Skip to content

Direction

Ursa has no Graph and DiGraph split. Every edge row is directed, from src to dst. Direction is a parameter of each operation.

The direction parameter

Operations that walk edges accept direction:

Value Follows
"out" (default) edges from the node: src → dst
"in" edges into the node: dst → src
"both" edges in either direction

degree, neighbors, hop, and shortest_path accept direction.

ur.degree(edges, direction="in")
ur.hop(edges, n=2, direction="both").from_([0])

Kernels with a fixed direction

Some algorithms are defined on directed graphs only, and some on undirected graphs only. These take no direction parameter.

Algorithm Reads the rows as
pagerank, betweenness, closeness, connected_components(mode="strong") directed
triangle_count, clustering_coefficient, connected_components(mode="weak"), louvain, label_propagation undirected

Undirected data

If your data is undirected and each edge appears once, the directed kernels see only one direction. The result is correct for the rows as stored, but it can differ from a published undirected result.

To get the undirected result from a directed kernel, store each edge in both directions. The example adds the reverse rows with pyarrow:

import pyarrow as pa
import ursa as ur

edges = ur.datasets.load_karate()
t = edges.collect().to_arrow()
both = ur.EdgeFrame(
    pa.concat_tables([t, t.select(["dst", "src"]).rename_columns(["src", "dst"])]),
    src="src",
    dst="dst",
)
print(ur.describe(edges).collect())
print(ur.describe(both).collect())

shape: (1, 5)
┌─────────┬─────────┬───────────┬────────────┬──────────────┐
│ n_nodes ┆ n_edges ┆ density   ┆ avg_degree ┆ n_components │
│ ---     ┆ ---     ┆ ---       ┆ ---        ┆ ---          │
│ int64   ┆ int64   ┆ double    ┆ double     ┆ int64        │
╞═════════╪═════════╪═══════════╪════════════╪══════════════╡
│ 34      ┆ 78      ┆ 0.0695187 ┆ 2.29412    ┆ null         │
└─────────┴─────────┴───────────┴────────────┴──────────────┘
shape: (1, 5)
┌─────────┬─────────┬──────────┬────────────┬──────────────┐
│ n_nodes ┆ n_edges ┆ density  ┆ avg_degree ┆ n_components │
│ ---     ┆ ---     ┆ ---      ┆ ---        ┆ ---          │
│ int64   ┆ int64   ┆ double   ┆ double     ┆ int64        │
╞═════════╪═════════╪══════════╪════════════╪══════════════╡
│ 34      ┆ 156     ┆ 0.139037 ┆ 4.58824    ┆ null         │
└─────────┴─────────┴──────────┴────────────┴──────────────┘

Tip

For degree, hop, and shortest_path, pass direction="both" instead. It gives the undirected result with no change to the data.

See Semantics vs NetworkX for the exact NetworkX call each kernel matches.