Input shuffling

Part 1 (2019)
1

Does fastai utilise input shuffling for training as a default?
I have images from a video that are labelled by frame and I want to know whether I need to shuffle them myself (as to avoid correlated images being batched together) or if fastai naturally does this for me?

2

Yes. By default the shuffle argument is set to true for the training dataset and false to both valid and test datasets.

You can check in the source code (basic_data.py)

On line 119, you can see (True,False,False,False) and that corresponds to the argument shuffle=s for the corresponding datasets (train_ds,valid_ds,fix_ds,test_ds)

   @classmethod
    def create(cls, train_ds:Dataset, valid_ds:Dataset, test_ds:Optional[Dataset]=None, path:PathOrStr='.', bs:int=64,
               val_bs:int=None, num_workers:int=defaults.cpus, dl_tfms:Optional[Collection[Callable]]=None,
               device:torch.device=None, collate_fn:Callable=data_collate, no_check:bool=False, **dl_kwargs)->'DataBunch':
        "Create a `DataBunch` from `train_ds`, `valid_ds` and maybe `test_ds` with a batch size of `bs`. Passes `**dl_kwargs` to `DataLoader()`"
        datasets = cls._init_ds(train_ds, valid_ds, test_ds)
        val_bs = ifnone(val_bs, bs)
        dls = [DataLoader(d, b, shuffle=s, drop_last=s, num_workers=num_workers, **dl_kwargs) for d,b,s in
               zip(datasets, (bs,val_bs,val_bs,val_bs), (True,False,False,False)) if d is not None]
        return cls(*dls, path=path, device=device, dl_tfms=dl_tfms, collate_fn=collate_fn, no_check=no_check)
3

Thanks