GraphEncoderOptions
Options & Configuration · AiDotNet.Models.Options
Shared hyperparameters for the graph encoder networks — GraphSAGE, the Graph Isomorphism Network and the Graph Attention Network.
Inheritance: ModelOptions → NeuralNetworkOptions → ModelHyperparameterOptions → GraphEncoderOptions
Known derived types: GraphAttentionNetworkOptions, GraphIsomorphismNetworkOptions, GraphSAGEOptions
For Beginners
These are the message-passing backbones a graph model is built from. Each has its own way of combining a node’s neighbours, but they all stack the same kind of layer, so the number of layers is the one setting they share.
How It Works
Deliberately separate from GraphModelOptions, which serves the graph TASK heads
(node classification, link prediction, graph classification). That base requires
HiddenDim, and these encoders do not all have one — GraphSAGE and the attention network
have none, and the isomorphism network calls its own MlpHiddenDim. Deriving them from a
base that requires a value they lack is precisely what made five earlier family bases throw at
their own members’ published defaults.
Properties
| Property | Summary |
|---|---|
int NumLayers { get; set; } |
Gets or sets the number of message-passing layers. |