Batching for GCNConv Layer

[kleingeo]

I am trying to use a graph inside an existing model. I have an adjacency matrix that I generated from networkx and converted using edge_index = adjacency_matrix.nonzero().t().contiguous(). The features for the GCNConv layer initial have a batch dimension, but I'm unsure if this is still necessary. The documentation for the GCNConv layer is |V|, F_in in shape, so does this mean that the different batches should be concatenated together in some way?

One way I have thought about this is applying adjacency_matrix = torch.block_diag(*[adjacency_matrix[b_idx] for b_idx in range(batch_size)]) to the different batched adjacency matries and flattening the input features to a correspond with this adjacency matrix. Is this the right approach to deal with the batch dimensions?