xgi-optimized 0.11

XGI extension for faster hypergraph operations

What has been done?:

• Metrics computation has been updated
• The implementation of bipartite graphs has been changed
• Application of graph and hypergraph metrics to improve ML models
• Cpp implementation of linegraph

Metrics computation has been updated

The computation of the linegraph vector centrality metric was accelerated by using the igraph library instead of networkx . For comparison, implementations using the networkx.Graph class were retained.

H_enron = xgi.load_xgi_data("plant-pollinator-mpl-034")
H_enron_cleaned = H_enron.cleanup(
    multiedges=False, singletons=False, isolates=False, relabel=True, in_place=False
)
start_time = time.time()

xgi.centrality.line_vector_centrality(H_enron_cleaned)

middle_time = time.time()

xgi.centrality.fast_line_vector_centrality(H_enron_cleaned)

end_time = time.time()

print(f"Время выполнения nx: {(middle_time-start_time):.6f} секунд")
print(f"Время выполнения ig: {(end_time-middle_time):.6f} секунд")

The implementation of bipartite graphs has been changed.

The implementation for working with bipartite graphs was modified: the igraph.Graph class is now used instead of networkx.Graph for converting from a Hypergraph to a bipartite graph and back to a Hypergraph .

hg = xgi.Hypergraph()
hg.add_edges_from(
    [[1, 2, 3], [3, 4, 5,8], [1, 4, 10, 11, 12], [7,0,8], [5,7,2]]
)

bpg = xgi.convert.to_bipartite_graph(hg)
plot(bpg,vertex_label=bpg.vs.indices,layout = Graph.layout_bipartite(bpg), bbox=(0, 0, 500,500))

Application of graph and hypergraph metrics to improve ML models

In the file for solving the classification problem, in addition to article embeddings, article identifiers, and author IDs used as features, metrics such as eigenvector centrality of the graph and vector centrality of the hypergraph were also utilized.

Cpp implementation of linegraph

• In progress...