bytes: number of nodes(int)
- 12 bytes: nodeA(int), nodeB(int), weight(float)
import graph_force
import struct
with open("edges.bin", "rb") as f:
n = 10
f.write(struct.pack("i", n))
for x in range(n-1):
f.write(struct.pack("iif", x, x+1, 1))
pos = graph_force.layout_from_edge_file("edges.bin", iter=50)
iter
, threads
and model
, initial_pos
are optional parameters, supported by layout_from_edge_list
and layout_from_edge_file
.
pos = graph_force.layout_from_edge_list(
number_of_nodes,
edges,
iter=500, # number of iterations, default 500
threads=0, # number of threads, default 0 (all available)
model="spring_model", # model to use, default "spring_model", other option is "networkx_model"
initial_pos=[(0.4, 0.7), (0.7, 0.2), ...], # initial positions, default None (random)
)
spring_model
: A simple spring model (my own implementation)networkx_model
: Reimplementation of the spring model from networkx