GraFPrintGraphBlockLayer<T>
Layers · AiDotNet.NeuralNetworks.Layers
Implements one GraFPrint Grapher + residual feed-forward block.
Inheritance: LayerBase<T> → GraFPrintGraphBlockLayer<T>
Implements: IShapeContract
How It Works
This follows the official encoder: 1x1 projection and batch normalization, a dynamic dilated k-NN max-relative graph convolution, a 1x1 projection back to the residual width, then a 4x expansion feed-forward network. Both the graph branch and FFN are residual.
Constructors
| Constructor | Summary |
|---|---|
GraFPrintGraphBlockLayer(int channels, int k, int dilation, double dropPathRate) |
Creates a paper-faithful GraFPrint graph block. |
Properties
| Property | Summary |
|---|---|
bool SupportsTraining { get; } |
|
double DropPathRate { get; } |
Gets the residual stochastic-depth probability. |
int Channels { get; } |
Gets the residual feature width. |
int Dilation { get; } |
Gets the dilated-neighborhood factor. |
int K { get; } |
Gets the graph neighborhood size. |
Methods
| Method | Summary |
|---|---|
IReadOnlyList<ILayer<T>> GetSubLayers() |
|
IReadOnlyList<OutputAxisContract> OutputAxesFor(int inputRank) |
Derived from this type’s [TensorLayout] declarations. |
void ResetState() |