Heterogenous graph

[negarmaleki96]

Hi all,
I am building a heterogeneous graph. I have 3 different node types with 9 additional features within each. Also, I have 3 different edge types with two features within each. I want to implement GNN models on my graph after creating it. My problem is that in the GNN models, they can get only one feature of the graph; for example, in the following code, the model gets x_dict which is common among all the nodes in the heterogeneous graph, but in my case, since I have 9 different features, how should I give it to the model?

class HGT(torch.nn.Module):
def init(self, hidden_channels, out_channels, num_heads, num_layers):
super().init()

    self.lin_dict = torch.nn.ModuleDict()
    for node_type in data.node_types:
        self.lin_dict[node_type] = Linear(-1, hidden_channels)

    self.convs = torch.nn.ModuleList()
    for _ in range(num_layers):
        conv = HGTConv(hidden_channels, hidden_channels, data.metadata(),
                       num_heads, group='sum')
    self.convs.append(conv)

    self.lin = Linear(hidden_channels, out_channels)

def forward(self, x_dict, edge_index_dict):
    for node_type, x in x_dict.items():
        x_dict[node_type] = self.lin_dict[node_type](x).relu_()

    for conv in self.convs:
        x_dict = conv(x_dict, edge_index_dict)

    return self.lin(x_dict['author'])

model = HGT(hidden_channels=64, out_channels=dataset.num_classes,
num_heads=2, num_layers=2)

My graph is like this:

HeteroData(
post={
otype=[147, 1],
created=[147, 1],
last_payout=[147, 1],
payout=[147, 1],
net_rshares=[147, 1],
abs_rshares=[147, 1],
vote_rshares=[147, 1],
author_rewards=[147, 1],
author_reputation=[147, 1]
},
user={ otype=[8394, 1] },
comment={
otype=[446, 1],
created=[446, 1],
last_payout=[446, 1],
payout=[446, 1],
net_rshares=[446, 1],
abs_rshares=[446, 1],
vote_rshares=[446, 1],
author_rewards=[446, 1],
author_reputation=[446, 1]
},
(user, authored, post)={
edge_index=[2, 514],
edge_attr={
otype=[514, 1],
time=[514, 1]
}
},
(user, vote, post)={
edge_index=[2, 13572],
edge_attr={
otype=[13572, 1],
time=[13572, 1]
}
},
(post, reply, comment)={
edge_index=[2, 446],
edge_attr={
otype=[446, 1],
time=[446, 1]
}
}
)

[rusty1s]

The idea here is that you group all your features of a single node type into one feature matrix, e.g., via:

data["post"].x = torch.cat([data["post"].otype, data["post"].created, ...], dim=-1)

[negarmaleki96]

Thank you so much. This is what I thought, too. So for the edge attributes, as they have 2 attributes each, should I do the same?

Also, is it okay that my nodes have a different number of features within? (e.g., post node has 9 features, user node has 1 feature, comment node has 9 features, edge attributes have 2 features each.)

[rusty1s]

Yes. In a heterogeneous graph, it is okay that features of different type have different number of features.