TensorDictBase.clone#
- TensorDictBase.clone(recurse: bool = True, **kwargs) Self#
Clones a TensorDictBase subclass instance onto a new TensorDictBase subclass of the same type.
To create a TensorDict instance from any other TensorDictBase subtype, call the
to_tensordict()method instead.- Parameters:
recurse (bool, optional) – if
True, each tensor contained in the TensorDict will be copied too. Otherwise only the TensorDict tree structure is copied: the sub-tensordicts are new objects, but the leaves are the same tensors as inself. Defaults toTrue.
Note
Unlike many other ops (pointwise arithmetic, shape operations, …)
clonedoes not inherit the original lock attribute. This design choice is made such that a clone can be created to be modified, which is the most frequent usage.Examples
>>> import torch >>> from tensordict import TensorDict >>> td = TensorDict({"a": torch.zeros(3), "b": {"c": torch.ones(3)}}, [3]) >>> td_clone = td.clone() >>> td_clone["a"] is td["a"] False >>> td_shallow = td.clone(recurse=False) >>> td_shallow["a"] is td["a"], td_shallow["b"] is td["b"] (True, False) >>> td.lock_().clone().is_locked False
Every tensordict class and every tensorclass has this method. For the other methods, see Methods by task.
