merge_tensordicts

class tensordict.merge_tensordicts(*tensordicts: T, callback_exist: Optional[Union[Callable[[Any], Any], Dict[NestedKey, Callable[[Any], Any]]]] = None)

Merges tensordicts together.

Parameters:

*tensordicts (sequence of TensorDict or equivalent) – the list of tensordicts to merge together.

Keyword Arguments:

callback_exist (callable or Dict[str, callable], optional) – a callable in case an entry exists in each and every tensordict. If the entry is present in some but not all tensordicts, or if callback_exist is not passed, update is used and the first non-None value in the tensordict sequence will be used. If a dictionary of callables is passed, it will contain the associated callback function for some of the nested keys in the tensordicts passed to the function.

Examples

>>> from tensordict import merge_tensordicts, TensorDict
>>> td0 = TensorDict({"a": {"b0": 0}, "c": {"d": {"e": 0}}, "common": 0})
>>> td1 = TensorDict({"a": {"b1": 1}, "f": {"g": {"h": 1}}, "common": 1})
>>> td2 = TensorDict({"a": {"b2": 2}, "f": {"g": {"h": 2}}, "common": 2})
>>> td = merge_tensordicts(td0, td1, td2, callback_exist=lambda *v: torch.stack(list(v)))
>>> print(td)
TensorDict(
    fields={
        a: TensorDict(
            fields={
                b0: Tensor(shape=torch.Size([]), device=cpu, dtype=torch.int64, is_shared=False),
                b1: Tensor(shape=torch.Size([]), device=cpu, dtype=torch.int64, is_shared=False),
                b2: Tensor(shape=torch.Size([]), device=cpu, dtype=torch.int64, is_shared=False)},
            batch_size=torch.Size([]),
            device=None,
            is_shared=False),
        c: TensorDict(
            fields={
                d: TensorDict(
                    fields={
                        e: Tensor(shape=torch.Size([]), device=cpu, dtype=torch.int64, is_shared=False)},
                    batch_size=torch.Size([]),
                    device=None,
                    is_shared=False)},
            batch_size=torch.Size([]),
            device=None,
            is_shared=False),
        common: Tensor(shape=torch.Size([3]), device=cpu, dtype=torch.int64, is_shared=False),
        f: TensorDict(
            fields={
                g: TensorDict(
                    fields={
                        h: Tensor(shape=torch.Size([]), device=cpu, dtype=torch.int64, is_shared=False)},
                    batch_size=torch.Size([]),
                    device=None,
                    is_shared=False)},
            batch_size=torch.Size([]),
            device=None,
            is_shared=False)},
    batch_size=torch.Size([]),
    device=None,
    is_shared=False)
>>> print(td["common"])
tensor([0, 1, 2])