This pull request introduces the GraphConvolutionalNeuralNetwork component as outlined in "Semi-Supervised Classification with Graph Convolutional Networks". Additionally, the pull request brings in a new processing block called TransformPropagateUpdate, designed to cover a broad spectrum of graph operations.
The GraphConvolutionalNeuralNetwork component applies the Laplacian normalization ($D^{-1/2}\hat{A}D^{-1/2}$ where $\hat{A} = A + I$) by default on the adjacency matrix. This normalization supports various forms of adjacency matrices, including square matrices, edge indices, and sparse matrices.
I have modified the unit tests to ensure that each change is thoroughly verified.
This pull request introduces the
GraphConvolutionalNeuralNetwork
component as outlined in "Semi-Supervised Classification with Graph Convolutional Networks". Additionally, the pull request brings in a new processing block calledTransformPropagateUpdate
, designed to cover a broad spectrum of graph operations.The
GraphConvolutionalNeuralNetwork
component applies the Laplacian normalization ($D^{-1/2}\hat{A}D^{-1/2}$ where $\hat{A} = A + I$) by default on the adjacency matrix. This normalization supports various forms of adjacency matrices, including square matrices, edge indices, and sparse matrices.I have modified the unit tests to ensure that each change is thoroughly verified.