This package implements python bindings for the ORbit Counting Algorithm.
The original source code was modified to avoid memory leaks upon repeated function calls and allow for parallel orbit counting when provided with several graphs (using OpenMP).
Install via pip:
pip install orbit-count
This package provides two functinos, namely node_orbit_counts and edge_orbit_counts. Both functions accept networkx graphs and return numpy arrays.
import networkx as nx
import orbit_count
graph = nx.erdos_renyi_graph(64, 0.2, seed=0) # Graph with N nodes and M edges
# Graphlet size may be 4 or 5
node_count = orbit_count.node_orbit_counts(graph, graphlet_size=4) # Returns numpy array of shape (N, 15)
edge_count = orbit_count.edge_orbit_counts(graph, graphlet_size=4) # Returns numpy array of shape (M, 12)
# To get a specific ordering of counts, pass node_list or edge_list
node_count = orbit_count.node_orbit_counts(graph, graphlet_size=5, node_list=list(graph.nodes))
edge_count = orbit_count.edge_orbit_counts(graph, graphlet_size=5, edge_list=list(graph.edges))
# Count orbits in multiple graphs in parallel
node_count = orbit_count.batched_node_orbit_counts([nx.erdos_renyi_graph(64, 0.2, seed=i) for i in range(32)], graphlet_size=4)
edge_count = orbit_count.batched_edge_orbit_counts([nx.erdos_renyi_graph(64, 0.2, seed=i) for i in range(32)], graphlet_size=4)The source (i.e., the code provided in this GitHub repository) is licensed under GPL v3, as detailed in LICENSE.md. The wheels may bundle an external implementation of OpenMP as a dynamically linked library. We refer to LICENSE.wheel.md for information on how these components are licensed.
