How to assign a unique graph id for each graph in the TUDataset?

[jiajunyu1999]

I want to assign a unique graph id for each graph in the TUDataset as follows:
dataset = TUDataset(root = args.datapath, name = args.dataset) dataset.data.graph_id = torch.arange(len(dataset)) dataset.graph_id = torch.arange(len(dataset))
However, I can add the graph id attribute for dataset.data, but cannot add it for dataset, the object of 'dataset' have only a fixed number of attributes? Because after feeding the object 'dataset' into 'dataloader', I want to make each batch has the graph id to identify the graph.
Thank you for your answer!

[akihironitta]

Hi @jiajunCAU, I'd pass pre_transform to the dataset constructor to add a new attribute to each data. I haven't tested this, but it'll look like this:

class AddGraphIdTransform:
    def __init__(num_graphs):
        self.graph_id = 0
        self.num_graphs = num_graphs

    def __call__(self, data):
        data.graph_id = self.graph_id
        self.graph_id += 1
        return data

transform = AddGraphIdTransform(999)
dataset = TUDataset(..., pre_transform=transform)

[jiajunyu1999]

Thank you! But after testing, it seems don't work😭, the attributes in the dataset have not changed.

[EdisonLeeeee]

Need to delete your processed/ dir and run it again.

[jiajunyu1999]

Yor're right! Thank you very much!!