Do we need random seed setting for DataLoader?

[doh16101]

Dear Luis @lrm22005 ,

I saw in the end of this Pytorch official tutorial:
https://pytorch.org/docs/stable/notes/randomness.html

They set the random seed for DataLoader as well:

def seed_worker(worker_id):
    worker_seed = torch.initial_seed() % 2**32
    numpy.random.seed(worker_seed)
    random.seed(worker_seed)

g = torch.Generator()
g.manual_seed(0)

DataLoader(
    train_dataset,
    batch_size=batch_size,
    num_workers=num_workers,
    worker_init_fn=seed_worker,
    generator=g,
)

According to Pytorch DataLoader documentation, the default shuffle is False.
https://pytorch.org/docs/stable/data.html#torch.utils.data.DataLoader

So, do we need the pass the generator = g in the DataLoader to provide reproducible results?

[lrm22005]

Yes, I'm using random seed on most of the models, some models don't require to define it because the library by default is generating it, some others require it.