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

Applied Network Science

Table 4 Chosen hyperparameters for GAT model

Dataset	Dropout	Learning rate	Regularization weight	Heads	Attention dropout
CORA-ML	0.4	0.01	0.0100	4	0.2
Citeseer	0.4	0.01	0.0100	4	0.2
Pubmed	0.2	0.01	0.0001	8	0.3
CORA-Full	0.2	0.01	0.0001	8	0.3
Actor	0.4	0.01	0.0100	4	0.3
Squirrel	0.4	0.01	0.0001	2	0.2
Texas	0.4	0.01	0.0100	4	0.3
Wisconsin	0.4	0.01	0.0100	4	0.3