TensorDictBase.pad#

TensorDictBase.pad(pad_size: Sequence[int], value: float = 0.0, inplace: bool = False, safe: bool = True) → Self#

Returns the tensordict padded along its batch dimensions with a constant value.

Parameters:

pad_size (sequence of int) – the padding, applied to the batch dimensions starting from the first. For a tensordict with ndim batch dimensions, it is a sequence of even length up to 2 * ndim: (dim0_left, dim0_right, dim1_left, dim1_right, ...).
value (float, optional) – the fill value. Defaults to 0.0.
inplace (bool, optional) –
if True, the tensordict keeps its identity and keys, its leaves are replaced one by one by their padded copies, and the call returns self. Peak memory stays close to 1x instead of 2x. See In-place operations. Defaults to False.

Warning

If inplace=True and the padding of a later leaf fails (for example, out of memory), the tensordict is left in an inconsistent state. safe=True catches the user errors before any leaf is changed.
safe (bool, optional) – if True, checks that the padding works for every leaf before any leaf is changed. Pass False to skip this check when the inputs are known to be valid. Defaults to True.

Returns:

The padded tensordict, or self with inplace=True.

Note

On a lazy stack whose tensordicts have no names, the result of inplace=False has no names, even when the stack dimension has a name. See pad().

Examples

>>> import torch
>>> from tensordict import TensorDict
>>> td = TensorDict({"a": torch.zeros(3, 4)}, batch_size=[3, 4])
>>> td.pad([0, 0, 0, 1]).batch_size
torch.Size([3, 5])
>>> td.pad([0, 0, 0, 1], inplace=True) is td
True
>>> td.batch_size
torch.Size([3, 5])

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

TensorDictBase.pad#

Docs

Tutorials

Resources