DGL Dataloader num workers > 0 problems

#1

Hi,

I am using the GAT model, with the standard batched graph classification framework in the examples. However, I am trying to use multiple workers for the pytorch dataloader to speed up the creation of batches. However, I run into problems, with this? See below…

dgl._ffi.base.DGLError: Cannot update column of scheme Scheme(shape=(256,), dtype=torch.float32) using feature of scheme Scheme(shape=(19,), dtype=torch.float32).

Does the creation of batch graphs along with the collate function support the data loader configured to use multiple processes?

#2

Yep we support multiple processes to load data, however could you please show your code so that we could find where the problem is?