Module gat_basic 

Graph Attention Networks (GAT) - Basic Implementation

Veličković et al. (2018) - ICLR “Graph Attention Networks”

Key innovation: learn attention coefficients between nodes and their neighbors, enabling the model to selectively focus on relevant structural information. Multi-head attention provides stability and richer representations.

Structs

Gat

Graph Attention Network embedding model

GatConfig

Configuration for a Graph Attention Network

GatEmbeddings

Output embeddings from GAT inference

Functions

gat_err

Convert to EmbeddingError

softmax

Softmax over a slice (re-exported for use in ab_test module)