Torch_geometric, HeteroData, GNN, Multi and different dimensional Node Features

81 Views Asked by At

I am experimenting with torch_geometric and HeteroData et al…

following specifically MovieLens link regression example… In the example user_features/movie_features are one-dimensional…ie, each user is represented by an one-dimensional vector…and HeteroData is created and all fine…

My question is WHAT IF my a particular node features are multi and different dimensional…ie, each node is represented by a matrix and these matrices have different dimensions…

eg: I have a list of ‘torch_geometric.data’ objects as following…

***tmp_node_features = [
Data(x=[28, 36]),
Data(x=[22, 36]),
Data(x=[32, 36]),
Data(x=[30, 36]),
Data(x=[35, 36]
]***

How do I use this ‘tmp_node_features’ to create the node features in a HeteroData set ie,

***data = HeteroData()

data[‘node_name’].x = ???***

Help much appreciated…

Regards

0

There are 0 best solutions below