GraphNeuralNetworks.jl
GraphNeuralNetworks.jl copied to clipboard
Allow for additonal features in GNNGraph
There are cases where the shape of the feature array might not have the last dim = number of nodes and edges. For example: In a mesh represented by a graph where the vertices are nodes. Currently, there is no way to include triangular faces, except for graph data (which they are not).
GNNGraph should be allowed to contain such features which are not either node or edge. cc: @CarloLucibello
How does PyG works in this setting?
PyG doesn't force checks on features.
Why we can't store them as graph data?
We can do it given our current implementation. But at the end of the day, they are not exactly graph level features.
closing as it not clear how to act on this