The exp result reproduce of the whole dataset on Table 2

[gegemy]

Hi, I wonder how to produce the whole dataset's experiment results shown in Table 2. Does it directly run the whole graph (like Cora) on a model (if yes, does it run on GCN?), without any reduction rate? or set the reduction rate as 1?

[ChandlerBang]

Hi,

Sorry for the late reply. This is indeed included in the train_coreset.py file:

GCond/train_coreset.py

Lines 61 to 70 in 9f2032c

	# Setup GCN Model
	device = 'cuda'
	model = GCN(nfeat=features.shape[1], nhid=256, nclass=labels.max()+1, device=device, weight_decay=args.weight_decay)
	model = model.to(device)
	model.fit(features, adj, labels, idx_train, idx_val, train_iters=600, verbose=False)
	model.eval()
	# You can use the inner function of model to test
	model.test(idx_test)

You may run train_coreset.py to get the result you want.

python train_coreset.py --dataset cora --r=0.01  --method=random
python train_coreset_induct.py --dataset flickr --r=0.01  --method=random