lattix

Knowledge graph substrate: core types, algorithms, and serialization formats.

lattix provides three graph representations (KnowledgeGraph, HeteroGraph, HyperGraph), centrality/community algorithms, neighbor sampling for GNN training, and RDF format support.

Usage

[dependencies]
lattix = "0.5.1"

Default features include formats (N-Triples, Turtle, N-Quads, JSON-LD, CSV) and algo (centrality, PageRank, random walks, sampling, label propagation).

use lattix::{Triple, KnowledgeGraph};

let mut kg = KnowledgeGraph::new();
kg.add_triple(Triple::new("Apple", "founded_by", "Steve Jobs"));
kg.add_triple(Triple::new("Apple", "headquartered_in", "Cupertino"));
kg.add_triple(Triple::new("Steve Jobs", "born_in", "San Francisco"));

let apple_relations = kg.relations_from("Apple");
assert_eq!(apple_relations.len(), 2);

Graph types

HeteroGraph and HyperGraph both convert to/from KnowledgeGraph.

Features

Disable defaults for a minimal build (just types + serde):

lattix = { version = "0.5.1", default-features = false }

Algorithms

Centrality (7 algorithms): degree, betweenness, closeness, eigenvector, Katz, PageRank, HITS.

Other: personalized PageRank (PPR), Node2Vec-style random walks, connected components, neighbor sampling (homogeneous and heterogeneous), label propagation community detection.

All iterative algorithms have convergence controls (max_iterations, tolerance). Sampling is deterministic under a given seed.

Formats

Reads and writes N-Triples, Turtle, N-Quads, and JSON-LD. CSV import is read-only. Parsing uses oxttl; the hand-rolled Triple::from_ntriples parser is always available (no feature gate).

Dependencies

petgraph is the graph backbone and is re-exported (lattix::petgraph) for advanced use. Algorithm dependencies (rand, rayon, graphops) are optional behind the algo feature. Format dependencies (oxttl, oxrdf, csv) are optional behind formats.

License

MIT OR Apache-2.0