Module graph_models 

Graph neural network embedding models (v0.3.0).

This module provides production-ready GNN implementations for knowledge graph and general graph embedding tasks:

• graphsage: GraphSAGE - inductive representation learning via neighbor sampling with multiple aggregator strategies.
• gat: Graph Attention Networks - multi-head attention for adaptive neighborhood aggregation.

Quick Start

GraphSAGE

use oxirs_embed::graph_models::graphsage::{Graph, GraphSAGEConfig, GraphSAGEModel};

let config = GraphSAGEConfig {
    input_dim: 16,
    hidden_dims: vec![32],
    output_dim: 8,
    ..Default::default()
};
let model = GraphSAGEModel::new(config)?;
// Build your graph, then:
// let embeddings = model.embed(&graph)?;

GAT

use oxirs_embed::graph_models::gat::{GATConfig, GATModel};

let config = GATConfig {
    input_dim: 16,
    output_head_dim: 8,
    num_layers: 2,
    ..Default::default()
};
let model = GATModel::new(config)?;

Re-exports

pub use gat::GATConfig;

pub use gat::GATEmbeddings;

pub use gat::GATModel;

pub use graphsage::AggregatorKind;

pub use graphsage::Graph;

pub use graphsage::GraphSAGEConfig;

pub use graphsage::GraphSAGEEmbeddings;

pub use graphsage::GraphSAGEModel;

pub use graphsage::LSTMAggregator;

pub use graphsage::MaxPoolAggregator;

pub use graphsage::MeanAggregator;

pub use graphsage::MeanPoolAggregator;

pub use graphsage::MiniBatchConfig;

pub use graphsage::MiniBatchGraphSAGE;

pub use graphsage::TrainingMetrics;

Modules

gat

GAT: Graph Attention Network Embeddings (v0.3.0)

graphsage

GraphSAGE: Inductive Representation Learning on Large Graphs