nvalchemi.training.configure_dataloader#

nvalchemi.training.configure_dataloader(dataset, *, batch_size, shuffle=None, sampler=None, batch_sampler=None, collate_fn=None, **dl_kwargs)[source]#

Thin wrapper around DataLoader.

Parameters:
  • dataset (Any)

  • batch_size (int)

  • shuffle (bool | None, optional) – Defaults to None, which resolves to True when no sampler is provided and False otherwise. Passing True with sampler raises ValueError.

  • sampler (Any, optional) – Optional sample-ordering object forwarded to DataLoader.

  • batch_sampler (Any, optional) – Optional batch sampler forwarded to DataLoader.

  • collate_fn (Callable | None, optional)

  • **dl_kwargs (Any) – Forwarded to DataLoader.

Return type:

torch.utils.data.DataLoader

Raises:

ValueError – If shuffle=True and sampler are both provided.