Finding benchmark dataset - node prediction + binary classes + high number of graphs

[kimzed]

Hello everybody,

For my research I would like to test my models and my training pipeline with a benchmark dataset with similar characteristics as my own dataset to be sure that the model architecture / training pipeline has no error bug. I looked into various torch_geometric.datasets but it is difficult to match what i need.

Does anybody knows about a benchmark dataset from torch geometric that matches these criterias?

node prediction, binary classes, high number of graphs

[rusty1s]

I am not aware of any datasets many datasets with these characteristics. The closest would probably be PPI dataset (but it is multi-label prediction on only 24 graphs), and the PATTERN and CLUSTER datasets from the GNNBenchmarkDataset (but they are multi-class classification).

[kimzed]

PATTERN really looks like the dataset that I need. thanks a lot, i used it and it works.