Dataloader question

[ZIKEYUAN]

Hi,
I have a question in Dataloader function. As we know, when we use Dataloader function, we will create some batch, and those graphs in each batch will seem like a big graph. For example , if we have two graphs G1 and G2, we feed them to a Dataloader function which batch_size = 2, if A1.edge_index = [[0,1],[1,0]] and A2.edge_index = [[0,1],[1,0]]. After this step, we will creat a Batch B2 which include those two graphs and B2.edge_index = [[0,1,2,3],[1,0,3,2]], well its ture. But for other feature in the graph, such as node_type, A1.node_type = [0,1], A2.node_type = [0,1] this feature just for a single graph and not a global feture. So my problem is when I use Dataloader function, B2.node_type will is [0,1,0,1] but it is not correct , I need the feature seem like to edge_index and the B2.node_type is [0,1,2,3], so how to solve this question?

[rusty1s]

You can achieve this by customizing the Data.__inc__ method, see here:

class MyData(Data):
    def __inc__(self, key, value, *args, **kwargs):
        if key == 'node_type':
            return int(node_type.max()) + 1
        return super().__inc__(key, value, *args, **kwargs)

[ZIKEYUAN]

Thanks for your help!