Class GraphConv
java.lang.Object
ai.beestream.jitnn.nn.Module
ai.beestream.jitnn.nn.GraphConv
- All Implemented Interfaces:
AutoCloseable
Graph convolution (Kipf and Welling, 2017), PyTorch Geometric's GCNConv: Â (x W^T) + b, with Â
the normalised adjacency
SparseMatrix.gcn(int, int[], int[]) of the graph, given with each call since one
layer serves many graphs. x [nodes, in] gives [nodes, out].-
Constructor Summary
ConstructorsConstructorDescriptionGraphConv(int in, int out, long seed) Graph convolution fromintooutfeatures: a linear map without bias initialised fromseed, and a bias of zeros. -
Method Summary
Modifier and TypeMethodDescriptionApplies the module tox.forward(Tensor x, SparseMatrix.Valued adjacency) Applies the layer to node features x [nodes, in] over the graph's normalised adjacency (seeSparseMatrix.gcn(int, int[], int[])).graph(SparseMatrix.Valued adjacency) Sets the graph the next forward calls use.weight()The weight [out, in].Methods inherited from class Module
buffer, buffers, child, close, isClosed, isTraining, loadStateDict, parameter, parameterCount, parameters, stateDict, train, zeroGrad
-
Constructor Details
-
GraphConv
public GraphConv(int in, int out, long seed) Graph convolution fromintooutfeatures: a linear map without bias initialised fromseed, and a bias of zeros.
-
-
Method Details
-
graph
Sets the graph the next forward calls use. -
forward
Applies the layer to node features x [nodes, in] over the graph's normalised adjacency (seeSparseMatrix.gcn(int, int[], int[])). -
forward
-
weight
The weight [out, in].
-