ISubgraphSampler<T>

Interfaces · AiDotNet.FederatedLearning.Graph

Samples neighborhoods from a client’s local subgraph for mini-batch GNN training.

For Beginners

GNNs learn by aggregating features from neighboring nodes (message passing). For large subgraphs, using all neighbors is too expensive. A sampler selects a fixed-size neighborhood at each hop, creating small “computation trees” for efficient training.

How It Works

Example: For a target node, with 2 hops and max 10 neighbors per hop:

Methods

Method Summary
SampledSubgraph<T> Sample(Tensor<T> adjacency, int[] targetNodes, Tensor<T> nodeFeatures) Samples a k-hop neighborhood around the specified target nodes.