GLobalMax pool for dense GCN

[shrimonmuke0202]

How can I apply global max-pool for DenseGCn?
because if I apply global max-pool on the output of DenseGCn then an error occurs of mismatched dimension as global max-pool layer wants the batch size.

[rusty1s]

In dense-mode, you can simply use x.max(dim=1)[0] for global max pooling, where x has shape [batch_size, num_nodes, num_features].

[shrimonmuke0202]

RuntimeError: batch1 dim 2 must match batch2 dim 1
this error occurs for applying DenseGCN on the adjacency matrix.

[rusty1s]

Can you show me an example on what you are trying to do and what fails?

[shrimonmuke0202]

x , edge_index, batch = data.x,data.edge_index,data.batch
adj = to_dense_adj(edge_index,batch)
print(adj.shape)
target = data.target

if self.k1 == 1:
  print("1")
  h1 = self.Conv1(x,adj)

[rusty1s]

You need to convert the x to a dense representation as well via x, _ = to_dense_batch(x, batch)

[shrimonmuke0202]

Thank you