Stacked Linear Algebra Graph

An embedded and in-memory graph using sparse linear algebra.

Capabilities

Architecture

The Stacked Linear Algebra Graph implements a directed graph with a weight on each vertex and edge. The graph models vertices and adjacency matrices as GraphBLAS sparse vectors and matrices respectively. The graph operates on its vertex vectors and adjacency matrices using GraphBLAS operators.

Indexing

The graph assigns an unsigned interger index index to each new vertex, vertex vector and adjacecency matrix. The graph may reuse indices after the index has been deleted beforehand.

The numerical vertex indices reference the same coordinates in all vertex vectors and adjacency matrices. All vertex vectors and adjacency matrices thus have compatible sizes.

Each combination of vertex vector and adjacency matrix thus defines a separate graph. All graphs share the same coordinates.

The graph automatically expands the size of the vertex vectors and adjacency matrices as new vertices are added. The graph cannot reduce their size.

Data types

The graph stores the following Rust primitive numeric types in its vertices and edges: bool; i8; i16; i32; i64; u8; u16; u32; u64; f32; f64; isize; usize

Type casting

Each vertex vector and adjacency matrix has a single data datatype. The data type is set upon adding the vertex vector or adjacency matrix to the graph.

Operations involving different value types will use type casting according to ANSI C, with the following exceptions:

• +-Inf or integer values outside their maximum range are clipped
• NaN casts to zero

Linear algebra operations

Graph operators apply to any applicable combination of vertex vector and adjacency matrix.

Transactions

The graph does not implement ACID database transactions.

Persistence

The graph resides in-memory and does not exist in persistent storage.

Minimum example

use graphblas_sparse_linear_algebra::operators::binary_operator::{Assignment, Plus};
use graphblas_sparse_linear_algebra::operators::index_unary_operator::IsValueEqualTo;
use graphblas_sparse_linear_algebra::operators::options::OperatorOptions;
use graphblas_sparse_linear_algebra::operators::semiring::PlusTimes;

use stacked_linear_algebra_graph::graph::edge::{DirectedEdgeCoordinate, WeightedDirectedEdge};
use stacked_linear_algebra_graph::graph::graph::Graph;
use stacked_linear_algebra_graph::operators::add::{AddEdge, AddEdgeType, AddVertex, AddVertexType};
use stacked_linear_algebra_graph::operators::apply_operator::ApplyIndexUnaryOperatorToVertexVector;
use stacked_linear_algebra_graph::operators::element_wise_multiplication::BinaryOperatorElementWiseVertexVectorMultiplication;
use stacked_linear_algebra_graph::operators::multiplication::VertexVectorAdjacencyMatrixMultiplication;
use stacked_linear_algebra_graph::operators::options::OptionsForOperatorWithAdjacencyMatrixAsRightArgument;
use stacked_linear_algebra_graph::operators::read::GetVertexValue;

fn main() {
    let mut graph = Graph::with_initial_capacity(&5, &5, &5).unwrap();

    let numbers_vertex_type_index: usize = AddVertexType::<i32>::apply(&mut graph).unwrap();
    let odd_number_sequence_edge_type_index = AddEdgeType::<i32>::apply(&mut graph).unwrap();

    // Add vertices
    let mut vertex_indices: Vec<usize> = Vec::new();
    for n in 0..12 {
        vertex_indices.push(
            graph
                .add_vertex(&numbers_vertex_type_index, n as u8)
                .unwrap(),
        );
    }

    // Define a sequence of subsequent odd numbers
    for i in [1, 3, 5, 7, 9] {
        let edge = WeightedDirectedEdge::new(
            DirectedEdgeCoordinate::new(
                odd_number_sequence_edge_type_index,
                vertex_indices[i],
                vertex_indices[i + 2],
            ),
            true,
        );

        graph.add_or_replace_edge_from_edge(edge).unwrap();
    }

    // Find the fourth number in the sequence, starting at 1
    let selected_vertices_index: usize = AddVertexType::<i32>::apply(&mut graph).unwrap();

    ApplyIndexUnaryOperatorToVertexVector::<u8>::by_index(
        &mut graph,
        &numbers_vertex_type_index,
        &IsValueEqualTo::<u8>::new(),
        &1,
        &Assignment::new(),
        &selected_vertices_index,
        None,
        &OperatorOptions::new_default(),
    )
    .unwrap();

    for _i in 0..2 {
        VertexVectorAdjacencyMatrixMultiplication::<u8>::by_index(
            &mut graph,
            &selected_vertices_index,
            &PlusTimes::<u8>::new(),
            &odd_number_sequence_edge_type_index,
            &Assignment::new(),
            &selected_vertices_index,
            None,
            &&OptionsForOperatorWithAdjacencyMatrixAsRightArgument::new_default(),
        )
        .unwrap();
    }

    BinaryOperatorElementWiseVertexVectorMultiplication::<u8>::by_index(
        &mut graph,
        &selected_vertices_index,
        &Plus::<u8>::new(),
        &numbers_vertex_type_index,
        &Assignment::new(),
        &selected_vertices_index,
        None,
        &OperatorOptions::new_default(),
    )
    .unwrap();

    assert_eq!(
        GetVertexValue::<u8>::vertex_value(&graph, &numbers_vertex_type_index, &vertex_indices[7])
            .unwrap(),
        Some(7u8)
    )
}

Requirements

Please make sure to meet the requirements for building graphblas_sparse_linear_algebra.

Contributing

Awesome, contributions are welcome. stacked_linear_algebra_graph and your contribution may be relicensed and integrated into commercial software in the future. Therefore, you will be asked to agree to the Contributor License Agreement when you make a pull request.

Licensing

stacked_linear_algebra_graph is licensed under Creative Commons Attribution Non Commercial 4.0 International. For other licensing options, please contact Sam Dekker.

Acknowledgements

Stacked Linear Algebra Graph is inspired by LAGraph and uses the same underlying GraphBLAS implementation from Timothy A. Davis.