The interplay between communities and homophily in semi-supervised classification using graph neural networks

Applied Network Science

Table 5 Chosen hyperparameters for all GNN models except GAT

	Dataset	Dropout	Learning rate	Regularization weight
GCN	CORA-ML	0.8	0.010	0.0100
	Citeseer	0.8	0.010	0.0100
	Pubmed	0.8	0.010	0.0100
	CORA-Full	0.2	0.005	0.0010
	Actor	0.8	0.010	0.0100
	Squirrel	0.4	0.010	0.0100
	Texas	0.8	0.010	0.0100
	Wisconsin	0.8	0.010	0.0100
SAGE	CORA-ML	0.8	0.010	0.0100
	Citeseer	0.8	0.010	0.0100
	Pubmed	0.8	0.010	0.0100
	CORA-Full	0.8	0.005	0.0010
	Actor	0.8	0.010	0.0100
	Squirrel	0.4	0.005	0.1000
	Texas	0.8	0.010	0.0100
	Wisconsin	0.8	0.010	0.0100
APPNP	CORA-ML	0.8	0.010	0.0100
	Citeseer	0.8	0.010	0.0100
	Pubmed	0.8	0.010	0.0100
	CORA-Full	0.2	0.005	0.0001
	Actor	0.8	0.010	0.0100
	Squirrel	0.2	0.010	0.0010
	Texas	0.8	0.010	0.0100
	Wisconsin	0.8	0.010	0.0100
SGC	CORA-ML	0.8	0.010	0.0100
	Citeseer	0.8	0.010	0.0100
	Pubmed	0.8	0.010	0.0100
	CORA-Full	0.2	0.005	0.0010
	Actor	0.8	0.010	0.0100
	Squirrel	0.4	0.010	0.0100
	Texas	0.8	0.010	0.0100
	Wisconsin	0.8	0.010	0.0100
CGCN	CORA-ML	0.8	0.001	0.1000
	Citeseer	0.8	0.001	0.1000
	Pubmed	0.6	0.010	0.0001
	CORA-Full	0.8	0.001	0.1000
	Actor	0.6	0.010	0.1000
	Squirrel	0.6	0.005	0.1000
	Texas	0.8	0.010	0.1000
	Wisconsin	0.6	0.005	0.1000