Stacking different inputs for the same graph to a batch dimension

[knshnb]

Hi! I would like to handle multiple different inputs (e.g., node features) of the same graph structure. I can handle them as different graphs, but it might be a little inefficient. Would it be possible to pass batched inputs for a single graph, such as (n_node, n_batch, n_feat), to graph conv?

[rusty1s]

There exists a few layers in PyG that support this kind of structure (see the "Static Graph" column here. With it, you can do

x = ... # [batch_size, num_nodes, in_channels]
edge_index = ... # edges of a single graph

out = conv(x, edge_index)  # [batch_size, num_nodes, out_channels]

[knshnb]

Thanks for the quick response! That answered my question.