ruvector/examples/graph/hybrid_search.rs

//! Hybrid Vector-Graph Search Example
//!
//! This example demonstrates combining vector similarity search
//! with graph traversal for powerful hybrid queries:
//! - Vector embeddings on graph nodes
//! - Semantic similarity search
//! - Graph-constrained vector search
//! - Ranking by combined scores

fn main() {
    println!("=== RuVector Graph - Hybrid Search ===\n");

    // TODO: Once the graph and vector APIs are integrated, implement:

    println!("1. Initialize Hybrid Database");
    // let db = GraphDatabase::open("./data/hybrid_db.db")?;
    // let vector_store = db.enable_vector_index("embeddings", 768)?;

    println!("   ✓ Created graph database with vector indexing");

    println!("\n2. Create Nodes with Vector Embeddings");
    // Create documents with text embeddings
    // let doc1 = db.create_node()
    //     .label("Document")
    //     .property("title", "Introduction to Rust")
    //     .property("content", "Rust is a systems programming language...")
    //     .embedding(embedding_model.encode("Introduction to Rust..."))
    //     .execute()?;

    // let doc2 = db.create_node()
    //     .label("Document")
    //     .property("title", "Advanced Rust Patterns")
    //     .property("content", "This article covers advanced patterns...")
    //     .embedding(embedding_model.encode("Advanced Rust Patterns..."))
    //     .execute()?;

    println!("   ✓ Created documents with vector embeddings");

    println!("\n3. Create Relationships");
    // Create citation relationships
    // db.create_relationship()
    //     .from(doc2)
    //     .to(doc1)
    //     .type("CITES")
    //     .execute()?;

    println!("   ✓ Created citation relationships");

    println!("\n4. Semantic Search (Vector Only)");
    // let query = "memory safety in programming";
    // let query_embedding = embedding_model.encode(query);

    // let results = db.vector_search()
    //     .index("embeddings")
    //     .query(query_embedding)
    //     .top_k(5)
    //     .execute()?;

    // println!("   Query: '{}'", query);
    // for (doc, score) in results {
    //     println!("     - {} (score: {:.4})", doc.property("title"), score);
    // }

    println!("\n5. Graph-Constrained Vector Search");
    // Find similar documents that are cited by "Advanced Rust Patterns"
    // let results = db.hybrid_query()
    //     .cypher(r#"
    //         MATCH (source:Document {title: 'Advanced Rust Patterns'})-[:CITES]->(cited:Document)
    //         RETURN cited
    //     "#)
    //     .vector_similarity(query_embedding, 0.7)  // minimum similarity threshold
    //     .execute()?;

    // println!("   Cited documents similar to query:");
    // for doc in results {
    //     println!("     - {}", doc.property("title"));
    // }

    println!("\n6. Ranked Hybrid Search");
    // Combine vector similarity and graph metrics
    // let results = db.hybrid_query()
    //     .vector_search(query_embedding)
    //     .graph_metric("pagerank")
    //     .combine_scores(|vector_score, graph_score| {
    //         0.7 * vector_score + 0.3 * graph_score
    //     })
    //     .top_k(10)
    //     .execute()?;

    // println!("   Top results (vector + graph ranking):");
    // for (doc, combined_score, details) in results {
    //     println!("     - {} (score: {:.4}, vector: {:.4}, graph: {:.4})",
    //         doc.property("title"),
    //         combined_score,
    //         details.vector_score,
    //         details.graph_score
    //     );
    // }

    println!("\n7. Multi-Hop Vector Search");
    // Find documents connected through multiple hops, ranked by similarity
    // let results = db.hybrid_query()
    //     .cypher(r#"
    //         MATCH path = (start:Document {title: 'Introduction to Rust'})
    //                      -[:CITES|:RELATED_TO*1..3]->
    //                      (end:Document)
    //         RETURN end, length(path) as distance
    //     "#)
    //     .vector_similarity(query_embedding, 0.6)
    //     .penalty_per_hop(0.1)  // Reduce score for each hop
    //     .execute()?;

    // println!("   Multi-hop search results:");
    // for (doc, score, distance) in results {
    //     println!("     - {} (score: {:.4}, hops: {})",
    //         doc.property("title"), score, distance
    //     );
    // }

    println!("\n8. Filtered Vector Search with Graph Constraints");
    // Complex hybrid query with multiple constraints
    // let results = db.hybrid_query()
    //     .cypher(r#"
    //         MATCH (doc:Document)-[:AUTHORED_BY]->(author:Person)
    //         WHERE author.institution = 'MIT'
    //           AND doc.year >= 2020
    //         RETURN doc
    //     "#)
    //     .vector_search(query_embedding)
    //     .top_k(5)
    //     .execute()?;

    // println!("   MIT papers since 2020 similar to query:");
    // for (doc, score) in results {
    //     println!("     - {} (score: {:.4})", doc.property("title"), score);
    // }

    println!("\n9. Performance Comparison");
    // Compare pure vector vs hybrid search
    // let start = std::time::Instant::now();
    // let vector_results = db.vector_search()
    //     .query(query_embedding)
    //     .top_k(100)
    //     .execute()?;
    // let vector_time = start.elapsed();

    // let start = std::time::Instant::now();
    // let hybrid_results = db.hybrid_query()
    //     .vector_search(query_embedding)
    //     .graph_filter(/* some constraint */)
    //     .top_k(100)
    //     .execute()?;
    // let hybrid_time = start.elapsed();

    // println!("   Vector-only: {:?}", vector_time);
    // println!("   Hybrid: {:?}", hybrid_time);
    // println!("   Overhead: {:.2}%", (hybrid_time.as_secs_f64() / vector_time.as_secs_f64() - 1.0) * 100.0);

    println!("\n=== Example Complete ===");
    println!("\nNote: This is a template. Actual implementation pending graph API exposure.");
}