rune-node2vec

Graph node embeddings via biased random walks and skip-gram.

What it does

rune-node2vec implements the Node2Vec algorithm (Grover & Leskovec, 2016) to learn continuous vector representations for every node in an undirected graph. The two hyperparameters p and q bias the random walk between local-neighbourhood (BFS) and global-exploration (DFS) strategies, making the embeddings useful for link prediction, node classification, and community detection.

Installation

[dependencies]
rune-node2vec = "0.1"

Usage

use rune_node2vec::Node2Vec;

// Two triangles with no edges between them.
let edges = vec![(0,1),(1,2),(2,0), (3,4),(4,5),(5,3)];

let result = Node2Vec::new()
    .embedding_dim(32)
    .num_walks(10)
    .walk_length(20)
    .n_epochs(5)
    .random_seed(42)
    .fit(6, &edges);

assert_eq!(result.embeddings.len(), 6);
assert_eq!(result.embeddings[0].len(), 32);

CLI

# basic use — one edge `u v` per line
rune-node2vec graph.edgelist

# stdin, custom parameters
cat graph.edgelist | rune-node2vec - --dim 64 --epochs 5 --p 0.5 --q 2.0

# all options
rune-node2vec graph.edgelist \
  --dim 128 --walk-length 80 --num-walks 10 \
  --window 10 --p 1.0 --q 1.0 --epochs 1 --lr 0.025 --neg-samples 5 --seed 42

Output

Each line is one node's embedding — tab-separated floats, one line per node in ascending index order. A summary is written to stderr.

0.012345	-0.034567	0.056789	...
-0.023456	0.045678	-0.012345	...
# 6 nodes → 128-dim embeddings

License

MIT