In your notebook, you follow the official pytorc_geometric tutorial regarding the GNN but you wrongly copy-pasted the weight_decay for the optimizer. Indeed weight_decay=5e-3 is very big and should be reduced to weight_decay=5e-4 which doubles the accuracy of the model.
In your notebook, you follow the official pytorc_geometric tutorial regarding the GNN but you wrongly copy-pasted the
weight_decayfor the optimizer. Indeedweight_decay=5e-3is very big and should be reduced toweight_decay=5e-4which doubles the accuracy of the model.