Format of edge_index

[hkim716]

Hi Matt, I would like to draw my mesh dataset using plotly.

https://plotly.com/python/3d-mesh/
Here in the 3D Mesh in Dash example to draw a sandal, it requires i=df.i, j=df.j, k=df.k, format.

I think in that example, i, j, k means vertices of a triangular mesh, and x, y, z is the coordinates of each node.
Assuming that I have triangular mesh like edge_index,
I want to know how I could transform edge_index to the format compatible with this example.

Also, I guess edge_index that is used in PyG is called COO format with shape [2, num_edges]. I think it's pretty clear to understand how it works. But what is CSR format then?

[rusty1s]

There is no easy way to convert the edge_index representation to a mesh representation of faces. Therefore, it is best to simply keep your mesh data for visualization while converting it to the edge_index representation for neural network processing.

You are right that PyG represents graphs in a COO format. CSR denotes a compressed representation of this, where values are sorted by source node index, and only node-wise offsets are stored. We support this format via the SparseTensor class.