I’m trying to train my network model and I need a dataloader for that. The dataloader is returning different results in the first iteration over the dataset compared to the second iteration. How is this possible?

I have worked through the “Node Classification with Graph Neural Networks” Example Notebook of Pytorch Geometric. In the further Example Section it was recommended to use data.val_mask of the Cora Set to increase the Model Accuracy by “selecting and testing the model with the highest validation performance”.