Can it be used with HeteroConv？

[huaizhenkou]

This is a great job! I tried to use it with HeteroConv, but it failed. Can the author offer any advice？

[Ryanfzhang]

This is a great job! I tried to use it with HeteroConv, but it failed. Can the author offer any advice？

Can you provide some details about the errors?

[huaizhenkou]

It seems the issue lies within my code. I encountered an error similar to the one discussed in pyg-team/pytorch_geometric#5652. I built a model as shown below, but neither ChebConv nor kanChebConv supports handling heterogeneous graphs. They require the input x to be a tensor, not a tuple. Do you know how to resolve this issue? I would greatly appreciate your help.

class HeteroCheb(nn.Module):
    def __init__(self, edge_types, in_channels, hidden_channels, out_channels, dropout=0.5):
        super(HeteroCheb, self).__init__()

        self.conv1 = HeteroConv({
            edge_type: ChebConv(in_channels, hidden_channels, K=2)
            for edge_type in edge_types
        }, aggr='sum')

        self.conv2 = HeteroConv({
            edge_type: ChebConv(hidden_channels, out_channels, K=2)
            for edge_type in edge_types
        }, aggr='sum')

        self.dropout = dropout
        self.name = 'HeteroGCN_without_pooling'

    def forward(self, x_dict, edge_index_dict):
        x_dict = self.conv1(x_dict, edge_index_dict)
        x_dict = {key: x.relu() for key, x in x_dict.items()}
        x_dict = {key: F.dropout(x, p=self.dropout, training=self.training) for key, x in x_dict.items()}
        x_dict = self.conv2(x_dict, edge_index_dict)
        return x_dict