I am working with geometric dataset in Pytorch using pytoch geometric library. I want to do train my data in batches. I observed that when I loaded my data to DataLoader in batches. When I loaded data to the dataloader in batches, it automatically increments the edgeindex tensor by the cumulated number of nodes of all graphs .
train_loader = DataLoader(train_dataset, batch_size=32, shuffle=Train)
train_dataset is a graph data containing node features and edge indices.
My data is a time series data.
I am concerned that by the modification of the edge index tensor across each batch might impact my performance. Does anyone has an idea about this on how to stop automatic increment. Otherwise, could I just train my graph data model and not bother about the automatic increment.
Thanks