Question about building our own dataloader

Questions
#1

Hi

We want to build a dataloader for our dataset.
We found that there are 3 abstract functions must be implemented for the subclass of DGLDataset according to this tutorial https://docs.dgl.ai/guide/data.html

These 3 abstract functions are process() , __getitem__(idx) and __len__().
We understand that __getitem__(idx) and __len__() are used for python iterator which is the same with PyTorch DataLoader usage.

We don’t understand the role or the purpose of process() here.
Will this function be executed only once (e.g., like the role of init function) or something else?

Thank you.

#2

process is called only once here. It is responsible for processing information needed for modeling, such as the masks for the dataset split and the graph for modeling. An example can be found here.

#3

I see. Thank you very much.