reference/generated/tensordict.TensorDictBase.empty

Run in Google Colab

Colab

Download Notebook

Notebook

View on GitHub

GitHub

TensorDictBase.empty

TensorDictBase.empty(recurse=False, *, batch_size=None, device=NO_DEFAULT, names=None) → Self

Returns a new tensordict of the same type, with no tensor leaves.

The result has the batch size, device and dimension names of the tensordict, unless the keyword arguments change them. It is not locked, even if the tensordict is.

Parameters:

recurse (bool, optional) – if True, the result has the nested tensordicts and the non-tensor data of the tensordict, without its tensor leaves. If False, the result has no entries. Defaults to False.

Keyword Arguments:

• batch_size (Size or list of int, optional) – the batch size of the result. Defaults to None, which keeps the batch size of the tensordict. With recurse=True, a nested tensordict whose batch size is the start of the new one gets the new one, a nested tensordict whose batch size starts with the new one keeps it, and any other nested tensordict raises a RuntimeError. A LazyStackedTensorDict whose members took the batch size before 0.15 (one stacked along dim 0, for instance) keeps that result, the empty stack of its members, with a FutureWarning, when a nested tensordict does not fit the new batch size or names has another length. Starting with TensorDict 0.17, the default will change to raising a RuntimeError or a ValueError, as TensorDict.empty and the names setter do.

• device (device, str or int, optional) – the device of the result. None gives a result with no device. By default, the result has the device of the tensordict.

• names (list of str, optional) – the names of the batch dimensions of the result, one per dimension. Defaults to None, which keeps the names of the tensordict for the dimensions that remain. Where empty() keeps names that the names setter would reject or drop (a list of another length, or a name used twice), it emits a FutureWarning. Starting with TensorDict 0.17, the default will change to applying them as the names setter does, with or without recurse.

Examples

>>> import torch
>>> from tensordict import TensorDict
>>> td = TensorDict({"a": torch.zeros(3), "b": {"c": torch.ones(3)}}, [3])
>>> empty = td.empty()
>>> list(empty.keys()), empty.batch_size
([], torch.Size([3]))
>>> list(td.empty(recurse=True).keys(include_nested=True))
['b']
>>> td.empty(batch_size=[]).batch_size
torch.Size([])

Every tensordict class and every tensorclass has this method. For the other methods, see Methods by task.