跳转至

PNTensor

danling.tensors.PNTensor

Bases: Tensor

Wrapper for tensors to be converted to NestedTensor.

PNTensor is a subclass of torch.Tensor. It implements three additional property as NestedTensor: tensor, mask, and concat.

Although it is possible to directly construct NestedTensor in dataset, the best practice is to do so is in collate_fn. PNTensor is introduced to smoothen the process.

Convert tensors that will be converted to NestedTensor to a PNTensor, and PyTorch Dataloader will automatically collate PNTensor to NestedTensor.

Source code in danling/tensors/nested_tensor.py
Python
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
class PNTensor(Tensor):
    r"""
    Wrapper for tensors to be converted to `NestedTensor`.

    `PNTensor` is a subclass of `torch.Tensor`.
    It implements three additional property as `NestedTensor`: `tensor`, `mask`, and `concat`.

    Although it is possible to directly construct `NestedTensor` in dataset,
    the best practice is to do so is in `collate_fn`.
    `PNTensor` is introduced to smoothen the process.

    Convert tensors that will be converted to `NestedTensor` to a `PNTensor`,
    and PyTorch Dataloader will automatically collate `PNTensor` to `NestedTensor`.
    """

    @property
    def tensor(self) -> Tensor:
        r"""
        Identical to `self`.

        Returns:
            (torch.Tensor):

        Examples:
            >>> tensor = torch.tensor([1, 2, 3])
            >>> pn_tensor = PNTensor([1, 2, 3])
            >>> bool((tensor == pn_tensor).all())
            True
            >>> bool((tensor == pn_tensor.tensor).all())
            True
        """

        return self

    @property
    def mask(self) -> Tensor:
        r"""
        Identical to `torch.ones_like(self)`.

        Returns:
            (torch.Tensor):

        Examples:
            >>> tensor = torch.tensor([1, 2, 3])
            >>> pn_tensor = PNTensor([1, 2, 3])
            >>> bool((pn_tensor.mask == torch.ones_like(pn_tensor)).all().item())
            True
        """

        return torch.ones_like(self)

    @property
    def contact(self) -> Tensor:
        r"""
        Identical to `self`.

        Returns:
            (torch.Tensor):

        Examples:
            >>> tensor = torch.tensor([1, 2, 3])
            >>> pn_tensor = PNTensor([1, 2, 3])
            >>> bool((tensor == pn_tensor).all())
            True
            >>> bool((tensor == pn_tensor.contact).all())
            True
        """

        return self

    def new_empty(self, *args, **kwargs):
        return PNTensor(super().new_empty(*args, **kwargs))

tensor property

Python
1
tensor: Tensor

Identical to self.

Returns:

Type Description
Tensor

Examples:

Python Console Session
>>> tensor = torch.tensor([1, 2, 3])
>>> pn_tensor = PNTensor([1, 2, 3])
>>> bool((tensor == pn_tensor).all())
True
>>> bool((tensor == pn_tensor.tensor).all())
True

mask property

Python
1
mask: Tensor

Identical to torch.ones_like(self).

Returns:

Type Description
Tensor

Examples:

Python Console Session
>>> tensor = torch.tensor([1, 2, 3])
>>> pn_tensor = PNTensor([1, 2, 3])
>>> bool((pn_tensor.mask == torch.ones_like(pn_tensor)).all().item())
True

contact property

Python
1
contact: Tensor

Identical to self.

Returns:

Type Description
Tensor

Examples:

Python Console Session
>>> tensor = torch.tensor([1, 2, 3])
>>> pn_tensor = PNTensor([1, 2, 3])
>>> bool((tensor == pn_tensor).all())
True
>>> bool((tensor == pn_tensor.contact).all())
True