About max_pool

[hkim716]

Hi Matt, I have some questions about the code below.

def forward(self, data):
    data.x = F.elu(self.conv1(data.x, data.edge_index, data.edge_attr))
    weight = normalized_cut_2d(data.edge_index, data.pos)
    cluster = graclus(data.edge_index, weight, data.x.size(0))
    data.edge_attr = None
    data = max_pool(cluster, data, transform=transform)

    data.x = F.elu(self.conv2(data.x, data.edge_index, data.edge_attr))
    weight = normalized_cut_2d(data.edge_index, data.pos)
    cluster = graclus(data.edge_index, weight, data.x.size(0))
    x, batch = max_pool_x(cluster, data.x, data.batch)

    x = global_mean_pool(x, batch)
    x = F.elu(self.fc1(x))
    x = F.dropout(x, training=self.training)
    return F.log_softmax(self.fc2(x), dim=1)

1. What is the difference between max_pool and max_pool_x?

2. Why should we reset data.edge_attr using data.edge_attr=None before max_pool?

[rusty1s]

1. max_pool will pool a given data object, i.e., it will take care of coarsening the graph and pool node features, while max_pool_x will only take care of pooling node features (e.g., in case edge_index is no longer needed at this point of time).
2. You don't have to, but here we compute new edge features (the spatial relations between nodes) given the new graph representation.