Class RandLambda

monai/transforms/utility/array.py:847–891

Randomizable version :py:class:`monai.transforms.Lambda`, the input `func` may contain random logic, or randomly execute the function based on `prob`. Args: func: Lambda/function to be applied. prob: probability of executing the random function, default to 1.0, with 100

class RandLambda(Lambda, RandomizableTransform):
    """
    Randomizable version :py:class:`monai.transforms.Lambda`, the input `func` may contain random logic,
    or randomly execute the function based on `prob`.

    Args:
        func: Lambda/function to be applied.
        prob: probability of executing the random function, default to 1.0, with 100% probability to execute.
        inv_func: Lambda/function of inverse operation, default to `lambda x: x`.
        track_meta:  If `False`, then standard data objects will be returned (e.g., torch.Tensor` and `np.ndarray`)
            as opposed to MONAI's enhanced objects. By default, this is `True`.

    For more details, please check :py:class:`monai.transforms.Lambda`.
    """

    backend = Lambda.backend

    def __init__(
        self,
        func: Callable | None = None,
        prob: float = 1.0,
        inv_func: Callable = no_collation,
        track_meta: bool = True,
    ) -> None:
        Lambda.__init__(self=self, func=func, inv_func=inv_func, track_meta=track_meta)
        RandomizableTransform.__init__(self=self, prob=prob)

    def __call__(self, img: NdarrayOrTensor, func: Callable | None = None):
        self.randomize(img)
        out = deepcopy(super().__call__(img, func) if self._do_transform else img)
        # convert to MetaTensor if necessary
        if not isinstance(out, MetaTensor) and self.track_meta:
            out = MetaTensor(out)
        if isinstance(out, MetaTensor):
            lambda_info = self.pop_transform(out) if self._do_transform else {}
            self.push_transform(out, extra_info=lambda_info)
        return out

    def inverse(self, data: torch.Tensor):
        do_transform = self.get_most_recent_transform(data).pop(TraceKeys.DO_TRANSFORM)
        if do_transform:
            data = super().inverse(data)
        else:
            self.pop_transform(data)
        return data


class LabelToMask(Transform):