proxfun_full

proxfun_full is an extensible Python implementation of link prediction scores inspired by R linkprediction::proxfun.

It accepts a NetworkX graph and an optional list of node pairs.

If pairs=None, the function scores non-existing edges.

Functional API

from relationalstats.linkprediction import proxfun_full

scores = proxfun_full(
    G,
    pairs=None,
    metrics=["jaccard", "adamic_adar"],
)

Parameters

Parameter	Description
G	Input networkx.Graph or networkx.DiGraph.
pairs	Optional iterable of (source, target) pairs. If None, non-edges are scored.
metrics	List of metrics to compute. If None, all supported metrics are computed.
directed	If False, scoring is performed on an undirected version of the graph.
return_dataframe	If True, returns a pandas.DataFrame.
beta_local_path	Parameter $\beta$ for the local path index.
beta_katz	Parameter $\beta$ for the Katz index.
rwr_alpha	Restart probability $\alpha$ for random walk with restart.

Estimator-style API

from relationalstats.linkprediction import ProxFun

result = ProxFun(metrics=["jaccard", "adamic_adar"]).fit(G)

df = result.to_dataframe()
top_pairs = result.top_k(k=20, metric="jaccard")

Result object

ProxFunResult exposes:

• scores_
• metrics_
• directed_
• to_dataframe()
• top_k()
• plot_score_distribution()

Notes

The function is designed for practical link prediction workflows. Metric definitions can vary across software packages, so validation against R linkprediction::proxfun should explicitly align graph direction, node ordering, pair selection, and metric parameters.