DBLP

[Yunbo-max]

If i wanna use those datasets like DBLP, then which code should I follow to reproduce the results from the paper considering there are some errors when I added commands to utils.py file.

[Suchun-sv]

Hi yunbo,

Could you show me the error your encountered? I could help you with solving it.

Best

[ChenglongMa]

Hi @Suchun-sv,

Thanks for your great work! The code about DBLP seems not implemented:

SGDD/utils.py

Lines 31 to 49 in 5ac77c1

	def get_dataset(name: str, normalize_features=False, transform=None, if_dpr=True):
	path = osp.join(osp.dirname(osp.realpath(__file__)), 'data', name)
	if name in ['cora', 'citeseer', 'pubmed']:
	dataset = Planetoid(path, name)
	elif name in ['ogbn-arxiv']:
	dataset = PygNodePropPredDataset(name='ogbn-arxiv')
	elif name in ['yelpchi', 'amazon']:
	name = 'yelp' if name == 'yelpchi' else 'amazon'
	dataset = FraudDataset(name, raw_dir=path)
	dataset = from_dgl(dataset[0], name=name, hetero=False)
	elif name.lower() == "sbm":
	num_nodes = [2000] * 3
	edge_probs = [[0.1, 0.05, 0.02],
	[0.05, 0.1, 0.02],
	[0.02, 0.02, 0.1]]
	dataset = StochasticBlockModelDataset(path, num_nodes, edge_probs, num_channels=32)
	dataset.name = "SBM"
	else:
	raise NotImplementedError

Could you advise how to reproduce the results?

Thanks!

Best,
Chenglong

[ChenglongMa]

Hi @Suchun-sv,

Actually, I am curious about the link prediction task.

According to the pseudocode of Algorithm 1: SGDD for Graph Condensation in your paper, it is only applicable for classification tasks.

How to apply SGDD to link prediction tasks or regression tasks?

Please correct me if I understand wrongly.
Thanks!

Best,
Chenglong