Research: Analyst Verbs at Notebook Scale

Scope: forge.rank / forge.cluster timings and write-back on SNAP ego-facebook (4,039 nodes, 88,234 edges).

Executive summary

Compiled analyst verbs complete the usual notebook centrality and community algorithms in well under a second on ego-facebook. Results match direct igraph runs on the same projection. Opt-in write_property persists scores for later Cypher without a per-node UNWIND loop.

Algorithm	Verb	Typical time (s)	Notes
Triangle count	`forge.rank(..., by="triangles")`	~0.07	Full-graph triangles
Betweenness	`forge.rank(..., by="betweenness")`	~0.16	Normalized scores
Closeness	`forge.rank(..., by="closeness")`	~0.11	Connected graphs
PageRank	`forge.rank(..., by="pagerank")`	~0.14	Stream or write-back
Louvain	`forge.cluster(..., by="louvain")`	~0.08	Community labels
Connected components	`forge.cluster(..., by="components")`	~0.06	Tree / component IDs
Degree	`forge.rank(..., by="degree")`	~0.06	Fast baseline
Clustering coefficient	`forge.rank(..., by="clustering_coefficient")`	~0.07	Mean high on ego-facebook

Pattern

from graphforge import GraphForge
from graphforge.datasets import load_dataset

forge = GraphForge()
load_dataset(forge, "snap-ego-facebook")

# Stream Arrow results
table = forge.rank("Node", by="pagerank")

# Or persist for Cypher follow-up
forge.rank("Node", by="pagerank", write_property="pagerank")
forge.cluster("Node", by="louvain", write_property="community")

top = forge.execute("""
    MATCH (n:Node)
    RETURN n.id AS id, n.pagerank AS score
    ORDER BY score DESC
    LIMIT 10
""")

Only rank and cluster accept opt-in write_property. Export to NetworkX or igraph remains available when you need algorithms outside the shipped catalog — see the use-case guide.

Correctness

On ego-facebook, PageRank and betweenness top-10 rankings from forge.rank match direct igraph rankings on forge.to_igraph() for the same undirected projection. Treat NetworkX/igraph as oracles when validating new notebook pipelines.

Recommendations

Prefer analyst verbs for the eight catalog algorithms above; reserve NX/igraph export for visualization and algorithms outside the catalog.
Use write_property when you will filter or join scores in Cypher later.
Restrict with via= / label filters when the graph is heterogeneous.