Convert graph to CSC

[atale480]

I'm trying to use CuGraphSAGEConv Layer. Here it says that I need CSC representation of graph, and for that use CuGraphModule.to_csc() method to convert an edge_index representation to the desired format. I'm working on 3D mesh data. Which contain positions and faces information. How to convert this information or the edge index information to CSC?

I came across one method called to_torch_csc_tensor. Does this outputs same that is needed as input to CuGraphSAGEConv layer? If not then what is correct way to get data in CSC format as per need of CuGraphSAGEConv.

Also when I try to initialize CuGraphSAGEConv as L = CuGraphSAGEConv( in_channels = 3, out_channels = 16 ) it gives error as ModuleNotFoundError: 'CuGraphSAGEConv' requires 'pylibcugraphops'.
I'm not sure what does this error mean. Any help would be great regarding this issues. Thanks

[rusty1s]

Since you are working on mesh data, you can directly use the FaceToEdge transform. Afterwards, you can call CuGraphModule.to_csc() from the given edge_index representation. There is nothing to do with pos. You can just use it as input features into CuGraphSAGEConv.

You also need to install pylibcugraphops from conda, see https://anaconda.org/nvidia/pylibcugraphops.